Semiconductor

Semiconductor Radiation
Detection Systems
2010 by Taylor and Francis Group, LLC
Devices, Circuits, and Systems

Series Editor
Krzysztof Iniewski
CMOS Emerging Technologies Inc., Vancouver, British Columbia, Canada
Internet Networks: Wired, Wireless, and Optical Technologies

Krzysztof Iniewski
Semiconductor Radiation Detection Systems
Krzysztof Iniewski
FORTHCOMING
Radio Frequency Integrated Circuit Design
Sebastian Magierowski
Semiconductors: Integrated Circuit Design for Manufacturability
Artur Balasinki
Electronics for Radiation Detection
Krzysztof Iniewski
Radiation Effects in Semiconductors
Krzysztof Iniewski
Electrical Solitons: Theory, Applications, and Extensions
in High Speed Electronics
David Ricketts
Integrated Microsystems: Materials, MEMs, Photonics, Bio Interfaces
Krzysztof Iniewski
Semiconductor Radiation
Detection Systems
Edited by Krzysztof Iniewski
Boca Raton London New York
CRC Press is an imprint of the

Taylor & Francis Group, an informa business
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works
Printed in the United States of America on acid-free paper
10 9 8 7 6 5 4 3 2 1
International Standard Book Number: 978-1-4398-0385-1 (Hardback)
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts
have been made to publish reliable data and information, but the author and publisher cannot assume
responsibility for the validity of all materials or the consequences of their use. The authors and publishers
have attempted to trace the copyright holders of all material reproduced in this publication and apologize to
copyright holders if permission to publish in this form has not been obtained. If any copyright material has
not been acknowledged please write and let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented,
including photocopying, microfilming, and recording, or in any information storage or retrieval system,
without written permission from the publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.
com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood
Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and
registration for a variety of users. For organizations that have been granted a photocopy license by the CCC,
a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used
only for identification and explanation without intent to infringe.
Library of Congress Cataloging-in-Publication Data
Semiconductor radiation detection systems / editor, Krzysztof Iniewski.
p. cm. -- (Devices, circuits, and systems)
A CRC title.
Includes bibliographical references and index.
ISBN 978-1-4398-0385-1 (hard back : alk. paper)
1. Semiconductor nuclear counters. I. Iniewski, Krzysztof. II. Title. III. Series.
TK9180.S448 2010
539.77--dc22
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
2009051391
Contents
Preface......................................................................................................................vii
About the Editor.........................................................................................................ix
Contributors...............................................................................................................xi
Chapter 1. Spatial and Spectral Resolution of Semiconductor Detectors in
Medical Imaging...................................................................................1
Bjrn J. Heismann
Chapter 2. Silicon Radiation Detectors with Three-Dimensional Electrodes
(3D Detectors)..................................................................................... 19
Gian-Franco Dalla Betta and Andrea Zoboli
Chapter 3. Cadmium Zinc Telluride Pixel Detectors for Hard X-Ray
Astrophysics........................................................................................ 67
Fiona A. Harrison, Walter R. Cook, H. Miyasaka, and R. McLean
Chapter 4. Hydrogenated Amorphous Silicon Radiation Detectors..................... 83
Matthieu Despeisse
Chapter 5. Novel X- and Gamma-Ray Detectors Based on Metamaterials........ 109
Paul Lecoq
Chapter 6. Multicell Geiger-Mode Avalanche Photodiodes: Silicon
Photomultipliers................................................................................ 127
Gabriela Llos
Chapter 7. Hybrid Photodetectors (HPDs) for Single-Photon Detection............ 151
Atsuhito Fukasawa
Chapter 8. High-Resolution CdTe Detectors and Application to GammaRay Imaging...................................................................................... 171
Tadayuki Takahashi, Shin Watanabe, and Shin-nosuke Ishikawa
vi
Contents
Chapter 9. Caliste: Microcamera for Hard X-Ray Astronomy........................... 193

Olivier Limousin, Aline Meuris, Olivier Gevin, and Francis Lugiez
Chapter 10. Hybrid Pixel Array Detectors for Photon Science............................ 217
Heinz Graafsma
Chapter 11. XPAD, a Photon-Counting Imager for X-Ray Applications............. 237
Patrick Pangaud, Pierre Delpierre, and Jean-Franois Brar
Chapter 12. Ultrafast Electron Beam Tomography.............................................. 263
Uwe Hampel
Chapter 13. Compton Imaging: Principles and Practice....................................... 281
Carolyn E. Seifert
Chapter 14. Multimodality Imaging with MR/PET and MR/SPECT.................. 315
Troy Farncombe
Chapter 15. Hard X-Ray Imaging Detectors Onboard the Balloon-Borne
High-Energy Focusing Telescope..................................................... 359
C. M. Hubert Chen, Finn E. Christensen, James C. Chonko,
Walter R. Cook, William W. Craig, Charles J. Hailey,
Fiona A. Harrison, Carsten P. Jensen, Jason E. Koglin,
Kristin Kruse Madsen, and Klaus Ziock
Index....................................................................................................................... 383
Preface
Semiconductor Radiation Detection Systems addresses the state-of-the-art in semiconductor detector and integrated circuit design in the context of medical imaging
using ionizing radiation. New exciting opportunities in X-ray detection, computer
tomography (CT), bone dosimetry, and nuclear medicine (PET, SPECT) are also
discussed. Emerging detector technologies, circuit design techniques, new materials,
and innovative system approaches are explored. The book is a must for anyone serious about detectors and electronics in a healthcare sector.
In addition to medical imaging, this book also addresses applications in other
fields that utilize X-ray and gamma-rays; particularly, in the rapidly growing field of
security applications. Luggage scanning, dirty bomb detection, and border control
are just a few examples of systems that are being deployed today.
The book is written by top-notch international experts in the industry (Siemens
Healthcare, Hamamatsu Photonics), leading international research institutes (CERN,
PNL, INFN, CEA, DESY, JAXA), and academia (University of Trento, FZ Dresden,
Hamilton Health Sciences). The intended audience is practicing engineers with some
electronics or radiation background. This book can be also used as a recommended
reading and supplementary material in graduate course curriculum.
vii
About the Editor

Krzysztof (Kris) Iniewski is managing research and development at Redlen
Technologies Incorporated, a startup company in British Columbia. His research
interests are in very large scale integrated circuits for medical and security applications. From 2004 to 2006, he was an associate professor at the Electrical Engineering
and Computer Engineering Department of the University of Alberta, where he
conducted research on low-power wireless circuits and systems. During his tenure
in Edmonton, he edited a book published by CRC Press, Wireless Technologies:
Circuits, Systems, and Devices.
From 1995 to 2003, he was with PMC-Sierra and held various senior technical and management positions. Prior to joining PMC-Sierra, from 1990 to 1994 he
was an assistant professor at the University of Torontos Electrical Engineering and
Computer Engineering Department. Dr. Iniewski has published over 100 research
papers in international journals and conferences. He holds 18 international patents
granted in the United States, Canada, France, Germany, and Japan. He received
his Ph.D. degree in electronics (honors) from the Warsaw University of Technology
(Warsaw, Poland) in 1988. Together with Carl McCrosky and Dan Minoli, he is an
author of Network Infrastructure and ArchitectureDesigning High-Availability
Networks (Wiley, 2008). He recently edited Medical Imaging: Principles, Detectors,
and Electronics (Wiley, 2009); VLSI Circuits for Bio-Medical Applications (Artech
House, 2008); and Circuits at Nanoscale: Communications, Imaging and Sensing
(CRC Press, 2008).
ix
Contributors
Jean-Franois Brar
Institute Neel, UJF-GNRS
Grenoble, France
Atsuhito Fukasawa
Hamamatsu Photonics K.K.
Shizuoka, Japan
C. M. Hubert Chen
California Institute of Technology
Pasadena, California
Olivier Gevin
CEA, Irfu
GIF-SUR-Yvette, France
James C. Chonko
Columbia University
New York, New York
Heinz Graafsma
Deutsches Elektronen Synchrotron
Hamburg, Germany
Finn E. Christensen
Danish National Space Center
Copenhagen, Denmark
Charles J. Hailey
Columbia University
New York, New York
Walter R. Cook
Uwe Hampel
Forschungszentrum Dresden-Rossendorf
Dresden, Germany
William W. Craig
Lawrence Livermore National
Laboratory
Livermore, California
Fiona A. Harrison
Gian-Franco Dalla Betta

Universit degli Studi di Trento
Trento, Italy
Bjrn J. Heismann
Siemens AG
Erlangen, Germany
Pierre Delpierre
Universit de la Mditerrane
Marseille, France
Shin-nosuke Ishikawa
Institute of Space and Astronautical
Science (ISAS), JAXA
Kanagawa, Japan
Matthieu Despeisse
Ecole Polytechnique Fdrale de
Lausanne
Neuchtel, Switzerland
Carsten P. Jensen
Copenhagen, Denmark
Troy Farncombe
Hamilton Health Services
Hamilton, Canada
Jason E. Koglin
Columbia University
New York, New York
xi
xii
Contributors
Paul Lecoq
European Organization for Nuclear
Research, CERN
Geneva, Switzerland
H. Miyasaka
Patrick Pangaud
Universit de la Mditerrane
Marseille, France
Olivier Limousin
CEA, Irfu
Gif-sur-Yvette, France
Gabriela Llos
IFIC, Instituto de Fsica Corpuscular
Valencia, Spain
Carolyn E. Seifert
Pacific Northwest National Laboratory
Richland, Washington
Francis Lugiez
CEA, Irfu
Tadayuki Takahashi
Kanagawa, Japan
Kristin Kruse Madsen

R. McLean
Aline Meuris
CEA, Irfu
Shin Watanabe
Kanagawa, Japan
Klaus Ziock
Lawrence Livermore National
Laboratory
Andrea Zoboli
Trento, Italy
and Spectral
1 Spatial
Resolution of
Semiconductor Detectors
in Medical Imaging
Bjrn J. Heismann
Siemens AG
Erlangen, Germany
Contents
1.1 Introduction.......................................................................................................1
1.2 Detector Physics................................................................................................3
1.2.1 Indirect and Direct Conversion Detectors.............................................3
1.2.2 Signal Transport Processes....................................................................6
1.3 Spatial Resolution..............................................................................................8
1.3.1 Definition of the Modulation Transfer Function....................................8
1.3.2 Simulation and Measurement of the MTF.............................................9
1.3.3 Properties of the MTF...........................................................................9
1.4 Spectral Resolution.......................................................................................... 10
1.4.1 Definition of the Detector Response Function (DRF)......................... 10
1.4.2 Comparisons of Detector Response Functions.................................... 12
1.4.3 Integrating Indirect Conversion Detectors.......................................... 13
1.4.4 Counting Direct Conversion Detectors................................................ 14
1.5 Conclusions...................................................................................................... 16
References................................................................................................................. 16
1.1Introduction
Medical imaging devices commonly use gamma and X-ray radiation to generate
internal images of the human body. Single-photon emission computed tomography
(SPECT) and positron emission tomography (PET) systems detect gamma emissions
of radionuclide tracers. Computed tomography (CT), radiography, and mammography systems measure the X-ray attenuation of the human body. Figure1.1 outlines
the modes of operation of SPECT, CT, and radiography devices.
Single photon emission

computed tomography
Computed tomography
X-ray source
Radiography
X-ray source
Patient
Source
Antiscatter collector
Patient
SPECT detector
CT detector
X-ray
fan
beam
Patient
X-ray
fan
beam
RAD detector
Figure 1.1 SPECT, CT, and radiography medical imaging devices. (From Heismann,
B. J., Henseler, D., Niederloehner, D., Hackenschimed, P., Strassburg, M., & Wirth, S. IEEE
Room Temperature Seminconductor Workshop, 2008. With permission.)
The image quality and dose usage of these systems is strongly influenced by the
employed radiation detectors. From the early stages, scintillator detectors based on
materials like NaI, BGO, LSO, GOS, and CsI performed the first step of radiation
detection.1 An increasing number of scientific and commercial activities have used
conversion semiconductor detectors for medical imaging. For lower X-ray energies,
amorphous selenium detectors are routinely employed in mammography detectors.
For higher X-ray and gamma-ray energies, cadmium zinc telluride (CdZnTe; CZT)
and cadmium telluride (CdTe) have come into focus. SPECT prototypes in cardiology, scintimammography, and small-animal imaging have been presented (see, e.g.,
Blevis et al.; Parnham et al.; Wagenaar et al.).24 The authors reported improved
spectral resolution and underlined the potential to perform dual-isotope imaging. For
CT, direct conversion counting electronics and prototype systems have been built
and evaluated.58 The high X-ray flux of more than 108 quanta per second and millimeter squared is found to be a major challenge. A main reason for this is attributed
to the dynamic material properties of CZT. It has been shown that defects like Te
inclusions and subsequent inferior hole mobility lead to polarization in CZT detectors under medical imaging X-ray fluxes.912 The main mechanism is seen in the
creation of a dynamic space charge in the semiconductor bulk, degrading the charge
transport properties. The potential benefits of semiconductor detectors in medical
imaging rely mainly on their spatial and spectral resolution.
In this chapter, we analyze the signal transport in both a scintillator and a semiconductor detector. As an application example, we focus on a CT detector. The pixel
geometry, scintillator material, and thickness as well as the electronic readout are
chosen accordingly. As figures of merit, we use the modulation transfer function
(MTF) to quantify the spatial resolution and the detector response function D(E,E) to
Spatial and Spectral Resolution of Semiconductor Detectors in Medical Imaging 3
analyze the spectral behavior. It should be noted that the results indicate an upper performance limit since degradations by, for example, material defects are not included.
1.2Detector Physics
1.2.1Indirect and Direct Conversion Detectors
The indirect conversion scintillation detector in Figure1.2 is based on a GOS scintillator bulk material. Each pixel is enclosed by an epoxy compound filled with backscattering TiO2 particles. Typical pixel dimensions of around 1 mm and below are
obtained. A registered photosensor detects the secondary light photons at the bottom
surface of each pixel. The primary interaction in a detector pixel is given by absorption of an incoming X-ray quantum by a gadolinium atom. The X-ray energy is
converted into light photons. The energy conversion rate is around 12%.13 Secondary
light photon transport takes place. Photons that reach the photosensor contribute
to the output energy signal E. Radiography and mammography detectors follow
similar designs. CsI is usually employed as a scintillator. Owing to its vertical needle
structure, it has the advantage of providing intrinsic light-guiding properties; thus,
no backscattering septa are required. This allows for improved detector resolution at
the expense of reduced stopping power and signal speed.
Two main physical effects influence the spatial and spectral resolution in pixelized
scintillator detectors: First, the primary energy deposition is not perfectly localized.
For the high-Z atom gadolinium, absorption is governed by the photoelectric effect.
This generates fluorescence escape photons with mean free-path lengths on the order
of several 100 m. They might be reabsorbed in the pixel volume, become registered in a neighboring pixel, or leave the detector volume completely. Second, light
X-ray quantum
of energy E
Reflector,
TiO2 based
Scintillator
Glue
Photodiode
array (Si with
SiO2 top layer)
Energy
deposition
Optical
photon
transport
X-ray
scattering
Detected output energy E
Figure 1.2 Schematics of a CT scintillation detector as an example for an indirect conversion detector. (Reproduced from Heismann, B. J. et al., Nucl. Instrum. Methods Phys. Res. A,
591, 2008. With permission from Elsevier.)14
Common
cathode
Direct
converter
Energy
deposition
Pixel
dened
by E-eld
Pixelized
anodes
Readout
electronics
Fluorescence
and scattering
+
Charge
transport
Diusion
Charge
sharing
Figure 1.3 Schematics of a common cathode CZT direct conversion detector. (From
Heismann, B. J., Henseler, D., Niederloehner, D., Hackenschimed, P., Strassburg, M., &
Wirth, S. IEEE Room Temperature Seminconductor Workshop, 2008. With permission.)
transport is affected by optical cross talk. Septa walls are designed with limited
thickness to optimize overall dose usage and light yield. As a consequence, a significant portion of the light is transferred to adjacent "false" pixels (see Figure1.2).
The direct conversion CZT scheme is shown in Figure1.3. A common-cathode
design with pixelized anodes on the bottom surface of the semiconductor bulk is
typically used. Pixels are established by funnel-shaped electrical fields of several
hundreds of volts per millimeter. The physics of the primary energy deposition are
comparable to the indirect conversion detector. However, the deposited energy is
converted to charges instead of optical photons. The holes and electrons are separated and accelerated by the electrical field. Electrical pulse signals are induced on
the electrodes. The main signal pulse is generated when the electrons follow the
stronger curved electrical field in the bulk region close to the anodes.
The main signal degradation mechanisms are comparable to indirect conversion
scintillator detectors: First, fluorescence scattering takes place. Owing to the lower
K-edge energy, the mean free-path lengths of fluorescence quanta in CZT are about
100 m. The smaller the pixel size, the more fluorescence cross talk will affect the
behavior of the detector. Second, the charge signal transport is affected by charge
sharing. The moving charge cloud also induces electrical pulses on neighboring pixels,14 again mostly at the bottom part of the pixel field configuration.
Figures1.4a1.4e summarize the main difference between an indirect conversion
scintillator detector and a direct conversion semiconductor detector. The scintillation
detector is an optical device that uses light photons as intermittent information carriers. A direct conversion detector omits the conversion to light and directly generates
charge carriers. It is an electrical device that employs electrons and holes to transfer
the event information to the electrodes.
Both detector types can be operated in an integrating or a counting mode. In
integrating mode, the charge information is sampled over an integration time and
Scintillator
pixel
Semiconductor
pixel
Common
cathode
X-ray
quantum
Pixelized
anode
Photo sensor
(d)
(a)
Light
photons
(b)
V
Holes
Electric
field
Electrons
(c)
Sampling
and ADC
Electrons
Pulse detection
and counting
(e)
Figure 1.4 Signal conversion steps in an indirect conversion detector (ac) and a direct
conversion detector (de).
converted to a digital signal. In counting mode, the total number of events is measured
by counting the charge pulses. In addition, the energy of each absorbed quantum can
be obtained by measuring the total charge or pulse amplitude of each quantum.
Counting detectors thus offer spectral resolution of the input quantum field.
The detector parameters for our comparison are summarized in Table 1.1. For
the scintillator detector, a 1.4-mm thick GOS with a pixel size of 1.2 mm has been
chosen. The direct conversion detector has a 2-mm thick CZT at 700 V bias with a
quadratic pixel size of 450 m2. For this pixel size, fluorescence cross talk contributes significantly. The choice reflects mostly the high-resolution case. The spatial
resolution is not directly comparable to the scintillator detector. The setting is chosen to investigate whether a direct conversion detector can provide improved spatial
resolution at a reasonable spectral resolution.
1.2.2Signal Transport Processes

Figures1.5a and 1.5b outline the cascaded system theory (CST) model of an indirect
conversion integrating and a direct conversion counting detector. CST models have
been applied to a number of detector evaluations, especially for flat-panel radiography and mammography detectors.15
The indirect conversion detector in Figure1.5a has the following signal conversion steps: First, the X or gamma quantum is absorbed, and its energy is converted
to light photons. Second, the light photons travel through the scintillator setup. Third,
light photons are detected as an electrical current by the photosensor. Finally, the
current signal is digitized by a sampling analog-to-digital converter (ADC). The
first step of the direct conversion detector in Figure1.5b also consists of the X-ray
energy deposition. A cloud of electrons and holes is generated. Second, the generated
charges travel to the electrodes, forming current pulses. Finally, the current pulses
are detected by a readout electrodes.
The signal transport can be simulated by a cascade of independent conversion
steps. For the indirect conversion detector, this can be done as follows:13

1. Primary energy deposition. The primary energy deposition is modeled by

a Monte Carlo simulation tool based on the GEANT4 (GEometry ANd
Tracking) particle interaction simulation framework.1618 For each incoming
X-ray quantum, the spatial energy deposit is simulated on a 10 10 10
subvoxel grid. Rayleigh and Compton scattering and escape processes generally lead to multiple deposition sites per event. Each portion of the primary energy deposit is converted into a number of optical photons, taking
into account the main scintillator photon emission energies. The energy-tooptical-photons conversion gain for Gd2O2S:Pr is taken from measurements
as EC = 0.12 with a standard deviation of (Ec) = 0.04. As a result of the
first step, we obtain a lookup table of individual energy deposition events. By
using a high number of events (106 and more), systematic errors are avoided.
2. Light transport. The second step describes the light transport to the photosensor pixels. A photon-tracking Monte Carlo simulation13 traces the photon paths in the entire detector system until they are detected at a pixel
of the photosensor array or lost by bulk absorption or scintillator escape.
Photon interaction processes like optical scattering, photon reabsorption,
and diffuse and specular reflection at pixel septa borders are included. The
corresponding optical parameters are taken from experimental results. For
a specific photon starting position, the average detection probability of an
optical photon in a photosensor pixel is obtained.
3. Light detection. The light photons that have reached a photosensor pixel are
converted to electrical charges. The wavelength-dependent quantum efficiency () of the photosensor is taken into account. As a result of the third
step, we obtain a photo current for each pixel.
4. Electronic readout. In the final step, the photo current is sampled to charge
and digitized. For medical X-ray applications, sigma-delta ADCs are common ADC designs. Direct current measurements by a charge-coupled
X-ray
intensity
1. X-ray energy deposition
2. Light transport
X-ray field
behind patient
Light
intensity
X-ray
intensity
1. X-ray energy deposition
2. Pulse generation
3. Light detection
4. Electronic readout
Current
Current
Sampling
and ADC
Pulse
detection
Digitized
charge
Counted
pulses
Measured
data
Measured
data
(a)
Figure 1.5 Cascaded detection models: (a) indirect conversion, (b) direct conversion.
3. Electronic readout
(b)
X-ray field
behind patient
oscillator are also employed. The electronic readout usually has limited
linearity and additional offset noise. For the results in this chapter, nonlinearity and electronic noise do not play a role and are neglected.
For a given detector geometry, X-ray quantum input spectrum, and field distribution, this scheme yields the average signal of the scintillator detector.
The signal chain of the direct conversion detector in Figure1.5b is modeled
as follows:19

1. Primary energy deposition. The primary energy deposition step is equivalent to the scintillator model. Instead of a GOS material, a CZT absorber
is used.
2. Pulse generation. A detailed charge transport model can be based on the
work of Eskin et al.20 A local weighting potential allows calculation of the
signal pulse shape for arbitrary charge starting positions in the detector.21 A
time-resolved pulse signal on the anode is obtained.
3. Electronic readout. Depending on the priority of spectral or spatial resolution, two main electronic design schemes for direct conversion detectors
can be selected. Spectrally resolving detectors in SPECT and PET require
a precise measurement of the energy of each quantum.
As a result of this, the anode signals are usually filtered with comparably long
shaping times. The signal is integrated and digitized. High-resolution detectors, on
the other hand, address applications in mammography, radiography, and CT. The
corresponding electronics employ shorter shaping times close to the primary pulse
duration. The filtered pulse signals are usually detected by amplitude threshold triggering.5,6 In the following, we assume the second case of a high-flux X-ray detector.
The threshold noise due to electronic noise contributions in the electronic readout is
included in the model.
1.3Spatial Resolution
The spatial resolution of X-ray detectors is mainly given by pixel pitch and aperture. The pixel pitch defines the Nyquist frequency. The smaller the pixel aperture, the larger the spatial resolution will become. However, in practical imaging
systems, defining the spatial resolution of a detector is a trade-off with dose usage
and detector cost. In particular, scintillator detectors are often limited by the
required septa walls and the cost of the required number of electronic digitization channels.
1.3.1Definition of the Modulation Transfer Function

The MTF(f) is commonly used to describe the spatial resolution of pixelized detectors.15 It is given by the normalized absolute value of the Fourier transformation of
the detector pixel point spread function. The MTF evaluation scheme is commonly
applied to pixelized scintillator and CZT detectors (see, e.g., Michel).21,22

Measured Pixel Signal
Simulated Pixel Signal
Figure 1.6 Measured and simulated slanted slit images for the indirect conversion GOS
detector.
1.3.2Simulation and Measurement of the MTF

For both detector types, the MTF is determined by the slanted slit method.
Figure1.6 shows a slanted slit image for the indirect conversion detector. A tungsten
plate with a slit 0.1 mm wide is placed on top of the scintillator array with a slit angle
of approximately 3 with respect to the fundamental directions of the pixel lattice,
here denoted as I and k. The slit is illuminated by an X-ray flat field. Summing the
image along the line direction yields the line spread function. It is an oversampled
representation of the point spread function. The procedure is repeated with various
angles of the slit toward the axes to obtain a two-dimensional MTF of the detector.
It has been shown that measured MTFs have an excellent agreement to simulated
MTFs for both indirect conversion detectors and direct conversion detectors.22 In
the following, we use the simulation framework described in the previous section to
obtain the slit images required for the MTF calculation.
1.3.3Properties of the MTF

Figure1.7 shows the MTF comparison between an indirect and direct conversion
detector. The straight lines are simulated curves, whereas the corresponding dashed
curves show the respective ideal sinc functions. The indirect conversion detector
shows a midfrequency drop in comparison to the ideal sinc function. This is mainly
due to optical cross talk, which leads to low-pass signal filtering in the detector. In
principle, the midfrequency drop can be recovered by appropriate inverse filtering at
the expense of amplified electronic noise in the signal. For high- and medium-flux
10

1
MTF indirect converter

sinc(f * aperture)
MTF direct converter
sinc(f * aperture)
0.9
0.8
0.7
MTF
0.6
0.5
0.4
0.3
0.2
0.1
0
0.5
1.5
2.5
Spatial frequency in 1/mm
Figure 1.7 Modulation transfer functions for indirect conversion 1.2-mm pixel detector
(red) and direct conversion 450 m detector (blue). Dashed lines reflect the ideal sinc functions. (From Wirth, S., Heismann, B. J., Niederloehner, D., Baetz, L., Metzger, W. & Pharm
Gia, K. IEEE Nuclear Science Symposium Conference Record, 2008. With permission.)
medical applications, this has no major impact. The signal-to-noise ratio is mainly
affected in low-flux screening applications.
In comparison to this, the direct conversion detector is close to the ideal sinc
behavior. The remaining deviations are mainly due to fluorescence escapes between
adjacent pixels. Despite the fact that the pixel aperture has been more than halved,
charge sharing plays only a minor role compared to the effects of optical cross talk.
Note that in both detector systems a small deviation in the zero-frequency position
is visible. This is due to the fact that fluorescence cross talk leads to smaller signal
contributions close to the pixel borders, effectively shrinking the pixel aperture.
1.4Spectral Resolution
In nuclear physics and medical imaging applications like PET and SPECT, the spectral resolution of the detector is commonly described by the pulse height spectrum
(PHS). A typical PHS of a CZT pixelized detector and a NaI Anger camera is shown
in Figure1.8. For X-ray applications, the detector has to register a whole range of
input energies. Figure1.9 shows 80- and 140-kV tungsten tube spectra. The generalization of the PHS to a range of input energies E leads to the detector response
function (DRF).13
1.4.1Definition of the Detector Response Function (DRF)

The DRF D(i,k)(E,E) yields a probability density to measure the output energy E
for an incoming quantum of energy E. The incoming quantum flux is directed at
a central reference pixel. Its lateral position is equally distributed across the pixel
Intensity
Nal
FWHM
CZT
50
100
150
200
E/keV
Figure 1.8 Pulse height spectrum (PHS) of CZT pixelized detector and a NaI Anger camera. FWHM, full width at half maximum. (From Wirth, S., Heismann, B. J., Niederloehner,
D., Baetz, L., Metzger, W. & Pharm Gia, K. IEEE Nuclear Science Symposium Conference
Record, 2008. With permission.)
0.06
U1 = 80 kV
U2 = 140 kV
Normalized Intensity
0.05
0.04
0.03
0.02
0.01
0
50
Energy in keV
100
150
Figure 1.9 The 80- and 140-kV tungsten tube spectra. (From Heismann, B. J., Henseler, D.,
Niederloehner, D., Hackenschimed, P., Strassburg, M., & Wirth, S. IEEE Room Temperature
Seminconductor Workshop, 2008. With permission.)
area. The output energy is detected at a photosensor pixel with the position (i, k). The
pair (0,0) marks the center position (1,0), the horizontal neighbors (0,1), the vertical
neighbors, and so on (see Figure1.11).
The DRF allows us to express the statistics of the microscopic signal transport
processes as a macroscopic probability function. We can simplify its variable dependencies for medical imaging applications. Here, the pixel-to-pixel variation of the
12
3000
120
2500
100
E in keV
E in Photons on Photodiode
3500
2000
1500
80
60
1000
40
500
20
20
40
60
80
100
120
140
20
40
60
80
E in keV
E in keV
(a)
(b)
100 120
Figure 1.10 Detector response function for (a) indirect conversion GOS detector and
(b) direct conversion CZT detector. (From Heismann, B. J., Henseler, D., Niederloehner, D.,
Hackenschimed, P., Strassburg, M., & Wirth, S. IEEE Room Temperature Seminconductor
Workshop, 2008. With permission.)
D(1,1)
D(0,1)
D(1,1)
D(1,0)
D(0,0)
D(1,0)
D(1,1)
D(0,1)
D(1,1)
Figure 1.11 Spatial indices (i, k) of the DRF. The symmetry arises for pixels unaffected
by border effects. (From Heismann, B. J. et al. Nucl. Instrum. and Meth. in Phys. Res. A, 591,
2008. With permission.)
projected anatomical input signal is usually below 1%. This is close to a flat-field
irradiation of the detector. In this case, the mean signal cross talk between pixels is
symmetrical. We realize the flat-field approximation by irradiating the detector surface homogeneously. The simplified D(E,E) function is used to describe the results.
1.4.2Comparisons of Detector Response Functions

The DRFs of the indirect and direct conversion detector setups are shown in
Figures1.10a and 1.10b. In both cases, the probabilities are normalized to 1 for each
input energy E. This leads to the respective color codings. Below 15 keV output
energy, E, the electronic noise in the counting direct conversion detector dominates
the output behavior. The respective range is omitted for clarity.
The indirect conversion D(E,E) in Figure1.10a consists of the following structures: Up to the gadolinium K-edge energy EK ~ 50.2 keV, a linear branch E ~ E
is visible. Its broadening is explained by the energy conversion gain variance. The
output energy peak has a tail toward higher output energies E for increasing input
energy E. This light tailing effect is due to the fact that the light transport yield
increases with the interaction depth, which in turn increases with the input energy
E. Above the K-edge energy, a secondary branch occurs. The events are formed by
absorption of the primary energy with a fluorescence energy loss to the surroundings.
The corresponding reabsorbed fluorescence events are found in the third, approximately vertical branch starting at around 50 keV output energy. Its slight inclination
is again due to the increase of the interaction depth with input energy. The overall
absorption probability of the quanta is reduced with increasing input energy E. The
low-energy output events, including Compton and Rayleigh scatter depositions, are
not shown; see Heismann et al.13 for a more detailed discussion of these effects.
The direct conversion detector D(E,E) in Figure1.10b has a more pronounced
linear branch. Its stronger relative signal content is explained by the about two times
higher intrinsic conversion gain of CZT and the reduced depth dependency due to the
small pixel effect. The fluorescence branches appear at the lower Cd, Zn, and Te fluorescence energies of 23 to 28 keV. The differential branches are consequently closer
to the main linear branch. Charge-sharing events create a low-energy tail increasing
toward lower output energies and overlapping with the fluorescence branches.
The spectral behavior described by D(E,E) has consequences for both detector
schemes. In the following, we consider the cases of an integrating indirect conversion detector and a counting direct conversion detector as prominent examples.
1.4.3Integrating Indirect Conversion Detectors

For integrating indirect conversion detectors, it has been shown that the output signal
variance increase leads to Poisson excess noise.13 Following the work of Rabbani
et al.,23 a formula for the noise amplification is established as

f (E ) =
SNR out
=
( E ) SNR in
1
( E)
( E)2 + 2 (e)
where (E) is the quantum detection efficiency, <E> is the average output energy,
and (E) is the output energy variance. f(E) is a generalized energy-dependent
Swank factor. The Poisson excess noise shown in Figure1.12 is most pronounced
around the K-edge. A noise increase of about 15% is visible. This is due to the fact
that the output signal variance increases strongly beyond the K-edge. For continuous
input spectra, a typical excess noise of 510% can be estimated, depending on the
input spectra and the patient attenuation.
14
f (E)
0.8
0.6
0.4
0.2
0
20
40
60
80
100
120
140
E in keV
Figure 1.12 Generalized Swank factor f(E). (Reproduced from Heismann, B. J. et al.,
Nucl. Instr. Methods in Phys. Res., 1, 591, 2008. With permission from Elsevier.)
1.4.4Counting Direct Conversion Detectors

For a counting direct conversion detector, we can distinguish between the full energy
resolution required in SPECT or PET and the binned energy resolution required for
dual-energy CT or radiography. In the case of full energy resolution, D(E,E) contains directly the normalized PHS for specific input energies.
In the following, we focus on the case of a two-bin energy resolution. Like shown
in Figure1.13, this is commonly achieved by using two threshold levels in the electronic readout. The first threshold Eth1 discards noise events. The second threshold
Eth2 separates the output energy range into two separate bins. The diagonal rectangular sections mark the quanta events that are correctly assigned. The lower right
region contains high-energy bin primary events that are falsely assigned to the lowenergy bin.
Figure1.14 shows the consequence of the low-energy shift. We have assumed a
140-kV tungsten X-ray tube input spectrum (see the shaded gray curve). The two
detected spectra in the respective energy bins are given by the two straight curves.
Two effects are visible: First, the low-energy detected spectrum loses low-energy
events. This is because fluorescence events can carry away enough energy from a primary event to reduce the detected energy below the first energy threshold. Second,
the two detected spectra overlap significantly. When we normalize both detected
spectra to 1, we obtain the system weighting functions of the two energy bins. The
overlap reaches around 60% and is thus comparably larger than the 4050% overlap
of dual kVp or dual-source CT.24 This indicates that the dual-energy measurement
capabilities of direct conversion detectors are significantly affected by low-energy
shift mechanisms due to fluorescence.

Energy Transfer Probability
0.08
0.07
120
0.06
E in keV
100
Ef h2
0.05
80
0.04
60
0.03
False low energy
bin counts
40
0.01
20
Ef h1
(noise cuto )
0.02
20
40
60
80
120
100
E in keV
Figure 1.13 Schematics of energy binning for the detector response function of Fig. 1.10a.
Two energy threshold levels Eth1 = 15 keV, Eth2 = 55 keV are used. (From Heismann, B. J.,
Henseler, D., Niederloehner, D., Hackenschimed, P., Strassburg, M., & Wirth, S. IEEE Room
Temperature Seminconductor Workshop, 2008. With permission.)
14
Input spectrum
Eective spectrum in low energy bin
Eective spectrum in high energy bin
12
Intensity
10
8
6
4
2
0
20
40
60
80
E in keV
100
120
140
Figure 1.14 Detected spectra for a 140-kV tungsten input spectrum (shaded gray). (From
Heismann, B. J., Henseler, D., Niederloehner, D., Hackenschimed, P., Strassburg, M., &
Wirth, S. IEEE Room Temperature Seminconductor Workshop, 2008. With permission.)
16
1.5Conclusions
From the MTF results, we can deduce that the spatial resolution of semiconductor
detectors is a clear potential benefit for medical imaging devices. The direct conversion of the primary X-ray field information into charge pulses omits the interpixel cross talk of scintillator detectors almost completely. The direct conversion
into charges demands a strict control of electrical semiconductor defects. Low-flux
applications in CT and other medical X-ray devices probably benefit the most from
the improved spatial resolution, since it requires less image filtering for the same
obtained image resolution.
The intrinsic energy resolution of a counting detector readout is a second potential benefit of a direct conversion semiconductor detector. For gamma-ray emission
applications like SPECT and PET, the registered charge is a direct measure for the
primary quantum energy. X-ray applications usually require only two or three energy
bins defined by threshold energies. The DRF results indicate that CZT semiconductor detectors are prone to a shift of quantum detection to lower energy bins due to
interpixel fluorescence cross talk.
The required spatial and spectral resolutions in semiconductor detectors are
defined by the targeted medical device. For mammography and radiography detectors, spatial resolution is essential. CT relies on the detection of high X-ray fluxes
at intermediate spatial and spectral resolution. SPECT and PET detectors mainly
require a superior energy resolution. For each of these applications, detector parameters like the pixel size and the electronic readout have to be balanced accordingly.
For mammography and radiography detectors, spatial resolution is essential. SPECT
and PET detectors mainly require a superior energy resolution. For each of these
applications, detector parameters like the pixel size and the electronic readout have
to be balanced accordingly.
References

1. G. F. Knoll, Radiation Detection and Measurement, 3rd ed. Wiley, New York, 2000.
2. I. M. Blevis, M. K. OConnor, Z. Keidar, A. Pansky, H. Altman, and J. W. Hugg, CZT
gamma camera for scintimammography, Phys. Med. Biol., 21 (Suppl. 1), 5659, 2006.
3. K. B. Parnham, S. Chowdhury, J. Li, D. J. Wagenaar, and B. E. Patt, Second-generation,
tri-modality pre-clinical imaging system, M0629, IEEE Nuclear Science Symposium
Conference Record, San Diego, CA: 2007.
4. D. J. Wagenaar, J. Zhang, T. Kazules, T. Vandehei, E. Bolle, S. Chowdhury, K. Parnham,
and B. E. Patt, In vivo dual-isotope SPECT imaging with improved energy resolution,
MR13, IEEE Nuclear Science Symposium Conference Record, San Diego, CA: 2007.
5. E. Kraft, P. Fischer, M. Karagounis, M. Koch, H. Krueger, I. Peric, N. Wermes, C.
Herrmann, A. Nascetti, M. Overdick, and W. Ruetten, Counting and integrating readout
for direct conversion X-ray imaging: concept, realization and first prototype measurements, IEEE Trans. Nucl. Sci., 54 (2), 383390, 2007.
6. D. Moraes, J. Kaplon, and E. Nygard, CERN DxCTA counting chip, Proceedings of
the 9th International Workshop on Radiation Imaging Detectors, Erlangen, Germany:
2007.

7. Y. Onishi, T. Nakashima, A. Koike, H. Morii, Y. Neo, H. Mimura, and T. Aoki, Material

discriminated X-Ray CT by using conventional microfocus X-ray tube and CdTe imager,
M272, IEEE Nuclear Science Symposium Conference Record, Honolulu, HI: 2007.
8. J. P. Schlomka, E. Roessl, R. Dorscheid, S. Dill, G. Martens, T. Istel, C. Baeumer,
C. Herrmann, R. Steadman, G. Zeitler, A. Livne, and R. Proksa, Experimental feasibility
of multi-energy photon counting K-edge imaging in pre-clinical computed tomography,
Phys. Med. Biol., 53, 40314047, 2008.
9. S. A. Soldner, D. S. Bale, and C. Szeles, Dynamic lateral polarization in CdZnTe under
high flux X-ray irradiation, IEEE Trans. Nucl. Sci., 54 (5), 723727, 2007.
10. E. Bolotnikov, N. Abdul-Jabber, S. Babalola, G. S. Camarda, Y. Cui, A. Hossain,
E. Jackson, H. Jackson, J. James, K. T. Kohman, A. Luryi, and R. B. James, Effects of
Te inclusions on the performance of CdZnTe radiation detectors, R272, IEEE Nuclear
Science Symposium Conference Record, Honolulu, HI: 2007.
11. G. S. Camarda, A. E. Bolotnikov, Y. Cui, A. Hossain, S. A. Awadalla, J. Mackenzie,
H. Chen, and R. B. James, Polarization studies of CdZnTe detectors using synchrotron X-ray radiation, R273, IEEE Nuclear Science Symposium Conference Record,
Honolulu, HI: 2007.
12. L. Abbene, S. D. Sordo, F. Fauci, G. Gerardi, A. L. Manna, G. Raso, A. Cola, E.
Perillo, A. Raulo, V. Gostilo, and S. Stumbo, Study of the spectral response of CZT
multiple-electrode detectors, N24298, IEEE Nuclear Science Symposium Conference
Record, Honolulu, HI: 2007.
13. B. J. Heismann, K. Pham-Gia, W. Metzger, D. Niederloehner, and S. Wirth, Signal transport in computed tomography detectors, Nucl. Instrum. Methods Phys. Res., 591, 2833,
2008.
14. T. Michel, G. Anton, M. Boehnel, J. Durst, M. Firsching, A. Korn, B. Kreisler, A. Loehr,
F. Nachtrab, D. Niederloehner, F. Sukowski, and P. T. Talla, A fundamental method to
determine the signal-to-noise ratio (SNR) and detective quantum efficiency (DQE) for
a photon counting pixel detector, Nucl. Instrum. Methods Phys. Res., 568, 799802,
2006.
15. I. A. Cunningham, Applied linear system theory, in: Handbook of Medical Imaging,
Vol. 1, J. Beutel, H. L. Kundel, and R. L. van Metter, Eds. SPIE, Bellingham, WA:
2000.
16. J. Giersch and J. Durst, Monte Carlo simulations in X-ray imaging, Nucl. Instrum.
Methods Phys. Res., 591, 300, 2008.
17. S. Agostinelli et al., G4a simulation toolkit, Nucl. Instrum. Methods Phys. Res., 506
(3), 250303, July 2003.
18. J. Allison et al., Geant4 developments and applications, IEEE Trans. Nucl. Sci., 53 (1),
270278, February 2006.
19. B. J. Heismann, D. Henseler, D. Niederloehner, P. Hackenschmied, M. Strassburg,
S. Janssen, and S. Wirth, Spectral and spatial resolution of semiconductor detectors in
medical X- and gamma ray imaging, R031, IEEE Room Temperature Semiconductor
Workshop, Dresden, Germany: 2008.
20. J. D. Eskin, H. H. Barrett, and H. B. Barber, Signals induced in semiconductor gammaray imaging detectors, J. Appl. Phys., 591, 647, 1999.
21. B. Kreisler, J. Durst, T. Michel, and G. Anton, Generalised adjoint simulation of induced
signals in semiconductor X-ray pixel detectors, J. Inst., 3, 11, 2008.
22. T. Michel, Energy-dependent imaging properties of the Medipix2 X-ray-detector,
Proceedings of Science on the 16th International Workshop on Vertex Detectors, Lake
Placid, NY: 2007.
23. M. Rabbani, R. Shaw, and R. van Metter, Detective quantum efficiency of imaging
sytems with amplifying and scattering mechanisms, J. Opt. Soc. Am. A, 4, 895901,
1987.
18
24. B. J. Heismann and S. Wirth, SNR performance comparison of dual-layer detector and
dual-kVp spectral CT, IEEE Medical Imaging Conference Record, pp. 32803822,
Honolulu, HI: 2007.
25. M. N. Wernick and J. N. Aarsvold, Emission Tomography, Elsevier Academic Press,
Amsterdam, 2004.
Radiation
2 Silicon
Detectors with ThreeDimensional Electrodes
(3D Detectors)
Gian-Franco Dalla Betta and Andrea Zoboli
Trento, Italy
Contents
2.1 Introduction.....................................................................................................20
2.2 Device Description..........................................................................................20
2.2.1 Basic Concept......................................................................................20
2.2.2 Technology Computer-Aided Design Simulations.............................. 23
2.3 Fabrication Technology...................................................................................26
2.3.1 Full 3D Detectors with Active Edge....................................................26
2.3.2 Alternative Approaches....................................................................... 31
2.4 Applications..................................................................................................... 33
2.5 Experimental Results....................................................................................... 35
2.5.1 Main Results for Full 3D Detectors (Stanford)................................... 35
2.5.2 Results for 3D Detectors from Other Manufacturers.......................... 39
2.6 3D Detector Developments in Trento.............................................................. 39
2.6.1 3D-STC................................................................................................ 39
2.6.1.1 Device Description and Simulations..................................... 39
2.6.1.2 Fabricated Devices and Electrical Properties....................... 42
2.6.1.3 Functional Characterization.................................................. 43
2.6.2 3D-DDTC............................................................................................ 45
2.6.2.1 Simulations...........................................................................46
2.6.2.2 Fabricated Devices................................................................ 49
2.6.2.3 Experimental Results from the First Batch.......................... 52
2.6.2.4 Experimental Results from the Second Batch...................... 59
2.6.3 Next Steps............................................................................................60
2.7 Conclusions...................................................................................................... 61
Acknowledgments..................................................................................................... 61
References................................................................................................................. 62
19
20
2.1Introduction
Silicon radiation detectors have been extensively used in a large variety of scientific,
medical, and industrial applications for many years.1,2 A major breakthrough in
silicon detector technology came in the early 1980s from Kemmer, who pioneered
the use of the planar fabrication process, derived from microelectronics;3 exploiting
the passivation properties of silicon dioxide and keeping the thermal budget to a
minimum, ion-implanted detectors, which allow for fine pitch segmentation of the
electrodes and very low leakage currents, became available. Since that time, silicon
detector technologies have been continuously advancing, and more complex and
reliable detectors could be obtained, featuring outstanding performance in terms of
energy, timing or position resolution, long-term stability, and radiation tolerance.
In the recent history of radiation sensors, another milestone was the introduction
of detectors with three-dimensional electrodes (3D detectors), which were first proposed in 1997 by Parker and collaborators.4 Unlike planar detectors, for which the
electrodes are confined to the wafer surfaces, in 3D detectors electrodes penetrate
entirely through the substrate perpendicular to the surface. This architecture offers a
number of substantial advantages with respect to the planar one, making 3D detectors ideal candidates for some critical applications, especially in high-energy physics
(HEP). Of course, this comes at the expense of a more complicated and expensive
fabrication process, which combines microelectronic and MEMS (micro-electromechanical system) technologies.
In this chapter, 3D detectors are reviewed, covering all relevant aspects: device
description and simulation, fabrication technology and design issues, application
fields, and selected experimental results. Alternative 3D detector concepts, aimed at
a simplification of the fabrication technology, are also addressed. Among these are
3D detectors developed in Trento, Italy, discussed in Section 6.
2.2Device Description
2.2.1Basic Concept
The 3D detectors consist of an array of columnar electrodes of both doping types
arranged in adjacent cells and oriented perpendicular to the wafer surface, penetrating entirely through a high-resistivity silicon substrate.4 Electric field lines begin at
one electrode type and end at the closest electrode of the opposite type in parallel
with the wafer surface. Like standard detectors, the strength of the electric field is
controlled acting on the bias voltage. Either single-column or multicolumn arrangements can be adopted for the electrodes, using diffusion or metal surface connections,
so that several types of detectors can be obtained (e.g., pixel, strip, pad, etc.).
This configuration offers many advantages over the planar one, as detailed in the
following with the aid of Figure2.1. In standard planar detectors, the electrodes are
implanted on the top and bottom surfaces of the wafer, so that the depletion region
grows vertically, and the full depletion voltage depends on the substrate thickness. On
the contrary, in 3D detectors the electrode distance L and the substrate thickness can
Silicon Radiation Detectors with Three-Dimensional Electrodes (3D Detectors) 21

p+
MIP
p+
n+
p+
MIP
n+
n+
h
L
(a) (b)
Figure 2.1 Schematic cross sections of (a) planar detector and (b) 3D detector, emphasizing the decoupling of active thickness and collection distance L in 3D detectors.
be decoupled:* The depletion region grows laterally between the electrodes, whose
distance is much smaller (about a factor of 10) than the substrate thickness, so that the
full depletion voltage can be dramatically reduced (about a factor of 100) with respect
to planar detectors. Apart from the related savings in terms of power consumption,
this property is extremely important in applications for which full depletion of planar detectors can be difficult to achieve because of junction breakdown or thermal
runaway problems, among them very thick detectors, which are of interest for X- and
gamma-ray detection, and heavily irradiated detectors in HEP applications.
As far as the charge collection properties are concerned, 3D architecture is superior to the planar one in several respects. Let us refer to a minimum ionizing particle
(MIP), traversing the detector and producing a uniform electronhole (e-h) pair density along its track, as shown in Figure2.1. The amount of generated charge is the
same for both detector types if they have the same substrate thickness. However,
the charge collection distance is much shorter in 3D detectors, and high electric
fields as well as carrier velocity saturation can be achieved at very low voltage, so
that the charge collection times can be much faster (on the order of a few nanoseconds, compared to a few tens of nanoseconds for planar detectors). In particular,
using Ramos theorem, one can estimate the signal current from the carrier velocity
and the weighting field.5 In planar detectors, each charge carrier is generated at a
different distance from the collecting electrodes, thus inducing its peak signal at different times. This effect is strongly attenuated in 3D detectors, in which all charges
along the ionization track are generated within a much shorter distance from the
electrodes, thus inducing a signal with faster rise time.6 In addition to easing applications requiring very high speed, this property can counteract charge-trapping effects
*
This assumption neglects constraints due to the maximum aspect ratio (i.e., depth/diameter) achievable
for columnar electrodes, which are discussed in Section 2.3.
22
due to high levels of radiation in HEP experiments.7 Finally, due to their peculiar
structure, which provides a self-shielding effect in each cell, in 3D detectors charge
sharing between adjacent electrodes is considerably reduced with respect to planar detectors with small pixel sizes, making them appealing for photon-imaging
applications.8
Apart from the technological complications, detailed in Section 2.3, 3D detectors
are also affected by some functional disadvantages with respect to planar detectors:
The signal response to particles is not spatially uniform because of the
existence of zero (or at least very low) field regions within the active volume, which are due to (a) the null points between electrodes of the same
doping type and (b) the electrodes themselves. As a result, charge carriers
generated in these regions have to diffuse until they reach a region with a
sufficient electric field, thus delaying the signal response and lowering the
efficiency in the charge collection process.
The short distance between the electrodes and their deep extension all
through the substrate cause the capacitance to be quite high, degrading the
noise performance at short shaping times that are of primary interest for
these fast detectors.
One important evolution of the 3D concept should also be mentioned, that is,
the so-called active edge. In planar detectors (see Figure2.2), the active region is
normally kept far away from the scribe line so the bulge of the edge of the electric
field in the depletion region does not reach the defects (e.g., cracks, chipped regions)
related to the saw cut, which could otherwise inject high currents. Moreover, additional space is typically required for guard rings aimed at evenly distributing the lateral voltage drop and enhancing the breakdown performance (see Da Rold et al.9 and
references therein). As a result, a dead region exists at the sensor edges, which can
extend for some hundred micrometers and up to about 1 mm in some designs. In 3D
detectors, heavily doped trenches (having the same characteristics as columnar electrodes except for the geometry) can be used to terminate the structures, thus minimizing the dead region area. A precursor to this solution, the wall electrode, has
indeed shown that the insensitive edge region width can be reduced to a few micrometers.10 The same approach can be employed to produce 3D-planar detectors, that
a
b
d
c
Figure 2.2 Schematic cross section of the edge region in a standard planar detector showing some of the reasons why an insensitive region exists: (a) Space is needed for guard rings;
(b) the saw cut is conductive and (c) contains chips and microcracks, so that (d) the bulge at the
edge of the electric field in the depletion region should be kept far away from it. (From Kenney,
C. J. et al., IEEE Trans. Nucl. Sci., 48, 6, 2001. Copyright IEEE 1999. With permission.)10
is, detectors with collection electrodes arranged by standard planar design but with
scribe lines performed as trenches.11
2.2.2Technology Computer-Aided Design Simulations

Deep insight into the behavior of 3D detectors can be gained by means of numerical
device simulations, for which several commercial software packages for technology
computer-aided design (TCAD) are now available (e.g., Silvaco and Synopsys).12 In
Parker et al.,4 simulation results relevant to the static and dynamic characteristics
of 3D detectors are reported. A typical unit cell in a 3D detector with 25-m pitch
between columns is shown in Figure2.3a. Because of symmetry considerations, the
simulation domain can be reduced to a quarter of the unit cell to limit the computational effort. The 3D structure of the electrodes allows the detector properties relevant
n+
n+
n+
n+
5 V, 1012
p+
n+
5.0
n+
p+
n+
n+
Distance (microns)
n+
50 m
0.0
50 m
10.0
15.0
20.0
300 m
n+
n+
25.0
0.0
5.0
10.0
15.0
Distance (microns)
20.0
25.0
(b)
0.0
p+
5 V, 1012
Distance (microns)
5.0
(a)
10.0
15.0
20.0
25.0
0.0
n+
5.0
10.0
15.0
Distance (microns)
(c)
20.0
25.0
n+
Figure 2.3 (a) 3D view of a unit cell in a 3D detector; (b) equipotential lines for onequarter of the unit cell with 1012 cm3 p-type substrate doping concentration and 5-V reverse
bias voltage; (c) drift lines in the same condition as in (b). (From Parker, S. I. et al., Nucl.
Instrum. Methods A, 395, 1997. With permission of Elsevier.)4
24
to the bulk (i.e., in the region not perturbated by surface effects) to be accurately
predicted by two-dimensional (2D) simulations performed along a plane parallel to
the wafer surface. In case of a detector with p-type substrate doping concentration
NA = 1012 cm3, simulations predict a full depletion voltage of just 1.6 V (including
the contribution from the built-in voltage). The full depletion voltage is increased to
8.8 V at NA = 1013 cm3, which represents the effective doping concentration expected
after 10 years of operation in the Large Hadron Collider (LHC) for pixel detectors
due to radiation damage.
As an example, Figure2.3b shows the equipotential lines in a quarter cell with
NA = 1012 cm3 and a reverse bias (Vrev) of 5 V, whereas Figure2.3c shows the corresponding drift lines. In both figures, low-field regions between two electrodes of
the same type can be observed, as also evident from Figure2.4, in which the electric
field profiles at different Vrev along two lines connecting the electrodes are shown.
In Figure2.4a (i.e., along a line from the p+ to the adjacent n+ electrode), the electric
field magnitude is high enough to provide fast drift, whereas the field peaks at the
electrode edges are still safely below the critical electric field in all bias conditions.
On the contrary, in Figure2.4b (i.e., along a line connecting two adjacent n+ electrodes) an almost-zero field region, about 2- to 3-m wide, is present at the midpoint
between the two electrodes.
These nonuniformities in the electric field configuration cause the carrier drift
time (and the signal shape) to be strongly dependent on the particle impact position.
1.00
p+ to adjacent n+
0.90
0.90
0.80
0.80
0.70
0.70
Electric Field (V/cm) 105
Electric Field (V/cm) 105
1.00
0.60
0.50
0.40
0.30
0.60
0.50
0.40
0.30
0.20
0.20
0.10
0.10
0.00
0.0
5.0
10.0
15.0
20.0
25.0
n+ to adjacent n+
0.00
0.0
5.0
10.0
15.0
Distance (microns)
Distance (microns)
(a)
(b)
20.0
25.0
Figure 2.4 Electric field profiles for the quarter cell of Figure 2.3a with 1012 cm3 p-type
substrate doping concentration and reverse bias voltages of 50, 40, 30, 20, 10, 5, and 0 V
(curves from top to bottom), along lines from (a) the p+ electrode to the adjacent n+ electrode
and (b) the n+ electrode to the adjacent n+ electrode. (From Parker, S. I. et al., Nucl. Instrum.
Methods A, 395, 1997. Copyright Elsevier. With permission.)4
With reference to the same cell as in Figure2.3a, electrons and holes generated at
the cell center are collected within 1 ns, whereas a longer time (up to 5 ns for holes)
is necessary if charge is generated at the null point between electrodes.4 This is
reflected in markedly different signal shapes, as highlighted in Figure 2.5, which
compares current pulses on the electrodes from particles hitting perpendicularly to
4.00
(AMPS/Micron) 108
Cell center
2.00
0.00
P+
2.00
4.00
0.00
0.50
1.00
1.50
2.00
2.50 3.00
Time (ns)
(a)
3.50
4.00
4.50
5.00
7.0
8.0
9.0
10.0
1.00
(AMPS/Micron) 108
Null point
0.50
0.00
0.50
1.00
0.0
P+
1.0
2.0
3.0
4.0
5.0
6.0
Time (ns)
(b)
Figure 2.5 Current pulses on the electrodes in response to an MIP with a track parallel to
the electrodes and passing (a) through the cell center and (b) through the null point in between
two n+ electrodes (NA = 1012 cm3, Vrev = 10 V). (From Parker, S. I. et al., Nucl. Instrum.
Methods A, 395, 1997. Copyright Elsevier. With permission.)4
26
the detector surface in two different positions. Nevertheless, also in the worst case,
signals are much faster than in planar detectors with much lower peak fields.
Simulations also predict the RC time constants related to the resistance and
capacitance of the electrodes to be in the order of 100200 ps,4 small enough for
most applications.
Finally, it should be mentioned that close to the top and bottom surfaces of 3D
detectors, both static and signal characteristics deviate from those observed in the
bulk. This is due to oxide fixed charge that results in an electron accumulation layer
at the interface and to the surface isolation structures (p-spray or p-stops13), which
prevent the n+ columnar electrodes from being resistively connected (these are strictly
necessary only if n+ column readout is used). Interface charge and surface structures
cause significant distortions in the equipotential lines and electric field profiles, also
affecting the charge collection properties. For these effects to be accurately predicted, 3D simulations are generally necessary. Alternatively, 2D simulations performed along a planar cross section including two half-columns were reported in
Parker et al.;4 as an example, Figures2.6a and 2.6b show the equipotential lines in a
structure with NA = 1012 cm3 and interface charge density of 1011 cm2 at 0 and 10 V
bias, respectively. At 0 V, the closest equipotential lines are almost parallel to the
surface, and their distribution starts approaching the ideal (vertical) one only several micrometers below the surface. This picture is only partially improved at 10 V
owing to the depletion of the surface electron layer at the interface. In fact, also at
higher reverse bias, the surface electron layer is not completely depleted close to the
n+ electrode, as shown in Figure2.6c. Of course, in case of higher interface charge
densities, as well as in the presence of surface isolation regions, deviations from the
ideal detector behavior are even more pronounced. Simulations also showed that
surface effects play a major role in decreasing the 3D detector breakdown voltage14
and in increasing the electrode capacitance,15 especially after irradiation.
2.3Fabrication Technology
2.3.1Full 3D Detectors with Active Edge
The fabrication of 3D detectors has become possible owing to the rapid development of MEMS technologies during the 1990s and, in particular, to the availability
of deep reactive ion etching (DRIE) equipment, allowing columnar holes with high
aspect ratio (depth to diameter) to be obtained. Etching rates higher than 3 m/min,
selectivity to masking materials higher than 70:1, very good etching profiles, and
nonuniformities lower than 5% across the wafer can be achieved.16 The so-called
Bosch process is currently a standard for DRIE;17 it repeats several times a two-phase
etching cycle based on fluorine compounds, alternating nearly isotropic plasma etching steps (by SF6) and sidewall passivation steps (by C4F8) to achieve high anisotropy
in the overall etching profiles (see Figure2.7). Aspect ratios on the order of 30:1 are
now possible with this technique; this value should be kept in mind when evaluating
the current limits in the 3D technology.
The key aspects relevant to the fabrication of 3D detectors were addressed in
Parker et al.4 and further detailed in Kenney et al.18 With reference to these articles,
0.0
1.0
1.0
2.0
2.0
3.0
3.0
4.0
5.0
6.0
7.0
5.0
6.0
7.0
8.0
9.0
9.0
5.0
10.0
15.0
20.0
25.0
18
4.0
8.0
10.0
0.0
10 V, 1012 cm2
Log (concentration (cm3))
0.0
Distance (microns)
Distance (microns)
10 V, 1012 cm2
10.0
0.0
17
16
15
n+
p+
0V
5V
10 V
14
20 V
13
12
11
5.0
10.0
15.0
20.0
25.0
10
0.0
5.0
10.0
15.0
20.0
Distance (mircons)
Distance (mircons)
Distance (mircons)
(a)
(b)
(c)
25.0
Figure 2.6 The 2D simulations along a planar cross section of the unit cell of Figure 2.3a from the center of the p+ electrode to the center of the
adjacent n+ electrode (interface charge density is 1011 cm2, NA = 1012 cm3). (a) Equipotential lines at Vrev = 0 V; (b) equipotential lines at Vrev = 10 V;
(c) net carrier concentration profile along a line 0.1 m below the surface at different bias voltages. (From Parker, S. I. et al., Nucl. Instrum. Methods A,
395, 1997. Copyright Elsevier. With permission.)4
0 V, 1012 cm2
28
ITC irst
SEI
5.0 kV
5,000
1 m
WD 6.3 mm
Figure 2.7 Top: Schematic representation of the etching sequence by the Bosch process:
(a) Patterning of the masking layer; (b) first shallow isotropic etching by SF6; (c) protective
layer deposition by C4F8; (d) removal of the protective film from the horizontal surfaces by
directional ion bombardment followed by second shallow isotropic etching. (Copyright IOP
2001, from Ayn et al.)16 Bottom: (e) Scanning electron micrograph showing a detail of the
column sidewalls after DRIE etching with the characteristic scalloping effect. (From Ayn,
A. A. et al., Smart Mater. Struct., 10, 2001. With permission.)
the main process steps are summarized as follows, which is a discussion of the parts
of Figure2.8:

a. Oxidation and wafer bonding. Oxidation is typically performed first in

any detector technology for surface passivation purpose. On the contrary,
wafer bonding is a relatively recent technology that has emerged mainly for

Resist
n+ polysilicon
Detector wafer
Oxide
Support wafer
(a)
(b)
Resist
(d)
(c)
Metal
p+ polysilicon
(e)
(f )
Figure 2.8 Schematic representation of the main steps in the fabrication process of 3D
detectors. See discussion in text.
MEMS applications (see Christiansen et al.19 and references therein). This

technique consists of the process of adhesion of two wafers by exploiting
hydrogen-bonding forces at room temperature, which are further enhanced
by high-temperature annealing. In the context of 3D detector fabrication,
a sacrificial support wafer is bonded to the detector wafer because (a) it
provides high mechanical resistance, preventing the detector wafer from
damage by cracks possibly developed during electrode-etching steps and
other stress-inducing steps, increasing the process yield; (b) as will be seen
at the end of this section, it is mandatory in the case of detectors with active
edge for silicon pieces to be held together after the trench etching; (c) it
allows the detector wafer to be back thinned to the desired thickness without affecting the process yield, thus increasing design flexibility.
b. Hole definition and etching. Geometries of the p+ columnar electrodes are
defined by lithography and oxide etching. Then, deep holes are etched in
silicon by DRIE using the oxide layer and a thick photoresist layer as a protective mask. Photoresist is then ashed at the end of the etching.
In the earlier fabrication attempts, the hole aspect ratio achieved was at
most 11.5:1, so that column diameters larger than 20 m were necessary for
250-m deep columns to be obtained.18 As mentioned, new DRIE equipment now enables aspect ratio values as high as 30:1, so that columns can be
etched all the way through wafers of standard thickness while maintaining
a diameter of about 10 m.
c. p+ hole filling and doping. Holes are completely filled by polysilicon deposited in a low pressure chemical vapor deposition (LPCVD) reactor with conformal coating provided by properly setting the temperature and pressure
conditions. As for the column doping, it is performed by thermal diffusion
30
from a boron source either before the polysilicon deposition or after a first
polysilicon deposition step, partially filling the holes, and later followed by
a second deposition.
The fact that holes are completely filled with polysilicon offers some
advantages: During the following lithography steps, photoresist can be uniformly spinned on the wafer surface and does not get trapped in the holes;
moreover, in case a particle traverses the polysilicon electrodes, at least part
of the generated charge could be collected, provided that carrier lifetimes
in polysilicon are not too short since the charge motion in the electrodes
is governed by diffusion. In particular, carrier lifetimes in polysilicon are
proportional to the size of grain boundaries (0.5 ns correspond to a 1-m
grain size), and a thermal treatment after polysilicon deposition can cause
recrystallization and grain size increase.18 Note that in Parker et al.,4 hole
filling by crystalline silicon using an epitaxial process was also mentioned;
this involves a solution that would ease charge collection from the electrodes due to the electric field sustained by doping concentration gradient,
but this solution is very difficult to implement in practice.
The problem with polysilicon filling of the holes is that the wafer surface
is finally covered by a thick polysilicon layer (10 m or more). Etching such
a thick layer of polysilicon is a difficult task; as an alternative, chemical
mechanical polishing (CMP) can be used.20 In all cases, this step further
increases the process complexity.
d. n+ hole definition and etching. Similar to step b, geometries of the n+ columnar electrodes are defined by lithography and oxide etching, and deep holes
are etched by DRIE.
e. n+ hole filling and doping. Similar to step c, holes are completely filled by
polysilicon and doped by thermal diffusion from a phosphorus source. The
thick polysilicon layer has to be removed from the surface one more time.
f. Metal deposition and definition. An oxide layer is deposited; then, contact
holes are defined and etched through the oxide, and metal is deposited and
patterned. A final passivation layer could then be deposited and etched only
in the probe/bonding pad regions.
Not shown in Figure 2.8 is the final removal of the sacrificial support wafer.
Moreover, this process sequence does not account for two additional ion implantation steps that would be necessary if the signals were read out from n+ electrodes:
the isolation implantations (p-spray or p-stop) on both the top and bottom sides of the
wafer.
The same process could be completed by a few additional steps for active edges
to be implemented:10
One of the two hole-etching steps could be used to etch border trenches at
the same time. Owing to the length of trenches along the sensor edge (at
least a few millimeters), the etching gas and the etching products can enter
and leave trenches more easily than they do for columnar holes. Thus, for

p
p
Sensor wafer
Support wafer
Oxide
(a)
p
p
Support wafer
Oxide
(b)
Figure 2.9 Schematic representation of the active edge process showing two adjacent
detectors still bonded to a support wafer: (a) sensors after the holes and trenches for the n
and p electrodes have been etched, doped, and filled; (b) sensors after the larger dicing trench
(reaching to the dashed white lines in the top diagram) have been etched to separate the sensors. (Kenney, C. J. et al., IEEE Trans. Nucl. Sci., 48, 6, 2001. Copyright IEEE 2001. With
permission.)10
the trench and hole vertical etch rates to be comparable, the trench width
should be smaller than the column diameter. Trenches should then be filled
and doped like columnar electrodes, as shown in Figure2.9a.
An additional etching is needed at the end of the process to remove all the
material surrounding the detector while leaving a few micrometers of polysilicon to protect the sensor from mechanical damage and contamination
from impurities, as shown in Figure2.9b.
2.3.2Alternative Approaches
The described fabrication process is indeed long and complicated since it involves
several nonstandard steps. As a matter of fact, the existing prototypes of full 3D
detectors with active edge were fabricated at a research laboratory (the Stanford
Nanofabrication Facility). The feasibility of large-scale production by an industrial
foundry has not been demonstrated yet, and fabrication yield and costs are major
concerns in view of future mass production of 3D detectors. However, it should be
mentioned that two industrial vendors have started the development of 3D detectors:
32
(a)
(b)
(c)
(d)
(e)
(f )
Figure 2.10 Schematic cross sections describing the modified 3D detector architectures
so far reported: (a) single-type column 3D detectors, also called semi-3D detectors, with
back-side ohmic contact, independently proposed by FBK-irst21 and VTT;22 (b) an alternate
version of (a) with passing-through column, proposed by FBK-irst;21 (c) single-type column
3D detectors with front-side ohmic contact, proposed by BNL/CNM;23 (d) and (e) doublesided, double-type column detectors with slightly different back-side configuration, independently proposed by FBK-irst25 and CNM, respectively; (f) single-sided, double-type column
detector proposed by BNL.23
SINTEF (Norway) and IceMOS Technology Limited (Northern Ireland). Moreover,

some other research institutes are involved in the development of modified 3D detector architectures, allowing for simplified fabrication technology.
The most interesting among the alternative structures so far proposed are summarized in Figure2.10. The device sketched in Figure2.10a has columnar electrodes
of one doping type only, and not passing through the entire substrate, with a uniform
ohmic contact on the back side. It has been independently proposed by Fondazione
Bruno Kessler (FBK-irst, Trento, Italy) with the name single-type column 3D detector (3D-STC)21 and by VTT (Helsinki, Finland) with the name semi-3D detector.22
The only differences between the two detectors is the column-doping type (n-type
for FBK-irst, p-type for VTT) and the fact that in FBK-irst detectors columns are
hollow, whereas they are filled with polysilicon for VTT. FBK-irst also proposed
another 3D-STC version with passing-through columns and back-side, grid-shaped
ohmic contact (see Figure2.10b),21 but this was not fabricated. Details about 3D-STC
detectors from FBK-irst are given in Section 2.6.1.
A 3D detector with single-type columns has also been proposed by BNL (Upton,
NY, USA) in collaboration with CNM (Barcelona, Spain);23 it differs from the FBKirst 3D-STC only because, instead of a blank p+ implant on the back side, the p+
regions are patterned and implanted on the front side (see Figure2.10c). This onesided technological solution makes the fabrication of detectors in the so-called stripixel configuration possible.23 Simulation results also suggested that the electric field
configuration is better than in 3D-STC detectors made by FBK-irst, thus improving
the charge collection properties.24 Nevertheless, no experimental results but some
preliminary leakage current measurements have been published for these devices.
Common to all these versions of 3D detectors with columnar electrodes of one type
only is a major simplification in the fabrication technology with respect to standard
3D detectors, but this comes at the expense of worse charge collection efficiency
(CCE) and radiation resistance.
Better performance is expected from double-sided, double-type column 3D detectors, which have been independently proposed by FBK-irst25 and CNM, in collaboration with the University of Glasgow26 (see Figures2.10d and 2.10e, respectively):
Columnar electrodes are etched from both wafer sides and stopping a short distance from the opposite surface. This solution still provides some advantages from
the viewpoint of process complexity. FBK-irst devices differ from CNM ones by the
facts that they feature an ohmic contact region made by a uniform doping layer
on the wafer back side in addition to the back-side columns, and that columns are
hollow rather than partially filled with polysilicon. Also for double-type column
detectors, BNL has proposed a one-sided alternative (see Figure2.10f), for which,
however, no results are available. For all these modified 3D detectors with doubletype columns, simulations predict overall performance comparable to that of full 3D
detectors provided that the gap between column tips and the opposite wafer surface
is small enough (on the order of 1020 m).14,2426 Further details about FBK-irst
3D-DDTC detectors are given in Section 2.6.2.
2.4Applications
In the past 15 years, the research and development in the field of radiation detectors has been strongly focused on the experiments to be performed at LHC at the
European Organization for Nuclear Research (CERN). The particle-tracking systems
employed at LHC are indeed mainly based on silicon detectors of microstrip, pixel,
and drift types. Because of the high luminosity (nominal value of 1034 cm2 s1), during the 10 years of operation of the accelerator, the detectors will be exposed to very
high radiation fluences, leading to severe performance deterioration. High-energy
particles (pions, neutrons, protons, etc.) cause displacement of silicon atoms from the
lattice of the crystal. Vacancies and interstitials created through this mechanism will
then migrate to form stable defects with the impurities, always present in the material, like oxygen and carbon, and with the dopants, like phosphorus and boron. The
main macroscopic consequences of these defects in the detector bulk are (a) changes
in effective doping concentration, which lead to an increase of the full depletion
voltage; (b) higher leakage currents due to the creation of generation/recombination
centers; and (c) deterioration of the CCE due to carrier trapping.27
The resulting degradation of the detector performance could compromise the
success of the entire experiment. This is the reason why the ATLAS experiment
has decided to replace part of its tracker after a few years of operation with radiation harder detectors, for which 3D technology is a strong candidate. Even more
challenging is the luminosity upgrade of the LHC (Super-LHC or SLHC) up to
1035 cm2 s1, corresponding to equivalent hadron fluences higher than 1016 cm2 in
the innermost detector layers after 5 years of operation.7 At such high fluences, standard planar detectors are not favored, mainly because of the dramatic decrease of
34
the collected charge (signal amplitude) due to charge trapping. Owing to its intrinsic
radiation hardness and speed, 3D technology is one of the most promising solutions
for the extreme radiation environment foreseen at the SLHC.28,29
Another key feature of 3D technology is the feasibility of active edge detectors,
which would allow for large-area seamlessly tiled detector matrices, that is, omitting
sensor overlap within the same layer. This greatly facilitates the layout, reduces the
material budget, and therefore improves the momentum resolution, as required by
other HEP applications at the next generation of machines, such as the International
Linear Collider30 and the Super B-Factory.31
The 3D and active edge detectors are also appealing for applications involving
high-energy photons. To this purpose, advantages of the 3D concept over planar
detectors include the possibility to use thicker substrates to improve the detection
efficiency while maintaining the low operating voltage, owing to lateral depletion between electrodes. As an example, a thick (~5 mm) silicon drift detector has
been proposed, featuring arrays of trenches penetrating the substrate with different
depths and acting as 3D drift cathodes.32 Moreover, in 3D detectors the reduced
charge-sharing effect between neighboring cells can improve the imaging and
differential spectroscopy performance in single-photon counting X-ray imaging
applications.33 In addition, active edges allow for modular detector assembly without
the need for substantial overlapping at the edges, thus offering important advantages for those imaging applications calling for large-area detection systems with a
minimal amount of dead area and dead material, like photon detection in medicine,
biology, and astrophysics. As an example, a pixel detector with active edges, aimed
at X-ray crystallography, was proposed in Parker et al.,34 whereas Kenney et al.11
reported on effective X-ray beam monitoring using novel active edge detectors.
Other innovative detectors for imaging applications can derive from the possibility of filling columnar electrodes with scintillators or other energy converter materials. In particular, the confinement of scintillator materials into the columns can
provide an intrinsically better spatial resolution with respect to the standard coupling
with planar detectors.35,36 Moreover, due to the peculiar properties of charge collection dynamics in 3D structures, detectors for fast imaging of hard X-rays or gamma
rays would be feasible. It should be noted that holes could be filled with materials
sensitive to neutrons, leading to a 3D detector for neutron radiography, which is
a technique complementary to X-ray radiography. While X-rays are attenuated by
heavier material like metals, neutrons allow the imaging of light materials such as
hydrogenous substances, a fact that is very important in a variety of applications,
among them explosive or land mine detection.
The 3D technology can also be useful for the implementation of optical sensors;
as an example, fully tileable photodiode arrays for medical imaging have been proposed, for which the diode contact pads are moved to the insensitive back side of
the detector owing to through-wafer interconnects (i.e., conductive columns passing
all the way through the silicon substrate)37 to allow for (bump or wire) bonding the
photodetectors to the readout electronics without blocking the light on the sensitive
side. Moreover, lateral PIN photodiodes that are CMOS (complementary metal oxide
semiconductor) compatible and based on deep trench electrodes have been proposed
as high-speed receivers for optical interconnects.38
2.5Experimental Results
2.5.1Main Results for Full 3D Detectors (Stanford)
Initial results related to the electrical characteristics and the response to infrared
(IR) light-emitting diode (LED) pulses of the first 3D detector prototypes, made
on 121-m thick p-type wafers, were reported in Kenney et al.18 They include good
leakage current densities, on the order of 1 nA/mm3, and breakdown voltages larger
than 60 V, to be compared to full depletion voltages of 5 and 8 V for the 100 and
200 m electrode pitches, respectively.
The first charge collection characteristics in response to X-rays and particles
were reported in Kenney et al.;39 the full width at half maximum (FWHM) energy
resolution at the manganese K line of a 55Fe source is 652 eV, a value that is well
explained by the combined effect of the detector capacitance and leakage current at
the considered shaping time (1 s). Measurements confirmed the low charge sharing
between adjacent cells characterizing 3D detectors. Using a 106Ru source and coupling the detectors with fast electronics, effective detection of particles in coincidence with a scintillator was also demonstrated.
The first radiation hardness tests were described in Parker and Kenney.40 Detectors
were irradiated with 24-GeV/c protons at a fluence of about 5 1014 cm2 and with 55
MeV protons up to a fluence of 1015 cm2. It was shown that the depletion voltage for
the 100-m electrode pitch increased to about 105 V after the largest fluence, and that
the leakage current increase was in good agreement with expectations, with a damage constant of (45) 10 17 A/cm, well within the generally accepted range. Values
of the capacitance (~0.1 pF/electrode) were close to those predicted by calculations.4
Devices from the same batch as those studied in Parker and Kenney.40 were also
irradiated and measured independently by a research group from the University of
New Mexico.41 A capacitance increase by 70% was shown for detectors after irradiation with 55-MeV protons at a fluence of 1015 cm2. 3D detectors implementing
a mix of cylindrical electrodes and wall electrodes were tested with an IR microbeam, showing that the active volume extends to within 5 m from the wall electrodes.10 This excellent result was also confirmed in planar detectors with active
edge using a 12.5-keV X-ray microbeam,11,42 for which the sensitive region was also
found to extend to within 5 m from the physical edge of the device (see Figure2.11).
Preliminary results relevant to MIP signals in two 3D detectors, before irradiation
and after irradiation with protons at 1015 cm2 were reported,43 showing a very fast
response and indicating that the CCE was still 60% after irradiation.
Other measurements were carried out using a synchrotron, 12.65-keV X-ray beam
with the aim of investigating the spatial response of 3D detectors.44 The charge collection properties were very good, with an energy resolution of 0.91 keV FWHM. Only
the electrodes were found to be dead areas. Charge splitting between adjacent cells
was confined within 20 m at boundaries, but this value was an overestimate due to
the approximately 10-m probe beam size, and there was no loss of charge if pixel
signals were summed. Additional studies with the X-ray beam actually demonstrated
that the electrodes were not completely dead regions, although a strong reduction of
the signal was observed in charge if generated inside them.45
36

Strip 1
1.6
Strip 2
1.4
Signal (a.u.)
1.2
1
0.8
0.6
0.4
0.2
0
50
100
150
250
200
Position (microns)
300
350
400
Figure 2.11 Part of an active edge planar detector and plot of the output signals from the
two segments near the edge of the device. (Kenney, C. J. et al., Nucl. Instrum. Methods A,
582, 2007. Copyright Elsevier. With permission.)42
The most recent and significant works demonstrated the suitability of 3D sensors as
tracking detectors for the innermost pixel layers at SLHC. IR laser tests on 3D detectors irradiated with neutrons up to a 8.6 1015 1-MeV eq. n cm2 and readout with a
fast transimpedance amplifier confirmed that these devices can withstand very large
radiation fluences.46 As an example, Figure2.12 shows the signals of a 3D detector
with 71-m interelectrode pitch in response to IR laser pulses for different irradiation
fluences. The corresponding values of the signal efficiency, which is defined as the
fraction of the original signal available before irradiation, are shown in Figure2.13
as a function of 24-GeV/c equivalent protons fluence. Noticeably, after the largest fluence, which corresponds to about 10 years of operation at the SLHC at 4 cm from the
beam, the signal efficiency was still about 38%. This remarkable figure can indeed be
further improved by using 3D detectors with narrower pitch between the electrodes.
Some 3D pixel detectors bump bonded to the ATLAS FEI3 front-end chip47 have
been measured in a 100-GeV pion beam at CERN Super Proton Synchrotron (SPS).48
The spatial resolution with binary readout corresponds to the theoretical expectations from a cell size of 50 400 m2; the hit efficiency is 95.9 0.1% for orthogonal
incidence of the particles, due to efficiency loss in the electrodes, and increases up
to 99.9 0.1% in the case of 15 track inclination, the last also causing larger cluster

0.002
Amplitude (V)
0
0.002
8.6E15 n/cm2
0.004
5.98E15 n/cm2
0.006
3.7E15 n/cm2
0.008
Nonirradiated
0.01
3 108 2 1081 108 0
1 108 2 108 3 108
Time (s)
Figure 2.12 Averaged (1,000 samples) oscilloscope traces of 3D detector signals in

response to fast IR pulses at 1,060-nm wavelength. Data relevant to different irradiation fluences are compared. (Da Via, C. et al., Nucl. Instrum. Methods A, 587, 2008. Copyright
Elsevier 2007. With permission.)46
100
Signal eciency (%)
80
60
40
20
5 1015
1 1016
1.5 1016
2 1016
Fluence (p/cm2)
Figure 2.13 Signal efficiency of 3D detectors irradiated with neutrons versus the fluence,
which was converted to proton equivalent using the Non Ionizing Energy Loss (NIEL) scaling to 24 GeV/c cm2. (Da Via, C. et al., Nucl. Instrum. Methods A, 587, 2008. Copyright
38

9,000
8,000
300
7,000
250
Entries
Entries
6,000
5,000
4,000
3,000
150
100
2,000
50
1,000
0
200
10
20
30
40
50
10
20
30
Charge (ke)
Charge (ke)
(a)
(b)
40
50
Figure 2.14 Measured distributions of cluster charge for 100-GeV pion tracks under different
inclination angles: (a) 0 angle (i.e., orthogonal to the detector surface) and (b) 15 angle. (From
Mathes, M. et al., IEEE Trans. Nucl. Sci., 55, 6, 2008. Copyright IEEE 2008. With permission.)48
sizes (i.e., larger charge sharing between adjacent pixels). As an example, Figure2.14
compares the measured Landau distributions of the cluster charge in the two cases
with tracks orthogonal to the detector surface (see Figure2.14a) and tracks impinging with an inclination of 15 (see Figure2.14b); low-charge entries affecting the 0
plot due to the electrode inefficiency are not present for 15 incidence, improving the
tracking efficiency. However, it should be noted that for 15 incidence the Landau
distribution is broader by about 20%, and that the lowest detected charge is 5 ke, very
close to the system threshold, thus representing a possible concern, especially for irradiated detectors for which the signal charge is lower. Further test beam results have
been reported,49 confirming the edge sensitivity of these devices (1012 m, probably dominated by tracking resolution and residual misalignment). After irradiation
with 24-GeV/c protons at a fluence of 1.0 1015 cm2, the overall efficiency was about
21% at a bias voltage lower than 5 V (owing to accidental damage from bad handling,
a high leakage current prevented the detector from being biased beyond 5 V).
The 3D sensors from Stanford, combined with LiF conversion material, were
also tested as neutron detectors. A system for neutron detection based on planar
pixel detectors coupled to the Medipix2 chip was proposed.50 An LiF conversion
layer a few micrometers thick, excited with neutrons, releases 2-MeV alpha particles
and 2.72-MeV 3H particles, which can be detected in a silicon detector. However,
the measured efficiency was just around 6% in the case of planar pixel detectors.
An experiment was later carried out with 3D detectors,51 filling the electrode holes
with LiF, and obtaining a detection efficiency of about 30%. The large improvement
with respect to the planar detectors was due to the increased surface between the
neutron converter and the silicon detector and because particles generated inside the
columns now had a higher probability of crossing the sensitive (high-field) volume of
the device. Moreover, in planar detectors either 2-MeV alpha particles or 2.72-MeV
3H particles can be detected as a result of the 180 emission angle, whereas in 3D
detectors both particles can be detected at the same time.
2.5.2Results for 3D Detectors from Other Manufacturers

Briefly, we recall here the results relevant to modified 3D architectures from VTT
and CNM. Results relevant to 3D detectors made in Trento are extensively reviewed
in Section 3.6.
Results from VTT semi-3D detectors22 include (a) leakage currents on the
order of a few picoamperes per column at 100 V (well beyond full depletion);
(b) capacitance at full depletion ranging from 40 to 90 fF/column, which is high
because of a large contribution from the MOS capacitance related to the metal
interconnection between columns at the surface; (c) charge collection tests with
an americium source, with a FWHM energy resolution of 7.7% for the 59.5-keV
peak. Samples were also irradiated with protons at CERN up to 1.0 1016 p/cm 2;
the depletion voltage was found to remain below 100 V even at a fluence 6.0
1015 p/cm 2 but could not be measured at the highest fluence because of problems
caused by very high leakage current. Additional results related to semi-3D pixel
detectors coupled to the Medipix2 readout chip were reported;52 sensors were
shown to effectively detect X-rays from a 109Cd source, but the energy resolution of about 1.63 keV did not allow resolving the K lines at 22 and 25 keV. A
pulsed IR laser system was used to examine the uniformity in the pixel response,
and reduced sensitivity from the pillars was observed. Finally, the good imaging
properties of the device were demonstrated using a W-target X-ray tube operated
at 35 kV.
The first results reported for CNM 3D detectors referred to double-sided, double-type column devices having 250-m deep electrodes on 300-m thick, n-type
silicon wafers. Initial electrical tests showed low depletion voltages (a few volts),
good leakage currents (~1 pA/column), and breakdown voltages higher than 60 V.53,54
Functional characteristics of pixel detectors bump bonded to Medipix2 readout have
been reported.55 X-ray tube tests confirmed the low depletion voltage (~2 V lateral
depletion, ~9 V full depletion); spectroscopic tests using monochromatic 15-keV
X-rays from a synchrotron showed a substantially reduced charge-sharing process
with respect to planar detectors (see Figure2.15). In addition, preliminary results
from the characterization of strip detectors irradiated with reactor neutrons at a fluence of 5.0 1015 cm2 and readout with LHC-speed electronics showed a charge
signal of about 12,800 electrons in response to an MIP, evidence of a high radiation
tolerance, comparable to that of full 3D detectors.56
2.63D Detector Developments in Trento

2.6.13D-STC
2.6.1.1Device Description and Simulations
Since 2004, FBK-irst of Trento (formerly ITC-irst) has been developing 3D detectors with modified architectures, aiming at process simplification. The first step in
this activity has been the 3D-STC detector. On one hand, this approach offers some
advantages from the technological point of view: (a) Columnar electrodes do not
penetrate all the way through the substrate so the sacrificial support wafer is not
40
Dierential count rate (arb. units)
12
106
Planar, 100 V
3D, 22 V
10
8
6
Half beam
energy
Charge
shared
4
2
0
Not
shared
0
10
15
20
Energy (keV)
Figure 2.15 Comparison of spectra from a 15-keV monochromatic X-ray beam measured
with planar and 3D detectors. (From Pennicard, D. et al., Nucl. Instrum. Methods A, 2009.
Copyright Elsevier 2009. With permission.)55
necessary; (b) columns are only of one doping type so etching and doping are performed only once; (c) columns are hollow (e.g., not filled with poly), thus avoiding
both the related deposition and removal steps. On the other hand, in terms of performance, 3D-STC detectors are not as good as full 3D detectors.
TCAD simulations, implementing process and geometrical parameters related to
the fabricated devices, enable insight into the electric field configuration and into the
charge collection mechanism.21 Figure2.16 schematically describes the two depletion
Back side (p+)

(a)
Columns (n+)
(b)
Silicon
(c)
Figure 2.16 Cross section of a region between two columns with different depletion conditions in a 3D-STC detector: (a) initial lateral depletion, (b) depletion toward the backplane,
(c) full depletion. (From Pozza, A. et al., Nucl. Instrum. Methods A, 560, 2007. Copyright

4E07
Current (A)
3E07
Simulated cell
2E07
Impact point
25 m
Na = 1E13 1/cm3
Na = 5E12 1/cm3
1E07
25 m
0E+00
0E+00
1E08
Current (A)
2E05
2E08
Time (s)
(a)
Na = 1E13 1/cm3
Na = 5E12 1/cm3
1E05
3E08
4E08
Simulated cell
10 m
10 m
Impact point
0E+00
0E+00
1E09
2E09
3E09
Time (s)
(b)
4E09
5E09
Figure 2.17 Simulation of the current signals induced by an MIP particle impinging
a 3D-STC detector in two different positions (see insets). (From Piemonte, C. et al., Nucl.
Instrum. Methods A, A541, 2005. Copyright Elsevier 2007. With permission.)21
phases in an STC detector: lateral depletion between columns and then vertical depletion toward the back-side electrode in a planar-like fashion.57 Since columns are all
of the same doping type, once lateral depletion is reached, the electric field between
columns cannot be further increased, and it depends only on the substrate doping concentration. As a result, low-field regions are present midway between two columns.
The signal induced by an MIP particle hitting a 3D-STC structure at different impact points has been simulated, and the induced currents are plotted in
Figure2.17.21,58 There is a fast signal component of a few nanoseconds due to electrons and holes drifting horizontally toward the nearest electrode and toward the center of the structure, respectively. The collection time of electrons strongly depends
on the impact position, as can be understood from the different peak position in the
two considered cases. Then, the induced signal has a long tail that lasts for microseconds due to the slow diffusion of holes toward the back plane. This part of the signal
42
does not depend on the impact position. During their motion to the back side, holes
start drifting only when they reach the region below column tips, where the vertical
electric field is greater than zero.
2.6.1.2Fabricated Devices and Electrical Properties
Three batches of 3D-STC detectors were fabricated on high resistivity p-type substrates and with n+ column readout.59 The column depths varied from 150 to 180 m
with a diameter of 10 m, whereas the substrates were 300-m thick CZ wafers
and 380- and 500-m thick FZ wafers, with different resistivities. The implemented
surface isolation was either p-stop or p-spray. The layout included planar test structures, such as diodes and MOS capacitors, aimed at parametric testing (i.e., substrate
doping concentration, oxide charge density, etc.), but 3D microstrip detectors with
different layout solutions covered most of the wafer area.60 There were long (1.8-cm)
microstrip sensors, with 1-cm2 active area, and short (1-mm) strip sensors with 5-mm2
active area. Each strip is formed by columns connected either by surface diffusion or
metallization; the column pitch varies from 50 to 100 m, alternating current (AC)
and direct current (DC) pads allow the bonding connection to the readout electrode.
All strips can be biased from a common bias ring by punch through. The layout also
includes 3D square diodes of 1-mm2 area, featuring 80- or 100-m column pitch.
Table2.1 summarizes the main electrical parameters of 3D-STC extracted from
current voltage (I-V) and capacitance voltage (C-V) measurements made on both
diode and strip structures. As can be seen, lateral depletion was achieved at very low
voltages, from 5 to 30 V depending on column pitch and on substrate doping concentration. The leakage current was very low: microstrip detectors of 1-cm2 area had
leakage currents between 5 and 20 nA. The breakdown voltage depended on the surface isolation technique; it was about 50 V for p-spray isolation and ranged between
150 and 200 V for p-stop isolation, as shown in Figure2.18. In case of p-spray isolation, the n+ and p regions overlap, and the critical electric field value was reached at
relatively low voltages, leading to early breakdown. For p-stop isolation, p+ and n+
regions were a few tens of micrometers apart, allowing a higher applied bias voltage
before reaching critical electric fields. After irradiation, the breakdown voltage for
structures with p-spray was expected to increase, since the radiation-induced damage on the oxide charge will compensate the p-spray doping. For p-stop isolation, on
Table2.1
Summary of the Electrical Parameters of 3D-STC Detectors
p-spray
Substrate doping concentration
Lateral depletion
Full depletion
Leakage current at full depletion
Breakdown voltage
Interstrip capacitance
Strip to back capacitance
p-stop
110 1012
330
3070
<1
4060
150200
46
57
<5
Unit
cm3
V
V
pA/column
V
pF/cm
pF/cm

1.0E04
Leakage Current (A)
IGR
IBL
1.0E05
1.0E06
p-stop
1.0E07
1.0E08
1.0E09
1.0E10
p-spray
0
50
100
150
200
250
Reverse Voltage (V)
Figure 2.18 Bias line IBL and guard ring IGR currents of a selection of 3D-STC microstrip
detectors. (From Ronchin, S. et al., Nucl. Instrum. Methods A, 583, 2007. Copyright Elsevier.
With permission.)60
the other hand, the increased oxide charge concentration will increase the electron
sheet layer concentration at the surface, leading to lower breakdown voltages. It is
worth mentioning that the isolation layers were not optimized for these fabrication
runs and that the best isolation solution has to take into account many parameters
given by each specific application of the sensor. For a more comprehensive understanding of the problem, refer to Piemonte.13
The interstrip capacitance values were higher than for planar detectors, as expected,
since the electrodes were strongly coupled to each other, and the capacitance value
was directly proportional to the column depth and inversely proportional to the intercolumn pitch. Since the noise of the readout channel is strictly correlated to the
interstrip capacitance value, 3D-STC detectors are intrinsically affected by a higher
noise with respect to planar detectors. On the other hand, the capacitance versus the
backplane electrode has comparable value with respect to a planar detector and basically depends only on the distance between column tips and the back surface. Finally,
C-V measurements also allow for an indirect calculation of the column depth based on
the extraction of the substrate-doping profile as a function of the depletion width.57
2.6.1.3Functional Characterization
Both diodes and microstrip detectors have been delivered to different groups for
functional tests such as transient current technique (TCT), charge collection, and
efficiency measurements. The 3D diodes have also been tested in Trento with a
pulsed laser source at different wavelengths to investigate the charge collection
time in case of charge generated at different depths into the substrate.61 TCT measurements, made by connecting the diode to a fast transimpedance amplifier and
stimulated by a 780-nm pulsed laser, have shown a slow response. The induced current peaks at 150 ns, much slower than the few nanoseconds expected. This can be
explained by the peculiar electrode arrangement in 3D diodes, in agreement with

0.5
Normalized signal (a.u.)
Normalized signal (a.u.)
44
0.0
Readout strip
0.5
1.0
10
20
30
Time (ns)
(a)
40
50
0.5
0.0
Readout strip
0.5
1.0
10
20
30
Time (ns)
40
50
(b)
Figure 2.19 TCT signal induced on the central strip by a 1,060-nm pulsed laser (black
spot) in the case of (a) a beam focused near the readout strip and (b) a beam focused near
the adjacent strip. (From Dalla Betta, G.-F. et al., Vertex 2007, Paper 23, 2008.59 Copyright
owned by the authors under the terms of the Creative Commons Attribution-NonCommercialShareAlike License. http://creativecommons.org/licenses/by-nc-sa/3.0/.)
TCAD simulation;58 since all the columns are shorted together to form the diode
electrode, electrons induce a signal of opposite polarity on adjacent columns, and
the fast component is canceled out. Only when the holes cross the column tips, while
drifting to the back electrode, do they induce a signal of the same polarity on all
readout electrodes.
The TCT measurements have been performed also at JSI (Ljubljana, Slovenia)62
with a system based on a 1,060-nm pulsed laser, focused to few micrometers on 3D
microstrip sensors and connected to fast transimpedance amplifiers and an oscilloscope, to study the induced signals on the readout strip and adjacent strips.
Figure 2.19a shows the induced signal when the laser beam was focused near
the readout electrode; it has a fast component of a few nanoseconds and a long tail
that lasts for a few microseconds, as expected from simulations. When the laser
was focused on the noncollecting electrode (see Figure 2.19b), a fast signal peak
of the opposite polarity was induced, and a long tail of the same polarity is still
observed for a few microseconds. This behavior is known also for planar strip detectors and can be explained with the aid of Ramos theorem. Since the induced signal
on the adjacent strip is not negligible, an increased spatial resolution could indeed be
obtained in 3D-STC detectors. On the other hand, if the charge is generated underneath a strip rather than in the interstrip region, the fast signal component would
be attenuated since a positive and a negative fast signal would be induced on two
columns of the same strip.
Sensors were irradiated with neutrons up to fluences of 5 1015 cm2, and exploiting the TCT technique, the signal induced on one readout strip was investigated for
different laser incident points. The figure shows only results for two points: the first
(point 1) is at a few micrometers from the readout electrode (Figure 2.20a) and the
second point (point 3) is at the center of a four-column cell (i.e., the region where
the electric field is very weak; Figure 2.20b). Both plots refer to the induced signal
integrated over 25 ns. Signal efficiency reached 40% after 5 1014 cm2 for point 1,
100
90
80
70
60
50
40
30
20
10
0
60
50
40
Nonirr.
1 1014
5 1014
5 1015
0
50
150
100
Ubias (V)
200
250
300
CCErel (%)
CCErel
30
20
10
Nonirr.
1 1014
5 1014
0
10
50
100
150
200
250
300
Ubias (V)
Figure 2.20 Relative charge collection efficiency (i.e., signal efficiency) versus reverse
bias in 3D-STC detectors irradiated with neutrons at different fluences for different incidence
point of the laser beam: (a) close to the readout electrode and (b) at the center of the cell. Data
were extracted by TCT measurements, normalizing the signals after irradiation to the signals before irradiation. (From Zavrtanik, M. et al. 2007 IEEE Nuclear Science Symposium.
Copyright IEEE 2007.)62
whereas for point 3 the signal efficiency was only 10% at the same fluence. This is a
confirmation of the nonuniform response of 3D-STC detectors, which can degrade
CCE, especially after high irradiation levels. At the largest fluence, the signal efficiency was less than 5% also for point 1, evidence that 3D-STC detectors are not
very radiation hard.
Similar measurements have been performed at the University of Freiburg, with
an 980-nm IR laser and connecting 3D-STC strip detectors to the ATLAS ABCD3T
readout.63 Results after irradiation with 26-MeV protons at 1 1015 1-MeV eq. n
cm2 are shown in Figure2.21.64 The induced signal is not uniform on the square
area delimited by four columns; charge trapping, together with low electric field, is
responsible for the lower charge collection from the central region.
Detectors have also been tested with a beta source setup before and after irradiation. The collected charge versus the bias voltage is shown in Figure2.22. Detectors
can be biased up to 500 V, to reach full depletion, and still collect about 2.1 fC as it
was before irradiation.65
2.6.23D-DDTC
The fabrication experience with 3D-STC detectors has proved the good process yield
at FBK and has set the basis for the development of 3D double-sided, double-type
column (3D-DDTC) detectors.25,66 This detector concept aims at overcoming the
performance limitation of the 3D-STC while keeping a simpler process complexity
compared to full 3D. These detectors have columnar electrodes of both doping types
etched from both wafer sides: the junction columns from the top side and the ohmic
columns from the back side (see Figure2.10d). All ohmic columns are connected
by surface diffusion and metallization. The columnar electrodes are not etched all
the way through the substrate, but they stop at a few micrometers from the opposite
??
80
60
40
20
0
??
??
??
??
??
??
11.64
11.62
11.6
m
11.58
nm
yi
11.56
11.54
8
.9
2
9
.9
x in mm
1
.0
3
.0
3
3
.0
3
.0
5
.0
06
3
.
??
(b)
(a)
vt50 in mV
?? ??
??
?? in mm
??
??
??
yi
n
m
m
??
.9
yi
n
??
11.58
11.56
11.54
150
100
50
0
50
??
12.68
.9
?? in mm
.0
.0
.0
.0
.0
.0
150
100
50
0
50
vt50 in mV
vt50 in mV
46
(c)
Figure 2.21 Results of a position-resolved scan over a square cell at a bias voltage of
110 V after proton irradiation at 1 1015 cm2: (a) and (b) represent the signals induced on
each single strip, whereas (c) shows the sum from both strips. (From Eckert, S. et al., Nucl.
Instrum. Methods A, 581, 2007. Copyright Elsevier. With permission.)64
surface, and columns are not filled with polysilicon (as for 3D-STC). This fabrication
approach allows for some advantages over standard 3D detectors concerning process
complexity: Since columns are not passing through the whole substrate, there is no
need for wafer bonding and for the consequent removal of the sacrificial wafer at
the end of the process; due to double-sided DRIE etching, there is no need to grow
a thick protection layer after the first DRIE step, and finally, polysilicon deposition
and CMP removal are avoided. This alternative 3D technology indeed has some
drawbacks: Columns are dead regions, and low-field regions exist between the column tips and the opposite surface. These regions could degrade the performance of
3D-DDTC with respect to standard 3D.
2.6.2.1Simulations
To investigate the characteristics of these detectors, 3D TCAD simulations on a
quarter of a unit cell have been performed with parameters representative of FBK
technology. Figure 2.23 reproduces a p-type detector with 80-m pitch between
readout columns, a typical layout for radiation detectors. A quarter-volume cell of
40 40 250 m3 has been simulated to save mesh points and computational time.
The substrate doping concentration was 2 1012 cm3; oxide charge and p-spray
isolation were included as well. Four different geometries have been compared to

2.5
Collected Charge (fC)
1.5
0.5
3D_long CZ p-spray unirradiated

3D_long CZ p-spray irradiated
0
0
100
200
300
400
500
VBias (V)
Figure 2.22 Collected charge versus bias voltage for 3D-STC long strip detector with
p-spray isolation before and after 26-MeV proton irradiation at 1 1015 cm2. (From Khn, G.
et al., IEEE Trans. Nucl. Sci., 55, 6, 2008. With permission.)65
investigate how the distance d affects the performance: Three structures reproduce
3D-DDTC detectors with d values equal to 25, 50, and 75 m, and one structure is
representative of standard 3D detectors with passing-through columns (d = 0). For
the last structure, p-spray isolation was also included on the back surface.
Figure2.24 shows the electric field configuration along the diagonal of the simulated cell of Figure2.23 at 16 V, that is, slightly beyond full depletion. The electric
field lines in standard 3D are horizontal and very homogeneous between columns,
apart from regions close to the surfaces, due to the effect of p-spray. Taking this
electric field configuration as a reference, among the three simulated 3D-DDTC
geometries, the one with d = 25 m shows comparable results. At higher d values,
the electric field becomes more distorted, and only in a small central region the field
lines are similar to those of standard 3D. Low-field regions can be observed near the
top and bottom surfaces of 3D-DDTC detectors, a fact that could degrade the CCE.
Unlike 3D-STC detectors, increasing bias voltages beyond the full depletion value
increases the electric field between columns and the carrier drift velocity, thus reducing the collection time.
Transient simulations have been performed of an MIP particle impinging the cell
at a few micrometers from the ohmic p+ column. Also, a 3D-STC geometry has been
simulated to have a direct performance comparison with the previous technology.
Figure2.25a shows the induced currents on the collecting electrode for all the investigated structures at a bias voltage of 16 V. As expected from the static electric field
simulations, the shorter the distance d was, the higher the current peak was and the
shorter was the collection time. The current signals have been postprocessed with
48

Oxide layer
p-spray
isolation
n+ column
p+ column
40
p-type
substrate
d
p+ ohmic
contact
Figure 2.23 Sketch of a 3D-DDTC detector on p-type substrate. The cell represents a
quarter of the typical pattern present in the layout. The pitch between columns of the same type
was 80 m. (From Zoboli, A. et al., IEEE Trans. Nucl. Sci., 55, 5, 2008. With permission.)66
an algorithm that reproduces a CR-(RC)3 filter with a shaping time of 20 ns, suitable for emulating the fast readout electronics of the ATLAS Semiconductor Tracker
(SCT) detector.63 The results are already expressed in terms of collected charge (see
Figure2.25b); the peak value of the collected charge for d = 25 m was only about
10% lower than for standard 3D and confirmed that comparable performance can be
achieved if d is kept small enough. 3D-STC geometry, due to the inefficiencies in the
charge collection mechanism, collects only a small fraction of the generated charge
within 20 ns.
Keeping in mind the expected radiation hardness of 3D detectors, simulations
accounting for high radiation damage have also been performed. The highest damage factor, foreseen after 5 years of operation at SLHC67,68 at a short radius from
the interaction point in the ATLAS experiment, has been simulated. Taking into
account the increase of the effective substrate-doping concentration and of the trapping probability7 after 1016 cm2 1-MeV eq. n fluence, full 3D and 3D-DDTC with
100
200
200
50
X
DDTC d = 50 m
0
200
50
X
DDTC d = 75 m
0
100
100
100
DDTC d = 25 m
0
Standard 3D
200
50
X
50
X
Figure 2.24 Electric field distribution of a simulated 2D cross section taken across the
diagonal of the quarter cell. The column junction is placed at x = 0, and the ohmic column
is at x = 40 2 m. The four plots (from left to right) refer to a full 3D and three 3D-DDTC
with d = 25 m, d = 50 m, d = 75 m. (From Zoboli, A. et al., IEEE Trans. Nucl. Sci., 55, 5,
2008. With permission.)66
d = 25 m have been simulated. Owing to a high increase of the substrate-doping

concentration, the full depletion voltage was 350 V, so the transient simulation has
assumed a 400-V bias condition. At such high irradiation levels, the trapping time
rather than the full depletion voltage was the limiting factor, since its value was
only about 0.3 ns, leading to drifting distances of only a few tens of micrometers.
Results were very similar between the two technologies: The collected charge,
calculated from the induced current with the same algorithm as before, confirmed
that 3D-DDTC detectors are nearly as efficient as full 3D, showing a collected
charge that was only 10% lower. Both architectures can collect about 1.7 fC within
the shaping time of the ATLAS SCT readout, which is half the charge collected
before irradiation.
2.6.2.2Fabricated Devices
Two wafer layouts have been designed and fabricated on high-resistivity n-type and
p-type FZ substrates, respectively.25 For both batches, an external service at IBS,
France, was used for DRIE because, at the time of fabrication, the DRIE equipment
was not available in house. The maximum column depth achievable with a diameter
of 10 m after etching tests was declared to be 180 m. The first batch was processed
on 300-m thick n-type substrates so that optimized columns could not be obtained,
whereas the second one was made on 200- to 220-m p-type substrates, promising
a smaller distance between column tips and the opposite surface.
50
Current (A)

3.0
Standard 3D
2.5
d = 25 m
2.0
d = 50 m
1.5
d = 75 m
1.0
0.5
STC
0.0
Time (ns)
(a)
3.5
Collected charge (fC)
3.0
Standard 3D
2.5
d = 25 m
2.0
d = 50 m
1.5
1.0
d = 75 m
STC
0.5
0.0
10
20
30
40
50
60
70
80
Time (ns)
(b)
Figure 2.25 Transient signals in 3D detectors with different geometries taken from
simulation at a bias of 16 V in response to an MIP particle: (a) induced current signal;
(b) equivalent charge signal at the output of a semi-Gaussian CR-(RC)3 shaper amplifier
with 20-ns peaking time. (From Zoboli, A. et al., IEEE Trans. Nucl. Sci., 55, 5, 2008. With
permission.)66
The first n-type batch contained mainly 3D diode test structures and 3D strip
detectors suitable to be read out with the ATLAS SCT front-end electronic (p-readout). The second batch contained mostly 3D pixel structures designed to fit the
ATLAS FE-I3 readout chip47 and CMS readout chip (n-readout). Several ATLAS
pixel detectors featuring different electrode configurations (e.g., different number of
columns per pixel) have been realized. Table2.2 summarizes the main features of
the two fabricated batches of 3D-DDTC detectors.
Figure 2.26 illustrates the main process steps for the fabrication of 3D-DDTC
detectors on n-type substrate; the following discusses the various parts of the figure:
Table2.2
Main Features of the Two Batches of 3D-DDTC Detectors
Batch
3D-DDTC-1
3D-DDTC-2
Substrate type
Substrate thickness
Maximum column depth
n-type
p-type
300 m
180 m
AC/DC coupled
200220 m
180 m
AC/DC coupled
80/100 m
ALICE, Medipix1
80/100 m
ATLAS, CMS
Strip design
Strip pitch
Pixel design
Oxide
Thin oxide
n+
n sub.
(a)
(b)
p+
(c)
(d)
Metal
Passivation
d
(e)
(f )
Figure 2.26 Main steps of the fabrication of 3D-DDTC detectors on n-type substrates.
See text for discussion. (From Zoboli, A. et al., IEEE Trans. Nucl. Sci., 55, 5, 2008. With
permission.)66
52
a. A thick oxide is grown to be used also as a mask for the first DRIE process
on the back side.
b. The thick oxide is etched from the back side, and phosphorus is diffused
from a solid source into the columns and at the surface. Later, a thin oxide
is grown to prevent the dopant from diffusing out.
c. The oxide layer on the front side is patterned, and the DRIE step is performed on the top surface, defining readout electrodes.
d. The thick oxide is removed from a circular region around columns, and
boron is diffused into the columns and on the surface to ease the contact
formation.
e. A thin oxide layer is grown to prevent the dopant out-diffusion. Contact
holes are defined and etched through the oxide; aluminum is sputtered and
patterned.
f. A final passivation layer is deposited and patterned to define the access
to the metal layer, while on the back side aluminum is sputtered to have a
metal back electrode.
The fabrication steps for the p-type batch are similar but with an inversion of
the column doping types and one additional process step to implement p-spray surface isolation between n-columns. If p-stop or a combination of p-stop/p-spray is
employed, an additional mask for the p-stop patterning is needed.
2.6.2.3Experimental Results from the First Batch
The electrical characterization of planar structures (diodes, MOS, gated diodes) has
been made to extract the relevant technology parameters.66 Leakage current of planar diodes was very low and reached saturation at very low voltages. From C-V
measurements, a full depletion voltage of about 10 V was obtained, corresponding to
a very low substrate-doping concentration of 1.4 1011 cm3. 3D test structures were
square diodes with an area of 2.56 mm2. The readout electrode was made either by
an array of 16 16 p+ columns with 100-m pitch or by a 20 20 array of p+ columns
with 80-m pitch. All p+ columns were shorted together either by a uniform surface
diffusion or by metal strip lines. Around the inner readout pad, a guard ring made
of two rows of p+ columns was present. The back electrode was made by n+ columns
all shorted together by the back diffusion. Also, 3D-STC diodes (i.e., without ohmic
columns) were available in different geometries with similar layouts. From C-V measurements on 3D-STC diodes,57 the junction column depth has been estimated to be
190 m, as expected from the DRIE characteristics.
Leakage currents of 3D diodes saturated at very low voltage; in full depletion
conditions, values were as low as 0.1 pA/column. No sign of breakdown appeared
in 3D-STC diodes, whereas the leakage currents on 3D-DDTC diodes showed a
nonuniform behavior (see Figure2.27a); in particular, some diodes showed a current
rise even at low voltage, which could be due to defects on the columns.
Figure 2.27b shows the capacitance as a function of the reverse voltage for
3D-DDTC diodes; the sharp decrease in the curves suggests that lateral depletion
was reached at a very low voltage (a few hundred millivolts), and diodes with a
higher pitch depleted laterally at slightly higher voltages. The main contribution to

10
Leakage Current (nA)
Pitch 80 m
Pitch 100 m
1
0.1
0.01
10
15
Reverse Voltage (V)
(a)
20
Capacitance (pF)
50
Pitch 80 m
Pitch 100 m
40
30
20
10
0
2
3
Reverse Voltage (V)
(b)
Figure 2.27 Leakage current (a) and capacitance (b) as a function of the bias voltage for
3D-DDTC diode test structures from the first batch (n-type substrate). (From Zoboli, A. et al.,
IEEE Trans. Nucl. Sci., 55, 5, 2008. With permission.)66
the total capacitance actually came from the overlap between the junction and the
ohmic columns. This was confirmed by the fact that the capacitance at full depletion
was higher for diodes with the shorter interelectrode distance, which also featured
a higher number of columns.66 Comparing analytical calculations and simulations
with capacitance measurements of 3D-DDTC diodes, it is possible to estimate the
ohmic column depth.
Owing to the geometries of the electrode, the capacitance between one junction
column and the surrounding ohmic columns can be approximated by the capacitance
of a cylindrical capacitor:

Ccyl = 2 Si
H

r2
ln
r1
(2.1)
54
Table2.3
Summary of the Electrical Parameters Evaluated
from 3D Diode Test Structures from the First Batch
Fabricated on n-Type Substrates
Parameter
Unit
Value (STC)
Value (DTC)
Junction column depth

Ohmic column depth
Lateral depletion voltage
Full depletion voltage
Leakage current at FD
Backplane capacitance
180190
n.a.
<0.5
~1.5
0.10.5
58
180190
160170
<0.5
~1.0
0.10.7
1820
m
V
V
pA/column
fF/column
where H is the overlap between columns, and r 2 and r1 are the outer and inner radii of
the cylinder, respectively. Fitting the capacitance measurements with this analytical
formula gives the ohmic column depth of 165 m, so that the overlap between ohmic
and junction columns was only 5060 m. Table2.3 summarizes the main electrical
parameters extracted from 3D diodes.
Functional characterization has been carried out on 3D strip detectors with 1-cm2
area featuring 102 strips and 102 columns per strip.69 Figure2.28a shows a detail of
the microstrip sensor under test: DC and AC pads are visible as well as the bias ring
and the guard ring. Figure2.28b shows the hybrid where the sensors, glued on a thermal baseboard, are connected to the ATLAS SCT ABCD3T readout module, with a
Re-bondable
fan-in
Detectors
Guard ring
Bias ring
DC pad
AC pad
(a)
(b)
Figure 2.28 (a) Photo of a small portion of a microstrip detector. The square region is the
area under investigation with the laser setup. (b) Two detectors glued on the carbon-carbon
support and connected to the ATLAS ABCD3T readout chip. (From Zoboli, A. et al., IEEE
Nuclear Science Symposium, 2008. With permission.)69
shaping time of 20 ns.63 Sensors have been stimulated with two separate systems. The
first one adopted a pulsed laser with 980-nm wavelength focused with a microscope
down to 2 m on the detector surface. The module could be moved under the beam
with a fast-precision motorized stage with 1-m step in x and y directions. With this
system, a very high precision map of the collected charge was obtained. The second
one was based on a 90Sr beta source system in which events are triggered by two
scintillators placed behind the sensor working in coincidence. This system allows for
efficiency measurements at a fixed threshold and absolute charge collection measurements because fast electrons from the source deposit charge in the sensor, like a MIP.
Figure2.29 shows a laser scan over a square region of 50 50 m2, with 2-m
steps. The investigated region corresponds to the quarter reproduced for TCAD
25
20
15
10
Signal in mV
30
5
50
40
??
um
30
0
20
10
20
30
x (um
y(
Signal in mV
10
40
50
50
40
??
10
20
20
x (um
30
um
30
0
y(
Signal in mV
220
200
180
160
140
120
100
80
60
40
20
Signal in mV
(a)
10
40
50
(b)
Figure 2.29 Laser scan made at (a) 0.5 V and (b) 40 V over a square region of 50 50
m2 with 2-m resolution step. (From Zoboli, A. et al., IEEE Nuclear Science Symposium,
2008. With permission.)69
56

2.5
Collected Charge (fC)
2
1.5
1
DDTC D2
DDTC D14
STC D18
0.5
0
10
20
30
Bias Voltage (V)
40
50
Figure 2.30 Charge collection as a function of the bias voltage for two 3D-DDTC and one
3D-STC microstrip detectors. Data refers to the measurements with beta setup. (From Zoboli,
A. et al., IEEE Nuclear Science Symposium, 2008. With permission.)69
simulation. The two plots refer to a bias voltage of 5 and 40 V, respectively. The strip
metal is clearly visible from the scan due to a total reflectivity of the aluminum trace.
At 5 V, the central region, featuring a lower electric field, shows a rather small charge
collection, but as the voltage was increased to 40 V the induced charge was more
uniform among the scanned area. Nevertheless, the central region still collected 20%
less charge with respect to other regions because of the nonoptimized column depth
for this batch.70
In Figure 2.30, referring to measurements with beta setup, the collected charge
versus the bias voltage of three detectors is plotted. As can be seen, the two 3D-DDTC
detectors had comparable behavior: They started collecting charge at very low voltages
and saturated at 40 V with a value of 2.4 fC. One STC detector was tested to have a
direct comparison with the previous technology; the collected charge was significantly
lower and reached 1.8 fC at full depletion. 3D-DDTC exhibited a net improvement
in terms of collected charge at fast shaping times with respect to the previous STC
technology. Unfortunately, due to the mentioned nonoptimized column depths, the
collected charge was lower than that expected from 300-m thick substrate (3.5 fC) .
The CCE at a threshold of 1 fC was 94% at full depletion.69 Owing to low-field
regions and hollow columns, 100% efficiency is probably not achievable in these
prototypes.
The samples were irradiated with 24-MeV protons at three different fluences up
to 2 1015 cm2, which is the highest fluence expected for short strips in ATLAS
at SLHC conditions after 5 years of operation. After irradiation, the samples were
annealed for 80 min at 60C to exploit beneficial annealing and to reach minimum
depletion voltage.
The laser scan shown in Figure2.31a refers to a 3D-DDTC detector irradiated to 5
1014 cm2 and reverse biased at 10 V. The induced signal was higher close to the back
??
40
40
60
x (
m)
y(
20
60
0
80
14
12
10
8
6
4
2
20
80
100
(a)
50
40
30
10
20
20
x (
m)
30
??
y(
Signal in mV
100
Signal in mV
Signal in mV
16
14
12
10
8
6
4
2
0
Signal in mV
10
40
50
(b)
Figure 2.31 (a) Laser scan made across a square region of 110 110 m, with a step of
5 m in a detector irradiated with 24 MeV at a fluence of 5 1014 cm2 and a bias voltage of
10 V. (b) Laser scan on a smaller region in the center of the cell with a resolution of 2 m.
(From Zoboli, A. et al., IEEE Nuclear Science Symposium, 2008. With permission.)69
electrode, so we can deduce the substrate was type inverted, and the main junction was
now on the back electrode. Nevertheless, there are still four signal peaks corresponding to the position of the front electrodes, evidence of the double-junction effect.71
The scan made with a finer step of 2 m of Figure2.31b proves that the depletion
region after irradiation indeed mainly proceeds from the back column toward the front
column; at 10 V, the sensor was not laterally depleted yet, as demonstrated by the fact
that the charge was collected only in the central region for a few micrometers around
the back column. The small circular region in the center of the scanned area, characterized by a lower charge collection, indicates exactly the location of the back column.
The irradiated sensors have been tested also with the beta setup; the whole system
was kept at 10C so that the sensors could be biased up to 200 V before reaching
58

4
DDTC unirradiated
DDTC = 5 1014 cm2
STC = 5 1014 cm2
3.5
3
2.5
2
1.5
1
0.5
0
50
100
150
200
Bias voltage (V)
250
300
Figure 2.32 Charge collection versus bias voltage for unirradiated 3D-DDTC detector and
for 3D-DDTC and 3D-STC detectors irradiated with 24-MeV protons at a fluence of 5 1014
cm2. (From Zoboli, A. et al., IEEE Nuclear Science Symposium, 2008. With permission.)69
high current values and therefore high noise levels. This was not possible in the laser
system, where the cooling was not as efficient as for the beta setup, and sensors could
be biased only up to 40 V.
Figure2.32 plots the collected charge of one 3D-DDTC and one 3D-STC irradiated at a fluence of 5 1014 cm2, along with the data taken from one 3D-DDTC
before the irradiation. The 3D-DDTC detector collected 1.5 fC at a full depletion
voltage of about 100 V, and it was laterally depleted at about 20 V. The lateral depletion value was in good agreement with the one calculated using the acceptor introduction rate gc = 0.02 cm1 taken from the literature.7 The 3D-STC started collecting
charge only at 80 V, and full depletion cannot be extracted from the plot. In fact,
keeping in mind that the substrate was type inverted, in 3D-STC the depletion proceeded upward from the back planar electrode, toward the front columns. The distance to be depleted was thus 110 m, whereas the electrode distance in a 3D-DDTC
is only 56 m. Furthermore, the readout electrodes in STC sensors (after type inversion) are far from the depleted region, and a small signal is induced. These are the
main reasons why a smaller amount of charge is collected in the 3D-STC compared
to 3D-DDTC.
Two 3D-DDTC detectors have been irradiated also at two higher fluences of 1
1015 cm2 and 2 1015 cm2, and the results are shown in Figure2.33. It should be
stressed that the annealing step reduces the effective doping concentration, as confirmed by the fact that, in both detectors, lateral depletion occurred at lower voltages
after the annealing. At these high fluences, due to trapping, the collected charge was
significantly reduced with respect to the preirradiation value. It is worth noting that
the trapping times were expected to be 2 and 1 ns, respectively, at the two investigated
fluences of 1 1015 cm2 and 2 1015 cm2;72 as a result, only the charge generated in

4
Unirradiated
= 1 1015 cm2 before annealing
= 1 1015 cm2 after annealing
= 2 1015 cm2 before annealing
= 2 1015 cm2 after annealing
3.5
3
2.5
2
1.5
1
0.5
0
50
100
150
200
250
300
Bias voltage (V)
Figure 2.33 Charge collection of two 3D-DDTC detectors irradiated with 24-MeV protons at 1 1015 cm2 and 2 1015 cm2 fluences. The figure includes data measured before and
after an annealing step at 60 for 80 min aimed at taking benefit from the short-term annealing. (From Zoboli, A. et al., IEEE Nuclear Science Symposium, 2008. With permission.)69
the region where columns overlap (5060 m) can be collected very quickly without
being trapped. On the contrary, a large part of the charge generated in the lowfield regions gets trapped, lowering the total induced signal, as observed from the
measurements. Nevertheless, despite columns not being optimized, the 3D-DDTC
detector can be favorably compared to planar detectors because of the short distance
between the electrodes, which allows low lateral depletion voltages. As an example,
p-type planar strips irradiated at the same fluence have to be biased up to 1,000 V73
to collect the same amount of charge as the 3D-DDTC detector.
At 150 V, sensors irradiated at fluences of 1 1015 cm2 and 2 1015 cm2 can
collect up to 1.5 and 1 fC, respectively. Beyond this voltage, all 3D-DDTC detectors
exhibit an exponential increase in the collected charge. This phenomenon could be
ascribed to avalanche multiplication triggered by particles; a detailed study with the
aid of TCAD simulations is currently being performed to investigate this effect.
2.6.2.4Experimental Results from the Second Batch
The electrical characterization of the second batch made on p-type substrate, with
a 200- to 220-m thickness, has been carried out, mainly on 3D diode structures.
Diode layout and geometries were exactly the same as for the first batch. Here, the
readout electrode was n+, and the ohmic electrode on the back side was p+. From C-V
curves of planar diodes, a substrate-doping concentration of 1 1012 cm3 has been
extracted. The leakage currents of both planar and 3D-STC diodes were very low:
~1 nA/cm2 for planar and 6 nA/cm2 for 3D-STC and 3D-DDTC.74 From C-V plots, a
lateral depletion of 4 V and a total depletion of 1520 V can be extracted. After C-V
measurements, using the same approach as described for the first batch, the junction
60
Table2.4
Summary of the Electrical Parameters Evaluated
from 3D Diode Test Structures from the Second
Batch Fabricated on p-Type Substrates
Parameter
Unit
Value (STC)
Value (DTC)
Junction column depth

Ohmic column depth
Lateral depletion voltage
Full depletion voltage
Leakage current at FD
Backplane capacitance
120
n.a.
23
15
<0.05
10
110120
200
34
15
<0.05
3035
m
V
V
pA/column
fF/column
and the ohmic column depths could be calculated. The back column was almost as
deep as the substrate thickness, about 220 m, as confirmed also by scanning electron microscopy (SEM) micrographs. On the contrary, due to a major problem that
occurred during the second DRIE step, the junction column was only 120-m deep.
Nevertheless, the overlap between columns was almost 120 m and should allow for
a sizable improvement in the performance compared to the previous batch because
of two concurrent factors: (a) n-side readout was intrinsically radiation harder than
p-side and (b) column overlap was higher. Table2.4 summarizes the electrical and
technological parameters evaluated from 3D diode structures.
In this batch, the wafer layout was mostly dedicated to pixel detectors suitable
to be connected to the ATLAS FEI3 readout chip.47 Several pixel sensors with different column configurations have been delivered to the Italian National Institute
for Nuclear Physics (INFN) group in Genova, responsible for the bump bonding to
the readout chip and for the detector functional characterization. After the bumpbonding process, made at SELEX SI, the sensors were working properly; the leakage
current was on the order of 200 pA per pixel; the breakdown voltage was typically
70 V (much higher than the depletion voltage); and the noise figures were good (~220
electrons rms). Functional tests are being carried out at INFN Genova and at CERN
with radioactive sources. Preliminary results are encouraging; in fact, the energy
spectrum and the corresponding induced charge are similar to those obtained from
standard 3D detectors fabricated at Stanford. The next important step will be the
characterization of these sensors in a test beam at CERN. Sensors will be exposed to
very-high-energy pions, and thanks to a telescope system, a very high spatial resolution of the impinging tracks will be obtained. This setup will allow a very precise
scan of the active pixel area and a 2D plot of the sensor efficiency.
2.6.3Next Steps
Owing to the in-house availability of the DRIE machine since 2008, process
developments for 3D detectors at FBK are evolving more rapidly. A fabrication
recycle of the second batch on p-type substrates (200-m thick) with optimized
column depths (180 m) will soon be completed. In addition, the fabrication of a

new batch of 3D detectors featuring passing-through columns has been started. To
this purpose, the technology has been slightly modified, still maintaining the double-sided approach that does not need a support wafer and is also suitable for the
recently proposed dual-readout concept.75 Process tests and TCAD simulations
are also being carried out to study the critical issues related to the active edge,
which we plan to implement first on a batch of planar detectors. The first batch of
full 3D detectors with active edge is scheduled to be available early in 2010.
2.7Conclusions
Since their introduction in 1997, 3D detectors are definitely emerging as one of the
most powerful technologies to cope with the challenging demands of future HEP
experiments. Results so far reported are very impressive in terms of signal efficiency, speed, and radiation hardness. New ideas, such as the recently proposed dual
readout, are still possible to fully exploit the enormous potential of these devices.
Owing to the active edge feature, sensitivity to within a few micrometers from the
physical edge opens the way to fully tileable detector modules of special interest
for imaging applications. The increased noise due to a higher electrode capacitance
has been shown to be overcompensated by a superior signal robustness, and the
opportunity to operate at relatively low voltages while maintaining a good detection efficiency brings along significant advantages also in terms of overall power
dissipation. The feasibility of large-volume productions of 3D detectors is still a concern due to the rather complicated fabrication process involving several nonstandard
steps. Nevertheless, besides the Stanford Nanofabrication Facility, several research
institutes and industrial vendors have started to develop their own 3D detector technologies, and initial results are quite encouraging. In a few years, 3D technology will
likely be reaching its maturity, with major benefits for most applications involving
radiation detection systems.
Acknowledgments
This work has been supported in part by the Provincia Autonoma di Trento and
in part by the Italian National Institute for Nuclear Physics (INFN), CSN5, Project
TREDI (20052008).
We are indebted to Dr. Maurizio Boscardin, Dr. Claudio Piemonte, Dr. Sabina
Ronchin, and Dr. Nicola Zorzi (Fondazione Bruno Kesslerirst, Trento, Italy) for the
technological development and the fabrication of 3D detectors.
Dr. Cinzia Da Via (Manchester University, UK); Prof. Luciano Bosisio (Univer
sity of Trieste and INFN, Trieste, Italy); Dr. Alessandro La Rosa (CERN, Switzer
land); Dr. Giovanni Darbo (INFN Genova, Italy); Dr. Hartmut Sadrozinski (SCIPP,
University of California Santa Cruz, USA); Dr. Ulrich Parzefall (University of
Freiburg, Germany); and Dr. Gregor Kramberger (JSI Ljubljana, Slovenia) are
warmly acknowledged for the fruitful collaboration.
62
References

1. G. F. Knoll, Radiation Detection and Measurement, 3rd ed., Wiley, New York, 2000.
2. G. Lutz, Semiconductor Radiation DetectorsDevice Physics, Springer-Verlag, Berlin,
2007.
3. J. Kemmer, Fabrication of low noise silicon radiation detectors by the planar process,
Nucl. Instrum. Methods, 169, 499502, 1980.
4. S. I. Parker, C. J. Kenney, and J. Segal, 3Da proposed new architecture for solid-state
silicon detectors, Nucl. Instrum. Methods A, 395, 328343, 1997.
5. S. Ramo, Currents induced by electron motion, Proc. IRE, 27. 584585, 1939.
6. A. Kok, G. Anelli, C. Da Via, J. Hasi, P. Jarron, C. Kenney, J. Morse, S. Parker, J. Segal,
S. Watts, and E. Westbrook, 3D detectorsstate of the art, Nucl. Instrum. Methods A,
560, 127130, 2006.
7. G. Kramberger, Recent results from CERN RD50 collaboration, Nucl. Instrum. Methods
A, 583, 4957, 2007.
8. V. A. Wright, W. D. Davidson, J. J. Melone, V. OShea, K. M. Smith, L. Donnohue,
L. Lea, K. Robb, S. Nenonen, and H. Sipila, 3-D Medipix: a new generation of X-ray
detectors, Nucl. Instrum. Methods A, 546, 319323, 2005.
9. M. Da Rold, N. Bacchetta, D. Bisello, A. Paccagnella, G. F. Dalla Betta, G. Verzellesi,
O. Militaru, R. Wheadon, P. G. Fuochi, C. Bozzi, R. DellOrso, A. Messineo, G. Tonelli,
and P. G. Verdini, Study of breakdown effects in silicon multiguard structures, IEEE
Trans. Nucl. Sci., 46(4), 12151223, 1999.
10. C. J. Kenney, S. I. Parker, and E. Walckiers, Results from 3-D silicon sensors with wall
electrodes: near-cell-edge sensitivity measurements as a preview of active-edge sensors,
IEEE Trans. Nucl. Sci., 48(6), 24052410, 2001.
11. C. J. Kenney, J. D. Segal, E. Weestbrook, S. Parker, J. Hasi, C. Da Via, S. Watts, and J. Morse,
Active-edge planar radiation sensors, Nucl. Instrum. Methods A, 565, 272277, 2006.
12. Silvaco, http://www.silvaco.com; Synopsys Advanced TCAD Tools, http://www.
synopsys.com/products/tcad/tcad.html.
13. C. Piemonte, Device simulations of isolation techniques for silicon microstrip detectors
made on p-type substrates, IEEE Trans. Nucl. Sci., 53(3), 16941705, 2006.
14. D. Pennicard, G. Pellegrini, M. Lozano, R. Bates, C. Parkes, V. OShea, and V. Wright,
Simulation results from double-sided 3-D detectors, IEEE Trans. Nucl. Sci., 54(4),
14351443, 2007.
15. J. Kalliopuska, S. Eranen, and R. Orava, 3D simulations of 3D silicon radiation detector
structures, Nucl. Instrum. Methods A, 568, 2733, 2006.
16. A. A. Ayn, R. L. Bayt, and K. S. Bruner, Deep reactive ion etching: a promising technology for micro- and nanosatellites, Smart Mater. Struct., 10, 11351144, 2001.
17. F. Laermer and A. Urban, Milestones in deep reactive ion etching, Proceedings of the
13th International Conference on Solid-State Sensors, Actuators and Microsystems
(TRANSDUCERS 05), Seoul, Korea, June 59, 2005, pp. 11181121.
18. C. J. Kenney, S. I. Parker, J. Segal, and C. Storment, Silicon detectors with 3-D electrode
arrays: fabrication and initial test results, IEEE Trans. Nucl. Sci., 46(4), 12241236, 1999.
19. S. H. Christiansen, R. Singh, and U. Gsele, Wafer direct bonding: from advanced substrate engineering to future applications in micro/nanoelectronics, Proc. IEEE, 94(12),
20602106, 2006.
20. S. Franssila, CMP: Chemical-mechanical Polishing, in Introduction to Microfabrication,
Wiley, Chichester, UK, pp. 165172, 2004.
21. C. Piemonte, M. Boscardin, G. F. Dalla Betta, S. Ronchin, and N. Zorzi, Development
of 3D detectors featuring columnar electrodes of the same doping type, Nucl. Instrum.
Methods A, A541, 441448, 2005.

22. S. Ernen, T. Virolainen, I. Luusua, J. Kalliopuska, K. Kurvinen, M. Erluoto,
J. Hrknen, K. Leinonen, M. Palvianen, and M. Koski, Silicon semi 3D radiation
detectors, 2004 IEEE Nuclear Science Symposium, Rome, Italy, October 1622, 2004,
Conference Record, Paper N28-3.
23. Z. Li, W. Chen, Y. H. Guo, D. Lissauer, D. Lynn, V. Radeka, and G. Pellegrini,
Development, simulation and processing of new 3D Si detectors, Nucl. Instrum. Methods
A, 583, 139148, 2007.
24. T. Grnlund, Z. Li, G. Carini, and M. Li, Full 3D simulations of BNL one-sided silicon
3D detectors and comparisons with other types of 3D detectors, Nucl. Instrum. Methods
A, 586, 180189, 2008.
25. G. F. Dalla Betta, M. Boscardin, L. Bosisio, C. Piemonte, S. Ronchin, A. Zoboli, and
N. Zorzi, New developments on 3D detectors at IRST, 2007 IEEE Nuclear Science
Symposium, Honolulu (USA), October 28November 3, 2007, Conference Record,
paper N18-3.
26. C. Fleta, D. Pennicard, R. Bates, C. Parkes, G. Pellegrini, M. Lozano, V. Wright,
M. Boscardin, G. F. Dalla Betta, C. Piemonte, A. Pozza, S. Ronchin, and N. Zorzi,
Simulation and test of novel 3D silicon radiation detectors, Nucl. Instrum. Methods A,
579, 642647, 2007.
27. G. Lindstrm, M. Moll, and E. Fretwurst (The RD48 Collaboration), Radiation hardness
of silicon detectorsa challenge from high energy physics, Nucl. Instrum. Methods A,
426, 115, 2001.
28. C. Da Via, S. I. Parker, and G. Darbo, Development, testing and industrialization of full3D active-edge and modified-3D silicon radiation pixel sensors with extreme radiation
hardness. Results, plans. ATLAS Upgrade Document, 2007, http://cern.ch/atlas-highlumi3dsensor.
29. D. Bortoletto, 3D detectors for inner pixel layers. CMS SLHC Upgrade Workshop,
CERN, May 2122, 2008, http://indico.cern.ch/conferenceDisplay.py?confId=28746.
30. R. D. Heuer, D. Miller, F. Richard, P. Zerwas, et al., TESLA Technical Design Report,
Part III: Physics at an e+e- Linear Collider, http://tesla.desy.de/new_pages/TDR_CD/start.
html.
31. M. Bona, J. Garra Tic, E. Graugs Pous, P. Colangelo, et al., SuperBa high-luminosity
asymmetric e+e Super Flavour Factory, Conceptual Design Report, INFN/AE 07/2,
SLAC-R-856, LAL 0715, March 2007, http://www.pi.infn.it/SuperB/?q=CDR.
32. M. Christophersen and B. F. Phlips, Thick silicon drift detectors, 2008 IEEE Nuclear Science
Symposium, Dresden (Germany), October 1925, 2008, Conference Record, Paper N34-5.
33. L. Tlustos, M. Campbell, E. Heijne, and X. Llopart, Signal variations in high-granularity
Si pixel detectors, IEEE Trans. Nucl. Sci., 51(6), 30063012, 2004.
34. S. I. Parker, C. J. Kenney, D. Gnani, A. C. Thomson, E. Mandelli, G. Meddeler, J. Hasi,
J. Morse, and E. M. Westbrook, 3DX: an X-ray pixel array detector with active edges,
35. P. Kleimann, J. Linnros, C. Frjdh, and C. S. Petersson, An X-ray imaging pixel detector based on scintillator filled pores in a silicon matrix, Nucl. Instrum. Methods A, 460,
1519, 2001.
36. X. Badel, J. Linnros, M. S. Janson, and J. sterman, Formation of pn junctions in deep
silicon pores for X-ray imaging detector applications, Nucl. Instrum. Methods A, 509,
96101, 2003.
37. M. Juntunen, F. Ji, K. Henttinen, I. Luusua, I. Hietanen, and S. Ernen, Fully tileable
photodiode matrix for medical imaging by using through-wafer interconnects, Nucl.
Instrum. Methods A, 580, 10001003, 2007.
64
38. J. D. Schaub, D. L. Rogers, M. Yang, S. M. Csutak, B. Yang, and J. C. Campbell, Lateral

PIN photodiodes for high-speed silicon optical receivers, IEEE LEOS Industrial Research
Highlights, December 2002, http://www.ieee.org/organizations/pubs/newsletters/leos/
dec02/industrial_research.html.
39. C. J. Kenney, S. I. Parker, B. Krieger, B. Ludewigt, T. P. Dubbs, and H. Sadrozinski,
Observation of beta and X rays with 3-D architecture silicon microstrip detectors, IEEE
Trans. Nucl. Sci., 48(2), 189193, 2001.
40. S. I. Parker and C. J. Kenney, Performance of 3-D architecture silicon sensors after
intense proton irradiation, IEEE Trans. Nucl. Sci., 48(5), 16291638, 2001.
41. J. E. Metcalfe, I. Gorelov, M. Hoeferkamp, and S. Seidel, Capacitance simulations and
measurements of 3D pixel sensors under 55 MeV proton exposure, IEEE Trans. Nucl.
Sci., 55(5), 26792684, 2008.
42. C. J. Kenney, J. Hasi, S. Parker, A. C. Thomson, and E. Westbrook, Use of active-edge
silicon detectors as X-ray microbeam monitors, Nucl. Instrum. Methods A, 582, 178181,
2007.
43. C. Da Via, G. Anelli, J. Hasi, P. Jarron, C. Kenney, A. Kok, S. Parker, E. Perozziello,
and S. J. Watts, Advances in silicon detectors for particle tracking in extreme radiation
environments, Nucl. Instrum. Methods A, 509, 8691, 2003.
44. J. Morse, C. J. Kenney, E. M. Westbrook, I. Naday, and S. I. Parker, The spatial and
energy response of a 3d architecture silicon detector measured with a synchrotron X-ray
microbeam, Nucl. Instrum. Methods A, 524, 236244, 2004.
45. A. Kok, C. Da Via, J. Hasi, S. Watts, G. Anelli, P. Jarron, J. Morse, S. Parker, C. Kenney, J.
Segal, and E. Westbrook, 3D detectorstate of the art, presented at the 13th International
Workshop on Vertex Detectors (VERTEX 04), Como, Italy, September 1318, 2004.
46. C. Da Via, J. Hasi, C. Kenney, V. Linhart, S. Parker, T. Slavicek, S. J. Watts, P. Bem,
T. Horazdovsky, and S. Pospisil, Radiation hardness properties of full-3D active edge
silicon sensors, Nucl. Instrum. Methods A, 587, 243249, 2008.
47. I. Peric, L. Blanquart, G. Comes, P. Denes, K. Einsweiler, P. Fischer, E. Mandelli, and
G. Meddeler, The FEI3 readout chip for the ATLAS pixel detector, Nucl. Instrum.
Methods A, 565, 178187, 2006.
48. M. Mathes, M. Cristinziani, C. Da Via, M. Garcia-Sciveres, K. Einsweiler, J. Hasi,
C. Kenney, S. I. Parker, L. Reuen, J. Velthuis, S. Watts, and N. Wermes, Test beam characterization of 3-D silicon pixel detectors, IEEE Trans. Nucl. Sci., 55(6), 37313735,
2008.
49. O. M. Rhne, Edge characterization of 3D silicon sensors after bump-bonding with the
ATLAS pixel readout chip, 2008 IEEE Nuclear Science Symposium, Dresden, Germany,
October 1925, 2008, Conference Record, paper N27-8.
50. J. Jakubek, T. Holy, E. Lehmann, S. Pospil, J. Uher, J. Vacik, and D. Vavrik, Neutron imaging with Medipix-2 chip and a coated sensor, Nucl. Instrum. Methods A, 560, 143147,
2006.
51. J. Uher, C. Frjdh, J. Jakubek, C. Kenney, Z. Kohout, V. Linhart, S. Parker, S. Petterson,
S. Pospil, and G. Thungstrm, Characterization of 3D thermal neutron semiconductor
detectors, Nucl. Instrum. Methods A, 576, 3237, 2007.
52. L. Tlustos, J. Kalliopuska, R. Ballabriga, M. Campbell, S. Eranen, and X. Llopart,
Characterisation of a semi 3-D sensor coupled to Medipix2, Nucl. Instrum. Methods A,
580, 897901, 2007.
53. G. Pellegrini, M. Lozano, M. Ulln, R. Bates, C. Fleta, and D. Pennicard, First double
sided 3-D detectors fabricated at CNM-IMB, Nucl. Instrum. Methods A, 592, 3843,
2008.
54. D. Pennicard, G. Pellegrini, M. Lozano, C. Fleta, R. Bates, and C. Parkes, Design, simulation, production and initial characterisation of 3D silicon detectors, Nucl. Instrum.
Methods A, 598, 6770, 2009.

55. D. Pennicard, C. Fleta, R. Bates, V. OShea, C. Parkes, G. Pellegrini, M. Lozano, J.
Marchal, and N. Tartoni, Charge sharing in double-sided 3D Medipix2 detectors, Nucl.
56. C. Fleta, R. Bates, D. Pennicard, C. Parkes, G. Pellegrini, M. Lozano, J. Marchal, and
N. Tartoni, Latest results from 3D Medipix pixel and strip sensors in Glasgow, presented
at the 4th Trento Workshop on Advanced Radiation Detectors (3D and p-type), Trento,
February 1719, 2009, http://tredi.fbk.eu.
57. A. Pozza, M. Boscardin, L. Bosisio, G. F. Dalla Betta, C. Piemonte, S. Ronchin, and
N. Zorzi, First electrical characterisation of 3D detectors with electrodes of the same
doping type, Nucl. Instrum. Methods A, 560, 317321, 2007.
58. C. Piemonte, M. Boscardin, L. Bosisio, G. F. Dalla Betta, A. Pozza, S. Ronchin, and
N. Zorzi, Study of the signal formation in single-type-column 3D silicon detectors,
Nucl. Instrum. Methods A, 579, 633637, 2007.
59. G.-F. Dalla Betta, M. Boscardin, L. Bosisio, M. Bruzzi, V. Cindro, S. Eckert,
G. Giacomini, G. Kramberger, S. Khn, U. Parzefall, M. K. Petterson, C. Piemonte,
A. Polyakov, I. Rachevskaia, G. Resta, S. Ronchin, H.F.-W. Sadrozinski, M. Scaringella,
C. Tosi, M. Zavrtanik, A. Zoboli, and N. Zorzi, Development of 3D detectors at FBKirst, Proceedings of Science16th Workshop on Vertex Detectors (Vertex 2007), 2008,
Paper 23.
60. S. Ronchin, M. Boscardin, C. Piemonte, A. Pozza, N. Zorzi, G. F. Dalla Betta,
G. Pellegrini, and L. Bosisio, Fabrication of 3D detectors with columnar electrodes of
the same doping type, Nucl. Instrum. Methods A, 573, 224227, 2007.
61. A. Zoboli, M. Boscardin, L. Bosisio, G. F. Dalla Betta, C. Piemonte, A. Pozza,
S. Ronchin, and N. Zorzi, Electro-optical measurements of 3D-STC detectors fabricated
at ITC-IRST, Nucl. Instrum. Methods A, 583, 149152, 2007.
62. M. Zavrtanik, V. Cindro, G. Kramberger, J. Langus, I. Mandic, M. Mikuz, M. Boscardin,
G.-F. Dalla Betta, C. Piemonte, A. Pozza, S. Ronchin, and N. Zorzi, Position sensitive
TCT evaluation of irradiated 3d-stc detectors, 2007 IEEE Nuclear Science Symposium,
Honolulu (USA), October 28November 3, 2007, Conference Record, paper N24-150.
63. F. Campabadal, C. Fleta, M. Key, M. Lozano, C. Martinez, G. Pellegrini, et. al., Design
and performance of the ABCD3TA ASIC for readout of silicon strip detectors in the
ATLAS semiconductor tracker, Nucl. Instrum. Methods A, 552, 292328, 2005.
64. S. Eckert, T. Ehrich, K. Jakobs, S. Khn, U. Parzefall, M. Boscardin, C. Piemonte, and
S. Ronchin, Signal and charge collection efficiency of a p-type STC 3D-detector irradiated
to sLHC-fluences, read out with 40MHz, Nucl. Instrum. Methods A, 581, 322325, 2007.
65. S. Khn, G. F. Dalla Betta, S. Eckert, K. Jacobs, U. Parzefall, A. Zoboli, and N. Zorzi,
Short strips for the sLHC: a p-type silicon microstrip detector in 3-D technology, IEEE
Trans. Nucl. Sci., 55(6), 36383642, 2008.
66. A. Zoboli, M. Boscardin, L. Bosisio, G.-F. Dalla Betta, C. Piemonte, S. Ronchin, and
N. Zorzi, Double-sided, double-type-column 3D detectors at FBK: design, fabrication
and technology evaluation, IEEE Trans. Nucl. Sci., 55(5), 27752784, 2008.
67. F. Gianotti, M. L. Mangano, T. Virdee, S. Abdullin, G. Azuelos, A. Ball, et al., Physics
potential and experimental challenges of the LHC luminosity upgrade, Eur. Phys. J.,
C39, 293333, 2005.
68. M. Bruzzi, J. Adey, A. Al-Ajili, P. Alexandrov, G. Alfieri, P. P. Allport, et al., Radiationhard semiconductor detectors for SuperLHC, Nucl. Instrum. Methods A, 541, 189201,
2005.
69. A. Zoboli, M. Boscardin, L. Bosisio, G.-F. Dalla Betta, S. Eckert, S. Khn, et al.,
Functional characterization of 3D-DDTC detectors fabricated at FBK-irst, 2008 IEEE
Nuclear Science Symposium, Dresden, Germany, October 1925, 2008, Conference
Record, paper N34-4.
66
70. A. Zoboli, M. Boscardin, L. Bosisio, G.-F. Dalla Betta, S. Eckert, S. Khn, et al., Laser
and beta source setup characterization of p-on-n 3D-DDTC detectors fabricated at FBKIRST, Nucl. Instrum. Methods A, 604, 238241, 2009.
71. E. Verbitskaya, V. Eremin, Z. Li, J. Hrknen, and M. Bruzzi, Concept of double peak
electric field distribution in the development of radiation hard silicon detector, Nucl.
72. G. Kramberger, M. Batic, V. Cindro, I. Mandic, M. Miku, and M. Zavrtanik, Annealing
study of effective trapping times in silicon detectors, Nucl. Instrum. Methods A, 571,
608611, 2007.
73. A. Affolder, P. Allport, and G. Casse, Liverpool charge collection results on RD50
p-type sensor sets irradiated with neutrons and protons, presented at the 3rd Workshop
on Advanced Silicon Radiation Detectors (3D and p-type Technology), Barcelona,
Spain, April 1416, 2008, http://indico.cern.ch/conferenceDisplay.py?confId=28165.
74. A. Zoboli, M. Boscardin, L. Bosisio, G. F. Dalla Betta, C. Piemonte, S. Ronchin, and
N. Zorzi, Initial results from 3D-DDTC detectors on p-type substrates, presented at
the 7th International Conference on Radiation Effects on Semiconductor Materials,
Detectors and Devices (RESMDD08), Florence, Italy, October 1517, 2008, to appear
in Nucl. Instrum. Methods A, 2009 (doi:10.1016/j.nima.2009.08.010).
75. C. Da Via, S. Parker, M. Deile, T.-E. Hansen, J. Hasi, C. Kenney, A. Kok, and S. Watts,
Dual readoutstrip/pixel systems, Nucl. Instrum. Methods A, 594, 712, 2008.
Zinc Telluride
3 Cadmium
Pixel Detectors for Hard
X-Ray Astrophysics
Fiona A. Harrison, Walter R. Cook,
H. Miyasaka, and R. McLean
Contents
3.1
3.2
3.3
3.4
Introduction..................................................................................................... 67
Detector Requirements.................................................................................... 68
CdZnTe Detector Architecture Considerations............................................... 70
The Caltech CdZnTe Pixel Detector................................................................ 71
3.4.1 Sensor Architecture and Material Selection........................................ 73
3.4.2 ASIC Readout...................................................................................... 74
3.4.2.1 Summary of Operation......................................................... 75
3.4.2.2 Additional Features............................................................... 77
3.5 Hybrid Detector Performance.......................................................................... 77
3.5.1 Spectral Resolution.............................................................................. 77
3.5.2 Depth Determination........................................................................... 79
3.6 Conclusion.......................................................................................................80
Appendix: Scaling of Readout Noise with Pixel Size.............................................. 81
References................................................................................................................. 82
3.1Introduction
The next generation of space-based hard X-ray (defined here as the energy range
from 10 to several hundred kiloelectron volts) telescopes for astrophysical observations will include both large-area coded aperture imaging systems for synoptic studies over a wide field of view and focusing telescopes to reach faint flux limits over
more limited regions of the sky. Both applications require detectors with good (a few
percentage points) energy resolution and two-dimensional position resolution of a
millimeter or better. In addition, space-based telescopes in this energy band benefit
from compact detector geometries that can be well shielded from the intense background radiation produced in the atmosphere and spacecraft by cosmic rays. The
ability to measure the depth of interaction is also desirable for background rejection.
67
68
Low power is necessary, especially for large-area applications, and complexity is

greatly reduced if focal planes do not require cryogenic operation.
For these reasons, a number of hard X-ray astrophysics missions currently under
study or development plan to employ cadmium zinc telluride (CdZnTe) pixel detectors. Relative to the alkali halide scintillators or germanium detectors employed by
previous-generation astrophysics experiments, CdZnTe offers the ability to implement position-sensitive focal planes in compact geometries without cryogenic cooling systems. CdZnTe can also, if properly implemented, provide reasonably good
spectral resolution of 24% across this energy range.
This chapter describes the requirements placed on imaging detector systems
by next-generation hard X-ray astrophysics telescopes and reviews the trade-offs
in possible detector architectures. The requirements of astronomical imaging systems for a range of telescope designs motivate a pixel detector geometry with a
direct-bonded readout as an optimal solution. This chapter describes in detail a pixel
detector developed for both focusing and coded aperture applications optimized for
space applications.
3.2Detector Requirements
Imaging telescopes for hard X-ray astrophysics use either coded aperture masks or
grazing incidence focusing optics. Coded apertures are based on a pinhole camera
concept, by which a mask of opaque and transparent elements is placed in front of
an imaging detection plane.1 Shadow patterns of the mask cast on the detector by
sources in the field of view encode information about the source distribution. Decon
volution of the shadow pattern produces an image of the source distribution. Coded
mask telescopes can achieve large fields of view (several hundred square degrees
or more) and can operate over a broad range of X-ray energies and even into the
gamma-ray band up to a few million electron volts if masks and detectors are made
sufficiently thick. They are employed for large-area surveys and detection and monitoring of short-lived and time-variable sources.
Achieving the sensitivity and angular resolution desired by next-generation coded
aperture telescopes requires large (>5,000 cm2) position-sensitive detection planes
with millimeter or better spatial resolution. A typical separation achievable between
mask and detector is 24 m, driven by the size of available expendable launch vehicles. The telescope angular resolution is set by the mask element (or hole) size
divided by the focal length. To achieve acceptable contrast, the detector must oversample the mask hole pattern by a factor greater than 2, implying 1- to 2-mm detector spatial resolution to achieve arc minute angular resolution.
Coded aperture telescopes offer the possibility of monitoring the flux from many
sources over a wide field of view, but they have limited sensitivity because the signal from a source is encoded over a large detector area, making noise from detector background events dominant. Table3.1 shows the detector requirements for the
Energetic X-ray Imaging Survey Telescope (EXIST) mission,2 being studied by the
National Aeronautics and Space Administration (NASA) for possible implementation in the next decade.
Cadmium Zinc Telluride Pixel Detectors for Hard X-Ray Astrophysics
69
Table3.1
Detector Requirements for the EXIST
Coded Aperture Survey Mission
Parameter
Requirement
Energy range
Detector area
Spatial resolution
Spectral resolution
Maximum count rate
Temporal resolution
5600 keV
4.5 m2
<700 m
<4% FWHMa
50 counts/cm2/s
<5 s
FWHM, full width at half maximum.
Technologies have been developed to extend true focusing optics to X-ray energies
above 10 keV. Focusing hard X-ray telescopes, in contrast to coded apertures, operate
over much smaller fields of view and achieve high sensitivity by concentrating the
X-ray flux from a point source on a small region of the detector. Extending grazing
incidence optics to X-ray energies above 10 keV requires long focal lengths of 510 m,
and special coatings3 are applied to the mirror surfaces to achieve an upper energy
limit of about 100 keV with a reasonable (~10 arcminutes on a side) field of view.
Hard X-ray focusing telescopes require imaging detectors with good spectral
resolution. The detector spatial resolution must be adequate to sample the optics
point spread function, which is determined by the level of perfection of the optics
figure and the coalignment of the optical elements. Current hard X-ray optics technologies4,5 achieve 2040 angular resolution (as characterized by the half-power
diameter or width of a spot containing half of the reflected X-rays). For a 10-m
telescope this requires a focal plane detector with 0.3- to 0.6-mm spatial resolution
to adequately sample the point spread function. The size of the detector required is
determined by the optics field of view (FoV; L = FoV*focal length); a 3-cm diameter
detector will fill a 10 arcminute field of view for a 10 m focal length. Table3.2 shows
Table3.2
Detector Requirements for the NuSTAR
Hard X-Ray Focusing Telescopes
Parameter
Requirement
Energy range
Focal plane size
Spatial resolution
Spectral resolution
Maximum count rate
580 keV
3.8 3.8 cm
<750 m
<1.6 keV FWHMa
200 counts/s/focal plane
FWHM, full width at half maximum.
70
the detector requirements for the Nuclear Spectroscopic Telescope Array (NuSTAR)
mission,6 a NASA mission that will deploy the first hard X-ray focusing telescopes
in orbit.
Because of the advantages it offers as a semiconductor material with a high atomic
number that can be instrumented with low-power application-specific integrated
circuit (ASIC) readouts, CdZnTe pixel detectors will be flown on the NuSTAR mission, and it is one of two detector options for the EXIST mission. In the following
sections, we consider some of the trade-offs inherent in selecting a detector architecture suitable for both very large area coded aperture applications and compact focal
plane arrays for which spectral resolution and spatial uniformity are critical.
3.3CdZnTe Detector Architecture Considerations

The applications described require position-sensitive detectors with 0.5- to 1-mm
spatial resolution and good energy resolution. Achieving good quantum efficiency
across the requisite energy range requires a detector thickness of 2 mm for focusing
systems to 5 mm for coded apertures. These requirements dominate the choice of the
appropriate detector architecture. The choices are either double-sided strip contacts
or a pixel architecture. For pixel detectors, the anode contact of the CdZnTe sensor is
segmented into pixels and connected to a readout. There are two options for making
this connection. The first is a direct-bonded approach, by which the ASIC readout
pitch matches the detector pixel pitch, and a direct connection is made between the
ASIC contact of the pixel and the anode contact of the sensor by a bump bond or
similar technique. In this case, the size of the ASIC chip matches that of the sensor.
Alternately, an interposer board can be used to route the pixel signals off the sensor
to a remote readout chip. In this case, the ASIC size need not match the sensor.
For CdZnTe, the charge transport properties are a dominant consideration in
choosing the architecture for applications demanding good spectral performance.
For an appropriate detector thickness of a few millimeters under typical bias voltages of several hundred volts, the electron mean free paths are on the order of 1 cm,
while hole mean free paths are more than an order of magnitude smaller.7 These values apply to typical material grown using a vertical Bridgeman method; however, an
imbalance in charge collection between electron and holes is evident in all material
that is readily available commercially. The consequence is that for detector thicknesses smaller than 5 mm, the holes are trapped close to the interaction site, while
the electrons are collected at the anode. Further, the charge collection properties for
both electrons and holes are spatially inhomogeneous due to extended defects and
tellurium inclusions, and isolated regions exist in typical material where the electron
collection is also poor.8
Charge collection considerations favor a pixel architecture with small pixels.
The poor hole mobility makes double-sided strip detectors difficult to implement9
since they rely on both hole and electron mobility. For detectors with a ratio of pixel
dimension to detector thickness larger than 1, the anode signal becomes dependent
on the depth of the X-ray interaction due to variation in the contribution of image
charge from trapped holes. As the aspect ratio becomes smaller (smaller pixels), the
anode signals become increasingly dependent only on electron transport, and the
71
deleterious effects of hole trapping are diminished. Spatial variations in the electron
transport properties due to extended defects and tellurium inclusions also broaden
energy resolution, and this effect will likewise be reduced for smaller pixels. The
advantage of smaller pixels is that they hold down sizes at which diffusion causes
charge to be predominantly split among numerous pixels, for which charge loss
between anode contacts becomes a dominant factor.10
Another consideration in determining the optimal architecture is the electronic
noise. If we assume the sensor is cooled to a level at which leakage current is a
negligible contribution to the electronic noise, the dominant concern is the input
capacitance. For a detector with anode contacts read out by a low-noise chargesensitive preamplifier, a straightforward but somewhat lengthy scaling argument
shows that the smaller the pixel, the smaller the contribution of electronic noise
to energy resolution for fixed power per unit detector area will be. We provide the
scaling argument in the Appendix, since it is not obvious that increasing the number of circuits per unit area accomplishes a gain in performance for the same total
power. The advantage of small pixels results from the lower input capacitance.
This consideration also favors a direct-bonded approach over a remote readout
because for the remote readout stray capacitance dominates the total input capacitance for pixels smaller than 1 mm. The advantage of smaller pixels from an
electronic noise perspective again ceases to hold when pixel sizes become small
enough that the majority of events have charge split among multiple pixels. For a
2-mm thick detector, this occurs for pixel sizes smaller than 0.4 mm.11
A potential disadvantage of direct-bonded geometry is that the area of ASIC
required is the same as the detector area, and for detectors of a few square centimeters or more, the ASIC cost could become comparable to the sensor cost unless yields
are high. In practice, for the implementation described in Section 3.4, the ASIC
yields are very high (typically better than 90% for the process and architecture discussed), significantly higher than the yield of high-quality spectroscopic detectors.
In practice, our experience showed that the detector cost dominates by a factor of 10
or more even for a direct-bonded approach.
After considering the trade-offs of the three architectures (strips, direct-bonded
pixels, and pixels with a remote readout), we conclude that for astronomical applications in the energy range from 5 to 500 keV at which optimal spectroscopic performance is required, a direct-bonded pixel detector with pixel size of about 500 m
is the best choice even for large-area applications and when the telescope resolution
does not demand this level of resolution. The direct-bonded architecture with pixels
of this scale has performance advantages because it minimizes effects of imperfect
charge collection and reduces electronic noise contributions for a fixed total power.
The 0.5-mm pixel size optimal from these considerations is also well matched to the
requirements of next-generation space missions.
3.4The Caltech CdZnTe Pixel Detector

In the next sections, we describe the design and performance of a CdZnTe hybrid
pixel detector optimized for space astrophysics applications. The detector was originally developed at Caltech for the High Energy Focusing Telescope (HEFT) balloon
72
program, and it has been improved for subsequent applications, including for homeland security devices and most recently for the NuSTAR space mission.
The primary objective of the Caltech program has been to develop a detector
that meets the performance requirements for space-based hard X-ray telescopes
that demand good spectral and spatial resolution with very low power per pixel for
moderate X-ray count rates of a few hundred per square centimeter per second. For
focusing telescopes, which only require about 10 cm 2 of detector area, low power
is required to allow the focal plane to be passively cooled to a level that minimizes
the contribution of leakage current to the electronic noise. For most space telescopes
in low-Earth orbit, detectors can typically be cooled passively to between 5 and
+5C if minimal power is dissipated locally. For large-area coded aperture arrays,
very low power per unit area is required to meet reasonable power levels that can be
generated onboard spacecraft. For the reasons described in the previous section, we
selected an architecture based on a custom low-noise ASIC chip directly bonded to
a CdZnTe sensor.
The pixel detector hybrid consists of a CdZnTe sensor with the anode segmented
into pixels with 605 m pitch bonded through a gold wire/epoxy interconnect to a
custom ASIC with low-noise charge-sensitive circuits laid out with the identical pitch
(Figure3.1). We were targeting a pixel size between 0.5 and 0.75 mm to meet the
spatial resolution requirements associated with NuSTAR and to optimize spectral
performance. The 605-m pitch ultimately resulted from the practical consideration
that the readout circuitry fit comfortably within this area. The pixels are laid out in a
32 32 square array, resulting in a total active area for the hybrid of 1.94 1.94 cm.
The array dimension was chosen to match both the maximum size possible using the
ASIC foundry process (AMInow ON semiconductorC5N) as well as the typical
a5V
reset(n)
precharge
charge(n)
read
pixel(m)
1p
a5V
(typ of 16 sample
caps; n = 0, 15)
adcclr
adcclk
read
pixel(m)
read
12 bit charge
balance ADC
Ireset
lockout
adcout
amp/disc/
latch
input(m)
15f
(typ of 1024 pixels;
m = 0, 1023)
250K
250K
read
pixel(m)
trigout
Figure 3.1 Schematic diagram of the ASIC readout circuitry associated with each pixel.
73
CdZnTe
Epoxy
Gold stud
ASIC
Figure 3.2 SEM photograph of the goldepoxy interconnect used to connect the ASIC
input pad to the CdZnTe anode contact. The height of 50 m minimizes the pixel input
capacitance.
dimension of CdZnTe sensors that, based on our experience, can be obtained with
relatively uniform quality.
The ASIC-to-pixel interconnect is designed to minimize the preamp input capacitance. The capacitance between CdZnTe contacts is fixed by the pixel size, and for
0.6-mm pixels it is about 300 fF. In order not to contribute significantly to this fixed
capacitance, the separation between CdZnTe anode contact and the ASIC ground
plane must be greater than 20 m. A standard indium bump bond technique will
not achieve this large a gap, so we use a gold wire attached to the ASIC input pad,
which is connected to the CdZnTe using a conductive epoxy dot. Figure3.2 shows
a scanning electron microscopic (SEM) image of the interconnect, which is 50 m
tall, making the contact to ASIC ground capacitance about 65 fF, so that it does not
contribute significantly to the total input capacitance.
NuSTAR requires a focal plane size greater than 1.8 cm on a side, requiring the
hybrid detector units to be tiled in a 2 2 array. For this reason, we have designed the
detector to have two critical edges, where two hybrids can be closely spaced, with a
typical gap of 100 m between units. The wire bonds routing signals off and onto the
ASIC chip all come out one edge of the device. Figure3.3 shows a 2-mm thick hybrid
detector mounted onto a carrier board, showing the two critical edges of the assembly.
3.4.1Sensor Architecture and Material Selection

We procure the CdZnTe sensors, grown using a high-pressure vertical Bridgeman
technique, with contacts from eV Products. The cathode contact is about 1,000
thick platinum, and the anode contacts are a combination of about 1,000 thick
platinum and about 3,000 thick gold. We designed the anode pattern to have a
guard ring on all four sides. On the critical edges, the guard ring is 50 m wide and
placed within 50 m of the physical detector edge. On the other two edges, the guard
74
Figure 3.3 Photograph of the Caltech hybrid detector on a carrier board.
ring is wider, with looser placement tolerances. Previous studies on the charge loss
between pixel contacts10 showed that the gap between pixels should be minimized,
even at the cost of larger input capacitance. Therefore, we have designed with the
minimum 50-m gap dictated by eVs fabrication process.
To obtain the most uniform material possible, we specify that the sensors have no
grain boundaries, and the CdZnTe wafers are imaged in the infrared by eV to select
uniform regions with small and uniform inclusion concentrations.
3.4.2ASIC Readout
The readout chip is an ASIC custom designed by Caltech and is optimized for determining the amplitude of charge pulses from CdZnTe detectors for moderate count
rates of about 100 cm2/s. The design incorporates a method of charge pulse amplitude measurement first used in an earlier Caltech chip, the HEFT chip,12 which minimizes power consumption while maintaining excellent low-noise performance. The
pixel readout circuitry also provides sensitivity to charge signals of both polarities
(due to electrons as well as that induced by holes) to provide depth of interaction
sensitivity, enabling both high-energy (EX-ray > 100 keV) spectral corrections and
background rejection capability, both of which are desirable for astrophysical applications. The chip includes rad-hard by design features, such as guard rings around
the Field Effect Transistors (FETs), to make the chip tolerant to the space radiation
environment.
The most recent iteration of the ASIC, the NuSTAR chip, is the result of a decade
of CdZnTe readout circuitry development, and the basic architecture has undergone
numerous upgrades motivated both by astrophysics and homeland security applications. The chip contains 1,024 identical readout channels in a 32 32 array with a
pitch between input pads of 605 m. Each channel includes a low-noise preamplifier,
16 sampling capacitors, a shaping amplifier, a discriminator, and a latch. The user
can select one of two methods of direct current (DC) feedback for the preamplifiers,
75
both of which allow direct detector coupling. The normal mode feedback, added
for homeland security applications and to provide for easy diagnostic operation
during ground testing for NuSTAR, handles leakage currents up to about 10 nA,
enabling operation at temperatures up to 40C for typical CdZnTe detector leakage
currents. The charge pump mode feedback achieves very low noise (near 200 eV
full width at half maximum [FWHM]) and handles leakage currents up to about
300 pA, making it suitable for cooled detector applications.
The direct coupling of detector and preamplifiers is important in compact applications, eliminating the need for bulky coupling networks. In particular, DC coupling allows the chip to be direct bonded to a detector of matching pixel pitch,
achieving the ultimate in low input capacitance and very low noise operation. Owing
to the sampling architecture (see Section 3.4.2.1), which eliminates the need for a
precision shaping amplifier and peak detector in each pixel, the power consumption
of the chip is very low, in the range 50 to 100 W per pixel, depending on the choice
of input FET currents, which are set by an external resistor.
The NuSTAR chip includes a 12-bit charge rebalance analog-to-digital converter (ADC). The voltage reference for the ADC is also included on the chip and
is designed to compensate the temperature coefficient of gain of the preamplifiers
down to a very low effective value near 20 ppm/C.
The overall gain of the readout circuits, the discrimination level, and other aspects
of chip operation are programmable via a serially loaded on-chip command register.
The circuit gain can be adapted over a wide range, yielding dynamic ranges suitable
for the NuSTAR astrophysical hard X-ray telescope (2150 keV), as well as groundbased radiation monitoring (up to 3 MeV).
3.4.2.1Summary of Operation
Figure3.1 shows a diagram of the readout circuitry. The output of each preamplifier is presented as a voltage signal to a shaping amplifier, which in turn drives a
discriminator input. A photon of energy above a threshold programmable from 2 to
40 keV will trigger the discriminator, set a latch, and signal off-chip logic to begin an
event-processing cycle. The output of each preamplifier is also presented as a current
signal to a bank of 16 sampling capacitors. Analog switches route the current signal
in turn to each capacitor, dwelling a fixed time (hundreds of nanoseconds) on each.
Each capacitor is reset to a fixed voltage prior to its integration period. The process
of acquiring the current-integrated preamplifier output samples proceeds continuously until an above-threshold event trigger occurs. At the time of photon detection,
the recent time history of the preamplifier output waveform is present on the bank
of 16 capacitors. After photon detection, the sampling process is allowed to continue
for about eight more samples, then is halted. At that point, the capacitor bank stores
approximately eight preevent samples and eight postevent samples. These samples
record the step-like response of the preamplifier to the photon event. Off-chip logic
circuitry then scans the chip to determine which of the 64 channels have been triggered. For the triggered channels (and optionally their nearest neighbors), the stored
charge on each of the 16 capacitors is measured using the on-chip ADC.
The readout sequence is designed to transfer the stored charges, one at a time, to
a charge-rebalance ADC. The charge transfer process returns the voltage on each
76
of the 16 sampling capacitors to their reset values. To first order the system transfer function is independent of the sampling capacitance values, such that variations
due to manufacturing tolerances do not produce noise on the preamplifier output record. This design feature greatly simplifies analysis of the preamplifier output
records since it is not necessary to store gain calibration information separately for
each sampling capacitor.
Analysis of the preamplifier output record (sixteen 12-bit numbers) is performed
off chip by a microprocessor, which extracts a single number proportional to the
photon energy. This extraction process performs a similar function as that of the precision pulse-shaping and peak detection circuits of a traditional pulse-height analysis
system. The key to the extremely low power consumption of the system is that these
traditional circuit elements are entirely eliminated and their functions replaced by
digital signal postprocessing.
Figure3.4 shows the performance of the ASIC when attached to a CdZnTe detector. The spectrum, from an 155Eu radioactive source, includes only events from one
pixel and further only includes those that trigger only that pixel (i.e., split-charge
events are excluded). Thus, it demonstrates performance achievable by the ASIC for
a highly uniform detector. The data were taken in charge pump mode with a 5-mm
thick detector cooled to 0C. The pulsar peak width is 340 eV FWHM, demonstrating
the very low electronic noise achieved by the design. The X-ray lines are broader and
increase in width with energy due to charge-trapping effects, ranging from 630 eV
FWHM at 41 keV to 1.1 keV at 86.5 keV.
H14, 500V, 0C
pixel 10,10
single pixel events only
no depth cut
common mode noise subtraction ON
a: pulser FWHM = 0.34 keV
b: 86.5 keV line FWHM = 1.10 keV
c: 60.0 keV line FWHM = 0.91 keV, centroid error = +0.017 keV(1)
d: 48.7 keV line FWHM = 0.68 keV, centroid error = 0.030 keV
e: 45.3 keV line FWHM = 0.63 keV, centroid error = 0.041 keV
(1) error in measured line energy using single point calibration
at 86.5 keV and assuming linear response with no oset.
d
c
Figure 3.4 Spectrum from an 155Eu gamma-ray source taken with the Caltech ASIC in
charge pump mode coupled to a 5-mm thick CdZnTe detector, demonstrating the low electronic noise (340 eV FHWM) achieved by the design. Only events from a single pixel are
included.
77
3.4.2.2Additional Features
A scope-out signal is provided for oscilloscope viewing of the preamp voltage
output waveforms, as well as the shaped signals that feed the discriminators, and a
DC signal from each pixel that is linearly related to that pixels leakage current. By
connecting an external ADC to the scope-out signal, the leakage current of every
pixel can be measured and tracked as a monitor of detector health.
Each preamplifier also has a separate precision test pulser. These test pulsers
produce a pulse with amplitude set by an externally applied DC voltage. The command register contains bits that define which pixels test pulser is configured to pulse
in response to an externally applied signal. These built-in pulsers allow a functional
check of all preamplifiers as well as precise measurement of the linearity, offset, and
electronic noise.
While the shaping amplifier/discriminator in each pixel is configured to trigger
only on negative input signals (electron signals), the sampling architecture allows the
recording of both positive and negative preamplifier step responses. The DC level
of the preamplifier output determines the amount of positive signal (hole signal)
dynamic range and is programmable via the command register.
3.5Hybrid Detector Performance

We have fabricated and evaluated hybrids with both 5-mm CdZnTe sensors and
2-mm thick material. The 5-mm devices were fabricated for homeland security
applications, for which the demands for spectral resolution are lower, and these sensors were not screened prior to fabrication at eV. The 2-mm sensors were screened as
described in Section 3.4.1. The performance evaluation demonstrates that good spectral resolution and uniform response can be achieved. This section presents results of
the spectral resolution and depth determination capability of the hybrid detectors.
3.5.1Spectral Resolution
Astrophysical sensors demand both near-unity detection efficiency and good spectral resolution. This requires that events not be discarded to achieve the desired
spectral performance; specifically, the majority of pixels must have good response,
and events with charge split among pixels should not be rejected. We have investigated both the extent of charge splitting and the level of corrections required to
achieve good spectral resolution with high quantum efficiency.
As described, the NuSTAR chip allows flexible readout of not only the triggered
pixels but also their nearest neighbors. We have found this is necessary for optimal
energy reconstruction (it is also used for depth determination as described in the next
section). In particular, we found that reconstructing charge from neighboring pixels
that do not trigger the discriminator but have a low-level signal present significantly
improves the performance. We also found that, due to charge loss between pixels as
well as nonuniformities in the detector material, corrections must be made based on
the distribution pattern of the charge splitting (an event topology correction) as well
as for the depth of interaction.
78
The event topology correction is critical to removing tailing, or the redistribution of events to an extended wing on the low-energy side of the spectral line.
We have performed a detailed study of the corrections that must be applied based
on the pattern of charge splitting using analysis of the 87-keV line from 155Eu. For
this analysis, we read out the pixel with the largest trigger as well as its nine nearest neighbors. If we applied a software trigger created by searching for an electron signal above the noise level (even if the hardware trigger is not present in that
pixel) in any of the nine pixels and include these, along with the hardware triggers as
hit pixels, we found that 97% of the X-ray events fell into 1 of 13 topologies. The
majority (~62%) were single-pixel events (no charge splitting), about 30% were one
of the four possible topologies of two-pixel events, about 4% were three-pixel events
(with four possible patterns), and the remainder were four-pixel events (with four possible patterns).
We found that all of these distinct patterns had a different gain, or charge collection efficiency, associated with them that must be corrected. If the dominant effect
of charge splitting is loss of charge between contacts but the detector is uniform,
we would expect that only four different gain factors would be needed. The fact
that topologically equivalent (identical symmetry) patterns required slightly different correction factors indicates that detector inhomogeneities are a significant factor
in broadening the line. This justifies our assumption that smaller pixels will result in
better spectral response, since these inhomogeneities can be removed to some degree
using pixel-specific corrections. This does, however, place a demand for significant
effort in detector calibration to determine all necessary corrections over all 1,024
pixels in the device.
In addition, to optimize resolution for the 87-keV line, we must apply a correction
based on the measured depth of interaction. The depth determination algorithm is
described in the next section. As a result of the variation in the contribution of the
hole signal imaged on the anode electrodes, the deeper the gamma-ray interaction
in the detector (i.e., the closer to the anode), the lower the measured signal is. This
again results in a tailing of the spectral lines above an energy at which the radiation penetrates a significant volume of the detector. A correction of the gamma-ray
energy is required above about 60 keV to compensate for this effect.
Figure3.5 shows a spectrum from a 155Eu gamma-ray source taken with a 2-mm
thick hybrid detector with the ASIC in charge pump mode. The spectrum is from
a single pixel selected at random, with all events, including those with charge split
among pixels, included. Only 1% of events were excluded based on a depth cut (those
deepest in the detector). Corrections have been made for the pattern of charge splitting
as well as for the depth of interaction. The FWHM resolution at 87 keV is 1.26 keV,
which is comparable to the 1.1 keV achieved for single-pixel events alone (i.e., those
with no charge splitting). In the surrounding region, pixels analyzed the same way
showed 67% with resolution better than 1.2 keV, with the worst measured resolution
close to 2 keV. For the few 2-mm detectors with highly screened material we have
studied in detail to date, the yield with spectral response of this quality for the majority of pixels was more than 50%, although more statistics are required to determine
79

5 104
FWHM = 1.26 keV at 86.5 keV
Data usage = 99.1%
Sigma/sqrt(N) = 0.00221 keV
Counts/keV
4 104
3 104
2 104
1 104
0
20
40
60
80
Energy (keV)
100
120
140
Figure 3.5 Spectrum from an 155Eu gamma-ray source taken with ASIC in charge pump
mode coupled to a 2-mm thick CdZnTe detector. The spectrum is from a single pixel selected
at random, with all events, including those split among pixels, included. Corrections have
been made for the pattern of charge splitting as well as for the depth of interaction. The
FWHM resolution at 87 keV is 1.26 keV.
the ultimate yield. The NuSTAR program will evaluate 50 detectors to select the best
8 for flight, which will provide a detailed evaluation of the material uniformity.
3.5.2Depth Determination
As described in Section 3.4.2, the ASIC circuitry is sensitive to both polarities of signals:
the negative signals resulting from the electrons and the positive signal induced on the
contacts by the holes. Reading out the triggered pixels and the nearest neighbors allows
the depth of interaction to be determined. In the untriggered pixels, the positive hole
signal dominates, while the triggered pixel signals provide a measurement of the electron signal. The ratio of the two depends on the interaction depth: The closer the trapped
holes are to the anode, the larger the induced signal on the neighboring pixels will be.
The ability to determine depth of interaction is useful not only for correcting
the energy signal, as described, but also for reducing the detector background for
astrophysical telescopes. Events produced in the detector by Compton scattering of
high-energy gamma-rays produced in the atmosphere and payload through cosmic
ray interactions are a dominant source of instrumental background. These will be
produced more or less uniformly throughout the detector volume, whereas X-rays
from astrophysical sources will be predominantly absorbed in the top of the detector,
at least in the energy range below 100 keV.
Figure 3.6 shows a plot of data taken by illuminating a 5-mm hybrid detector
with a 57Co gamma-ray source. The y-axis shows the sum of pulse heights in the
surrounding pixels (hole signal), and the x-axis shows the signal in the central pixel
80

20
0.0
0.5
8 Surrounding (keV)
1.0
20
2.0
40
Detector thickness = 2 mm
HV = 300 V
e = 1.0 102 cm2/V
h = 1.5 105 cm2/V
60
80
1.5
20
40
60
80
100
120
140
Center (keV)
Figure 3.6 Plot of data taken by illuminating a 5-mm hybrid detector with a 57Co gammaray source. The plot shows the sum of the surrounding pixels (hole signal) versus the signal in
the central pixel (electron signal). The solid line shows a fit to the depth profile using a charge
transport model. The best-fit values for the mu-tau products is ee = 1.0 10 2 cm2/V and
hh 1.5 10 5 cm2/V.
(electron signal). The data are selected to show only events with a trigger in one central pixel. The curvature of the 122-keV line with increasing magnitude of the hole
signal indicates the loss of signal due to hole trapping occurring closer to the anode.
The ratio of the two signals is a measure of the depth of interaction. The dark line
shows a fit to the depth profile using a charge transport model. The best-fit values
for the - products are ee = 1.0 10 2 cm2/V and hh 1.5 10 5 cm2/V. For more
complex patterns of charge splitting, the algorithm must be adjusted to include only
those pixels without an electron signal above the software threshold, but similar
depth resolution can be achieved.
To make a correction to line spectra, we fit the depth profiles (like that shown in
Figure3.6) with a second-order polynomial and correct the pulse heights based on
that fit, which is of course a function of X-ray energy. As mentioned, the depth signal
also allows background rejection. A cut can be made at a given hole signal that is
a function of the energy signal. Simulations, as well as data taken in flight with the
HEFT balloon experiment, showed that about 60% of the background above 40 keV
can be rejected for a shielded experiment with a 2-mm sensor in low-Earth orbit.
3.6Conclusion
Future astronomical hard X-ray missions require position-sensitive image plane detectors with good spectral resolution. For these applications, the optimal configuration is
81
a pixel detector with 0.4- to 0.7-mm pixels because it minimizes electronic noise and
deleterious effects of detector inhomogeneities. Larger pixels suffer from increased
electronic noise and detector inhomogeneities; for smaller pixel sizes, the majority
of events will have charge collected on multiple pixels, and the performance will
degrade as a result. We have shown that there is no advantage to larger pixels from
a perspective of power per area as long as good spectral resolution is a requirement.
At Caltech, we have developed a custom ASIC readout and sensor architecture that
can meet the spectral resolution requirements of next-generation focusing and coded
aperture telescopes. Data taken in the laboratory showed that good spectral resolution and depth sensitivity can be achieved if the detector material is appropriately
selected and screened and if a yield loss of about 50% can be accepted.
Appendix: Scaling of Readout Noise with Pixel Size

The scaling argument that favors small pixels for a fixed power per unit detector
area is as follows: Consider a detector with a single-pixel readout by a low-noise
charge-sensitive preamplifier. Now, divide the detector into N identical pixels and
scale the preamplifier such that the new preamps operate the same way as the original one (i.e., have the same bandwidth). To accomplish this, the width of each FET
is divided by N, each feedback capacitance is divided by N, but the circuit topology
remains unchanged. The result is that all the currents through the circuit divide by
N, while the bandwidths (resistivecapacitive; RC) all stay the same. The noise at
each node increases by ( N ) , such that if the circuits are recombined by connecting
the corresponding nodes, the noise would sum in quadrature back to the original
value. In particular, the noise at each preamp output increases in proportion to ( N ) .
However, since all the capacitances, including the input capacitance and the feedback capacitance, have decreased by 1/N, each preamplifiers gain [proportional to
1/(Feedback capacitance)], has increased by N. This means that, provided the input
signal is confined to a single pixel, the signal to noise has increased by N/ ( N ) or
( N ) . The power consumption has remained constant since all the currents have
dropped 1/N, while the number of circuits increased by factor N.
For large enough N, the scaling argument can break down in various ways. The
pixel might become so small that charge is split among pixels. Also, for very small
pixels, the pixel capacitance will cease to drop in proportion to 1/N. As the preamp
FETs are divided, they can become small enough that their internal capacitances no
longer scale as 1/N, and stray wiring capacitances within the preamp, which may not
scale at all, start to dominate for large enough N, eventually decreasing the preamp
bandwidth. However, for the pixel detector implementation described in this chapter,
the simulations showed the scaling argument holds for a range of pixel sizes greater
than 0.4 mm.
The detectorASIC interconnect method determines the stray input capacitance
and hence the degree of pixel size reduction for which these scaling arguments apply.
The lowest input capacitance method is that for which the detector and ASIC, with
matching pitch, are directly bonded together. In this case, the stray input capacitance that ultimately limits the scaling argument is near 100 fF, while the same
82
capacitance for other ASICdetector interconnect methods is 500 fF to 1 pf or even

higher. In practice, this means that with direct bonding the scaling argument applies
down to about 500-m pitch, while for other attachment methods stray capacitance
dominates at much larger pitches in the several millimeter range.
References

1. J. B. Stephen, Adv. Space Res., 11, 407, 1991.

2. J. E. Grindlay, New Astron. Rev., 49, 436, 2005.
3. P. H. Mao, F. A. Harrison, D. L. Windt, and F. E. Christensen, Appl. Opt., 38, 4766, 1999.
4. B. D. Ramsey, C. D. Alexander, J. A. Apple, C. M. Benson, K. L. Dietz, R. F. Elsner,
D. E. Engelhaupt, K. K. Ghosh, J. J. Kolodziejczak, S. L. ODell, C. O. Speegle, D. A.
Swartz, and M. C. Weisskopf, 568, 432, 2002.
5. J. E. Koglin, H. C. Chen, F. E. Christensen, J. Chonko, W. W. Craig, T. R. Decker, M. A.
Jimenez-Garate, C. J. Hailey, F. A. Harrison, C. P. Jensen, M. Sileo, D. L. Windt, and H.
Yu, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series,
vol. 4851, Society of Photo-Optical Instrumentation Engineers (SPIE) Conference
Series, ed. J. E. Truemper and H. D. Tananbaum, 673683, 2003.
6. F. A. Harrison, F. E. Christensen, W. Craig, C. Hailey, W. Baumgartner, C. M. H. Chen,
J. Chonko, W. R. Cook, J. Koglin, K.-K. Madsen, M. Pivavoro, S. Boggs, and D. Smith,
Exp. Astron., 20, 131, 2005.
7. Y. Eisen, Nucl. Instrum. Methods Phys. Res. A, 380, 431, 1996.
8. P. J. Sellin, A. W. Davies, S. Gkoumas, A. Lohstroh, M. E. Ozsan, J. Parkin, V. Perumal,
G. Prekas, and M. Veale, Nucl. Instrum. Methods Phys. Res. B, 266, 1300, 2008.
9. O. Tousignant, L. A. Hamel, J. F. Courville, J. R. Macri, M. Mayer, M. L. McConnell,
and J. M. Ryan, IEEE Trans. Nucl. Sci., 45, 413, 1998.
10. A. E. Bolotnikov, W. R. Cook, F. A. Harrison, A.-S. Wong, S. M. Schindler, and A. C.
Eichelberger, Nucl. Instrum. Methods Phys. Res. A, 432, 326, 1999.
11. C. M. H. Chen, S. E. Boggs, A. E. Bolotnikov, W. R. Cook, F. A. Harrison, and S. M.
Schindler, IEEE Trans. Nucl. Sci., 49, 270, 2002.
12. W. R. Cook, S. E. Boggs, A. E. Bolotnikov, J. A. Burnham, M. J. Fitzsimmons, F. A.
Harrison, B. Kecman, B. Matthews, and S. M. Schindler, IEEE Trans. Nucl. Sci., 47,
1454, 2000.
4 Hydrogenated
Amorphous Silicon
Radiation Detectors
Matthieu Despeisse
Ecole Polytechnique Fdrale de Lausanne

Neuchtel, Switzerland
Contents
4.1 Introduction..................................................................................................... 83
4.2 Hydrogenated Amorphous Silicon.................................................................. 85
4.2.1 Order in the Disorder........................................................................... 85
4.2.2 Electronic and Optical Properties....................................................... 86
4.3 A-Si:H Deposition Techniques and Associated Technologies......................... 88
4.3.1 Plasma-Enhanced Chemical Vapor Deposition and Hot-Wire
Chemical Vapor Deposition Systems.................................................. 88
4.3.2 a-Si:H Devices and Applications........................................................ 89
4.4 Hydrogenated Amorphous Silicon Detectors.................................................. 91
4.4.1 Hydrogenated Amorphous Silicon p-i-n Detectors............................. 91
4.4.2 Thin or Thick Films?........................................................................... 93
4.4.3 Signal Formation and Detection Efficiency......................................... 95
4.4.4 Potential of a-Si:H Detectors...............................................................97
4.5 Thin Film on ASIC Detectors.........................................................................99
4.5.1 Detector Segmentation........................................................................99
4.5.2 Recent Developments and Results..................................................... 101
4.6 Toward a-Si:HBased Radiation Detector Solutions..................................... 104
References............................................................................................................... 106
4.1Introduction
High-energy physics (HEP) experiments have already been important driving forces
for the development of detector technologies, and planar crystalline silicon sensors
developed for track finding and particle analysis are now mature detector technologies commercialized in many areas of science and medicine.1 Alternatives to the
state-of-the-art hybrid pixel silicon detectors are now developed to greater maturity
to improve detection performance, to profit from different technology specificities,
or to lower costs for future HEP experiments2,3 and for other detection applications. Since the late 1980s, alternative detector technologies based on amorphous
83
84
semiconductors such as hydrogenated amorphous silicon (a-Si:H) have gained a lot

of attention. Since its first application as a solar cell reported in 1976 by Carlson and
Wronski,4 a-Si:H has been widely studied, leading to a present coherent physical
comprehension of the material. The disorder associated with the amorphous phase,
however, leads to some controversial issues, so that research is still active for this
material. The specificities of amorphous silicon are first presented in Section 4.2.
In the past decades, a-Si:H has become a successful material technology and is
now widely industrialized for flat-panel displays and photovoltaics. The deposition
technique specificities of a-Si:H allow for various important technological advantages over many technologies. The opportunity to do uniform depositions over large
areas for low cost and the abilities to create field effect transistors, diodes, and other
semiconductor devices have already strongly established a-Si:Hbased devices as
successful imagers and X-ray flat-panel detectors.57 The thin-film silicon solar cell
industry, which uses a-Si:H in single-junction or multijunction cell structures, has
also gone into mass production of large-area modules, occupying an increasing market share in the already rapidly growing market of photovoltaics, thanks to the targeted low costs of manufacture of thin-film silicon solar modules. This amorphous
material can be deposited on flexible substrates, at low temperatures (150C for different novel applications and inventions.
Different research groups have built and characterized a-Si:H radiation sensors
that could profit from the deposition technique advantages. These works have demonstrated the potential of this material for single-particle detection,812 also suggesting
high radiation hardness of a-Si:H.13,14 However, different challenges are associated
with the development of any a-Si:Hbased detector technology for single-particle
detection. The sensors to be developed need to be thick in comparison to standard
a-Si:H devices, implying different technological issues. Sensors with a maximum
thickness of about 40 m can be efficiently used, so that a-Si:H sensors are thin in
comparison to standard silicon detectors (which have thicknesses above 150 m).
Fewer charges are therefore expected to be created by an ionizing radiation in an
a-Si:H detector than in a standard silicon detector, resulting in the most important
limitation and challenge of a-Si:H technologies for single-particle detection applications. The a-Si:H radiation sensor structure and the different issues for single-particle
detection are detailed in Section 4.4.
The low temperature (below 250C) of deposition of a-Si:H has led to the development of a new technology based on the direct deposition of an a-Si:H sensor on top of
a processed integrated circuit (IC). This novel technology was introduced in 199215
and is generally referred to as thin film on application-specific integrated circuit
(ASIC) (TFA),16 thin film on complementary metal oxide semiconductor (CMOS)
(TFC), or even above IC (AIC) technology. The first TFA sensors were developed as
image sensors with thin a-Si:H layers (about 1 m),15 but the technology has also initiated in the past years an increasing interest in the HEP community to combine the
works done on thicker a-Si:H diodes (~30 m) deposited on glass substrates and this
novel technology to build semimonolithic a-Si:Hbased pixel detectors. The progress in microelectronics and the possibility of direct deposition of the sensor permit
improving the sensitivity of a-Si:H detectors, and the different advancements toward
a vertically integrated a-Si:H sensor technology are presented in Section 4.5.
85
Hydrogenated Amorphous Silicon Radiation Detectors
4.2Hydrogenated Amorphous Silicon

4.2.1Order in the Disorder
The first depositions of silicon in its amorphous phase were done in the late 1960s
by Chittick,17 followed by the first n-type and p-type doping18 and the demonstration of defect passivation in pure amorphous silicon by hydrogen,19 leading to the
development of the higher quality a-Si:H. It is now accepted that while a-Si:H does
not show the periodic structure of crystalline silicon, it is not totally random. Even
though the material shows a long-range disorder, it still holds a short- or mediumrange order20,21 for an atom and its nearest neighbors (up to few nanometers).21 The
SiSi bond length and angle in a-Si:H show fluctuations due to the amorphous structure, but their average values are close to that of crystalline silicon, at respectively
2.35 and 109 28. The bond length distortions induce bond strains, which can
result in broken SiSi bonds, leaving the two silicon atoms with a coordination of
three (the coordination of an atom being defined as the number of atoms to which it
is bonded) and forming two dangling bonds. Pure amorphous silicon shows a high
dangling bond density of about 1019 cm3, which can for the most part be passivated
by hydrogen atoms, so that a-Si:H can show a dangling bond density down to about
2 1015 cm3. The a-Si:H structure is schematically illustrated in Figure4.1. Some
voids can be embedded in the structure, but device quality a-Si:H has a void fraction
of 1% with a typical diameter of 1 nm.22 The density of a-Si:H is therefore slightly
lower than that of its crystalline counterpart and is at about 2.15 g/cm3.
In amorphous silicon, the potential energy is not periodic, resulting in an uncertainty in k, the momentum of the electron, similar to its magnitude,23 so that the
electronic properties are described by the density of states distribution N(E). Valence
and conduction bands of semiconductors can be equivalently described by the
Coordination
4
3
1
Hydrogen passivated
defect
Silicon
atoms
Hydrogen
atoms
Coordination defect
dangling bond
Figure 4.1 A continuous random network model of hydrogenated amorphous silicon,

with silicon atoms threefold or fourfold coordinated, dangling bonds, and hydrogen atoms.
86
splitting of the bonding and antibonding states of the covalent bond, and the band
gap holds true for amorphous semiconductors.24 The extended states corresponding
to the fully coordinated covalent bonds are delocalized, and transitions from one
state to another are easy in these equivalent conduction and valence bands. In a-Si:H,
the bond length fluctuations lead to stretched or strained bonds, for which the associated energy states are well separated in the lattice site. Low tunneling probability
occurs between these sites so that the states become localized, with no conduction
expected to occur when excited electrons occupy them. The boundaries between the
localized and extended states are called the mobility edges (EC and EV, respectively),
defining the so-called mobility gap, which is 1.6 to 1.8 eV at 300 K. The localized
states induced by the spatial fluctuations of the bonds are in the mobility gap, in the
so-called band tails, which decay exponentially into the band gap, with the valence
band tail exhibiting more states than the conduction band tail. Moreover, atoms
showing deviations from their optimal coordination lead to coordination defects
with electronic states deep within the band. The most prominent intrinsic defect is the
dangling bond, which results from a broken SiSi strained bond. The defect density
is not fixed after deposition, since external perturbations can lead to a new equilibrium. This material metastability is an important feature, since light exposure leads
to an increased density of dangling bonds, degrading the electronic properties.25
This degradation on light exposure is called the StaeblerWronski effect (SWE) and
the induced defects are metastable, since they can be removed by thermal annealing
of the material. It is generally accepted that the recombination of free carriers taking
place at SiH bonds induces these defects,26 and the exact mechanisms and kinetics
are still subjects of research. The overall typical density of states in a-Si:H is shown
in Figure4.2.
Impurity atoms can be incorporated in the amorphous structure of a-Si:H with their
optimal coordination, so that it could prevent doping the material, since boron (B) or
phosphorus (P) are threefold coordinated. However, the fourfold doping configuration is low enough in energy to be present in reasonable concentrations. Part of the
doping B and P atoms is therefore integrated in their threefold configuration, and
part of the atoms are ionized and present as P+ (n-type doping) and B (p-type doping) in a fourfold doping configuration. The doping efficiency decreases with the
square root of the total concentration of doping atoms, so that the Fermi level cannot
be moved too close to the bands. Moreover, we have to consider ionized donors and
acceptors, thus inducing compensating defects taking up the excess charge, so that
doped layers in a-Si:H exhibit a high defect density.
4.2.2Electronic and Optical Properties

The conductivity in amorphous silicon involves the standard transport model in the
extended states of the bands, based on sharp mobility edges.27 Tunneling or hopping
conductivity in the localized states can also be observed, but except at low temperatures this contribution remains small. For temperatures above 200 K for holes and
50 K for electrons,23 carrier conduction occurs in the extended states. Carriers are
then consecutively trapped into localized states below the band edges, where they
87

Energy
Extended states
Conduction band
Mobility gap
Localized states
Defect states
H
EC
Thermalization
Multiple trapping
conduction
Band tail states

( from deviation
in bond length
and angle)
Nonradiative
recombination
Si
Radiative
recombination
Dangling bond
EV
EV
Valence band
Extended states
Density of states
Figure 4.2 Left: Density of state distribution in hydrogenated amorphous silicon. Right:
Typical thermalization, multiple-trapping conduction, and recombination of free electrons in
a-Si:H.
are immobile and then reemitted in the bands, leading to a multiple-trapping conduction.28 The release time of trapped carriers is a crucial parameter in determining the
conduction. It depends on the trap energy level, on temperature, and on the material
attempt-to-escape frequency , as defined in Tiedje and Rose.28 The localized
states located in the band tails are defined by the characteristic temperature T V and
by the exponential distribution slope of the states (kT V)1, which leads to a broad
distribution of states and thus of carrier release times. However, for temperatures
above T V , which ranges from 220 to 270 K for the conduction band tail (for electrons)
and from 400 to 450 K for the valence band tail (for holes), the transport is considered nondispersive as the multiple-trapping processes take place mainly in shallow
states. Thermal detrapping is a fast process for these states so that they can be well
described by a single trapping time. The mobility of carriers for such temperatures
(this is the case for electrons at room temperature) can therefore be defined by the
mobility 0 in the extended states and by a single release time constant, which has an
impact on mobility that is equivalent to an apparent reduction of mobility from 0 to
the constant drift mobility drift. Electron drift mobility is considered to be at about
15 cm2 V1 s1 at room temperature29 for a 0 of about 10 cm2 V1 s1 and therefore is
much lower than in crystalline silicon (~1,350 cm2 V1 s1).
For temperatures below T V , the transport of the carriers is dispersive because of
the broad distribution of release times from the different traps. The mobility of a
packet of the carriers therefore decreases with time as the mean carrier distribution
88
falls deeper into localized states with time. The average drift mobility is usually
defined in that case as power law time dependent:30

( )
d (t ) = (1 ) 0 t
( 1)
where is defined as the dispersion parameter and is equal to the ratio T/T V (0 <
< 1). The transport of holes is dispersive at room temperature, and ~ 0.65. Its
mobility can be considered to be about 0.01 cm2 V1 s1 and therefore is considerably
lower than in crystalline silicon (~480 cm2 V1 s1).
The lifetime of the charge carriers corresponds in a-Si:H to the average time
needed for the carrier to be recombined in a deep localized state. The two dominant
recombination mechanisms are radiative transitions between band tails, which dominate at low temperatures (<100 K), and nonradiative transitions from band edges
to defect states, which govern recombination at higher temperatures. The carrier
lifetime for both carriers is usually found to be about 10 6 s in intrinsic a-Si:H, in
comparison to 10 5 to 10 3 s in crystalline silicon.
Doping of a-Si:H layers and conduction in the amorphous structure are therefore
possible, but with much lower carrier mobility and lifetime (i.e., poorer electronic
properties than in crystalline silicon). However, a-Si:H exhibits an absorption of
visible light an order of magnitude higher than for crystalline silicon (c-Si), since
the optical transitions are considered semidirect in a-Si:H. Consequently, the typical thickness (<1 m) of an a-Si:H light sensor is only a fraction of that of a c-Si
cell, partly compensating the poorer electronic properties while developing imaging
sensors.
4.3A-Si:H Deposition Techniques

and Associated Technologies
4.3.1Plasma-Enhanced Chemical Vapor Deposition and
Hot-Wire Chemical Vapor Deposition Systems
Hydrogenated amorphous silicon films are prepared by chemical vapor deposition
(CVD) through the decomposition of gases containing silicon. The dominant deposition techniques are plasma-enhanced CVD (PECVD), which consists of silanebased glow discharges induced by radio-frequency (RF) voltages. The standard RF
frequency is 13.56 MHz, the frequency allotted for industrial processes by international authorities. Depositions at other frequencies have also been explored extensively, from direct current (DC) to very high frequencies (VHF) of 20150 MHz or
microwave frequency (W) of 2.45 GHz. Depositions by VHF-PECVD31 and by an
alternative hot-wire (HW) catalytic CVD32 technique proved to have a high potential
for application for high-throughput a-Si:H device manufacturing.
In a typical RF PECVD deposition system, as shown in Figure4.3, mixtures
of silane (SiH4) and hydrogen (H2) gases flow into the deposition chamber and
are evacuated by a pump, together with additional gases such as trimethylboron
89
Gas flux (shower)

RF electrode
Gas flux
(side)
Gas phase reaction
+
+
Pumping
Heated substrate
Ground electrode
Vacuum chamber
Figure 4.3 Schematic representation of a PECVD deposition chamber; gases flux from
the side or from the electrode shower head.
(TMB) or phosphine (PH3) to allow doping of the layers. The gases flow between
two electrode plates on which RF voltages are applied. A plasma can therefore be
initiated and maintained, decomposing the gases and generating radicals and ions
in the chamber. Various substrates can be mounted on one or both of the electrodes, and thin films grow on the substrate as the radicals diffuse into them. The
substrates have to be heated to achieve optimum film quality because of thermally
activated surface diffusion of atoms on the growing film. The film growth in a
PECVD process therefore consists of source gas diffusion, electron impact dissociation, gas-phase chemical reaction, radical diffusion, and deposition. These processes can be done at low temperatures of the substrate, between 150 and 350C.
Temperatures below 150C exacerbate silicon polyhydride powder formation,
which contaminates the growing silicon films, and for temperatures higher than
350C the quality of the material degrades due to a loss of hydrogen passivation
of dangling bonds.
Hot-wire chemical vapor deposition (HWCVD), as developed in Mahan et al.,32
can also be used to produce thin films of a-Si:H at high rate. The deposition system
involved is comparable to that of RF PECVD except that the RF electrode is replaced
with a heated metal filament, foil, or grid. The gas mixture can be catalytically
excited or decomposed into radicals or ions by heating the metal to a high temperature (around 1,8002,000C). The silicon radicals then diffuse inside the chamber
and deposit onto a substrate placed a few centimeters away and heated to temperatures of 150450C.
4.3.2a-Si:H Devices and Applications

The first practical a-Si:H device was the 2.4% efficient solar cell reported by Carlson
and Wronski4 in 1976. Since then, interest in a-Si:H has been growing rapidly, and
it is now widely used for different applications. This amorphous material has the
high technological advantage of the ability to be deposited in a cost-effective way
uniformly over large areas. The deposited thin films of silicon moreover have
90
Figure 4.4 Thin-film silicon solar modules over large areas. Left: Flexible a-Si:H solar
modules from Flexcell.35 Right: Large-area modules on glass from Oerlikon Solar.33
semiconducting properties and high optical absorption and can be doped in p-type
and n-type, allowing for the deposition of diodes, field effect transistors, and other
semiconductor devices over large areas. Amorphous silicon has become the material of choice for the active layer in thin-film transistors (TFTs), which are most
widely used in large-area electronics applications, mainly for liquid crystal displays
(LCDs). Also, a-Si:Hbased photovoltaic devices have evolved from their widely
commercialized applications as small-area solar cells for pocket calculators to massproduced large-area solar modules (1.4- to 5.7-m2 modules.)3335 Some examples of
a-Si:H modules under production are shown in Figure 4.4. Because of the shortage of silicon feedstocks, the prospects for significant cost decrease made possible
by thin-film silicon technologies, and the recent availability of complete turnkey
production lines provided by established manufacturers, large-area solar modules
integrating a-Si:H seem likely to play an important role in the future of solar electricity generation.36
Defect creation is associated with doping of the layers in amorphous silicon, so
that free carriers generated in p-doped or n-doped layers cannot diffuse over long
distances. Therefore, a-Si:H photovoltaic devices are all based on an intrinsic layer
absorber (150 to 500 nm) between thin doped layers (~10 nm), which provide the
internal electric field of the p = doped, i = intrisinc, d = doped (pin) structure, so
that generated carriers in the i-layer drift and induce a signal on the device electrodes. The top and bottom contacts of the p-i-n structures are principally realized in
the form of transparent conductive oxides (TCOs) for photon detection applications.
Because of the high defect density of the doped layers, an a-Si:H detecting device is
therefore always based on a p-i-n structure. The diode structure can then be used in
forward biases for photovoltaic applications, with the reverse current and the forward
voltage generated by sun illumination depending on the cell resistive load. For detection applications, the p-i-n device can then be reverse biased to allow depletion of
the intrinsic layer and drift and collection of free carriers generated by an incoming
ionizing radiation.
The potential opportunities to deposit both p-i-n diodes and thin-film transistors made of a-Si:H on large areas and industrially have strongly motivated the
91
development and the commercialization of flat-panel X-ray detectors to assess the

need for entirely electronic X-ray imaging equipment. These imagers integrate scintillators in contact with an a-Si:H active matrix array. X-rays are converted into
visible light by means of the scintillators, and the visible light is detected by a pixel
made of a p-i-n a-Si:H photodiode addressed by an a-Si:H TFT.1214
Hydrogenated amorphous silicon is therefore now widely used and commercialized,
but novel technologies and novel applications of the material are also being developed
and will certainly appear in the near future. For example, the deposition on flexible
substrate will certainly lead to novel applications, and the possibilities for radiation
detection with thin layers of silicon on large areas also show attractive potential.
4.4Hydrogenated Amorphous Silicon Detectors

4.4.1Hydrogenated Amorphous Silicon p-i-n Detectors
The different specificities offered by the deposition techniques of a-Si:H have motivated research groups to develop a-Si:Hbased radiation detectors and to study
radiation interactions and signal formations in such sensors. These studies were further motivated by first demonstrations of a potential higher radiation hardness of
this amorphous material in comparison to its crystalline counterpart, since thermal
annealing of defects could be observed after proton, neutron, photon, or electron
irradiations.10,11,25 The sensitivities of p-i-n a-Si:H devices to alpha particles were
demonstrated,5 and since then a few research groups59 have carried out substantial
work in trying to demonstrate that a-Si:H can be used for reliable sensors in HEP.
The a-Si:H detecting devices are mostly developed as p-i-n diodes, with the i-layer
as the sensitive volume. The doped layers permit the establishment of an electric field
profile in the i-layer and have to be kept thin (<20 nm), since their high defect density
prevents free carriers generated in these layers from contributing to signal creation.
The substrate of the detector can be glass, flexible substrate such as polymer, or even
metallic foils. The construction of the detector consists first of deposition on the
substrate of a conductive metal layer that will act as the detectors bottom contact.
Segmentation of the detector into different pixels, strips, or large structures can be
done at this stage, for example, via laser scribing, as it is done industrially on largearea a-Si:H solar modules, or by different masks during the bottom contact development. Consecutive depositions of the different a-Si:H layers are then done. A further
deposition of top contact is then carried out, which can also be patterned in a similar
way as for the bottom contact to segment the constructed detector. An example of
the structure of an a-Si:H detector deposited on a coated glass substrate is shown in
Figure4.5.
Under reverse bias across the p-i-n a-Si:H sensor, thermal ionization of the deep
defects occurs, leaving behind a positive space charge, so that part of the deep defects
behaves like donor states. The i-layer can therefore be considered as slightly n-doped
under reverse bias. In a first approximation, the depleted i-layer space charge can
be considered determined only by the density of ionized dangling bonds, so that the
electric field profile in the i-layer can be retrieved from the Poisson equation:37
92

Metal top contact - detector
segmentation
p-doped
Active
i-layer
n-doped
+
+
+
+
+
+
+Vbias
Laser scribing patterning
Jleak
1.06 1016
F/m2
d(in m)
Metal contact
Substrate
Figure 4.5 Schematic representation (left) and picture (right) of a typical p-i-n structure
of an a-Si:H sensing device deposited on a metal-coated glass substrate.
d 2V
qN db*
=
=
0a-Si
dx 2 0
where V is the electric potential; is the electric charge density; 0 and aSi are the
dielectric constants of vacuum and of a-Si:H, respectively; q is the electron charge;
and Ndb* is the density of ionized defects. This density of ionized dangling bonds
defines the electric field and at the same time varies itself with the electric field. In
high-potential regions, Ndb* can however be considered constant, so that the voltage
required to deplete an i-layer of thickness dtot can be estimated by considering a uniform density of ionized defects:

Vd =
qN db*d 2
20a-Si
For a typical dangling bond density of 2 1015 cm3, and if roughly 30% of the
dangling bonds are ionized in high-potential regions,37,38 the depletion voltage Vd can
be estimated as Vd ~ 0.45 d (in m). Owing to the positive space charge in the i-layer,
the electric field extends predominantly from the pi interface, where it is maximum,
into the i-layer. It then decreases linearly with the distance x in the i-layer from the pi
interface until it becomes very low at the end of the depleted region (for an underdepleted detector) or until it reaches the n-layer for an overdepleted sensor. The density of
ionized defects actually varies linearly with the electric field in low-potential regions,
as is the case in nondepleted regions. It results in an exponentially dropping off electric
field in these regions39,40 and into a low-amplitude electric field profile in the nondepleted region of the sensor. The depletion region in a-Si:H can therefore rather be seen
as the sensor volume in which high electric fields are induced by the reverse voltage.
Free carriers are generated in the depleted i-layer of the p-i-n detecting device by an
interacting radiation drift under this internal electric field and induce a corresponding
current signal on the electrodes. The basic principle of operation of an a-Si:H p-i-n
93

Negative
voltage
High energy particle
TCO top contact Electrode

p-layer
Depleted i-layer
E(V/cm)
+
+
Nondepleted i-layer
(slightly n-doped)
n-layer
Field
distribution
Position
Positive
voltage
Figure 4.6 Schematic representation of the interaction of a high-energy particle in a p-i-n

a-Si:H detector operated under reverse bias.
radiation sensor is shown in Figure4.6, with electrons drifting toward the n-layer or
toward the nondepleted low-field regions for the underdepleted sensor, where they
recombine, and with holes drifting toward the p-layer, where they recombine.
4.4.2Thin or Thick Films?

Charged particles create free carriers all along their track into the sensor through
direct Coulomb interactions, and the particle collision stopping power dE/dx in the
sensor depends on the particle energy and type and on the interacting material (as
defined in the BetheBloch formula). The interacting particle loses energy along its
track, so that the deposited energy varies in the depth of interaction, leading to a distribution of energy loss in the sensor. The mean electronhole pair creation energy
W then permits estimation of the average density of the free carriers as generated in
the sensor and its profile. In the case of electromagnetic radiations, the interactions
are localized, and the probability that a photon interacts after travelling a distance d
is an exponentially decaying function of d. The location of interaction in the sensor
of a single photon is therefore statistic. The interaction leads to the creation of an
electron for an incoming photon with an energy higher than the material bandgap.
This secondary electron can have a release kinetic energy of up to the interacting
photon energy minus the bandgap and can then induce further ionizations in the sensor through Coulomb interactions. The number of pairs created is then here again
defined by the secondary electron energy loss and by W.
The mean pair creation energy W can be estimated from the material bandgap,
and following calculations of Klein,41 a material with a bandgap of 1.8 eV (mobility bandgap of a-Si:H) should exhibit W = 6 eV. However, the notion of bandgap in
a-Si:H is not precisely defined because of the broad distribution of states, so that gaps
down to about 1.5 eV can be considered, for which we find W ~ 4.7 eV. The mean pair
94
creation energy in a-Si:H was studied by a few groups, and results confirmed this
possible range of values for W, as it was measured from 4.8 to 6 eV,6,9,42 in comparison to about 3.6 eV for crystalline silicon.
The minimum charge that can be created by a particle going all through the sensor is generated by minimum ionizing particles (MIPs). The energy loss distribution
of a MIP varies very little in the sensor layer because of its high initial energy, and
it is typically about 0.36 keV/m in a-Si:H. The sensor active i-layer has therefore to
be designed thick in comparison to a-Si:H typical submicron devices, since less than
6075 electronhole pairs will be generated on average per micron for MIP detection. The amplitude of the induced signal will depend on the depleted thickness of
the detecting device, so that a-Si:H layers of at least tens of microns are necessary
to obtain a sufficient signal. A first challenge in the development of a-Si:Hbased
direct radiation sensors is therefore the high technological challenge to grow the
thickest diode for which the full depletion condition can be achieved while keeping
the leakage current with acceptable bounds. Such development involves many novel
issues unique to thick a-Si:H sensors, such as high-field mechanisms negligible in
commercial a-Si:H products, requiring dedicated developments and optimizations
of the a-Si:H different layers. Progress has been made in developing thick a-Si:H
layers up to 50100 m, and the technological limit currently resides in the difficulty of fully depleting such thick sensors while keeping low leakage currents.
The density of ionized defects is found to be about 6 1014 cm3, so that while a
sensor of 30 m still requires about 400 V for depletion, sensors of 50 or 100 m
would require voltages of more than 1.1 and 4.4 kV, respectively, for full depletion.
Moreover, a strong increase of the thermally generated current in a-Si:H sensors has
been demonstrated at high electric fields.43 The dominant leakage current source in
good quality a-Si:H standard p-i-n devices can be considered as the thermal generation of trapped carriers Jth, which was shown to depend on the defect density Ndb and
on the material bandgap:44

J th = q d N db
1
exp ( EC EFD ) /kT
gen
where d is the intrinsic layer thickness, gen is the time constant governing the thermal generation, and EC EFD is the energy between the conduction band and the
traps involved in the generation. Low values of Jth in the range of 1 to 10 pA/cm2
were demonstrated for 1-m thick diodes.38 However, in the presence of an electric
field, the effective activation energy of the ionized defects is decreased by the energy
qE1/2, where E is the electric field and is the PooleFrenkel constant with a theoretical value of about 2.25 10 4 eV V1 cm1/2. The high reverse bias thermally
generated leakage current can then be written as

( (
I I 0 exp EC EFD E
) kT )
95
q
0aSi
The field-enhanced thermal generation will moreover be stronger at the pi interface, where the electric field is maximum, and because this region usually exhibits a
higher defect density. The field-enhanced thermal generation of trapped carriers in
the vicinity of the pi interface was therefore shown to be the dominating process at
room temperature, and reductions of activation energies from 0.9 eV at low biases to
0.4 eV for electric fields in the range of 105 V/cm were demonstrated.9 Typical leakage currents on the order of 10 8 to 10 6 A/cm2 were measured in 20- to 32-m thick
a-Si:H sensors,9,38 therefore about four orders of magnitude higher than for one order
of magnitude thinner cells. It is therefore clear that the thicker the a-Si:H detector
is, the higher the required reverse bias will be and the higher the electric field in the
i-layer will be exponentially increasing the device leakage current.
The maximum thickness of the developed sensors is consequently determined
now by the typical minimum density of ionized dangling bonds in a-Si:H, so that sensors with thicknesses of up to 3040 m only seem reasonable when considering the
required depletion voltage and the corresponding electric field values in the device.
Sensors with a thickness of about 40 m are thick in comparison to standard
submicron a-Si:H devices, but they are thin when considering typical semiconductor
radiation detectors, since crystalline silicon sensors usually have a thickness higher
than 150 m. In light-imaging devices, the thin absorber of the a-Si:H devices is
partly compensated by a higher optical absorption, but this is not the case for particle
detection, since the mean pair creation energy W is higher in a-Si:H than in c-Si. The
a-Si:H radiation sensors, thin absorbers, and higher W of the a-Si:H radiation sensors
therefore lead to the most important challenge and limitation in the development of
such detectors for high-energy physics applications, since fewer charges will be created by a particle going through an a-Si:H sensor than through a crystalline silicon
detector.
4.4.3Signal Formation and Detection Efficiency

Charge carriers are generated in the active i-layer of an a-Si:H sensor by an incident
radiation drift under the internal electric field E at a velocity defined as

vdrift = E
where is the carrier mobility. This relation is positive for holes and negative for
electrons and is valid until carriers reach saturation velocity (about 107 cm/s in silicon). The free carriers drift directly after their creation in the sensor, and their
motion induces a current on the detector electrodes. The induction of current for moving charges was first introduced in Ramo45 for free carriers in a vacuum, and the
concept was confirmed for semiconductor detectors46 and for detectors with resistive
96
elements such as nonfully depleted a-Si:H sensors.47 The current induced by qN

moving charges in a p-i-n a-Si:H detector can therefore be determined by the total
number of charges, by their velocity, and by the so-called weighting field EW , which
is a measure of the electrostatic coupling between the moving charges and the electrodes of the detector:45

I = q N vdrift Ew
For a p-i-n a-Si:H sensor, electrons and holes, respectively, drift toward the n-layer
and p-layer under reverse bias, and a corresponding positive current is induced on
the electrode connected to the p-layer, while a negative current is induced on the
electrode connected to the n-layer.
In a parallel plate configuration with infinite electrodes separated by a distance
dtot, the weighting field is equal to 1/dtot and is perpendicular to the electrodes. In
the case of devices with many electrodes with finite sizes, the weighting field is
more complicated and extends onto the electrode sides. It is determined by applying a unit potential on the sensing electrode of interest and zero potential on all
other electrodes, with all fixed charges neglected. The resulting map of the weighting field is consequently different from that of the electric field. In a-Si:H detectors,
the nondepleted region of the sensor corresponds to a region with low conductivity
(<10 10 S/cm), while in crystalline silicon the nondepleted region presents a higher
conductivity (~10 4 S/cm). This particularity of a-Si:H detectors leads first to a constant capacitance of the sensor that depends only on the i-layer thickness and does
not change with the applied voltage, in opposition to the capacitance in crystalline sensors, which is directly dependent on the depleted thickness of the sensor.
This leads to difficulties in determining the depletion condition of an a-Si:H sensor
through standard capacitance versus voltage (C-V) measurements. However, alternative techniques based on the signal induction theory were proposed.9
The effects of the high resistivity of the nondepleted layer in an a-Si:H sensor are
also observed when considering the weighting fields of the sensors. In the simple
case of parallel plates, for a sensor with a thickness dtot and a depleted thickness d,
the weighting field is defined as 1/d for a crystalline silicon sensor, while it will be
reduced in the case of an amorphous silicon sensor to a constant 1/dtot. This particularity has a direct impact on the detector efficiency. If N charges are generated in the
depleted region of the sensor and if we consider the perfect case of charges drifting all the way respectively to the p-layer and to the nondepleted zone, the overall
charge Qtot induced on the detector electrode can be calculated, and Qtot = qNdEW .
It is therefore clear that, in the case of a crystalline silicon sensor, the full charge is
retrieved on the detector electrode, while in the case of an a-Si:H sensor, only the
fraction defined by d/dtot of the total charge will be induced. For an a-Si:H radiation
detector, the nonfull depletion of the sensor has thus a double detrimental impact on
the performance, since it not only reduces the active volume of the sensor but also
reduces by d/dtot the induced currents and total induced charge.
The mobility of the free carriers in the sensor active volume will also determine the
induced current. The transport of carriers in amorphous silicon is nondispersive for
97
electrons at room temperature, and a constant release time from the shallow traps
can be considered, so that a drift mobility drift-e of 1 to 5 cm2 V1 s1 can be expected.
For holes, the transport is dispersive at room temperature and more complicated to
determine, but it is agreed that the mobility is on the order of 0.01 cm2 V1 s1. In
the case of thick a-Si:H devices and high electric fields, enhanced drift mobilities
were measured,48 demonstrating that the reduction of the activation energy of the
traps through PooleFrenkel mechanisms also has an impact on the mobility. This
field-assisted release from the traps will be important for a dispersive transport but
negligible for a nondispersive transport. At room temperature, the transport of holes
implies trapping in centers with a wide energy range, and the release energies will
be varied by the local electric field. The transport of holes is therefore much more
complicated to model, but the field effect can still be taken into account by using the
mobility with an attempt-to-escape frequency defined as48

= 0 exp E /E0
where E is the electric field and E0 is a constant parameter (E 0 ~ 5.7 104 V/cm).
Even though charge carriers exhibit three to four orders of magnitude lower
mobility in amorphous silicon than in crystalline silicon, the high reverse biases
and the high internal electric fields induced in the active sensor volume, together
with the reduced detector thickness, still permit operation of an a-Si:H detector
with reasonable carrier drift time. Typical electric fields in an a-Si:H sensor will
be on the order of 105 V/cm, so that typical drift times in 20-m thick sensors are
expected to be 20 ns for electrons and 1 s for holes. At such fields, the mean drift
length is in the millimeter range for electrons and down to a few tens of microns
for holes.
The timing properties of the induced signal in an a-Si:H radiation detector were
confirmed by specific measurements carried out in Despeisse et al.9 A laser pulse
was used to photogenerate a charge packet close to the pi interface of 13- to 32-m
thick a-Si:H detectors, so that the induced signal was constructed principally by the
electron motion in the whole depleted thickness. A typical full time of drift of electrons in thick a-Si:H sensors was demonstrated below 18 ns, only slightly varying
with the sensor thickness and voltage, since longer drift lengths correspond to higher
electric field and higher carrier velocity, resulting in a globally constant drift time.9
It was evaluated that for a readout time of 20 ns, all the charge induced by motion of
the generated electrons can be integrated, and that about only 515% of the charge
induced by the generated holes can be integrated.
4.4.4Potential of a-Si:H Detectors

For applications requiring an event measurement with high counting rate, as in HEP
experiments for which the separate bunch crossing time can be 25 ns (Large Hadron
Collider, LHC) down to 12 ns (Super-LHC), the signal will be principally determined by the free electrons created by the interacting particle in an a-Si:H sensor.
Applications with a longer time of integration will have an increasing integrated
98

TCO electrode
PE-CVD
deposition
of n-i-p
a-Si: H layers
Metal rear
electrodes
Processed
ASIC
Figure 4.7 Schematic representation of the consecutive depositions of n-doped, intrinsic,

p-doped a-Si:H films and of a metallic top contact directly above an integrated circuit to build
a TFA monolithic pixel detector.
charge originating from the hole transport, but not complete because of the few tens
of microns average drift length of holes before recombination. The readout of the
induced signal on the electrode connected to the n-doped layer will be advantageous
as electrons drift toward this electrode, and very small pixels with a typical size
below the sensor thickness could permit focusing the weighting field at the electrode
vicinity, thus leading to most of the signal created by the drift of the electrons (small
pixel effect).
For single-particle detection, most of the low charge induced by the detection
of an ionizing particle will consequently be created in less than 20 ns. The low
induced current needs to be amplified to be processed by further digital electronics circuitry, and the detection efficiency will be determined by both the signal
amplitude and the noise level of the detection system. An optimum signal from the
a-Si:H detector is expected for fully depleted 40-m thick detectors, and signals of
up to 1,400 electrons are expected in such sensors for a MIP 9 when read out with
20-ns shaping time. The electronic has therefore to be designed with an equivalent
noise charge below about 150 electrons rms. The a-Si:H detector main limitation for
use in HEP is this low induced charge, and the possible deposition of the detecting
device directly above the electronic readout has in the past permitted reduction of
parasitic capacitances and lowering the electronic readout noise. This thin film on
ASIC technology is discussed in Section 4.5, as it refocused efforts on the development of a-Si:H as a low ionizing radiation sensor. A schematic of its structure is
shown in Figure4.7.
The a-Si:H detector technology offers other unique potentials for different applications. Even though MIP detection is difficult to achieve, high ionizing radiation
can be well detected, and large-area detectors can be devised to be built on square
meters on glass substrate, metallic foil, or plastic substrates. The radiation hardness
of a-Si:H was studied in the context of testing solar cells for space applications, and
it was shown that proton, neutron, and electron irradiations11,12 or photon irradiation
99
(involving the StaeblerWronski effect)25 all led to the creation of metastable deep
defects that can be thermally annealed out. Full recovery of proton-induced defects
was observed after subsequent room temperature thermal annealing,11 while effective annealing of defects created by room temperature light irradiation was observed
for annealing temperature above 120C. The annealing temperature and full recovery time were shown to depend on the irradiation conditions (temperature, type of
radiation), and systematic studies still need to be done on this material to define its
potential higher radiation hardness compared to crystalline silicon. However, measurements have already been done with 20-m thick a-Si:H diodes in severe beam
conditions for the direct readout of a 24-GeV/c proton beam,49 demonstrating the
possibility to read out proton spill-induced currents and a factor 4 reduction of signal
after a fluence of 1.6 1016 protons/cm2 accumulated in a few days (without time
for thermal annealing). This signal was again complete after subsequent thermal
annealing of the sensors. This shows the high potential of a-Si:H detectors deposited
on different substrates and in very thin or thicker layers for beam monitoring and for
beam tracking. Different a-Si:H detector structures could therefore show interesting
potential applications not only for HEP but also for medicine at therapy accelerator
centers and for any other accelerator applications.
4.5Thin Film on ASIC Detectors

4.5.1Detector Segmentation
Hydrogenated amorphous silicon can be deposited through PECVD at temperatures
ranging from 150 to 250C, while the maximum postprocessing temperature for
crystalline silicon CMOS-processed wafers with aluminum-based interconnections
is generally considered to be slightly above 450C, the temperature at which the
aluminum connections start to melt. It is thus possible to deposit an a-Si:H detecting
device directly on top of a processed integrated circuit and to build an a-Si:Hbased
intelligent sensor in which the signal processing is integrated with the detecting
device. This TFA technology was developed for the first time as image sensors.15,16 It
has also initiated an increasing interest in the HEP community to combine the works
done on thicker (~30 m) a-Si:H diodes deposited on glass substrates and this novel
technology to build semimonolithic a-Si:H-based pixel detectors.9 The a-Si:H sensor is built on top of the ASIC by consecutive depositions of n-doped, intrinsic, and
p-doped a-Si:H films. The doped layers are very thin (~30 nm) because of their high
defect density, and the n-i-p diode thickness is equivalent to the i-layer thickness. The
a-Si:H sensor bottom contacts (anodes) are defined by metal structures integrated in
the electronic circuit and designed in the topmost metal layer of the technology, as
shown in Figure4.8. These metal pads are directly connected to the pixel electronic
readout and define the detector segmentation. They are interconnected through the
sensor n-layer, and to avoid further patterning of this layer, it is specially designed
with low conductivity to provide pixel isolation higher than 10 M. A metallic top
contact is finally deposited on the sensor p-layer, defining the detectors common
top electrode (cathode). A reverse-bias voltage is applied on the common cathode,
with the anode potential defined by the electronic circuitry input voltage. Carriers
100

Ionizing radiation
Detector common top electrode

p-doped a-Si:H
i-layer a-Si:H
n-doped a-Si:H
+
+
+
Detector pixel
+
electrode
+
ASIC 3 metal layers
ASIC
passivation
layers
n-well
Crystalline silicon ASIC p-substrate
Figure 4.8 Schematic cross section of a TFA detector at a pixel site. The n-i-p a-Si:H
detector is deposited on top of an ASIC with three metal layers, with passivation layers inside
the pixel metal pad defined in the topmost metal level of the technology.
generated by photons or particles in the sensor-depleted region drift under the electric field, and a corresponding current signal is induced on the pixel metal pad integrated in the circuit and located close to the particle track or photon interaction. The
induced current is then directly processed by the integrated electronic readout built
underneath the pixel-sensing area.
The TFA technology therefore offers a high integration level of the detecting
device and of the readout electronics. It has the advantages of monolithic sensors to
decrease the costs of construction in comparison to hybrid pixel detectors, which use
bump bonding connections from the detector array to the electronic readout. The
deposition process is low cost, with the price of a TFA detector slightly more than
the ASIC price itself. Together with this monolithic advantage, the TFA technology
presents the same flexibility as hybrid pixel detectors: The detecting device and the
electronics can be separately designed and optimized, and the detector can be biased
independently from the ASIC and at high voltages. The integration is also vertical, so
that the electronic and the sensing devices do not compete in the pixel area, allowing
for a geometrical fill factor (ratio between the active area of the sensor and the total
sensor area) close to 100%. These properties of pixel detector construction render
this technology attractive, but several important challenges have to be overcome for
its development.
The first limitations are the technological challenges to deposit on top of an integrated circuit thick sensors with leakage currents in acceptable bounds. As explained
in Section 4.4, a MIP creates a peak charge of about 700 electrons in a 20-m thick
a-Si:H detector,9,50 which corresponds to about 35 e m, with this low charge originating from a mean electronhole pair energy creation in a-Si:H of up to 6 eV and
101
from a partial collection of holes due to its dispersive transport in a-Si:H. These
characteristics lead to different challenges for the design of the integrated circuit,
since low-noise preamplifiers able to read out the small signal induced by a single
particle in the thin a-Si:H detecting device have to be designed.
4.5.2Recent Developments and Results

The feasibility of the TFA concept for particle detection was first demonstrated
by depositing a-Si:H sensors with thicknesses up to 32 m on a CERN (European
Organization for Nuclear Research) available high-speed (5 ns peaking time) and lownoise (450 electrons rms of equivalent noise charge) electronic circuit.9 Depositions
were done without affecting the integrated electronic readout, and the first efficient
detections of low-energy electrons (from about 15 to 50 keV) were demonstrated,
with a signal of a few thousand electrons created in a 20-m thick sensor. The TFA
detectors developed on this circuit were also used to study the timing properties of
particle detection in a TFA detector thanks to the low peaking time of the electronic
readout. These studies demonstrated that even though electrons have low mobility,
the high electric fields (104 105 V/cm) permit achieving a full collection of generated
electrons in less than 18 ns.9 The dispersive and slow hole transport leads to a partial signal from holes strongly depending on the applied reverse voltage and on the
electronic readout shaping time. For a shaping time of 25 ns, the signal from holes
represents only about 515% of the total signal expected from holes.
A first generation of electronic circuits for TFA detectors was then designed at
CERN, integrating preamplifiers exhibiting from 70 electrons rms noise for a 25-ns
peaking time to 35 electrons rms noise for a 150-ns peaking time. TFA detectors
built on these first-generation circuits are shown in Figure4.9 for sensor thicknesses
up to about 30 m. These circuits permitted great advances in the understanding of
the technology limits, and specificities and are shown in Figure4.9.
a-Si:H test output metal pad ring
a-Si:H sensor
Bottom electrodes
integrated in a-Si:H test
Figure 4.9 First-generation TFA detectors. Left: 16-mm2 a-Si:H test integrated circuit
with 20-m thick a-Si:H sensor deposited on top of different metal structures integrated in
the center of the ASIC, allowing full characterization of the technology.9 Right: TFA pixel
detector prototype.
102

Cross section image inside a pixel of the TFA detector
Pixels rows
a-Si:H
6.5 m
a-Si:H
6.5 m
a-Si:H sensor
~55
ASIC pads
Metal pad
Acc.V Spot Magn Det WD
Polyimide
5.5 m
8 m
10 m
Figure 4.10 Left: First TFA detectors developed for particle detection. Right: Scanning
electron microscopic image of the TFA detector cross section at the site of a pixel.
The direct deposition of the a-Si:H sensor on the ASIC was first shown to lead
to different material and electronic problems in the deposited sensor. In standard
CMOS technologies, integrated circuits are protected with 4- to 6-m thick passivation layers that cover its surface. Openings in these layers are made to contact metal
pads and are made inside a metal pad, as shown in Figure4.9. The surface of the
circuit is consequently uneven and exhibits 4- to 6-m falling edges at each metal
pad (i.e., at each of the electrodes of the future TFA detector). The deposition of the
a-Si:H layers on this uneven surface leads to an a-Si:H sensor morphology that follows the ASIC surface, as shown from a scanning electronic microscopic image in
Figure4.10 for a 6-m thick sensor. The growth of the a-Si:H films was shown to
lead to cracks in the material at the pixel edges and corners, as demonstrated for the
first time in Despeisse et al.51
This topological effect on the sensor leads to very high electric fields in these
regions, creating very high additional field-enhanced leakage currents at the pixel
periphery and corners. The first developments of TFA detectors demonstrated an
increase of leakage current of about two orders of magnitude for a diode deposited
on an ASIC than for a similar diode deposited on glass. Further tests were conducted on one of the first-generation circuits developed for the TFA technology,
which was integrating pads with openings in the passivation layers all around the
metal pads, at respectively 6 and 20 m from the pad sides.9 These tests demonstrated a respective reduction by a factor 3 and 10 of the current density compared to
structures with an opening inside the metal pad. This clearly demonstrated again the
high leakage currents created at the falling edges of the a-Si:H sensor. When these
edges are moved away from the pixel metal pad (even a few micrometers), the additional leakage currents are reduced and can be cancelled. These first studies therefore permitted identification of a strong dependence of the a-Si:H sensing element
leakage current on the pad definition at the ASIC level, so that the reduction of
the dark current has to be taken into account during the design phase of the integrated circuit.
For second-generation circuits for TFA technology, a flat substrate is thus required
for a fully optimized sensor. A first method can consist of the design of metal pads at
the ASIC level with minimal separation (for minimal dead area) and with openings
103
in the passivation layer made all around the full active area. A second technique
consists of additional process steps for the ASIC. Metal pads can be designed with
standard openings, and the detector electrodes can then be constructed by first filling with metal the different openings to obtain a rather flat surface. The second step
could consist in extending and patterning metal pads on top of this flat surface, with
each designed electrode connected to its electronic readout through the passivation
openings filled by metal. This second method though requiring a further process
step, would lead to lower parasitic capacitances than the first method, allowing the
design of lower noise electronic readout.
These additional leakage current sources were strongly limiting the performance
of the first-generation TFA radiation detectors, since as full depletion of sensors with
a maximum thickness of 20 m only could be achieved. Thicker sensors could be
deposited, but their partial depletion has the double detrimental effect on the signal
creation efficiency to limit the active volume and to reduce the total charge induced.
The identification and reduction of these additional leakage current sources was
crucial for TFA detector development. Moreover, overdepletion would also permit
improved detector efficiency, with higher electric fields increasing the signal created
by holes for fast shaping time applications.
Detections of 1550 keV electrons were demonstrated on 6.6-m wide strips
separated by a minimum distance of 4 m.9 These detections permitted estimation
of a mean pair creation energy W of 5.4 0.8 eV, in concordance with the 4.66 eV
expected range. Moreover, the thin strip structure permitted demonstration of very
low charge sharing in the TFA detector. Even through detecting low-energy electrons, which have a tortuous path in matter, very low lateral induction of signal was
observed with strip structures separated by 4 m, since electrons having a track on
top of a strip were not inducing any significant signal on the neighboring strips. This
can be explained by the fact that the low diffusion coefficient of about 10 2 to 10 3
cm2/s for electrons and about 10 4 cm2/s for holes in a-Si:H23 are negligible because
of the high electric field and the short drift times of less than 20 ns for electrons.
Lateral diffusion of an electron packet is small, and it is less than 1 m rms for a
charge packet created at the pi interface in a 20-m thick sensor, so that even on the
thin-strip structures, low charge sharing is seen. This represents a very interesting
potential for the design of TFA detectors with high spatial resolution.
X-ray detection was also demonstrated with the developed TFA detectors, with
detections down to 6 keV X-rays, with a maximum signal created at about 1,000 electrons and a peak charge at about 660 electrons in a fully depleted 15-m thick sensor.9
However, no clear proof of MIP detection could be obtained up to now on the
first-generation TFA detectors. Tests were only carried out with a Sr90 beta source.
The source low activity and the difficulty of triggering MIP events on small pixels
required a signal-over-noise ratio higher than 10 to obtain clear signal distribution,
so that even though MIP signals of 500 electrons as expected in a 15 m thick sensor could be detected and identified, the MIP signal distribution was too close to the
circuit noise to be fully extracted.
Better MIP detection can be achieved by an increase of the sensor thickness.
The different segmentation techniques suggested by the characterizations carried
out on these first-generation TFA detectors permit thinking of operating 40-m
104
thick sensors under full depletion. Second-generation electronic circuits also need
to implement leakage current compensation circuits so there is no impact on the
biases of the preamplifiers. A 40-m thick sensor fully depleted could lead to an
estimated charge of about 1,400 electrons for a MIP, which could be read out by the
preamplifiers already developed. The design of very small pixels (below 100 m 2)
could also permit increasing the signal following the small pixel effect. For pixels
with dimensions lower than the sensor thickness, the weighting field for the electrode
integrated in the ASIC is not constant anymore in the sensor but becomes concentrated close to the electrode. Charges drifting in the vicinity of this electrode will
therefore determine the induced signal, and in the n-i-p configuration used electrons
drift down to the electrode. The induced signal will thus be determined mostly by
the motion of electrons and less by the holes, maximizing the charge induced during the integration time of the preamplifier. Finally, further design optimizations
of second-generation circuits can also lead to reduced equivalent noise charge. All
these possible improvements will certainly lead in the near future to conclusions on
the MIP detection and on the viability of TFA detectors for HEP applications.
4.6Toward a-Si:HBased Radiation Detector Solutions

Different research groups have built and characterized a-Si:H radiation sensors since
the late 1980s to profit from the possibility to do uniform deposition of this material
over large areas for low cost. These works have demonstrated the potential of this
material for single-particle detection,812 also suggesting high radiation hardness of
a-Si:H,13,14 and have permitted significant progress in the understanding of the material specificities and in the description of a-Si:H sensor particularities. The sensors
need to be thick in comparison to standard a-Si:H devices, implying different technological issues. Sensors with a maximum thickness of about 40 m were shown to
be reasonable to allow the sensor full depletion while keeping the leakage current
in acceptable bounds, so that a-Si:H sensors are thin in comparison to standard silicon detectors (which have thicknesses above 150m). Fewer charges are therefore
expected to be created by an ionizing radiation in an a-Si:H detector, resulting in the
most important limitation and challenge of a-Si:H technologies for single-particle
detection applications. The lack of straightforward integration of the sensor with
the readout electronics together with the success encountered by crystalline silicon
technologies for detection and electronics hindered a-Si:H detection application in
high-energy physics in the 1990s. However, the progress made in the understanding
of the material-specific properties associated with its amorphous phase and in its
deposition techniques has led to a wide industrialization of successful products using
a-Si:H in different applications.
The uniform depositions over large areas for low costs and the abilities to create
field effect transistors, diodes, and other semiconductor devices have strongly established a-Si:Hbased devices as successful imagers and X-ray flat-panel detectors.
The thin-film silicon solar cell industry has entered an impressive phase of mass
production, with turnkey production lines provided by established manufacturers, so
105
that amorphous silicon-based large-area solar modules will certainly play an important role in the future of solar electricity generation. In parallel to this high expansion
of the material in light-sensing applications, the demonstrated possibility to deposit
an a-Si:H sensor directly on top of integrated circuits has triggered the development
of TFA technology, creating an increasing interest in the HEP community to combine the works done on thicker (~30 m) a-Si:H diodes deposited on glass substrates
and this novel technology to build semimonolithic a-Si:Hbased pixel detectors. All
these different studies have led to a better understanding of the interesting particularities offered by a-Si:H for radiation detection.
The main advantages of a-Si:H are technological, since ionizing radiation sensors can be built at low cost on large areas, flexible substrates, very thin substrates,
metallic foils, and so on. Even though MIP detection is difficult to achieve on a-Si:H
devices, higher ionizing radiation can be detected well, and direct readout of beams
of particles have been demonstrated thanks to supposed high radiation hardness.
This shows a high potential for various a-Si:H detector structures and devices, either
as extremely thin sensors when deposited thinly on thin substrates or as thicker layers to operate in counting mode. The standard a-Si:H detectors therefore appear to
be interesting solutions for beam monitoring and for beam tracking, for applications
in HEP or in medicine at therapy accelerator centers, and for any other accelerator
applications.
The possible direct deposition of the a-Si:H sensor on integrated circuits also
demonstrates the high technological benefit of combining the advantages of monolithic detectors and of hybrid pixel detectors. This TFA technology also permits the
development of detectors with improved signal-over-noise ratio thanks to the direct
interconnection between the pixel-sensing element and its electronic readout, so that
MIP detection could be thought of being achieved with the future generations of
TFA detectors. The first generation of TFA detectors has demonstrated that signals
are for the most part constructed from electron drift in a time less than 18 ns, that the
complicated dispersive transport of holes only leads to a partial collection, and that
specific care has to be taken when designing the integrated circuit of a TFA detector
to render its surface flat so it does not induce lateral pixel effects of high additional
leakage currents. Finally, the first-generation detectors have also demonstrated a high
potential of the technology for detection at high spatial resolution thanks to reduced
charge sharing between close-in space electrodes. Future generations of TFA detectors need to be developed to demonstrate the viability of the technology for particle
detection in HEP and to indicate the possibility of detecting MIPs and the detection
efficiency of TFA detectors. Further studies are still also required to indicate the
exact specificities of a-Si:H material radiation resistance, which are still not included
in a general model. However, the inner properties and performance of the technology already demonstrated show interesting potential for different ionizing radiation
detection applications. This technology also has the advantage of providing an extra
process step possibility in the construction of integrated circuits, with the vertical
integration of different possible semiconducting devices, which could permit widening the performance and possibilities of ASICs.
106
References

1. N. Wermes, Pixel detectors for tracking and their spin-off in imaging applications. Nucl.
Instr. Methods A, 541, 150165, 2005.
2. J. E. Brau, The science and challenges for future detector development in high-energy
physics, SNIC Symposium, Stanford, CA, 2006.
3. M. Moll et al., Development of radiation tolerant semiconductor detectors for the SuperLHC. Nucl. Instr. Methods A, 546, 99107, 2005.
4. D. E. Carlson, C. R. Wronski, Amorphous silicon solar cell. Appl. Phys. Lett., 28(11),
671673, 1976.
5. S. N. Kaplan, J. R. Morel, and T. A. Mulera, Detection of charged particles in amorphous silicon layers. IEEE Trans. Nucl. Sci., NS-33, 351, 1986.
6. V. Perez Mendez et al., Hydrogenated amorphous silicon pixel detectors for minimum
ionizing particles. Nucl. Instr. Methods A, 273, 127, 1988.
7. J. Dubeau et al., Response of amorphous silicon p-i-n detectors to ionizing particles.
Nucl. Instr. Methods B, 54, 458471, 1991.
8. F. Foulon et al., Nuclear radiation detectors using thick amorphous silicon MIS devices.
Nucl. Instr. Methods A, 456, 647648, 2001.
9. M. Despeisse et al., Hydrogenated amorphous silicon sensor deposited on integrated
circuit for particle detection. IEEE Trans. Nucl. Sci., 55(2), 802811, 2008.
10. J. R. Srour et al., Damage mechanisms in radiation-tolerant amorphous silicon solar
cells. IEEE Trans. Nucl. Sci., 45, 2624, 1998.
11. N. Kishimoto et al., Stable photoconductivity in metastable a-Si:H under high-energy
proton irradiation. J. Non-cryst. Solids, 227230, 238242, 1998.
12. R. A. Street et al., High performance amorphous silicon sensor arrays. J. Non-cryst.
Solids, 227230, 13061310, 1998.
13. R. A. Street et al., Amorphous silicon sensor arrays for X-ray and document imaging.
Thin Solid Films, 296, 172176, 1997.
14. J. P. Moy, Large area X-ray detectors based on amorphous silicon technology. Thin Solid
Films, 337, 213221, 1999.
15. H. Fischer et al., Thin film on ASICA novel concept for intelligent image sensors.
Mater. Res. Soc. Symp. Proc., 285, 11391145, 1992.
16. B. Schneider, P. Rieve, and M. Bohm, Image sensors in TFA (thin film on ASIC) technology. Handbook of Computer Vision and Applications, Vol. 1, 237270, Boston:
Academic Press, 1999.
17. R. C. Chittick, J. H. Alexander, and H. F. Sterling, The preparation and properties of
amorphous silicon. J. Electrochem. Soc., 116, 7781, 1969.
18. W. E. Spear and P. G. Lecomber, Substitutional doping of amorphous silicon. Solid State
Commun., 17, 11931196, 1975.
19. H. Fritzsche, Electrons in disordered systems, scaling near the mobility edge. Proceedings
of the Seventh International Conference on Amorphous and Liquid Semiconductors,
p. 301, Edinburgh, 1977.
20. K. Laaziri et al., High energy X-ray diffraction study of pure amorphous silicon. Phys.
Rev. B, 60(19), 1352013533, 1999.
21. R. K. Dash et al. A quantitative measure of medium-range order in amorphous materials from transmission electron micrographs. J. Phys.-Condens. Mat., 15, S2425S2435,
2003.
22. M. J. Van Den Boogaard et al., The influence of the void structure on deuterium diffusion in a-Si:H. Mater. Res. Soc. Symp. Proc., 258, 407, 1992.
23. R. A. Street, Hydrogenated Amorphous Silicon. New York: Cambridge University Press,
1991.
107
24. D. Weaire and M. F. Thorpe, Electronic properties of an amorphous solid. I. A simple

tight binding theory. Phys. Rev. B, 4, 2508, 1971.
25. D. Staebler and C. Wronski, Reversible conductivity changes in discharge-produced
amorphous Si. Appl. Phys. Lett., 31, 292, 1977.
26. H. M. Branz, Hydrogen collision model of light-induced metastability in hydrogenated
amorphous silicon. Solid State Commun., 105(6), 387391, 1998.
27. N. F. Mott and E. A. Davis, Electronic Processes in Non-Crystalline Materials. Oxford,
England: Clarendon Press, 1979.
28. T. Tiedje and A. Rose, A physical interpretation of dispersive transport in disordered
semiconductors. Solid State Commun., 37, 49, 1980.
29. B. Equer and A. Karar, Amorphous semiconductors for particle detection: physical and
technical limits and possibilities. Nucl. Instrum. Methods A, 275, 558563, 1989.
30. J.-H. Zhou. Estimation of the mobility-lifetime products in amorphous silicon in the
presence of dispersive transport. J. Appl. Phys., 33, 16551658, 1994.
31. H. Curtins, N. Wyrsch, A. Shah, High rate deposition of amorphous hydrogenated
silicon-effect of plasma excitation-frequency, Electron. Lett., 23, 228230, 1987.
32. A. Mahan, J. Carapella, B. Nelson, R. Crandall, and I. Balberg, Deposition of device
quality, low H content amorphous silicon, J. Appl. Phys., 69, 67286730, 1991.
33. Oerlikon Solar, http://www.oerlikon.com/solar
34. Applied Materials, Thin Film Silicon, http://www.amat.com/products/solar_3.html
35. Flexcell, VHF-Technology, http://www.flexcell.ch/index.php
36. Y. Hamakawa, Thin-Film Solar Cells, Next Generation Photovoltaics and Its Applica
tions. New York: Springer, 2004.
37. W. S. Hong, A. Mireshgi, J. S. Drewery, et al., Charged particle detectors based on high
quality amorphous silicon deposited with hydrogen or helium dilution of silane. IEEE
Trans. Nucl. Sci., 42(4), 240246, 1995.
38. N. Wyrsch, M. Despeisse, G. Anelli, et al., Thin-film silicon detectors for particle detection. Phys. Status Solid, 1(5), 12841291, 2004.
39. S. Qureshi, V. Perez-Mendez, S. N. Kaplan, et al., Signal generation in a hydrogenated
amorphous silicon detector. IEEE Trans. Nucl. Sci., 36(1), 194198, 1989.
40. D. Fischer, Electric field and photocarrier collection in amorphous silicon p-i-n solar
cells, effects of light-induced degradation and of low-level i-layer doping. Thesis.
Neuchatel, Switzerland: Universite de Neuchatel, 1994.
41. C. Klein, Bandgap dependence and related features of radiation ionizing energies in
semiconductors. J. Appl. Phys., 39(4), 20292038, 1968.
42. S. Najar, B. Equer, and N. Lakhoua, Electronic transport analysis by electron-beaminduced current at variable energy of thin-film amorphous semiconductors. J. Appl.
Phys., 69(7), 39753985, 1991.
43. J. B. Chevrier and B. Equer, High electric-field amorphous silicon p-i-n diodes: effect of
the p-layer thickness. J. Appl. Phys., 76(11), 74157422, 1994.
44. F. Meillaud, A. Shah, C. Droz, et al., Efficiency limits for single-junction and tandem
solar cells. Solar Energy Mater, 90(1819), 29522959, 2006.
45. S. Ramo, Currents induced by electron motion. Proc. IRE, 27, 584, 1939.
46. G. Cavalleri, E. Gatti, G. Fabri, et al., Extension of Ramos theorem as applied to induced
charge in semiconductor detectors. Nucl. Instrum. Methods, 92, 137140, 1971.
47. W. Riegler, Extended theorems for signal induction in particle detectors. Nucl. Instrum.
Methods A, 535, 287293, 2004.
48. Q. Gu, E. A. Schiff, S. Grebner, et al., Non-gaussian transport measurements and the
Einstein relation in amorphous silicon. Phys. Rev. Lett., 76(17), 31963199, 1996.
49. M. Despeisse, P. Jarron, D. Moraes, et al., Preliminary radiation tests of 32 m thick
hydrogenated amorphous silicon films. Nucl. Instrum. Methods A, 552, 8892, 2005.
108
50. R. Aleksan, T. Bolognese, B. Equer, et al., Observation of single minimum ionizing particles with amorphous silicon diodes. Nucl. Instrum. Methods A, 305, 512516, 1991.
51. M. Despeisse, S. Saramad, P. Jarron, et al., Characterization of a thick layer a-Si:H
pixel detector with TFA technology using a scanning electron microscope. J. Non-cryst.
Solids, 352(920), 18321836, 2006.
X- and Gamma5 Novel

Ray Detectors Based
on Metamaterials
Paul Lecoq
European Organization for Nuclear Research, CERN

Geneva, Switzerland
Contents
5.1
5.2
5.3
5.4
Introduction................................................................................................... 109
Mechanisms of Energy Conversion............................................................... 110
Heavy Crystal Fiber Technology................................................................... 115
Photonic Crystals and Quantum Dots........................................................... 117
5.4.1 Materials Based on Quantum Dots................................................... 117
5.4.2 Photonic Crystals for Light Transport and Light Extraction............. 119
5.5 Detector Concept........................................................................................... 120
5.6 Conclusion..................................................................................................... 124
Acknowledgment.................................................................................................... 125
References............................................................................................................... 125
5.1Introduction
In the majority of X- and gamma-ray conversion detector heads, there is generally
a trade-off between the spatial and the energy resolution, since good spatial resolution requires high segmentation, whereas good energy resolution is obtained in a
large enough detector volume to contain all the cascade interactions generated by
the incoming particle. The quest for better spatial resolution in all three dimensions
for the majority of applications (high-energy physics and particle detectors, spectrometry of low-energy gamma quanta, medical imaging, homeland security, space
applications) may lead to a huge increase in the number of readout channels, with all
the associated problems of connectivity, detector integration, and heat dissipation.
This chapter explores the potential of recent progress in the field of crystallogenesis, quantum dots, and photonics crystals toward a new concept of X- and gammaray detector based on metamaterials to simultaneously record with high precision
maximum information concerning the cascade conversion process, such as its direction, the spatial distribution of the energy deposition, and its composition in terms of
electromagnetic, charged, and neutral hadron contents (for high energy).
109
110
Detection of X-rays, gamma rays, or ionizing particles plays an important role in a

wide range of applications, such as high- and medium-energy physics and astrophysics detectors, spectroscopy, medical imaging, industrial nondestructive control systems, and homeland security. Present X- and gamma-ray detectors are mostly based
on inorganic scintillating materials or direct conversion semiconductor materials.
They generally aim to measure the coordinate of the conversion point and the total
energy released in the detector volume. If information on the direction of the incoming particle is desired, it is obtained by packing several detector planes.
These ionizing radiation detectors convert the energy of an incoming particle into
light in the case of a scintillator or into electronic carriers in the case of a solid-state
detector. In both cases, these materials must be dense enough to allow an efficient
X- or gamma-ray conversion in a small detector volume, have a fast response time to
sustain high acquisition rates, and have a linear response as a function of the energy
deposit. Good energy and position resolution is generally required in the majority
of the applications. However, it is difficult to have both at the same time, and most of
the detector designs have to compromise between these two parameters. Moreover,
present systems provide global information on the total energy deposit in the detector but have no particle identification capability and give no details on the cascade
mechanism of this energy deposition and on the physical signal generation process.
New scintillator production processes and recent advances in nanotechnologies,
in particular in the domain of photonics crystals and nanocrystals, open interesting
perspectives for developing new detector concepts capable of delivering much richer
information about X- or gamma-ray energy deposition. They are based on metamaterials to simultaneously record with high precision the maximum information
of the cascade conversion process, such as its direction and the spatial distribution
of the energy deposition and its composition in terms of electromagnetic, charged,
and neutral hadron contents for the case of high-energy incoming particles.
5.2Mechanisms of Energy Conversion

Charged and neutral particles interact with absorbing materials following the wellknown mechanisms of radiation interactions in matter described by many authors.1,2
A scintillator is an absorbing material that has the additional property of converting
into light a fraction of the energy deposited by ionizing radiation. Charged particles
continuously interact with the electrons of the medium through Coulomb interactions, resulting in atomic excitation or ionization. Neutral particles will first have
to undergo a direct interaction with either an electron or a nucleus, producing secondary electrons, recoil protons, or fission fragments, which then will transfer their
energy to the medium in the same way as primary charged particles.
The rate of energy loss dE/dx for charged particles is strongly energy dependent.
It is well described by the Bethe-Bloch formula for incoming particles in the range
of megaelectron volts to gigaelectron volts, with atomic shell corrections at lower
energy and radiative loss corrections at higher energy:

2m v 2
v2 v2
dE 4 e 4 z 2
0
ln
ln
=
NZ
1 2 2
dx
I
m0 v 2
c c
Novel X- and Gamma-Ray Detectors Based on Metamaterials
111
where v and ze represent the velocity and the electric charge of the primary particle, respectively; m 0 and e are the rest mass and the electric charge of the electron,
respectively; N and Z are the atom number density and atomic number of the absorbing material, respectively; and I is the mean excitation energy of the absorber.
For heavy materials currently used as scintillators with a density of 6 to 8 g/cm3,
the energy loss is typically on the order of 10 MeV/cm for a minimum ionizing particle, but it can be a factor up to 100 more at very low or very high energy.
In the case of X-rays or gamma rays, the three fundamental mechanisms of electromagnetic interaction are3
Photoabsorption
Compton scattering
Electronpositron pair production
The dominant process at low energy (up to a few hundred kiloelectron volts for
heavy materials) is the photoelectric absorption. The interacting photon is completely
absorbed and transfers its energy to an electron from one of the electron shells of the
absorber atom (usually from a deep shell). The resulting photoelectron is ejected
with a kinetic energy corresponding to the incident photon energy minus the binding
energy of the electron on its shell. This is followed by a rapid reorganization of the
electron cloud to fill the electron vacancy, which results in the emission of characteristic X-rays or Auger electrons. The photoelectric process has the highest probability
when the incident photon has an energy comparable to the kinetic energy of the electron on its shell. This is the origin of the typical peaks observed in the cross-section
curve corresponding to resonances for the different electron shells (Figure5.1). The
general trend of this cross section is a rapid decrease with energy and a strong dependence on the atomic number Z of the absorber, explaining the preponderance of highZ materials for X-rays or gamma rays shielding and detecting materials:
ph
Z5
E7 / 2
At energies above a few hundred kiloelectron volts, Compton scattering becomes

predominant. In this case, the incident photon transfers only part of its initial energy
E to an electron of the atomic shells and is scattered at an angle with respect to its
original direction. The recoil electron is then rapidly absorbed by the scintillator and
releases an energy according to the formula

Ee = E E Eebinding
where E is the energy of the scattered photon given by (with m 0 the rest mass of the
electron)

E =
E
1+
E
m0 c 2
(1 cos )
112
Cross section (barns/atom)
1 Mb
(b) Lead (Z = 82)

- experimental tot
p.e.
Rayleigh
1 kb
g.d.r.
1b
10 mb
10 eV
Compton
1 keV
nuc
1 MeV
Photon energy
1 GeV
100 GeV
Figure 5.1 Energy dependence of photon total cross sections in lead. (From Particle
Data Group.)
The energy released in the scintillator by the recoil electron is distributed on

a continuum between zero and a maximum up to E m 0 c2/2 = E 256 KeV (for
gamma energy that is large compared to the rest mass of the electron).
The probability of Compton scattering is related to the electron density in the
medium and increases linearly with the atomic number of the absorber. Since this Z
dependence is much smaller than for the photoelectric effect, high-Z materials will
therefore be favored when a high fraction of photoelectric conversion is required, as
is the case for positron emission tomographic (PET) scanners, for instance.
Above a threshold of 1.02 MeV (twice the rest mass of the electron) the mechanism of e+e pair production can take place, predominantly in the electric field of the
nuclei and to a lesser extent in the electric field of the electron cloud (respectively,
nuc and e in Figure5.1). Similar to photoabsorption and Compton scattering, this
process has a higher probability for high-Z materials as the cross section is approximately given by the formula4

pair Z 2 ln(2 E )
As long as the energy of particles is high enough for multiple scattering and
e lectronpositron pair creation, their energy is progressively distributed to a number of secondary particles of lower energy, which form an electromagnetic shower.
Below the threshold of electronpositron pair creation, electrons will continue to
lose energy mainly through Coulomb scattering.
113
In the case of an ordered material like a crystal, another mechanism takes place
at this stage. In the process of energy degradation within a shower, the electrons in
the kiloelectron volt range start to couple with the electrons and atoms of the lattice and excite the electrons from the occupied valence or core bands to different
levels in the conduction band. Each of these interactions results in an electronhole
pair formation. If the energy of the electron is high enough to reach the ionization
threshold, free carriers are produced, which will move randomly in the crystal until
they are trapped by a defect, collected on electrodes (semiconductors), or recombine
on a luminescent center (scintillators). If the ionization threshold is not reached, the
electron and hole release part of their energy by coupling to the lattice vibration
modes until they reach the top of the valence band for the hole and the bottom of
the conduction band for the electron. They can also be bound and form an exciton
whose energy is in general slightly smaller than the bandgap between the valence
and the conduction bands. At this stage, the probability is maximum for their relaxation on luminescent centers through an energy or a charge transfer mechanism.
For a material to be a scintillator, it must contain luminescent centers. They are
either extrinsic, generally doping ions, or intrinsic (i.e., molecular systems of the
lattice or of defects of the lattice, which possess a radiative transition between an
excited and a lower energy state). Moreover, the energy levels involved in the radi
ative transition must be smaller than the forbidden energy bandgap to avoid reabsorption of the emitted light or photoionization of the center.
In a way, a scintillator can be defined as a wavelength shifter. It converts the
energy (or wavelength) of an incident particle or energetic photon (ultraviolet [UV],
X-ray, or gamma-ray) into a number of photons of much lower energy (or longer
wavelength) in the visible or near-visible range, which can be easily detected by current photomultipliers, photodiodes, or avalanche photodiodes.
Owing to the complexity of the conversion mechanism, there is generally a tradeoff between the spatial and the energy resolution, since a good spatial resolution
requires a high segmentation, whereas a good energy resolution is obtained in a
large enough detector volume to contain all the cascade interactions generated by the
incoming particle. The quest for an increasing demand for better spatial resolution
in all three dimensions for the majority of applications may lead to a huge increase
of the number of readout channels, with all the associated problems of connectivity,
detector integration, and heat dissipation.
Moreover, present detectors provide no or very little information about the way the
energy is released in the detector: spatial distribution of the cascade events and physics mechanisms of the energy loss (ionization, multiple scattering, nuclear interaction, Cerenkov). This is particularly important for high-energy physics experiments,
for which high-precision measurement of hadrons and jets resulting from the decay
of heavy short-lived particles produced in high-energy collisions is one of the detector challenges at future high-energy colliders. Figure5.2 illustrates the complexity of
a 3-TeV electron collision producing pairs of W and Z bosons. The simulated event
shows the distribution of the multiple tracks in jets in the central tracking section of
the detector and the showering of these tracks in the calorimeter section.
One of the difficulties is related to the lateral extension of the showers associated with the individual tracks in a jet. These showers strongly overlap and produce
114
Figure 5.2 Simulated WW/ZZ event resulting from a 3-TeV electron collision at the
future linear collider CLIC. (Courtesy of C. Grefe.)
a signal that is dependent on the relative amount of electromagnetic and nuclear

interaction events in the energy conversion process. Indeed, these two mechanisms
are characterized by different mechanisms of nonradiative energy losses in the
majority of known scintillators and therefore have a different scintillation efficiency.
As a consequence of large event-to-event fluctuations in the relative contribution
of these two mechanisms, the resulting energy resolution is deteriorated. One possible way is to measure independently the scintillation light produced in an active
medium, which is proportional to the total energy deposited by the shower particles,
and the Cherenkov light, which is only produced by the charged, relativistic shower
particles. Since the latter are almost exclusively found in the electromagnetic (em)
shower component (dominated by 0s produced in hadronic showers), a comparison
of the two signals makes it possible to measure the energy fraction carried by this
component fem event by event. As a result, the effects of fluctuations in this component, which are responsible for significant loss of performance in the majority of
calorimeters (nonlinearity, poor energy resolution, non-Gaussian response function),
can be eliminated or at least reduced to improve substantially the energy resolution
of hadronic showers in general and of jets in particular. This leads to an important
improvement in the hadronic calorimeter performance.
Although apparently less complex, the conversion events at low energy are also
far from punctual, even for the case of a photoelectric conversion producing a recoil
electron and a cascade of X-rays and Auger electrons as a result of the reorganization
of the atomic electron cloud after the initial event. Indeed, one of the limitations of
currently available PET and single-photon emission computed tomography (SPECT)
cameras is poor spatial resolution (5 to 6 mm in whole-body clinical cameras), which
is partly related to the necessity to integrate on a sufficiently large detector pixel the
115
Figure 5.3 Example of a 511-KeV gamma conversion in three crystal pixels of a PET camera, with two successive Compton interactions followed by a photoelectric event. Note that a
large fraction of energy escapes the central crystal, where the first interaction took place.
primary Compton scattering interactions (70% of the cases in the currently used
conversion materials such as lutetium orthosilicate [LSO]) and the final photoelectric interaction to reconstruct the full energy of the event with sufficient precision.
In spite of this, a large number of Compton events still escape the pixel, where the
primary interaction took place. These events are generally rejected by the image
reconstruction algorithm, therefore reducing by a large amount the overall sensitivity
of the camera. The possibility to make use of this large fraction of rejected events,
for which at least one Compton interaction took place, would considerably increase
the sensitivity of the camera. The same image quality could therefore be obtained in
a shorter time or with a lower dose injected to the patient. Similar arguments apply
for the control of industrial systems or for homeland security: A detailed recording
of the whole Comptonphotoelectric interaction chain would allow separation of the
spatial resolution from the energy resolution requirement and have a strong impact
on both the reconstructed image quality and the sensitivity of the imaging device.
An example of a 511-keV conversion event in a PET detector made of several pixel
crystals (typically 2 2 10 mm3) is shown in Figure5.3. Even for such a simple
event, it appears that a large fraction of the energy escapes the central pixel where
the first Compton interaction took place.
5.3Heavy Crystal Fiber Technology

Shaped crystalline fibers grown from the melt have a number of advantages compared
to more traditional technologies of bulk crystal growth followed by cutting and polishing. First, fibers can be grown with cross sections approaching the final geometry,
therefore minimizing the work and cost of mechanical processing. Furthermore, the
structural quality of the surface is not damaged by the use of abrasives. The growth
rate is generally high, which has an impact not only on the production costs but also
on the homogeneity of the distribution of the doping ions. Such methods generally
induce fewer thermal stresses with a resulting smaller probability of cracks.
The analysis of the potential of different melt growth techniques shows that micropulling-down technology from a shape-controlled capillary die allows the production of elongated crystals with dimensions that are not accessible using traditional
cutting and polishing of bulk crystals grown by the more standard Czochralski or
116
Bridgeman methods. The size of the melting zone in the pulling-down technique is
up to one order of magnitude smaller than that observed in the Czochralski method.
Therefore, it is believed that the pulling-down process can be considered a good way
to achieve stationary pulling conditions and can facilitate the growth process, allowing, for instance, much faster growth and higher concentration of doping ions, even
for those with a high segregation coefficient.
Since 1993, several groups have contributed to impressive progress in the development of the micropulling-down scintillating fiber technology (see Ricard; Lebbous
and Boulon and references therein),4,5 mainly driven by laser rod production. New
applications in the field of ionizing radiation detection are now pursued in the frame
of the Crystal Clear collaboration,6 particularly for particle physics experiments
and for medical imaging instrumentation. High-quality fibers of LuAG (lutetium
aluminum garnet) have been grown in collaboration with the company Fibercryst
(Lyon, France) and the Laboratoire de Physico-Chimie des Matriaux Luminescents
(LPCML) from University Claude Bernard (Lyon). Other well-known heavy scintillating crystals such as bismuth germanate (BGO), lutetium or lutetium-yttrium orthosilicates (LSO, LYSO7), and yttrium and lutetium aluminum perovskites (YAP and
LuAP) have also been grown in fibers of different sizes, and lead tungstate (PWO) is
under study. These high-quality scintillating fibers can open attractive possibilities
for the design of future detectors for high-energy physics, medical imaging, or other
applications. Table5.1 summarizes the most relevant parameters of these crystals.
In the present state of the art, the micropulling-down technique allows growth
of fiber- (rod-) shaped crystals with a controlled diameter between 0.3 and 3.0 mm
and up to 2 m in length. By modifying the shape of the capillary die, it is also possible to produce elongated crystalline materials with more complex noncylindrical
cross section (square, rectangular, hexagonal) for easier integration of the crystal in
complex detectors. The procedure based on micropulling-down technology was
improved at the LPCM Laboratory and Fibercryst Company in Lyon to grow both
single-crystal fibers and shaped bulk crystals. The crystals are produced from the
melt obtained at the capillary die positioned at the center bottom of a cylindrical iridium crucible as illustrated in Figure5.4. Once the melt drop is formed, the growth
process is initiated after connection of the seed with the drop at the bottom of the
Table 5.1
Some Parameters of Crystalline Fibers Grown by Fibercryst
Crystal
BGO
GSO:Ce
YSO:Ce
LYSO:Ce
YAG:Ce
LuAG:Ce
LuAG:Pr
Light Yield
(ph/MeV)
Decay Time
(ns)
Peak Emission Wavelength

(nm)
Density
(g/cm3)
8,000
14,000
14,000
25,000
20,000
20,000
20,000
300
60
37 and 82
40
70
70
20
480
460
420
420
550
535
290350
7.13
6.71
4.45
7.40
4.57
6.73
6.73
117
RF heater
First drop
Ceramic holder of melt
Crucible
Pulled fiber
Pulling direction
Oriented
seed
Orientation
control
BGO
YAG:Ce
YAG 0.05% Ce
LYSO:Ce
YAP:Ce
= 400 m
= 1 mm
= 2 mm
= 2 mm
Figure 5.4 The micropulling-down crystal growth technology. (Courtesy of Fibercryst.)
crucible (capillary die). Then, the seed is pulled down continuously with a pulling
rate ranging from 0.1 to 0.5 mm/min (about 10 times faster than Czochralski and
50 times faster than Bridgeman methods). Ongoing developments are investigating
a multiple capillary die crucible allowing growth of several fibers in parallel. The
industrial optimization and cost-effectiveness study of this process still needs to
be made, but this should lead to comparable or even lower cost per unit volume of
grown crystal compared to standard crystal growth approaches.
5.4Photonic Crystals and Quantum Dots

Similarly impressive progress in the field of microtechnology and nanotechnology
today offers new perspectives for the development of novel detector designs based on
metamaterials. The microstructuration of bulk materials and of their surfaces allows
consideration of macroscopic structures, in which light production and propagation
are governed by quantum effects, giving them unusual and very attractive properties (negative refractive index, control of the direction of propagation, possibility
of gating, perfect antireflection at all angles, etc.). This is the domain of nanocrystals, quantum dots, and photonics crystals. Coupled to the parallel developments in
microelectronics and nanoelectronics, these new approaches will undoubtedly generate revolutionary changes of paradigms in detector concepts.
5.4.1Materials Based on Quantum Dots

Quantum dots are nanoparticles that have attracted widespread interest, particularly
in biology and medicine, because of their unique optical and electronic properties.
Quantum dots are made of semiconductor materials. The electrons in quantum dots
118
have a range of energies, and the concepts of energy levels, bandgap, conduction
band, and valence band still apply as in any semiconductor material. However, there
is a major difference. Excitons (bound electronhole pairs) have an average physical separation between the electron and hole, referred to as the exciton Bohr radius,
given by the following formula:

rB =
h 2
mee 2
where is the permittivity of the medium, h is the Planck constant, and me and e are
the mass and electric charge of the electron, respectively.
In bulk, the dimensions of the semiconductor crystal are much larger than the
exciton Bohr radius, allowing the exciton to extend to its natural limit. However,
if the size of a semiconductor crystal becomes small enough that it approaches the
size of the materials exciton Bohr radius (typically a few nanometers for most of
the known semiconductors), the electron energy levels become discrete through the
effect of quantum confinement. This has large repercussions on the absorptive and
emissive behavior of the semiconductor material, which can be easily tuned by the
radius of the quantum dots.8,9 Indeed, an interesting property of quantum dots is very
fast emission (in the nanosecond range) at a wavelength directly coupled to the size
of the nanosphere (Figure5.5). Moreover, the quantum efficiency of this emission
can reach as much as 7080% of absorbed energy, compared to about 15% for the
majority of known scintillators.
New technologies to prepare transparent ceramics from nanopowders can
be exploited to determine the conditions for which these optical properties can
Figure 5.5 UV-excited quantum dot solutions.
119
be maintained in a partially synthesized material. Another approach is based on the

choice of a host material, inorganic or organic, as a substrate for quantum dots. They
are suspended in an organic solvent so that they are castable from solution (e.g., by
spin-coating techniques) onto such a substrate without necessarily requiring lattice
matching. An interesting avenue to explore is to dope heavy host materials (such as
fluoride glasses10) with quantum dots of different diameters with a known spatial
distribution within the detection block. An appropriate distribution of quantum dots
of different size (and therefore different emission wavelengths) in the host volume
can help encoding the X- or gamma-ray conversion point coordinates with the emission wavelength of the quantum dots. The recording of the emission spectrum from
this detector, when hit by ionizing radiation, would then give an exact representation
of the three-dimensional distribution of the energy deposit within this block, opening the way for an imaging calorimeter concept with a minimum number of readout channels. This approach is challenging but has probably the highest potential in
terms of density of information such metamaterials can achieve.
5.4.2Photonic Crystals for Light Transport and Light Extraction

Photonic crystals are periodically structured electromagnetic media, generally possessing photonic bandgaps, which are ranges of frequencies in which light cannot
propagate through the structure. This periodicity, whose length scale is proportional
to the wavelength of light in the bandgap, is the electromagnetic analogue of a crystalline atomic lattice, where the latter acts on the electron wave function to produce
the familiar bandgaps, well known by solid-state physicists. Photonics crystals are
usually made of a block of transparent dielectric material that contains a number
of tiny airholes arranged in a lattice pattern. To be able to create photonic crystals
for optical devices, state-of-the-art semiconductor microfabrication techniques are
used. The patterned dielectric material will block light with wavelengths in the photonic bandgap while allowing other wavelengths to pass freely.
One interesting application of photonic crystals concerns the coupling of optical
propagation modes to extraction modes at the interface between two media with
different indices of refraction. This approach is actively pursued in our group11 to
improve the light extraction efficiency from heavy and therefore highly refractive
bulk materials, such as the majority of commonly used scintillators, for which the
light collection efficiency rarely exceeds 30% with standard techniques. The basic
idea is to exploit three important characteristics of two-dimensional periodic photonic crystal slab geometries. The first is the possibility of designing a photonic bandgap within a range of frequencies for the in-plane guided modes of the system. The
second is the possibility of coupling to resonant modes above the bandgap. The third
is an intrinsic upper cutoff frequency for waveguiding in the plane. In all cases, these
frequency ranges can be used to forbid radiation from propagating into the dielectric
slab, thus forcing the light into the light cone of the air region.
By combining the properties of quantum dots and of photonics crystals in properly engineered microstructured dielectric materials, it is indeed possible to tune the
light emission parameters and the light propagation properties in a bulk material at
a small scale and to guide the light through complex structures made of assembled
120
10 m
5 m
Figure 5.6 Photonic crystal fiber.
scintillating fibers or nanocrystals. In other terms, it means that high segmentation of

a bulk system and spatial encoding of the light signals would be possible, allowing a
high level of granularity without having to multiply the number of readout electronics channels.
To extract these encoded signals from complex metamaterial structures, the concept of photonic crystal fibers is of particular interest. Also called hollow-core bandgap fibers, they consist of a regular lattice of air cores running along the fiber, which
can then transmit a wide range of wavelengths without suffering from dispersion
(Figure5.6). It is made by packing a series of hollow glass capillary tubes in a material with a different dielectric constant. This structure is then heated and stretched to
create a long fiber that is only a few microns in diameter. The fiber has the unusual
property of transmitting a single mode of light, even if the diameter of the core is
very large. Unlike standard fibers, in which light is guided in a core of higher refraction index than for the cladding, photonic crystal fibers transmit light through the
hollow cores. The bulk of the fiber is only a support for the hollow-core lattice and
does not participate in the transfer of the electromagnetic energy. Such fibers are
therefore ideal in a high-radiation environment, since the light transmission is not
deteriorated by radiation damage.
The development of photonics crystal fibers is a very active field in modern pho
tonics, since the market perspectives of this segment are large for telecommunications and chip-to-chip interconnect for highly integrated electronics. Several
manufacturers propose providing such fibers at a moderate price, which is likely to
significantly decrease, if the market opens as expected.
5.5Detector Concept
The design of the majority of X- and gamma-ray detectors today results from a tradeoff between spatial and energy resolution requirements. It does not provide any detail
121
about the complexity of the energy conversion mechanism, which carries important
information on the spatial distribution as well as on the energy loss processes of the
shower resulting from the interaction of the incoming particle with the detection
medium. The basic idea to improve this situation is to structure the standard detector block or pixel in such a way as to extract more information than the total energy
deposit in the block. In electromagnetic calorimeters, the dimensions of the unit
detection block are typically 1 Moliere radius (1 to 3 cm for commonly used scintillators) and 25 radiation lengths (20 to 30 cm) in lateral and longitudinal directions,
respectively. For medical imaging devices, the pixel size is generally one order of
magnitude smaller, a few millimeters in section and 1 to 3 cm in length.
The proposed approach is based on scintillating fibers packed together to form
trunks of cables. The variety of fiber section shapes that can be produced allows
tuning the detector design as a function of the needs: hexagonal fibers if ultimate
homogeneity is required, rectangular section for different granularity requirements
in two directions, cylindrical fibers if free channels are needed for photonic crystal
fiber-based light collection systems or for other services. Notice also that the relative mechanical flexibility of these scintillating fibers (depending on the material)
allows twisting them in the cable in a similar way as in a rope, therefore minimizing
the impact of interfiber gaps for incoming particles. Moreover, various scintillators
can be selected to build these cables, having different emission wavelengths and
different scintillation yields and decay times, so that a direct encoding of the light
can be made as a function of the point of emission. A single photodetector can then
decode this complex signal and provide much more complete information about the
X- and gamma-ray conversion. For high-energy calorimeters, materials with different UV transmission cutoff can be selected so that the fraction of Cerenkov emission in the detected light can be determined. The direct extraction of the Cerenkov
signal in a scintillator from the pulse shape or wavelength analysis may however be
difficult if the scintillating signal is much higher than the Cerenkov one, which is
usually the case. Moreover, Cerenkov and scintillation signals are not independent,
since part of the UV Cerenkov emission is absorbed in the crystal to excite the
scintillation activator ions. An alternative approach is to select scintillating materials activated by a doping ion instead of self-activated scintillators such as BGO or
PWO. Cerium-doped LSO or LYSO, for instance, both very fast (40 ns decay time)
and dense (7.4 g/cc), or LuAG (70 ns decay time and 6.73 g/cc density), or LuAP
(17 ns decay time and 8.34 g/cc density) are excellent candidates for mixing ceriumdoped fibers, which would then behave as scintillators with undoped fibers of the
same material, which would only produce Cerenkov light. One could then obtain
a very homogeneous, dense, and compact calorimeter with a uniform radiation
length, Moliere radius, and interaction length in the whole volume of the detector.
This detector would have the additional feature of sampling the shower with a number of Cerenkov fibers conveniently distributed in the cable to directly measure the
electromagnetic-to-hadronic ratio of a shower on an event-to-event basis. Moreover,
the assembly of the fibers can be organized in a flexible way, allowing a multitude
of detector geometries.
If the detection of neutrons proves to substantially improve the overall jet energy
resolution, neutron-sensitive scintillation fibers can also be inserted in the cables.
122
SiPMTs
MOEMS diractive
optics
light concentrator
MOEMS diractive
optics
light concentrator
SiPMTs
Figure 5.7 Concept of a metacable for calorimetry at future linear colliders.
No attempts have been made yet to grow such fibers, but scintillators like lithium tetraborate LBO (Li2B4O6), lithium fluoride-based materials like LiCAF (LiCaAlF6),
or more generally the cerium-doped elpasolite family (Cs2-xRbxLiMX6, where X =
Sc, Y, La, Lu and X = Cl, Br, I) are very attractive candidates because of the presence of high neutron capture cross-section lithium and boron and of their low density
(2.42 for LBO), which makes them rather insensitive to gamma conversions.
Figure5.7 shows a possible concept for such a detector block presently developed
for future linear collider calorimeters.12,13 It is made of three types of fibers (scintillating, Cerenkov, and neutron sensitive) arranged in a cable, possibly in a twisted
configuration to minimize cracks seen by the incoming particles. The fiber section is
hexagonal to allow compact packing with minimum dead space between the fibers.
Such a configuration is quasi-homogeneous, allowing the best possible energy resolution for electromagnetic calorimetry. Moreover, it also provides useful information
about the composition of hadronic showers (charged light and heavy particles, neutrons), allowing a significant improvement for the resolution of jets. The light produced by each type of fiber is collimated by diffractive optics or microlens plates on
three small-size solid-state photodetectors (avalanche photodiode [APD] or geigermode silicon photomultiplier [SiPMT]) at both sides of the detector block to further
provide depth of interaction information, with a precision of typically 1/10 of the fiber
length. The optical transfer system from the fibers to the photodetectors requires special attention, since it must allow high specificity in multiplexing and measuring the
different components with the minimum of cross talk and be as compact as possible.
It must be noticed that similar optical systems are under development for adaptive
optics devices to be used in future large telescopes or telescope arrays. Fortunately
we can take advantage here of the large-scale academic and industrial ongoing
effort to transfer microelectronics technologies to the development and production
of MOEMSs (micro-optoelectro-mechanical systems). Microlens arrays are a vital
part of todays optical systems in a large range of domains, from telecommunications
123
to machine vision. Todays technology allows production of either refractive or diffractive lenses as small as a few tens of microns in diameter arranged in large-size
arrays with a position accuracy of a few microns only. Standard integrated circuit
production methods, such as optical lithography, are used to mass produce diffractive optics components with a high level of functionality and integration from photomasks imprinted by holographic techniques. A master component is created, from
which a negative mold is fabricated. The mold is then used for embossing or injection
molding of plastic plates with methods similar to those used for making holographic
Christmas wrapping papers or duplicating CD-ROMs. Such microlens arrays are
therefore thin and mechanically flexible, allowing easy coupling to the fiber bundle, and can be mass produced at low cost. Furthermore, there is a large variety of
mechanical supports available, which is important for high-energy physics applications, for which reasonably radiation-hard materials must be selected.
An alternative approach is being developed in our group, more oriented to low
X- and gamma-ray energies, below the pair production threshold. The objective is to
design a detector head for a PET scanner on the basis of photonic crystals and quantum dots. PET scanners measure in coincidence two 511-keV gamma rays resulting
from the decay of a positron emitted by a radioactive isotope (generally 18F) labeling
a molecule involved in some metabolic process. Once injected to the patient, this molecule will concentrate in the organs, where this metabolic function is active (i.e., cancer cells concentrate sugar more than healthy cells). The recording of the two 511-keV
gamma rays associated to each 18F positron emission allows the three-dimensional
reconstruction of the functionally active parts of the organs under examination.
The detector block is made of a matrix of heavy scintillating fibers. The scintillating material is selected as a function of its intrinsic light yield, which has to be
high enough to allow splitting of the light through different readout channels and
minimizing the photostatistics contribution to the energy resolution. Moreover, a
good linearity of the response is requested to achieve good energy resolution. We
are presently exploring different materials, among which LuAG is a good candidate
with a light yield of 20,000 ph/MeV (about one-half of sodium iodide), a decay
time of 70 or 20 ns whether it is cerium or praesodymium doped, and reasonable
response linearity. The technology to grow such fibers with the micropullingdown technology is also well mastered, and fibers several meters long with a diameter ranging between 300 m and 3 mm can be grown with consistent quality. The
diameter of the fibers is defined as a function of the desired spatial resolution and of
the detector ability to identify the first interaction point in the conversion process.
The angular distribution of photons diffused by Compton scattering of 511-keV
incident gamma-rays being strongly peaked forward, a precise determination of the
depth of each interaction in the conversion chain associated to one 511-keV gamma
conversion should allow determining with high precision where the first interaction
took place.
If the scintillating fibers have a cylindrical shape, the tiny gaps between them
can be filled with photonic crystal fibers a few hundred microns in diameter. As
explained, photonic crystal fibers transmit light through a regular lattice of holes
arranged in a material of higher dielectric constant than air. This supporting material
124

Quantum dot
loaded polymer
with a gradient
of QD radius
along the ber
Photonic bers
= 100 m
Heavy scintillating bers
(LuAG, LSO ...)
= 500 m
Figure 5.8 Concept of a metacable for low-energy X-rays and gamma rays.
can be loaded with quantum dots with a different diameter range from fiber to fiber
or distributed with a gradient of their diameter along each individual fiber. The spatial encoding is provided by the emission wavelength of these quantum dots. The
principle is to organize the readout through two different channels, one dedicated
to energy deposit measurement with the best possible resolution and the second one
optimized for three-dimensional spatial resolution of the different components of the
conversion cascade. A conversion event, photoelectric or Compton, in a scintillating
fiber will produce an isotropic emission of scintillating photons. The photons emitted
in the forward or backward direction will propagate along the scintillating fiber and
eventually be collected at both ends and summed on a sufficiently large number of
fibers to measure the total energy deposited in the cascade of primary and secondary
events produced by the conversion of the incident X-ray or gamma ray. On the other
hand, some of the photons emitted laterally will escape the scintillating fiber and
excite the quantum dots of the photonic crystal fiber, allowing identification of the
fiber hit and determining the depth of interaction with high precision by measuring
the spectrum of the light collected on miniaturized spectrophotometers. Diffractive
optics components will be used in the same way as for the previous detector to collect the light from the different types of fibers and to collimate it on specific photodetectors. A schematic of this low-energy detector block is shown on Figure5.8.
5.6Conclusion
The underlying idea of the work presented in this chapter is to develop metamaterials for the purpose of extracting the maximum information from the conversion of
X-rays or gamma rays in a material. Two proof-of-concept systems were described
and are aimed at different application domains. The first one proposes a concept of a
dual-readout (scintillation plus Cerenkov) or even triple-readout (with a possibility to
measure neutrons) imaging calorimeter aiming at an excellent jet energy resolution
for high-energy physics experiments. The second one is oriented to low-energy X-ray
or gamma-ray detection, particularly for finely three-dimensional granulated PET
detector heads. But, the two approaches are complementary, and the related generic
125
work on heavy scintillating fibers, photonic crystals, quantum dots, and refractive
optics is expected to open useful perspectives for the development of novel meta
material-based X-ray or gamma-ray detectors in a wide range of energy.
Acknowledgment
This contribution could not have been possible without many exchanges of view and
discussions with all my colleagues, physicists, and medical doctors from the Crystal
Clear collaboration and from the CERIMED project.
References

1. B. Rossi, High Energy Particles, Prentice-Hall, Englewood Cliffs, NJ, 1952.

2. U. Fano, Penetration of protons, alpha particles, and mesons, Annu. Rev. Nucl. Sci., 13,
1, 1963.
3. J. D. Jackson, Classical Electrodynamics, chap. 13, 2nd ed., New York, 1975.
4. J. Ricard, USA Patent. Processes for the continuous preparation of single crystals, patent
number 4565600, 1986.
5. K. Lebbou and G. Boulon, Fiber crystals growth from the melt, in Advances in Materials
Research Series, T. Fukuda, P. Rudolp, and S. Uda, Eds., Springer Verlag, Berlin,
pp. 219254, 2003.
6. R&D proposal for the study of new fast and radiation hard scintillators for calorimetry
at LNC, Crystal Clear Collaboration, CERN/DRDC/P27/91-15, Project RD18, CERN,
Geneva, Switzerland, 1991.
7. P. Anfr, C Dujardin, J. M. Fourmigu, B. Hautefeuille, K. Lebbou, C. Pedrini,
D. Perrodin, and O. Tillement, Evaluation of fiber-shaped LYSO for double readout
gamma photon detection, IEEE Trans. Nucl. Sc., 54(2), 391397, April 2007.
8. Y. Shen, C. S. Friend, Y. Jiang, D. Jakubczyk, J. Swiatkiewicz, and P. N. Prasad,
Nanophotonics: interactions, materials and applications, J. Phys. Chem. B, 104,
75777587, 2000.
9. Y. Kayakuma, Quantum size effects of interacting electrons and holes in semiconductor
microcrystals with spherical shape, Phys. Rev. B, 38, 9797, 1988.
10. E. Auffray et al., Cerium doped heavy metal fluoride glasses, a possible alternative for
electromagnetic calorimetry, Nucl. Instr. Meth. A, 380, 524536, 1996.
11. M. Kronberger, E. Auffray, P. Lecoq, X. Letartre, C. Seassal, and J. L. Leclercq, Probing
the concepts of photonic crystals on scintillating materials, IEEE Trans. Nucl. Sc., 55(3),
11021106, June 2008.
12. P. Lecoq, New crystal technologies for novel calorimeter concepts, Proceedings of
CALOR2008, International Conference on Calorimetry in High Energy Physics, Pavia
Journal of Physics: Conference Series 160 (2009), 012016, LSSN 17426588, May
2008.
13. E. Auffray and P. Lecoq, Dual readout calorimetry with heavy crystal fibers, IEEE
NSS/MIC Conference Records, Dresden, N55-6, 2009.
Geiger-Mode
6 Multicell
Avalanche Photodiodes
Silicon Photomultipliers
Gabriela Llos
IFIC, Instituto de Fsica Corpuscular

Valencia, Spain
Contents
6.1 Introduction................................................................................................... 127
6.2 Working Principle.......................................................................................... 129
6.3 Detector Characteristics................................................................................ 133
6.3.1 Photon Detection Efficiency.............................................................. 133
6.3.2 Dynamic Range................................................................................. 135
6.3.3 Noise.................................................................................................. 136
6.3.4 Time Resolution................................................................................. 138
6.3.5 Temperature Dependence.................................................................. 140
6.3.6 Insensitivity to Magnetic Fields......................................................... 141
6.3.7 Radiation Damage............................................................................. 141
6.4 Applications................................................................................................... 142
6.4.1 High-Energy and Space Physics........................................................ 143
6.4.2 Medical Imaging................................................................................ 145
Acknowledgments................................................................................................... 148
References............................................................................................................... 148
6.1Introduction
Multicell Geiger-mode avalanche photodiodes, generally known as silicon photomultipliers (SiPMs), are a new type of photodetector that has experienced fast development in the last few years. The name silicon photomultiplier is applied to devices
with the same working principle, also called metal-resistor-semiconductor avalanche
photodiodes (MRS APDs), multipixel photon counters (MPPCs), solid-state photo
multipliers (SSPMs), avalanche microchannel photodiodes (AMPDs), micropixel
avalanche photodiodes (MAPDs), and so on.
The work that gave origin to the development of SiPMs dates to the 1950s1960s1,2
with the studies of avalanche formation in microplasmas. However, it was only in
the 1990s with the invention of the MRS APD structure35 that previous work led
to the fabrication of operative devices. Since the development of the first samples,
127
128
Figure 6.1 SiPMs, 1 1 mm and 3 3 mm, from different manufacturers. From left to
right: FBK-irst (first two columns), Hamamatsu, and Zecotek.
they have experienced fast development aimed at the improvement of performance,

in particular the photon detection efficiency (PDE) for short wavelengths, and the
reduction of noise. They are currently produced experimentally or commercially
by several research institutes and companies, including CPTA/Photonique (Russia/
Switzerland), MEPhI/Pulsar (Russia), JINR/Micron (Russia), SensL (Ireland), FBKirst (Italy), MPI (Germany), Hamamatsu (Japan), Zecotek (Canada), and RMD (USA),
among others. Figure6.1 shows some SiPMs from different manufacturers.
The performance of current SiPMs is comparable to that of photomultiplier tubes
(PMTs) in terms of PDE.6,7 In addition, they present many other advantages compared to this and other types of photodetectors. They have high gain, fast timing
properties, and high rate capabilities. They operate at low bias voltage (below 100 V)
and have low power consumption (below 50 W/mm2). Their high signal-to-noise
ratio (SNR) makes it unnecessary to use low-noise electronics. They are compact
and rugged and insensitive to magnetic fields, and their cost is potentially low for
large production volumes, having already decreased significantly from the first commercial samples. At present, their main drawback is their small area, up to 5 5 mm.
The production of large-area devices is limited mainly by noise, which scales with
the detector area. However, significant improvements have also been made in this
aspect from the initial 1 1 mm devices.
Their good performance and their potential for further development have raised
enormous interest, and many research groups are testing this new kind of detector
for application in different fields as an alternative to other types of photodetectors.
In some fields, such as calorimetry or medical imaging, they already present an
advantageous alternative to photodetectors previously employed, and they offer better performance in many aspects.
129
Multicell Geiger-Mode Avalanche Photodiodes
6.2Working Principle
An SiPM consists of an array of several hundreds to thousands of detector structures
commonly known as microcells or micropixels in a common bulk. Each microcell is
a single photon Geiger-mode APD with a typical size of tens of microns (20100 m)
and a few microns deep. When a microcell is hit by a photon, an avalanche can be
triggered that is independent of the number of photons that arrive at the microcell.
All the microcells in an SiPM are connected in parallel though the common bulk and
the metal structure that connects the microcell outputs. The sum of the output signals
of all the microcells that compose the SiPM provides the SiPM output. If the number of photons is not excessively high in comparison with the number of microcells
in the detector, each photon is detected by a different microcell; therefore, the SiPM
output is proportional to the number of photons that reach the detector surface.
A common microcell structure consists of an n+/p junction in a low-doped (p-)
epitaxial layer of 3 to 5 m, all in a highly doped p substrate about 300500 m thick
(see Figure6.2).8,9 The doping concentration determines the breakdown voltage.
All the microcells are in the common epitaxial layer and connected to the top
metal layer through a quenching resistor in series (generally made of polysilicon)
that is necessary for the passive quenching of the avalanche. A guard ring around the
microcell serves to lower the electric field at the edges. In some cases, the microcells
are surrounded by optical trenches to ensure optical isolation between them and
prevent optical cross talk.
When reverse bias voltage (Vbias) is applied a few volts above the breakdown voltage (known as excess bias voltage or overvoltage), a high-field region is created in
the n+/p region where the avalanche is produced. The electric field is lower in the
epitaxial drift region, which is fully depleted, and it dies in the p+ region.
The microcell behavior can be described by an electric model1,2,10 (Figure6.3).
The p/n junction can be modeled as a capacitance CD in series with the quenching
Quenching resistor
Metal contact
SiO2
n+
p
p
~ 35 microns
p+
~ 500 microns
Figure 6.2 Schematic representation of the microcell structure in an SiPM. The dimensions are not to scale.
130
Rq
Cd
Rs
Vd
Vbias
Vbr
Microcell model
Bias circuit
Figure 6.3 Electrical model of the microcell behavior as in Oldham.10
resistance RQ. CD is determined by the geometry of the microcell and ranges from
about 50 to 500 fF.7
Photons that interact in the microcell produce carriers that traverse the high-field
region and have a certain probability, known as turn-on or triggering probability,
of initiating an avalanche by impact ionization. This state is modeled as a voltage
source Vbr with a series resistance RS. When the avalanche is triggered, the diode
capacitance CD discharges through the series resistance RS with a time constant D =
CD RS. Initially, the discharge current is limited by the avalanche formation process.
These two factors contribute to the fast leading edge of the SiPM signal, which is on
the order of hundreds of picoseconds. The avalanche spreads across the microcell
area by two mechanisms: lateral diffusion and photon emission. The lateral diffusion is caused by a nonequilibrium condition between the region where an avalanche
is produced and the regions that are not yet in breakdown. In addition, photons that
are emitted during the avalanche process11 can be absorbed nearby, initiating a new
avalanche in another region of the microcell and increasing the propagation speed.
As the voltage on CD decreases, the current flowing through the quenching resistor tends to the asymptotic value

Imax = (Vbias Vbr)/(RQ + RS)
which determines the peak of the SiPM signal. If RQ is high, over some hundreds
of kiloohms, the voltage drop in the resistor reduces the voltage in the diode
below the breakdown value. The electric field in the junction becomes insufficient to fulfill the breakdown conditions, and the multiplication breaks off. This
occurs when the diode current becomes low enough that a statistical fluctuation
brings the instantaneous number of carriers flowing through the high-field region
to zero, quenching the avalanche. The probability of such a fluctuation is significant when the current (known as latching or latchup current) is below 1020 A.
The quenching resistor must therefore be high enough to lower the current to this
131
value. Typical values of the quenching resistor are some thousand kiloohms to
1 megaohm. Under these conditions, two processes take place: recombination
of carriers and charge of the diode capacitance. CD starts to recharge to the bias
voltage with a time constant Q = CD RQ, giving the slow exponential decay
of the SiPM signal. This determines the recovery time of the microcell, which
is the time needed to recharge the diode up to 99% of the bias voltage and is
therefore given by about 5. Typical values of the recovery time range from about
20 to 500 ns. In practice, the possible existence of parasitic capacitances (e.g., due
to the presence of the quenching resistor on top of the microcell) modifies the
ideal behavior, and effective capacitance CD + CP needs to be considered to reproduce the real microcell behavior.12
The gain of the SiPM can thus be defined as the charge developed during one
Geiger avalanche divided by the electron charge, given by

G = ImaxQ /qe = (Vbias Vbr)Q /(RQ + RS) = (Vbias Vbr)CD /qe
where Q is the decay time, and qe is the electron charge. The gain is generally in the
range 105 to 107 for most devices, and it increases linearly with the bias voltage for a
few volts above breakdown until other effects begin to take place.
The gain fluctuations in SiPMs are small compared to APDs, and the individual
cells exhibit very small signal spread. However, due to afterpulsing and optical cross
talk (explained in this chapter), multiplication noise is not negligible, and it can be
expressed in terms of an excess noise factor F, which can be calculated from the
width of the single electron peak as6

F = 1 + 2/G 2
where is the variance of the Gaussian fit, and G is the SiPM gain.
The SiPM microcell structure strongly determines its performance. Although the
n-on-p structure is the one originally developed and most widely employed, other structures have been developed or are under investigation to improve the performance of
the SiPM in different aspects.9,13 The p-on-n structure is the preferred solution for the
detection of blue light given the higher PDE in that wavelength region and lower noise.
Other possible solutions are the buried junction, in which the high electric field region is
at a depth of 13 m, or back-illuminated structures. In the back-illuminated drift (BID)
SiPMs, the radiation enters from the back of a fully depleted wafer, and the generated
charges drift toward a small point-like avalanche region located on the front side. This
configuration results in full geometrical acceptance of the radiation entrance window.14
The study of an SiPM in the absence of light provides valuable information on its
characteristics. The currentvoltage I-V curve obtained by measuring the current for
different values of the reverse bias voltage allows determining essential information
on functionality of the SiPM.12 Figure6.4 shows a typical I-V curve of an SiPM. In
the first part of the curve, for bias voltages below the breakdown voltage, the current measured is the surface leakage current, which increases with the bias voltage.
132

IV Curve
Current (nA)
103
102
10
101
15
20
25
30
Reverse Bias Voltage (V)
35
40
Figure 6.4 Reverse currentvoltage (I-V) curve of an SiPM. The increase of the current
is linear with the bias voltage applied before the breakdown point and grows quadratically
after breakdown.
Above the breakdown voltage, the current is the sum of the leakage and the breakdown currents. The breakdown current increases with the gain and dark rate, which
both depend linearly on the bias voltage. As a result, the breakdown voltage grows
quadratically with the bias voltage, and it can thus be fitted with a parabola. The
breakdown voltage is indicated by the change in the response curve. The response of
the SiPM is given by the combination of the responses of all the microcells that fire
in each event and will therefore reproduce the microcell behavior, with the variations from one microcell to another plus the effects due to the interaction between
different microcells. The uniformity among all the microcells in an SiPM will be
an essential parameter in the SiPM characteristics. The I-V curve of the SiPM thus
reflects the behavior of all the microcells, each with slightly different parameters,
breakdown voltage, cell capacitance, and so on. Better uniformity of these parameters will result in a better defined SiPM response.
One of the characteristic features of SiPMs is their ability to detect extremely low
photon fluxes, down to the single photon. Single-photon detection is possible provided
signal photons can be distinguished from the noise background. Figure6.5 shows a
typical single-photoelectron spectrum of an SiPM at room temperature. The spectrum is obtained by illuminating a 1 1 mm SiPM with a light-emitting diode (LED)
at low light intensity levels and triggering the data acquisition system simultaneously
to the activation of the LED. The peaks corresponding to 1 to 10 photoelectrons
after the pedestal are clearly separated, showing the excellent single-photoelectron
resolution of these devices even at room temperature. Cooling the devices and crosstalk suppression result in a significant reduction of the noise, making it possible to
distinguish a high number of individual peaks.15
133

Single Photoelectron Spectrum
10,000
8,000
6,000
4,000
2,000
0
50
100
150
Signal area (a.u.)
200
250
Figure 6.5 Single-photoelectron spectrum of an SiPM obtained at room temperature illuminating the device with an LED at low light intensity. The pedestal and peaks corresponding
to 110 photoelectrons can be clearly distinguished.
6.3Detector Characteristics
The fundamental characteristics of the photodetector PDE and noise, dynamic range,
and the factors affecting them are reviewed in this section.
6.3.1Photon Detection Efficiency

The PDE of an SiPM is given by the product of three factors: the quantum efficiency (QE), the triggering probability Pt, and the geometrical efficiency or fill factor
G.16 The PDE varies with the wavelength of the incident radiation, the overvoltage
applied to the device, and the temperature.

PDE = QE Pt G
The QE is the probability that a photon that arrives at the detector surface interacts with it. It depends on the intrinsic QE or probability of photoabsorption, which
for a given wavelength decreases exponentially with the attenuation coefficient and
the depth of interaction x as

QEint = exp(x)
The attenuation coefficient is higher for shorter wavelengths (1.94 104 cm1 for
450 nm, 3.75 103 cm1 for 600 nm);17 therefore, the interaction takes place closer to
the detector surface. The intrinsic QE is over 95% for optical photons.
134
The QE is also affected by the reflectivity of the detector surface R, which can
be optimized for a given wavelength range by means of an antireflective coating.
Therefore, the QE for a given wavelength will be given by

QE = (1 R)QEint
The triggering probability Pt is the probability that the carriers created in the photon interaction initiate an avalanche. It is given by a combination of the individual
triggering probabilities of electrons and holes (Pe and Ph, respectively), and Pe and
Ph depend on the respective ionization coefficients e and h:10
Pt = Pe + Ph Pe Ph

with

dPe/dx = (1 Pe)e[Pe + Ph PePh]
dPh/dx = (1 Ph)h[Pe + Ph PePh]
The ionization coefficients increase with the electric field, and the dependence is
approximately linear at the electric field intensities found in SiPMs. Therefore, the
triggering probability increases linearly with the overvoltage.
Since the ionization coefficient is higher for electrons than for holes (roughly
e = 2h at 6 105 V/cm), the probability that the electrons initiate an avalanche is
higher. For a given detector structure and doping profiles, Pt depends strongly on the
position where the photon interacts and therefore on the photon wavelength, since
this determines whether it is the electrons of the holes that trigger the avalanche.
Given the different attenuation coefficients, the photons with short wavelengths
(blue) interact closer to the SiPM surface, while the photons with long wavelengths
(red) penetrate deeper. In n-on-p structures, if the interaction takes place beyond the
high-field region, the electrons will be the ones that drift to the high-field region and
produce an avalanche, while holes drift to the opposite side of the detector. Shortwavelength photons that interact close to the detector surface might interact in the
nondepleted n+ region, where the probability for the generated carriers to recombine
is high. If carriers reach the avalanche region, they will most probably be the holes.
Since the triggering probability of electrons is higher than that of holes, the overall
triggering probability will be better for longer wavelengths, and as a consequence,
the PDE will be higher in the red-green wavelength region. In p-on-n structures, the
situation is the opposite; therefore, the PDE will be higher for short wavelengths.
The geometrical efficiency or fill factor G is the ratio of the active to total
area of the microcell. A nonsensitive area around the microcell is due to the
presence of structures such as the guard ring, quenching resistor, or trenches that
prevent optical cross talk. The fill factor depends on the microcell structure and
size. A higher G can generally be obtained for a larger microcell size, thus leading to a higher PDE. However, it should be taken into account for a fixed SiPM
135
size that increasing the microcell size also results in a reduction of the dynamic
range of the device. The values typically range from 20% to about 80% and can
be even higher.
The recovery time of the microcell can also result in a reduction of the PDE in
case of intense light fluxes. This effect is better explained in the next section.
The PDE of SiPMs can be measured by two different methods. One consists of
the determination of the number of photons detected by measuring the direct current (DC) when the device is illuminated with a light source at a given wavelength
and subtracting the DC in dark conditions. The number of photons that arrive at the
SiPM Nph has to be determined with a calibrated detector. The gain of the SiPM G
must be previously known with high precision.

PDE = (Iillum Idark)/(G qe Nph)
where qe is the electron charge. With this method, the PDE can be overestimated,
since the current measured includes cross talk and afterpulse events. This effect
can be corrected by estimating an effective gain for the calculation that takes into
account these effects.
The second method consists of counting the rate of pulses above a threshold and
subtracting the dark count rate. This method does not require the measurement of
the gain. The agreement of the two methods indicates that the effects of afterpulses
and cross talk are within the errors. SiPMs with different structures have a PDE up
to 3045% in the 550- to 700-nm wavelength region or up to 2530% in the 400- to
550-nm region.6,7
6.3.2Dynamic Range
The dynamic range of an SiPM is determined by its PDE and by the number of
microcells and their recovery time. The ideal SiPM response R follows a Poisson
distribution given by3

R ~ Nfired = Ncells (1 e(Nph PDE/Ncells))
where Nfired is the number of microcells that trigger, Nph is the number of incident
photons in the SiPM surface, and Ncells is the total number of microcells of the SiPM.
The response is close to linearity if the number of photons that arrive at the detector
surface is much smaller than the number of microcells available. If the number of
photons is high compared to the number of microcells, the probability that a second
photon arrives at a microcell that has been previously hit and remains undetected
increases. The previous equation is represented in Figure 6.6 as a function of the
number of photons that arrive at the detector for 500 and 1,000 microcells and for
three different values of the PDE (10%, 20%, and 30%). The point at which the SiPM
saturation is significant depends mainly on these two parameters. The signal deterioration begins to occur for NphPDE 0.6 Ncells.18 SiPMs with a higher PDE will need a
higher number of microcells to have a linear dynamic range.
136

SiPM Response Curve
Number of microcells fired
700
600
1,000 microcells
500
500 microcells
PDE 30%
400
300
PDE 20%
200
PDE 10%
100
0
500
1,000
1,500
2,000
2,500
3,000
Number of incident photons
Figure 6.6 Theoretical response curve of an SiPM for a different number of microcells
and three values of the PDE. Saturation is observed for a high number of photons compared
to the number of microcells.
The microcell size is an important factor in the dynamic range of the SiPMs. As
explained, large microcell sizes favor a higher PDE, but this results in a smaller number of microcells per unit area and therefore in a reduced linear dynamic range.
In addition to the number of microcells, the recovery time has to be taken into
account. If a microcell is triggered when it is not totally recovered, the signal produced will be smaller, and the proportionality to the incoming number of photons
will be lost. Therefore, SiPMs with longer microcell recovery time will also have a
shorter dynamic range.
6.3.3Noise
Three different processes contribute to the noise in an SiPM due to the generation
of signals in the absence of light (dark rate). The main contribution, sometimes
known as the primary dark rate, comes from the generation of carriers in the device.
Afterpulses and optical cross talk can also contribute to the noise, up to 1020%
depending on the overvoltage and microcell geometry. The first two processes take
place in each microcell, while optical cross talk is a consequence of the interaction
between different microcells. The term dark rate is commonly employed to refer
to the carrier generation or primary dark rate, which is the dominant component.
However, the measurements generally include all types of noise. For most types of
devices, dark rate values at room temperature range from hundreds of kilohertz to
some megahertz per square millimeter at the single-photoelectron level and drop
rapidly to kilohertz at higher thresholds.
The primary dark rate is due to the generation of carriers in the depleted region
around the junction by thermal or field-assisted (tunneling) processes. Silicon is
137
an indirect semiconductor; therefore, the generation and recombination of carriers

from the conduction band to the valence band and vice versa are unlikely. However,
the existence of impurities introduces intermediate energy levels that facilitate the
process, increasing its probability. The energy is exchanged in the form of lattice
vibration, or phonon, exchanging thermal energy with the material. This process,
known as ShockleyHallRead (SHR), is the dominant generation and recombination process in silicon. The generation rate per unit volume is given by

G = ni/g
where ni is the intrinsic carrier concentration, and g is the generation lifetime, given
by 1 over the number of generation centers. A typical value of g of 10 ms in n-on-p
structures results in a dark rate of 2 MHz.
In the high-field regions, a tunneling effect takes place between bands and defect
energy levels or between the conduction and the valence band, which contributes to
the generation of carriers. The carriers generated can drift to the high-field region
and trigger an avalanche that cannot be distinguished from the signals generated by
the interaction of optical photons.
Since the number of thermally generated carriers is expected to be independent of the bias voltage, the variation of the primary dark rate with the bias voltage will be due to the variation of the triggering probability. The linear increase
of the dark rate with overvoltage thus confirms the linear variation of the triggering probability.
The dark rate depends strongly on the purity of the silicon in the epitaxial layer,
and it can be reduced by minimizing the number of generation-recombination centers, impurities, and crystal defects. Gettering processes can be employed to reduce
the number of impurities. The dark rate scales with the detector area, and it is the
main constraint in the operation of large-area devices.
The SiPM structure also has an effect on the dark rate. In n-on-p structures, the
avalanche is mainly triggered by the electrons produced in the depletion region, while
in p-on-n structures, it is triggered by the holes. Since the triggering probability is
higher for electrons, the dark rate will in general be higher in n-on-p structures.
Afterpulsing is the consequence of carriers that are trapped during an avalanche
and released shortly after it, triggering another avalanche. The probability of afterpulsing is related to the trap levels in the energy bandgap, and it decreases quickly with
time. More than 90% of the secondary pulses take place in some tens of nanoseconds
after the main pulse. The amplitude of the secondary pulse is lower than that of the
main pulses if the microcell is not fully recharged when the event takes place.12
The probability of having an afterpulse follows the equation
Pc e
Pafterpulse (t ) =
dt Pt
where Pc is the trap capture probability, which depends on the number of traps and
on the number of carriers flowing across the junction during an avalanche (provided
138
not all the traps are filled during a discharge) and increases linearly with the bias
voltage applied to the device; is the trap lifetime, which depends on its energy
level; Pt is the triggering probability, which depends on the recovery condition of
the microcell and increases linearly with the bias voltage. As a result, the afterpulse
probability depends quadratically on the overvoltage.
Optical cross talk is due to the generation of photons during the avalanche.11
Some of the photons generated can reach neighboring cells (either directly or after
a reflection at the back of the device)19 and might trigger a second avalanche
that takes place simultaneously with the original one. The production of photons
during an avalanche is due to three main processes: recombination of carriers,
bremsstrahlung of hot carriers in the Coulomb field generated by a charge impurity, and intraband transitions. Carrier recombination can explain the emission of
photons with energies above the energy gap, while intraband transitions can be
responsible for the lower energies. On average, three photons with energy higher
than 1.14 eV are emitted per 105 carriers crossing the junction. Since the number
of carriers flowing during the discharge increases with the bias voltage, the optical cross-talk probability also increases linearly with this parameter.
The emitted photons can reach neighboring cells, where they can trigger an avalanche. The geometry of the microcell has an influence on cross talk since small
microcells have less probability that a photon coming from another microcell interacts in them. In some cases, optical trenches are etched between the microcells to prevent the photons generated during the avalanche from reaching neighboring cells.
In general, the noise increases with the bias voltage applied to the detector and with
temperature. The temperature dependence of noise is explained in Section 6.3.5.
Figure 6.7 shows measurement of the noise of a 1 1 mm SiPM at different
voltages, obtained by counting the number of discriminated signals per unit time at
different values of the discriminator threshold. The three plateaus correspond to the
first, second, and third photoelectron levels.
6.3.4Time Resolution
SiPMs have an excellent intrinsic timing response due to the physical process of
the signal formation and to the reduced size of the detection region. The timing
resolution in these devices is determined by three parameters. The first parameter
is the avalanche propagation time, which is the time from the start of the avalanche
process until the entire junction is in breakdown. This time will depend on the position in which the avalanche initiates and whether this takes place in the center of the
microcell or in one corner, where it will take longer for the entire junction to be in
breakdown. However, given the small size of the microcell, the differences are small,
and the total avalanche propagation time is on the order of tens of picoseconds. The
second factor is the drift of the carriers in the depleted region. Given the high fields
in this region (>105 V/cm), the drift velocity is similar for electrons and holes, close
to 107 cm/s. In a region of a few microns, the drift time is around 10 ps. In addition,
one has to consider the diffusion of carriers generated in the nondepleted regions,
produced by a deep interaction of photons, which can also reach the drift region.
139

Dark Rate
V 4 V
V 3 V
V 2 V
Dark rate (Hz)
106
105
104
103
102
100
200
300
400
500
600
Threshold (mV)
Figure 6.7 Measured dark rate of a 1 1 mm SiPM at room temperature for different
values of the overvoltage. The three plateaus correspond to the first, second, and third photo
electron levels.
This effect contributes exponentially to the timing resolution, with a decay constant
given by

= L2/2 D
where L is the effective neutral-layer thickness, D is the diffusion coefficient, and it is

more significant for long-wavelength photons that penetrate deeper in the substrate.
This results in an excellent single-photoelectron resolution, about 120-ps full
width at half maximum (FWHM) at the single-photoelectron level.18,20 The time
response improves with the number of photons Nph as 1/sqrt(Nph). This behavior has
been verified up to 100 photons. Figure6.8 shows the variation of the intrinsic timing resolution with overvoltage for photons with 400- and 800-nm wavelength. The
lower curve indicates the estimated contribution of the electronics noise.
The excellent intrinsic timing resolution of SiPMs makes them the optimum candidates for timing applications, such as time-of-flight (TOF) measurements in highenergy physics or TOF positron emission tomography (PET) in medical imaging (see
Section 6.4.2). Timing measurements for TOF applications have been carried out
with a 3 3 mm SiPM coupled to a plastic scintillator BC418 of 3 3 40 mm size
and 1.4-ns decay time. The measurements were carried out with 3-GeV electrons.
The timing resolution obtained with such a detector was 75-ps FWHM, confirming
the possibility of using SiPMs for TOF experiments.21
Coincidence timing studies have also been performed with 3 3 10 mm
lutetium-yttrium orthosilicate (LYSO) scintillator crystals (40-ns decay time)
wrapped with white Teflon tape, coupled to 3 3 mm SiPMs and employing a 22Na
140

300
250
Sigma (ps)
200
150
100
50
3
4
Overvoltage (V)
Figure 6.8 Variation of the intrinsic timing resolution with overvoltage for photons of
400- (circles) and 800-nm (squares) wavelength. The lower curve indicates the estimated
contribution of electronics noise. (Reproduced from Collazuol, C. et al., Nucl. Instrum. A,
581, 2007. With permission from Elsevier.)
source to detect the two 511-keV photons. A coincidence timing resolution of 250-ps
FWHM has been achieved (i.e., about 177 ps per device).22
6.3.5Temperature Dependence
Temperature variations have a strong effect in SiPM performance. An increase of
the temperature increases the mobility of the carriers, which have a shorter mean
free path. The probability of scattering processes in which energy is dissipated is
higher; therefore, a higher electric field is necessary for the carriers to reach the necessary conditions to initiate the avalanche by impact ionization. This effect results in
an increase of the breakdown voltage with temperature, which modifies the overvoltage applied to the detector, and therefore in all its performance characteristics. The
increase of the overvoltage is about 0.3%/K.23
Therefore, for a fixed bias voltage applied to the detector, a higher temperature
results in a lower overvoltage, with a consequent decrease of the gain and the PDE.
Temperature variations are commonly expressed as a variation of the gain, since this is
the apparent consequence of the temperature changes. However, the microcell capacitance does not change with temperature; therefore, the gain is also temperature independent. If the variation in breakdown voltage is considered and the gain is expressed
in terms of real overvoltage at a given temperature, the measured gain is the same at
any temperature for a fixed overvoltage. The effects of the temperature variations will
therefore depend on the variations of the SiPM characteristics with overvoltage. The
141
reported values for the variation of the gain with temperature as a consequence of the
change in the breakdown voltage range from 0.3% to 4% per degree Kelvin.
Another significant effect of the increase of the temperature is the increase of the noise,
as mentioned. The variation observed in the dark current is given by the expression24

I ~ T 2 exp(Eg/2k BT)
where Eg is the bandgap energy, k B is Boltzmanns constant, and T is the temperature. For a fixed voltage, the observed change in current is about a factor of 2 for a
temperature variation of 8 degrees. The variation of optical cross talk with temperature is not significant, while the afterpulse probability decreases with temperature
due to the decrease of the trap lifetime. In those applications in which the detection
of low light level fluxes is required, a cooling system is necessary to reduce the noise
to levels that do not interfere with the photon detection.
Other properties can also be affected by temperature changes. The recovery time
of the microcells can also change with temperature due to the decrease of the resistance of the polysilicon quenching resistor.
6.3.6Insensitivity to Magnetic Fields

The working principle and small size of the microcells makes SiPMs insensitive to
magnetic fields. Tests have been carried out in static magnetic fields and with variable fields.
Tests with static magnetic fields up to 4 T have been performed for two orientations of the field lines. No variations in the stability of the SiPM or changes in their
characteristics were found to 1% accuracy.21
Tests have also been carried out by placing the SiPMs in a magnetic resonance
system, and data have been taken with the system on, and therefore with the SiPM
affected by rapidly changing gradient fields, and compared to the response when gradient fields were not active. Data were taken both at the single-photoelectron level and
coupling a scintillator crystal and acquiring an energy spectrum. In both cases, no difference was found when the SiPMs were affected by the variable magnetic fields.25
6.3.7Radiation Damage
Radiation hardness of SiPMs is an essential factor for accelerator and space physics applications. Radiation is known to produce defects in silicon; therefore, it is an
essential issue to determine how the characteristics of SiPMs change during irradiation.26 Radiation damage in silicon is dependent on the type and energy of the
radiation, with the two main effects surface damage (ionizing damage in the Si/SiO2
interface) and bulk damage (defects in the crystal lattice due to displacement of
silicon atoms). The surface damage is due to electronhole pairs created by ionizing
particles that can be trapped in very deep levels associated with the defects in the
oxide. Their emission back into the conduction or valence band is very unlikely at
room temperature, and the accumulation of positive charges in the SiO2 and Si/SiO2
interface can lead to the creation of parasitic fields. This can modify the breakdown
142
voltage, causing an early breakdown or unrecoverable damage, and can result in a

reduction of the PDE or an increase in the surface currents. The bulk damage scales
linearly with the amount of nonionizing energy loss (NIEL), which is very dependent
on the particle type and its energy (NIEL 1-MeV gammas ~ 10 5 NIEL 1-MeV neutrons). It can produce the formation of midgap states, which facilitate the transition
of electrons from the valence to the conduction band, with a consequent increase of
the dark current generated in the silicon bulk. The ionizing radiation can also cause
damage in the dielectric layer and as a consequence the failure of the device.
Studies have been performed with different types of devices exposed to several types of radiation.27,28 Irradiation with gamma rays (60Co, 240-Gy total dose)
resulted in a slight increase of the dark current up to 200 Gy, and above this dose a
drastic increase due to local high-amplitude dark pulses, which recovers after a short
time. This effect can be due to a local drop of the breakdown voltage due to chargetrapping effects in the SiO2 layer. With protons and neutrons, a significant increase
of the dark rate was observed after irradiation due to the induction of generation
centers. The afterpulse probability also increased due to the introduction of trapping
centers, contributing to the gradual loss of the single-photoelectron resolution until
it was completely lost (after 21-Gy proton irradiation). The variation of the gain and
PDE was less significant (<15%). The irradiation studies with both 53.3-MeV protons
up to a total fluence of 2.8 1010 protons/cm2 (42 Gy) and 212-MeV protons up to
3 1010 protons/cm2 showed that the leakage current and dark rate increased linearly
with fluence, and that the devices annealed at room temperature but did not recover
the original condition. In general, the devices remained functional as photon detectors after irradiation in the cases specified. In the case of irradiation with neutrons,
no significant changes were observed up to 3.3 107 neutrons/cm2, and for doses
higher than 1 1010 neutrons/cm2, the variations in the dark current were similar to
the proton case.
6.4Applications
SiPMs have many potential applications in physics, accelerator physics, space physics, neutrino physics, medical imaging, and so on, each with different requirements.
An essential advantage of SiPMs is the flexibility in their design, thanks to which
they can easily be adapted to the requirements of a particular application. As seen in
the previous sections, the SiPM structure and geometry influence its performance.
SiPMs are still a very active field of research that is ongoing in many aspects.
Among the main objectives are further enhancements of PDE in all wavelengths,
maximizing the fill factor, and developing structures that increase the triggering
probability. Another goal is the increase of the active area of the devices, together
with a reduction of the dark rate and cross talk. Existing SiPMs vary in size and
geometry (typical sizes range from 1 1 mm to 5 5 mm, with microcell sizes from
20 to 100 m), and SiPMs from different manufacturers have different characteristics and performance.7
An important aim is the development of one-dimensional (1D) and two-dimensional (2D) arrays of SiPMs that can cover large areas while providing position
information. Some manufacturers commercialize arrays of single SiPMs packed
143
Figure 6.9 Monolithic SiPM matrix composed of 4 4 elements of 1 1 mm fabricated

by FBK-irst.
together and develop methods to take the readout contact to the back of the detector
to simplify the coupling to scintillator crystals and readout electronics and to place
the SiPMs as close as possible to maximize the packing fraction. In 2006, the first
monolithic 2D arrays of SiPMs, consisting of four (2 2) SiPM elements of 1 1 mm
size in a common substrate were fabricated.29 Monolithic devices allow minimization of the dead space between the SiPM elements. More recently, other geometries,
with 1 4, 2 2, 4 4 (see Figure6.9), and 8 8 SiPMs of 1 1 mm or 3 3 mm
have been fabricated, with the same characteristics as single SiPMs and very good
uniformity among all pixels.30
The development of suitable electronics to fully exploit the benefits of this new
technology and that allow processing a high number of readout channels31,32 is also
a very active research field and is necessary for the further development and use of
these detectors.
There are already many applications in which SiPMs are the photodetector of choice
since their use can bring significant advantages over other types of photodetectors.
6.4.1High-Energy and Space Physics

A large number of experiments are now testing and considering the use of SiPMs, for
example, for calorimetry, neutrino experiments, fiber tracking, TOF measurements,
Ring Imaging Cherenkov (RICH) detectors, liquid xenon scintillation, Cherenkov
telescopes, and so on. In this section, some examples of the first applications that
have employed SiPMs and that have significantly contributed to their development
are mentioned.
144
One of the first applications of SiPMs has been their use in calorimeters for
high-energy physics experiments. SiPMs are being tested in tile calorimeter prototypes for the International Linear Collider (ILC).21 The physics program of the
ILC requires reconstructing hadronic final states of heavy boson (W, Z, H) decays
in multijet events. A jet energy resolution better that 30% sqrt (E) is necessary for
this purpose, which can be obtained if each particle in a jet is measured individually. To achieve this, the calorimeter must have very fine longitudinal and transverse
segmentation. For hadron shower separation, a transversal segmentation of about
3 3 cm is needed, and every sampling layer should be read out individually.
The required granularity can be achieved by the use of scintillator tiles with
wavelength-shifting fibers that are inserted in a groove of the tile and collect the light.
The fiber emission is in the green wavelength region and can be read out individually
by an SiPM on one side. A mirror on the other side of the fiber minimizes the light
losses. The first test prototype minical was successfully constructed and operated in
the German research center for particle physics, Deutsches Elektronen Synchrotron
(DESY) positron test beam at energies from 1 to 6 GeV. The CALICE collaboration,
an R&D group that develops new high performance detectors for high energy e+e
experiments, has constructed a tile hadron calorimeter prototype, HCAL, composed of
7,608 scintillator tiles, each individually read by a SiPM, and tested it in electron and
hadron beams at European Organization for Nuclear Research (CERN). The prototype is made of 38 layers of a plastic-scintillator-steel sandwich structure with a lateral
dimension of about 1 m2. Each layer consists of 1.6-cm thick steel absorber plates
and a plane of 0.5-cm thick plastic scintillator tiles housed in a steel cassette. The tile
sizes are 3 3 cm in the center of the module, 6 6 cm in an intermediate region,
and 12 12 cm in the outer region.33 The prototype has been exposed to beams of
electrons, hadrons, and pions of different energies. High-intensity muon beams were
also employed for calibration. The calibration of the prototype includes the study of the
response to minimum ionizing particles (MIPs) and the correction for SiPM saturation
and temperature variations. A MIP corresponds to about 15 photoelectrons. A 0.5-MIP
threshold for noise suppression results in 93% efficiency and a SNR of 9%. The test
beam results after calibration and correction showed a linear response within 4% up to
50 GeV. The energy resolution can be characterized as E/E = 19.6%/E 1.8% 0.27
GeV/E, in reasonable agreement with simulations. The granularity of HCAL allowed
observation of a rich substructure of hadronic showers. The test beam data made it
possible to study the capability of separating neutral hadron showers from the energy
deposition of nearby charged particles, which is essential for the reconstruction of jet
energies in the particle flow approach, and to compare the results to the simulations.
Another interesting application of SiPMs is their use in astroparticle experiments, either ground based such as Major Atmosferic Gamma Imaging Cherenkov
(MAGIC), or space missions like EUSO.34 MAGIC is the largest air Cherenkov
telescope, dedicated to the study of very-high-energy gamma-ray sources (active
galactic nuclei, supernovae remnants, gamma-ray bursts, and pulsars), from tens of
gigaelectron volts up to tens of teraelectron volts. Gamma rays interact with the
atmosphere, producing an electromagnetic cascade. Electrons propagating faster than
light emit Cherenkov photons, which are detected by the MAGIC telescope. The flux
145
is about 100 photons per square meter with wavelengths between 300 and 550 nm
for a 1-TeV gamma ray and scales up approximately linearly with the gamma-ray
energy. Therefore, large collection areas of fast, high-PDE detectors are necessary.
EUSO is a space mission that will aim at the detection of extreme energetic par
ticles (E > 1019 eV) through the detection of air showers produced in the atmosphere.
In the air showers, nitrogen molecules that may emit fluorescence photons of
wavelengths between 330 and 400 nm when decaying back to the ground state are
excited.
In both cases, a significant improvement of the SNR could be obtained by employing detectors with high PDE (>40%) together with high gain, fast timing, and singlephoton detection capability. The use of such detectors would allow setting a lower
energy threshold and improve the energy resolution. In addition to making it possible
to take data overlapping with other experiments, this would allow MAGIC to explore
the still-unobserved region below 30 keV, down to some gigaelectron volts.
The requirements mentioned make SiPMs an excellent option for these applications. Their compactness and low power consumption are also an enormous advantage in space experiments. With this aim, large-area SiPMs (currently 5 5 mm and
aimed at 10 10 mm and a combination of several devices) are being developed
and tested. Low noise is an essential requisite. Optical cross talk can be kept low by
maintaining a low gain and implementing optical trenches between the microcells,
while to keep the dark rate low, moderate cooling to 50C will be necessary in
ground-based experiments. The enhancement of the PDE above 50% is also foreseen
by the development of SiPMs with p-on-n structure with big microcells to increase the
fill factor and of backplane-illuminated SiPMs. As a test, four groups of four SiPMs
each have been mounted in the focal plane of the MAGIC telescope, and the analysis
of the signals indicated that Cherenkov light had been successfully recorded.15
6.4.2Medical Imaging
Medical imaging is a field in which the use of SiPMs can result in significant
improvements compared to current systems employing PMTs. In particular, PET has
been one of the first fields to consider SiPMs as an advantageous alternative.35 In this
application, a radiotracer containing a positron emitter such as the glucose analog
fluorodeoxyglucose (18F-FDG) is administered to the patient (or the animal in the
case of animal imaging for preclinical research), and it accumulates in the organ to
be imaged. The two 511-keV photons generated by the positron decay are detected
in time coincidence in two opposite detectors of the PET ring. The detection of the
position interaction and energy of both photons allows determining the region of
origin of the photons by image reconstruction methods.
In the development of PET detectors, the use of SiPM matrices composed of small
elements that provide position information can result in an excellent spatial resolution,
beyond the state of the art, both in preclinical and in clinical applications. Noise is not a
concern in PET since a high number of photons is detected in each event, and the fact that
the detectors are operated in time coincidence also contributes to reduce the noise. SiPM
saturation is also of no concern as long as the photo peak can be well separated from the
Compton continuum for noise reduction. The energy resolution obtained to date with
146
SiPMs coupled to LYSO crystals is about 1014% FWHM, comparable to that obtained
with PMTs. In the case of SiPM saturation, this effect must be taken into account for
an accurate measurement of the energy resolution. The potential low cost of SiPMs is an
essential requirement for the future commercialization of the instrumentation.
Several groups are investigating the use of SiPMs as an alternative to PMTs and
APDs and in novel configurations. Investigation is taking place mainly in the field of
small-animal PET and in some cases also for clinical scanners. Current PET detectors are based on pixellated scintillator crystals, with LYSO the one most commonly
employed, coupled to PMTs. Silicon photomultipliers can be employed to replace
PMTs both in the classical block detector configuration with light sharing among the
different photodetector elements and with one-to-one scintillator-photodetector coupling.36,37 However, the improvement of the resolution in pixellated configurations is
based on the reduction of the pixel size, which results in a loss of efficiency. A different approach that is gaining interest is the use of continuous scintillator blocks read
out by finely pixellated photodetectors.38 The use of SiPMs in this novel configuration
can result in an excellent spatial resolution, below 1 mm for small-animal PET. The
high granularity of the photodetector allows determination of the interaction position
with high resolution employing maximum likelihood or neural network methods,
and the use of continuous scintillators ensures high efficiency. In addition, the compactness of the scintillators makes it possible to stack several layers of detectors to
enhance the efficiency while providing discrete depth-of-interaction information.
A small-animal PET prototype with such characteristics has been proposed, with
a PET detector head composed of a stack of three continuous LYSO detector layers
read out by an SiPM matrix structure39 (Figure6.10). SiPM matrices are being tested
for this purpose with continuous LYSO crystals. Figure6.11 shows a 22Na energy
spectrum obtained with a 4 4 5 mm LYSO crystal coupled to an SiPM matrix
of 4 4 elements of 1 1 mm size (Figure6.9). The spectrum has been obtained
by adding the energy measured by all the elements that compose the matrix for
each event. The energy resolution obtained was 14.7%. Position determination studies have also been performed, resulting in an intrinsic spatial resolution of about
0.6-mm FWHM36 with center of gravity algorithms.
A highly innovative approach is the axial PET (AX-PET) project, a three-dimensional (3D) axial PET scanner that allows measurement of the interaction position
with very high precision.40 The detector heads are made of a matrix of long LYSO
crystals oriented in the axial direction, each coupled to an SiPM array. Wavelength
shifter (WLS) strips are mounted orthogonally and interleaved between the crystals
to provide the axial coordinate. The strips are also read out by SiPMs, and the position is obtained by the weighted average of the signals. The achievable resolution is
mainly driven by the dimensions of the LYSO crystals and WLS strips. The possibility of identifying Compton interactions in the detector will contribute to enhance
the detector efficiency. The detector components perform satisfactorily, and a demonstrator composed of two matrices of 8 6 LYSO crystals and 312 WLS strips is
under construction.
147
LYSO Crystals
SIPM Matrices
Figure 6.10 PET detector head consisting of three detector layers, each composed of a
continuous LYSO crystal and an SiPM matrix structure.
16-pixel SiPM matrix+ LYSO crystal. Na-22 energy spectrum

500
450
Energy resolution 14.7% FWHM
400
350
300
250
200
150
100
50
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
Sum ADC channel
Figure 6.11 The 22Na energy spectrum obtained with an SiPM matrix (Figure 6.9) coupled to a LYSO crystal. The signals of all the elements in the matrix are summed to obtain
the total energy of the event.
148
Their fast timing properties make SiPMs excellent detectors for TOF PET
applications. In this clinical PET modality, the difference in the arrival time
of the two photons is employed to have some information on the annihilation
region of the event, instead of only the line of response of the event, and thus
to reduce the noise in the reconstructed images. A 35-cm ring diameter system
with a timing resolution of 500-ps FWHM would be able to restrict the positron
emission position to a 7.5-cm line and thus to reduce the statistical variance by a
factor about 5,41 and for 200-ps timing, the line is restricted to 3-cm length. The
coincidence timing resolution of 250-ps FWHM achieved with LYSO crystals
makes these devices extremely promising candidates for the implementation of
this technique.
The compactness and insensitivity to magnetic fields of the SiPMs are basic
characteristics for the combination of PET and magnetic resonance (MR) imaging
modalities. PET is a molecular imaging technique with excellent sensitivity, but it
is not able to provide anatomical information. The combination of PET and computed tomography (CT) that provides anatomical images has already shown a highly
improved diagnostic value. MR can provide anatomical information complementary
to PET, with a better resolution and higher soft tissue contrast, while avoiding the
radiation doses of CT. The combination of PET and MRI has been hindered because
of the sensitivity to magnetic fields of PMTs. Different solutions have been developed with light guides that carry the light of the scintillators to the PMTs placed
outside the magnetic field or employing APDs. Different groups are already investigating SiPMs to make use of their advantages in this field in both preclinical and
clinical applications. The HyperImage project42 is developing a system for simultaneous PET/MR imaging for humans with MR-insensitive detectors employing silicon
photomultipliers.
The good performance and significant advantages achieved with SiPMs, together
with their potential for further development, make these photodetectors excellent
candidates for many applications and provide extremely interesting opportunities in
different research fields.
Acknowledgments
I would like to thank Gianmaria Collazuol and Claudio Piemonte for their help in
understanding silicon photomultipliers, and Prof. Alberto Del Guerra and Giuseppina
Bisogni for their support.
References

1. R. J. McIntyre, Theory of microplasma instability in silicon, J. Appl. Phys., 32(6),

983995, 1961.
2. R. Haitz, Model for the electrical behavior of a microplasma, J. Appl. Phys., 35(5),
13701376, 1964.
3. Z. Y. Sadygov, I. M. Zheleznykh, N. A. Malaknov, V. N. Jejer, and T. A. Kitillova,
Avalanche semiconductor radiation detectors, IEEE Trans. Nucl. Sci. NS-43(3),
10091013, 1996.

149
4. G. Bondarenko, B. Dolgoshein, V. Golovin, A. Ilying, R. Klanner, and E. Popova, Limited

Geiger-mode silicon photomultiplier with very high gain, Nucl. Phys. Proc. Suppl. 61B,
347352, 1998.
5. V. Saveliev and V. Golovin, Silicon avalanche photodiodes on the base of metal-resistorsemiconductor (MRS) structures, Nucl. Instrum. Methods A, 442, 223229, 2000.
6. Y. Musienko, Advances in multipixel Geiger-mode avalanche photodiodes (silicon photomultipliers), Nucl. Instr. Methods A, 598, 213216, 2009.
7. N. Dinu, Z. Amara, C. Bazin, V. Chaumat, C. Cheikali, G. Guilhema, V. Puill, C. Sylvia,
and J. F. Vagnucci, Nucl. Instr. Methods A, 610, 423426, 2009.
8. V. Golovin and V. Saveliev, Novel type of avalanche photodetector with Geiger mode
operation, Nucl. Instr. Methods A, 518, 560564, 2004.
9. C. Piemonte, A new silicon photomultiplier structure for blue light detection. Nucl.
10. W. Oldham, R. R. Samuelson, and P. Antognetti, Triggering phenomena in avalanche
diodes. IEEE Trans. Elec. Dev., ED-19(9), 10561060, 1972.
11. A. L. Lacaita, F. Zappa, S. Bigliardi, and M. Manfredi, On the bremsstrahlung origin
of hot-carrier-induced photons in silicon devices. IEEE Trans. Electron Dev., 40, 577,
1993.
12. C. Piemonte et al., Characterization of the first prototypes of silicon photomultiplier
fabricated at ITC-irst, IEEE Trans. Nucl. Sci., 54(1), 236244, 2007.
13. Z. Sadygov, A. Olshevski, I. Chirikov, I. Zhelznkh, and A. Novikov, Three advanced
designs of micro-pixel avalanche photodiodes: their present status, maximum possibilities
and limitations, Nucl. Instrum. Methods A, 567, 7073, 2006.
14. J. Ninkovic et al., The avalanche drift diodea back illumination drift silicon photo
multiplier, Nucl. Instr. Methods A, 580, 10131015, 2007.
15. R. Mirzoyan et al., The crosstalk problem and the SiPMs for the 17m diameter MAGIC
Telescope Project, Nucl. Instrum. Methods A, 61, 131134, 2009.
16. D. Renker, Geiger-mode avalanche photodiodes, history, properties and problems, Nucl.
17. R. Hull (Ed.), Properties of Crystalline Silicon, DataReviews Series No. 20, INSPEC,
IEE, London, 1999.
18. P. Buzhan, B. Dolgoshein, et al., An advanced study of silicon photomultiplier, ICFA
Instrum. Bull., 23, 28, 2000.
19. I. Rech et al., Optical crosstalk in single photon avalanche diode arrays: a new complete
model, Opt. Express, 16, 83818394, 2008.
20. G. Collazuol et al., Single photon timing resolution and detection efficiency of the IRST
silicon photo-multipliers, Nucl. Instrum. Methods A, 581, 461464, 2007.
21. Status report on silicon photomultiplier development and its applications, Nucl. Instr.
Methods A, 563, 368376, 2006.
22. C. L. Kim, G. C. Wang, and S. Dolinsky, Multi-pixel photon counters for TOF PET
detector and its challenges, 2008 IEEE NSS MIC Conf. Record, CD-ROM M023,
35863590, 2008.
23. S. Cova, M. Ghioni, A. Lacaita, C. Samori, and F. Zappa, Avalanche photodiodes and
quenching circuits for single-photon detection, Appl. Optics, 35(12), 19561976, 1996.
24. H. Kanbe, G. Grasskopf, O. Mikasmi, S. Machida, Dark current noise characteristics and
their temperature dependence in germanium avalanche photodiodes, IEEE J. Quantum
Elec., QE-17(8), 15341539, 1981.
25. R. Hawkes et al., Silicon photomultiplier performance tests in magnetic resonance pulsed
fields, 2007 IEEE NSS-MIC Conf. Record, CD-ROM, M18118, 34003403, 2007.
26. Y. Musienko, D. Renker, S. Revcroft, R. Scheuerminn, A. Stoyko, and J. Swain,
Radiation damage studies of multipixel Geiger-mode avalanche photodiodes, Nucl.
150
27. P. Bohn et al., Radiation damage studies of silicon photomultipliers, Nucl. Instrum.
Methods A, 598, 722736, 2009.
28. T. Matsumura for the KEK Detector Technology project, Radiation hardness of multipixel photon counters (MPPC), Presented at IEEE NSS-MIC, Honolulu, HI, 2007.
29. G. Llosa et al., Novel silicon photomultipliers for PET applications, Presented at 2006
IEEE NSS MIC and published in IEEE Trans. Nucl. Sci., 55(3), 877881, 2008.
30. N. Dinu et al., Characterization of a prototype matrix of silicon photomultipliers (SiPMs),
presented at NDIP08 conference, Nucl. Instrum. Methods A, 61, 101104, 2009.
31. F. Corsi et al., Modelling a silicon photomultiplier (SiPM) as a signal source for optimum front-end design, Nucl. Instrum. Methods A, 572, 416418, 2007.
32. M. Bouchel, F. Dulucq, J. Fleury, G. Martin-Chassard, and L. Raux, SPIROC (SiPM
integrated readout chip): dedicated very front-end electronics for an ILC prototype hadronic calorimeter with SiPM read-out, 2007 IEEE NSS MIC Conf. Record, CD-ROM,
N295, 18571860, 2007.
33. N. Wattimena, The CALICE Tile hadron calorimeter prototype with SiPM readout: design, construction and first test beam results, Proceedings of 10th ICATPP
Conference on Astroparticle, Particle, Space Physics, Detectors and Medical Physics
Applications, 2007.
34. A. N. Otte, B. Dolgoshein, J. Hose, S. Klemin, E. Lorenz, G. Lutz, R. Mirzoyan,
E. Popova, R. H. Richer, L. W. J. Struder, and M. Teshima, Prospects of using silicon
photomultipliers for the astroparticle physics experiments EUSO and MAGIC, IEEE
Trans. Nucl. Sci., 53(2), 636640, 2006.
35. D. Herbert et al., First results of scintillator readout with silicon photomultiplier, IEEE
Trans. Nucl. Sci., 53(1), 389394, 2006.
36. G. Llosa et al., Evaluation of the first silicon photomultiplier matrices for a small animal
PET scanner, 2008 IEEE NSS MIC Conf. Record, CD-ROM (2008).
37. M. Gtlich, Application of Multi-Pixel Photon Counter to Positron Emission Tomography,
2008 IEEE NSS MIC Conf. Record, CD-ROM. M021 (2008).
38. S. Tavernier et al., A high-resolution PET detector based on continuous scintillators,
39. S. Moehrs et al., A detector head design for small animal PET with silicon photomultipliers (SiPM), Phys. Med. Biol., 51, 11131127, 2006.
40. A. Braem et al., Wave length shifter strips and G-APD arrays for the read-out of the
z-coordinate in axial pet modules, Nucl. Instr. Methods A, 586, 300308, 2008.
41. W. W. Moses, Recent advances and future advances in time-of-flight PET, Nucl. Instrum.
Methods A, 580, 919924, 2007.
42. HyperImage simultaneous PET-MR imaging, http://www.hybrid-pet-mr.eu.
Photodetectors
7 Hybrid
(HPDs) for SinglePhoton Detection
Atsuhito Fukasawa
Hamamatsu Photonics K.K.

Shizuoka, Japan
Contents
7.1 Introduction................................................................................................... 151
7.1.1 Photon Detectors for Low-Light Detection....................................... 152
7.1.2 Operational Principle of the HPD..................................................... 153
7.1.3 History of HPD Devices.................................................................... 154
7.2 High-Speed HPD........................................................................................... 156
7.2.1 Design Concept.................................................................................. 156
7.2.2 Structure............................................................................................ 156
7.2.3 Basic Characteristics......................................................................... 157
7.2.3.1 Photocathode Quantum Efficiency..................................... 157
7.2.3.2 Gain Characteristics............................................................ 158
7.2.3.3 Temperature Characteristics............................................... 160
7.2.3.4 Pulse Height Distribution.................................................... 161
7.2.3.5 Time Response.................................................................... 162
7.2.3.6 Timing Resolution.............................................................. 162
7.2.3.7 Characteristics of Afterpulse.............................................. 165
7.2.3.8 Lifetime Characteristics...................................................... 166
7.2.4 Application Examples........................................................................ 166
7.3 Other Types of HPD...................................................................................... 166
7.4 Summary....................................................................................................... 168
References............................................................................................................... 168
7.1Introduction
Research and development in photon detection are rapidly spreading. In radiation
and fluorescence detection for biological applications, attention is being focused on
photodetectors that can detect single photons with accuracy and high speed. In this
chapter, various hybrid photodetectors (HPDs) developed as highly sensitive photodetectors are discussed.
151
152

Photocathode
Glass bulb
Stem
e
Photons
Secondary electrons
Dynode chain
Anode
Figure 7.1 Sectional view of a photomultiplier tube (PMT).
7.1.1Photon Detectors for Low-Light Detection

Under low-light conditions, a photomultiplier tube (PMT) is used as a common photodetector. A PMT consists of a photocathode, dynode chain, and anode. A structure
of a conventional PMT is shown in Figure7.1.1 The package of a PMT is made of
glass, in the case of bialkali photocathode; a thin layer of potassium and cesium is
evaporated on the inside of the glass bulb. When photons enter the photocathode,
electrons are emitted from them into the vacuum. Electrons are focused first on the
dynode by the electric field formed in the vacuum; then, electrons are multiplied in
the following dynode chain by a factor of 106 to 107 and are finally taken out as an
output signal. PMTs are useful devices for various applications to detect single photons because of their excellent performance, such as fast time response in the time
scale of nanoseconds, enough high gain to detect a single photon by a conventional
oscilloscope, large effective area, and low dark current. However, it is difficult to
make multipixel-type PMT detectors because of the complicated structure of the
dynode. In addition, PMTs also have a disadvantage in that they cannot be used in a
strong magnetic field.
On the other hand, the development of photodetectors based on semiconductor
devices has advanced in recent years. A typical detector of this type is a photodiode
(PD), and it is used for detection of light with high intensity. For low-light-intensity
conditions, an avalanche photodiode (APD) is used, since it has additional gain due
to the avalanche effect within the depletion layer of the diode. The sectional outline of the APD is shown in Figure7.2. Generated electronhole pairs are accelerated in the high electric field creating electronhole pairs. This avalanche process is
repeated; as a result, the number of original electronhole pairs is multiplied up to
the order of hundreds.
Silicon multipliers (SiPMs) operating under Geiger mode that can detect a single
photon as a PMT have been developed. These photodetectors are relatively cheap
compared to PMTs, since their manufacturing process is more suitable for mass
production. In addition, the SiPM detectors can be operated under strong magnetic
fields, making them promising candidates for magnetic resonance imaging/positron
emission tomography (MRI-PET). On the other hand, their effective area is small
and dark current is high compared to PMTs.
153
Hybrid Photodetectors (HPDs) for Single-Photon Detection

Photons
p+
n+
Figure 7.2 Sectional view of an avalanche photodiode (APD).
7.1.2Operational Principle of the HPD

The typical structure is shown in Figure7.3.2
Like the PMT, the HPD has both photocathode and multiplier, but it also has a PD
or an avalanche diode (AD) as a multiplier for electrons instead of the dynodes used
in PMTs. The HPT represents merging of vacuum tube and semiconductor technologies, hence the name hybrid photodetector.
A ceramic or glass bulb is used as an insulator between the photocathode and
AD because acceleration of electrons needs voltage as high as 10 kV to achieve
enough device gain. Emitted electrons from the photocathode are accelerated
to about 10 keV by the static electric field formed in the tube and are multiplied
inside the AD when they are bombarded on it. When electrons lose their energy
by sudden deceleration, new electronhole pairs are generated. This process
is called electron bombarded multiplication. In the case of silicon diodes, one
Photocathode
window glass
Output pin
Photons
Photoelectron
Signal out
Photocathode
Vacuumed package
Figure 7.3 Sectional view of a hybrid photodetector (HPD).
Avalanche diode
154
electronhole pair is generated by every 3.6-eV energy loss. The gain obtained
in this process is called the electron bombarded gain G eb, and it is given by the
following formula:
Geb = (HV Vth)/3.6
Here, HV is the photocathode applied voltage, and Vth is the threshold voltage determined by the thickness of the dead layer in the electron input surface of the AD, with
its value typically a few kilovolts.
If a target for electrons is an AD, electrons generated by electron bombarded
multiplication drift to the avalanche region, where the electric field is strong
(~3 108 V/cm), and then electrons are multiplied by the avalanche multiplication
process. The total gain of an HPD is given by the following formula:
G total = Geb Gad
where Gad is a gain by the avalanche multiplication process.

In the case of Hamamatsu R9792U-40 (18 mm in diameter gallium arsenic phosphorus [GaAsP] photocathode, which incorporates a 3-mm diameter AD), the electron bombarded gain is about 1,500, and the avalanche gain is about 50100, with a
photocathode voltage of 8 kV and AD reverse bias voltage less than the breakdown
voltage applied. As a result, total gain is 75,000150,000. Considering the typical
PMT gain of 106, the gain of HPDs is 1/10 compared to the PMTs. For this reason,
it is important for detection of a single photon to combine a HPD with a low-noise
amplifier. Another advantage in a combination of a photocathode and a PD is that
there is no degradation of time response due to the photodetector having a large
effective area. In other words, even if the photodetector has an effective area as large
as 1020 inches, the inherent characteristics of the PD can be preserved by gathering
electrons emitted from the large area by an electron lens and subsequently smashing
electrons into an AD having a small effective area.
There are additional advantages resulting from the simple structure of the HPD
compared to a conventional photodetector. For example, the vacuum inside the tube is
very high, since there are fewer parts compared to PMTs. As a result, there is low afterpulse probability and long life. Moreover, cost can be low. Another advantage of the
HPD is excellent pulse height resolution. Since the gain in the first stage (electron bombarded gain) is high, the fluctuation during the multiplication is very small. As a result,
photoelectron peaks for the irradiation of multiple photons can be clearly identified.
7.1.3History of HPD Devices

A history of HPDs is well summarized in the doctoral thesis written by M. Suyama:
Development of a Multi-pixel Photon Sensor with Single-Photon Sensitivity. The
following text is excerpted from the thesis.
A development of a HPD having the above characteristics can be traced back to 1960s.
In early 1960s, it was known that semiconductor device has sufficient sensitivity for
155
charged particles such as electrons. These particles generate electronhole pairs as

they lose their energy in the device. When the incident particles are low energy electrons, most of their kinetic energy is absorbed in the semiconductor device, and the
process generates electronhole pairs corresponding to their energy and the number.
Using this electron multiplication mechanism, a photo detector with combination of
a photocathode and a semiconductor device was proposed by F. A. White and J. C.
Sheffield in 1962.4
The first HPD with silicon (Si) diode was developed by R. Kalibjian in 1965.5 In this
case 5 kV voltage was applied to the photocathode, emitted electrons from photocathode were accelerated, and were focused on the Si diode of 9.5 mm in diameter by using
electrostatic field. The goal of this development was to achieve wide dynamic range,
which PMTs could not achieve due to space-charge effects in a vacuum. Although
detection of single photon was not object of studies in those days, P. Chevalier detected
single photon in 1967.6 In his experiment, the output signal corresponding to 13
photoelectron was clearly observed.
In 19717 and 1972,8 a HPD called Digicon with multi pixel diode was reported by
E. A. Beaver and C. E. McIlwain. This HPD had a diode of 38-channel linear array
with pixel size of 8 mm 0.1 mm, and side wall with multi layered structure by piled
electrode to accelerate electrons and glass of ring by turns. Emitted electrons from
photocathode were accelerated to 1530 keV and they were magnetically focused on
the diode by using a solenoid coil. In 1975, R. G. Tull et al. also reported this type of
HPD with monolithic self-scanned diode of 1024-channel.9 Output signal corresponding to single photon was clearly observed in these HPD. In spite of such excellent
characteristics, this type of HPD was not widely used in the world, because of complicated system required for extremely high operating voltage and solenoid coil for
electron focusing.
About 10 years later, R. DeSalvo re-developed HPD for particle physics experiments in 1987.10 L. K. Greest and K. W. Stoop also developed HPD of an electrostatic
focusing type, which had a diode of 12 mm in diameter, by using image intensifier technology.11 An excellent S/N ratio for multiplication of electrons was observed
in HPD with electrostatic focusing, and reported by C. Datema et al.12 Johansen and
Johnson also reported HPD developed for particle physics experiments in 1993.13
Subsequently, M. Suyama et al. developed small size HPD, which had photocathode
of 8 mm in diameter.14
In addition to pure HPD development several devices integrated with AD diodes
were also reported. In 1977, J. P. Choisser demonstrated advantage for the integrated
HPD-AD device.15 In 1993, P. Cushman evaluated proximity focusing type of the HPD
with beveled-edge type of AD.16 This HPD-AD had extremely high avalanche gain of
about 1000. In 1997, a HPD with single pixel AD was developed by M. Suyama et al.14
The HPD had multialkali photocathode of 8 mm in diameter and AD of 3 mm in diameter. Total gain was 65000, where the photocathode voltage of 8 kV and the AD reverse
bias voltage of 150 V were used. Rise and fall time were 1.2 ns and 13 ns, respectively.
In 2000, R. Mirzoyan et al. reported possibility of using this HPD for application in
Cherenkov telescope.17 S. Matsui et al. also reported possibility for time-of-propagation (TOP) counter in BELLE experiment.18 In recent development, Fukasawa et al.
developed a new HPD with high speed and high timing resolution in 2008, and they
reported that this HPD had high timing resolution match for MCP-PMT.19 Moreover,
X. Michalet et al. reported that this device had extremely low after pulse probability
compared to single photon counting APD operating under Geiger mode.20
156
7.2High-Speed HPD
7.2.1Design Concept
In various experiments in physics, chemistry, biology, and medicine, photodetectors are expected to detect extremely weak and fast optical signals. Various types of
photodetectors with fast response and timing resolution for single-photon detection
have been developed.
Multipixel photon counters (MPPCs) and SiPMs,21,22 developed based on semiconductor technology, are hot research areas for the high-energy physics field because
of their good timing resolution (on the order of a few hundred picoseconds) and
pulse height resolution for single photons. Since this type of detector can be used
in a strong magnetic field, it is expected to be used in MRI-PET as well. However,
their effective area is relatively small compared to the PMT, and the dark noise rate
is inherently high (on the order of 100 kHz at room temperature).
As a photodetector having excellent timing resolution, microchannel plate (MCP)
PMTs, which incorporate the MCP instead of dynodes, have been developed. Owing
to the small transit time spread, this type of detector shows extremely high timing
resolution (about 10 ps), although its maximum count rate and the detection efficiency are limited.
A third type of photodetector with excellent performance for single-photon detection is the HPD. As mentioned, an HPD incorporates a diode or AD in a vacuum
tube as a target for emitted photoelectrons from the photocathode. This device offers
both fast time response and a large effective area. The fast response time is due to
direct integration of a low-capacitance AD and a multiplying tube, and adoption of
an electrostatic focusing lens leads to a large effective area. By making full use of its
characteristics, this HPD can be used in various applications, such as light detection and ranging (LIDAR), PET scanning in nuclear medicine, and time-correlated
single-photon counting (TCSPC).
7.2.2Structure
The high-speed HPD is constructed from a photocathode, an AD, a cylindrical
ceramic sidewall, and a stem as illustrated in Figure7.4. In the example shown, a
bialkali photocathode 6 mm in diameter was fabricated on a plano-concave faceplate, where the curvature was selected to adjust the transit time of electrons from
the photocathode to the AD to be almost equal over the photocathode region. In
addition, a GaAsP photocathode was developed to achieve high quantum efficiency
for visible light around 500-nm wavelength. In the case of the GaAsP photocathode,
a flat faceplate was used because it was difficult to fabricate the photocathode on a
plano-concave faceplate. To maintain high timing resolution, the effective area was
limited to 3 mm in diameter.
For achievement of high speed, an AD with very low capacitance was developed.
The capacitance is estimated to be 3.4 pF for the effective area of 1 mm in diameter
and suggests a rise-and-fall time of 0.4 ns into a 50- load. A sub miniature type A
(SMA) connector was used so the fast signal did not deteriorate. A photograph of

Photons
157
Electron
Bialkali photocathode
Plano-concave faceplate
Electrode for electron lens
Cylindrical ceramic
Electron bombarded gain
Avalanche diode
Avalanche gain
Stem
Photocathode
voltage
SMA connector
AD reverse
Bias voltage
Output
Figure 7.4 High-speed HPD consisting of a faceplate, an AD, a cylindrical ceramic sidewall, and a stem. To produce an electron lens in the vacuum for focusing electrons to the
small AD, 8 kV is applied to the photocathode. (Reproduced from Fukasawa, A. et al., IEEE
Trans. Nucl. Sci., 55, 2, 2008. Copyright 2008 IEEE; with permission.)19
the potted tube is shown in Figure7.5. In device operation, 8 kV is supposed to be

applied to the photocathode and about 400 V to the AD. In response to incident light,
photoelectrons are emitted from the photocathode, accelerated, and focused on the
AD of grand potential by an electrostatic focusing lens made by the electrode in the
tube. The number of electrons is multiplied by 104 105 in the AD through electron
bombarded multiplication and the avalanche multiplication process.
7.2.3Basic Characteristics
7.2.3.1Photocathode Quantum Efficiency
Most photocathodes are made from alkali metals with low work function; several
types are available on the marketplace. In the case of high-speed HPDs, two kinds
of photocathode types have been developed. One is a bialkali photocathode, which
consists of two alkali materials, such as potassium or cesium. Because of its high
sensitivity in blue light, a typical application of this photocathode is the detection
of radiation using an NaI (TI) scintillator. A GaAsP photocathode is the other;
the GaAsP crystal is activated by cesium in a vacuum. It has the advantage of a
high quantum efficiency of about 50% for visible light around 500-nm wavelength.
The GaAsP photocathode is suitable for applications such as fluorescence lifetime
imaging (FLIM) and fluorescence correlation spectroscopy (FCS). The quantum
158

SMA connector
HV cable
AD cathode pin
AD anode pin
20 mm
Figure 7.5 A rear view of high-speed HPD. To achieve a fast time response, an SMA connector is hermetically sealed to the stem. (Reproduced from Fukasawa, A. et al., IEEE Trans.
Nucl. Sci., 55, 2, 2008. With permission.)19
50
Quantum Eciency (%)
45
40
35
30
25
20
15
10
5
0
200
300
400
500
600
Wavelength (nm)
700
800
Figure 7.6 Quantum efficiencies for bialkali (dotted line) and GaAsP photocathodes
(solid line) as a function of wavelength. The peak quantum efficiencies are 34% at 350 nm
with the bialkali photocathode and 46% at 500 nm with GaAsP photocathode. (Reproduced
from Fukasawa, A. et al., IEEE Trans. Nucl. Sci., 55, 2, 2008. With permission.)19
efficiency of the photocathode, as a function of wavelength, was measured by the

photocathode current in response to calibrated input light. The results are shown in
Figure7.6. Those measured efficiencies were 34% at 350 nm and 46% at 500 nm for
the bialkali and GaAsP, respectively.
7.2.3.2Gain Characteristics
As mentioned, there are two gain stages: electron bombarded gain and avalanche
gain. The total gain of an HPD is the product of these two gains. The electron bombarded gain was measured as the ratio of the photocathode current to input current
and output current of the AD at the unity avalanche gain. The result is shown in
159
Electron Bombarded Gain
2000
1500
1000
500
0
10
Photocathode Applied Voltage (-kV)
Figure 7.7 The electron bombarded gain as a function of the photocathode voltage. The
gain was approximately 1,600 at the photocathode voltage of 8 kV. (Reproduced from
Fukasawa, A. et al., IEEE Trans. Nucl. Sci., 55, 2, 2008. With permission.)19
Figure7.7 as a function of the photocathode voltage. The electron bombarded gain is

determined by energy of electrons and is proportional to it beyond the threshold voltage. The threshold voltage is determined by the dead layer existing on the surface of
the AD and was evaluated to be approximately 3 kV in this case. The electron bombarded gain was measured as about 1,600 at the photocathode voltage of 8 kV.
The avalanche gain was measured as a ratio of the output current of the AD to that at
the unity avalanche gain by varying the applied bias voltage to the AD. The avalanche
gain was determined by the AD bias voltage, since the avalanche multiplication depends
on the strength of an electric field formed inside the diode. Measured avalanche gain is
350
Avalanche Gain
300
250
200
150
100
50
0
100
200
300
400
500
AD Reverse Bias Voltage (V)
Figure 7.8 The avalanche gain as a function of the AD voltage shows a gain of approximately 110 at 405 V. The total gain is the product of the electron bombarded gain and the
avalanche gain and reaches 180,000. (Reproduced from Fukasawa, A. et al., IEEE Trans.
160
shown in Figure7.8. In the given sample, it was measured to be about 110 at the reverse
bias voltage of 405 V, which is the maximum applicable voltage within the breakdown
voltage for this AD. As a result, the total gain was approximately 180,000 as the product of the electron bombarded and avalanche gains. In real operation, the temperature
coefficient of the avalanche gain should be taken into account. The temperature coefficient of the AD is large when operated under high avalanche gain. To avoid this effect,
it is preferable to precisely control the temperature of the AD.
7.2.3.3Temperature Characteristics
The avalanche gain and breakdown voltage of an AD are influenced by the ambient temperature of the HPD. As the ambient temperature becomes high, the lattice
vibration inside the AD increases, leading to a decrease in the avalanche gain as
illustrated in Figures7.9 and 7.10. The change of the avalanche gain is high, especially at high reverse bias voltage. Therefore, great care should be taken not to break
90
320 V
360 V
382 V
404 V
80
Avalanche gain
70
60
50
40
30
20
10
0
10
10
20
30
Temperature (degrees centigrade)
40
50
Figure 7.9 The avalanche gain as a function of the ambient temperature.
Breakdown voltage (V)
430
425
420
415
410
405
400
10
10
20
30
40
50
Temperature (degrees centigrade)
Figure 7.10 The breakdown voltage as a function of the ambient temperature.
161
the avalanche diode. A module of the HPD with compensation for temperature characteristics is now under development in Hamamatsu Photonics K.K.
7.2.3.4Pulse Height Distribution
The output pulse height distribution of the HPD was measured to evaluate the distribution at an incident light level of approximately 10 photons per pulse on average (a few
photoelectrons from the photocathode). This measurement was made under the total
gain of 75,000, for which the photocathode and AD reverse bias voltages were 8
kV and 380 V, respectively. The light source was a light-emitting diode (LED) with
wavelength of 470 nm. The output signal from the HPD was fed to the charge-sensitive
amplifier (580 K, Clear-Pulse, charge sensitivity 2 V/pC), followed by the shaping
amplifier (3100-02, Canberra, 500-ns shaping time) to be analyzed by the multichannel analyzer (MCA; 2100C, Laboratory Equipment Corp.). The experimental setup
for pulse height resolution is shown in Figure7.11. Up to six photoelectron peaks were
clearly identified, as shown in Figure7.12. This is typical for the performance of HPDs.
As indicated by this result, gain fluctuation of HPDs is much less than that of PMTs.
LED
HPD
Amp1
Amp2
MCA
PC
Figure 7.11 Experimental setup for pulse height resolution measurements.

300
Frequency
250
200
150
100
50
0
500
1000
1500 2000 2500

Output Pulse Height
3000
3500
4000
Figure 7.12 The pulse height spectrum of the HPD for multiple photons clearly shows
peaks corresponding to up to six photoelectrons. Voltages of 8 kV and 380 V were applied
to the photocathode and the AD, respectively. (Reproduced from Fukasawa, A. et al., IEEE
Trans. Nucl. Sci., 55, 2, 2008. With permission.)19
162

60
Output Voltage (mV)
50
40
30
20
10
0
10
Time (ns)
Figure 7.13 The output waveform for impulse light shows rise and fall times of 350 and
460 ps, respectively. The load impedance of the oscilloscope was 50 , and the bandwidth
was 20 GHz. (Reproduced from Fukasawa, A. et al., IEEE Trans. Nucl. Sci., 55, 2, 2008. With
permission.)19
7.2.3.5Time Response
Time response of the developed high-speed HPD was measured using a picosecond
light pulser (PLP10-040, Hamamatsu) as a light source, which has a pulse width of
30 ps (one sigma) at 405 nm. This pulse width is fast enough to be regarded as an
impulse for the high-speed HPD. The output pulse from the HPD was measured by
a high-speed oscilloscope (TDS8000B sampling oscilloscope and 80E03 sampling
module, Tektronix) with a bandwidth of 20 GHz. The rise and fall times were 350
and 460 ps, respectively, as shown in Figure7.13. Time response of the high-speed
HPD is determined by the 3.4-pF capacitance of the AD, which leads to the estimated value of rise and fall times Trf of 370 ps with a load resistance of 50 .
7.2.3.6Timing Resolution
The MCP-PMT is a better photodetector than the PMT in terms of timing resolution.
As discussed, the MCP-PMT consists of a photocathode, MCP, and anode. Electrons
emitted from the photocathode smash into the channels of the MCP, and electrons
are multiplied by a secondary electron emission process; then, electrons are taken
out as signals. Since strong electric fields are formed in parallel from photocathode
to the MCP input and MCP output and anode, the emission angle distribution and
initial velocity distribution are almost negligible. The electron transit time of the
MCP-PMT is extremely short compared to the PMT. The measured timing resolution
is about a few tens of picoseconds in the MCP-PMT with the alkali photocathode.1
From the operation principle, the timing resolution of the HPD is determined by
three elements. First, it is the distribution of the difference of electron transit time
inside the photocathode, the time difference between the time generating electrons
for incident light, and the time of emission from the photocathode. Second, it is the
difference of electron transit time in a vacuum from the photocathode to the AD.
Third, it is the difference of electron transit time in the process that the bombarded
Mirror
163
fs-Laser
ND filter
HPD
Amp
CFD
Pin-PD
TAC
MCA
Figure 7.14 Experimental setup for timing resolution. Laser beam from a Ti-sapphire
laser was split into two for the HPD and a PIN photodiode, the outputs of which were used
as start and stop signals for the timeamplitude converter (TAC). The TAC output was then
analyzed by a multichannel analyzer (MCA). (Reproduced from Fukasawa, A. et al., IEEE
electron is multiplied inside the AD. In the alkali photocathode, the difference of
the electron transit time is extremely small,23 so it is negligible compared to the
jitter of the measurement system. And, since electrons bombarded to the AD are
first multiplied by more than 1,000 times in the electron bombarded multiplying
process, the difference of the electron transit time inside the AD is also negligible.
Therefore, the timing resolution of the high-speed HPD is determined by the difference of the electron transit time when electrons emitted from the photocathode fly in
the vacuum. In the developed high-speed HPD, since an input window with a planoconcave faceplate is used, electron transit time from photocathode to AD is almost
constant at every point of the photocathode. Consequently, the timing resolution of
the high-speed HPD is excellent.
The timing resolution of the high-speed HPD for single photons was measured using
a Ti-sapphire laser (REGA9000, Coherent) having a pulse width of less than 50 fs at
a 405-nm wavelength. As shown in Figure7.14, the laser beam was split into two by a
half mirror, one for the HPD attenuated by neutral density filters to the single-photon
state and another for a PIN PD (S597301, Hamamatsu). The output signal of the HPD
was used as a starting signal of the timeamplitude converter (TAC) after being amplified by the amplifiers (C5594-12, Hamamatsu, 36-dB gain, 50-kHz to 1.5-GHz bandwidth; and HP8447F, Hewlett-Packard, 22-dB gain, 100-kHz to 1.3-GHz bandwidth)
and a constant fraction discriminator (9307 Pico-Timing Discriminator, ORTEC). For
the stop signal, the output of the PIN PD was used. In this measurement, the HPD with
a bialkali photocathode was used; voltages of 8 kV and 405 V were applied to the
photocathode and the AD, respectively, to obtain the total gain of 160,000.
As shown in Figure7.15, the timing resolution for a single photon of the highspeed HPD with the bialkali photocathode depended on the irradiated spot size of the
164
Frequency (Normalized)
1.2
1
0.8
0.6
0.4
0.2
0
50
100
Time (ps)
150
200
Figure 7.15 The measured timing resolutions using the irradiated area on the photocathode as a parameter (black circle, 1 mm in diameter; open square, 3 mm in diameter; black
triangle, 5 mm in diameter; open triangle, 8 mm in diameter). The resolution was less than 15
ps in sigma at the irradiated area less than 5 mm in diameter, including electronics jitter. It
deteriorated to 26 ps at full illumination. (Reproduced from Fukasawa, A. et al., IEEE Trans.
incident light on the photocathode. For an irradiated area less than 5 mm in diameter,
the measured standard deviation was 15 ps, and it was 26 ps at full illumination. The
simulation for the electron transit time spread between the bialkali photocathode and
the AD is shown in Figure7.16. Therefore, the difference of these experimental results
can be explained by the difference of transit time in the incident light position.
In HPD or MCP-PMT with a GaAs or GaAsP crystal photocathode type, the
difference of the electron transit time is not negligible, so it is a dominant factor
for determination of the timing resolution of the HPDs having such photocathodes.
Timing Resolution (ps)
30
25
20
15
10
5
0
4
6
Spot Size (mm)
10
Figure 7.16 Simulation of the electron transit time spread (open triangle) and the results
of the experiment (black circle) as a function of irradiated area on the photocathode. The
deterioration in 8 mm in diameter can be explained by the deterioration of the electron transit
time spread from the photocathode to the AD. (Reproduced from Fukasawa, A. et al., IEEE
165
In the case of the GaAsP photocathode, the timing resolution was estimated to be
approximately 40 ps for 3 mm (full illumination on the photocathode). The difference of the timing resolution between the GaAsP and bialkali photocathodes can be
explained by the transit time spread in the photocathode and the GaAsP photocathode with a thickness of about 1 m.
7.2.3.7Characteristics of Afterpulse
In the case of pulse measurement using photodetectors, pulses with a delay of time
for a main signal can be occasionally observed. Such subpulses, which do not correspond to incident photons, are called afterpulses. Since an afterpulse causes problems for various applications, a low afterpulse characteristic is desired. Generally,
in the afterpulse characteristic of PMTs, there are two cases to be distinguished: an
afterpulse with short time delay from the main signal (a few nanoseconds to a few
tens of nanoseconds) and an afterpulse with a long time delay (up to a few microseconds). Pulses with short time delay can be dealt with, since they are covered with a
time constant of the subsequent signal-processing system. On the other hand, pulses
with long delay, so-called ion feedback, can cause problems. The usual PMT has
many components, 10 dynodes in chain or so, in the vacuum, whereas the main components of the HPD are only the photocathode, AD, and cylindrical ceramic or glass
to isolate them. From the point of view of noise, fewer components are desirable. In
the multiplying process, electrons smash into residual molecules in the vacuum, and
ionized residual molecules smash into the photocathode or a dynode for electrostatic
fields formed inside the tube, and then secondary electrons as afterpulses are generated by the impact. Especially in the dynodes, since one photoelectron emitted from
a photocathode is multiplied to millions of electrons, the probability of ionized molecules is increasingly higher. However, since HPDs multiply electrons inside the AD,
electrons flying in vacuum are only one photoelectron for incidence of one photon;
therefore, it is overwhelmingly less than for PMTs. This is why the HPD has excellent afterpulse characteristic performance compared to that of the PMT.
Another noise source in HPDs is X-ray feedback noise. X-ray photons are generated as high-energy electrons smash into a material. Generated X-rays go back to
the photocathode, and they are absorbed in the photocathode, creating photoelectron noise. The delay from the main signal is very small, typically estimated to be
approximately 500 ps, almost the same as the transit time of electrons from the
photocathode to the AD.
Noises caused by X-rays have two components. One is X-rays created by the
bremsstrahlung effect, and the second is characteristic X-rays. In the case of feedback of bremsstrahlung X-rays, the output pulse height for the noise does not have
a peak value. On the other hand, since the characteristic X-ray has an energy of
1.7 keV, for which the AD material as a target for electrons is silicon, noises having
approximately the same pulse height are generated for X-ray feedback. For example,
in the case of feedback of one 1.7-keV photon, if the quantum efficiency of the GaAsP
photocathode was 50%, about 100 photoelectrons are emitted from the photocathode. Although the X-ray feedback is the dominant noise in HPDs, it is known from
experience that the probability of the noise is about 1,00010,000 for single-photon
irradiation, a manageable problem for most applications.
166
7.2.3.8Lifetime Characteristics
Since an HPD has a photocathode placed in a vacuum, photocathode sensitivity is
deteriorated by ion feedback, similar to the other photodetectors using photocathodes. GaAs or GaAsP photocathodes especially tend to deteriorate compared to
alkali photocathodes. However, since the ion feedback of an HPD is less than that
of a PMT, the life characteristic is superior to PMT. The lifetime of the HPD with a
GaAsP photocathode was evaluated as 1,500 hours under a photocathode current at
200 pA (1.25 109 photoelectron/s).
7.2.4Application Examples
High-speed HPDs have various strong points compared to the other photodetectors
for single-photon detection. Applications making use of high timing resolution are
time-of-flight (TOF) counter, LIDAR, or FLIM using TCSPC. The application that
has attracted a great deal of attention in recent years is FCS. In FCS, a laser beam is
focused on a very small area of a sample, and fluorescence from the sample is monitored by a photodetector. The diffusion time of molecules in the region of focused
light can be found by calculating the autocorrelation function of arrival time for each
signal. Since the existence of afterpulses distorts the autocorrelation function, it is a
large problem for this application. Xavier Michalet of the University of California at
Los Angeles and colleagues reported that the amount of the afterpulse of high-speed
HPD is less than for the single-photon counting avalanche PD (SPAD) generally used
for this application.20
7.3Other Types of HPD

In addition to the high-speed HPD, various special HPDs have been developed in the
field of high-energy physics and astronomy. Several of these devices are discussed
in this section.
Masahiro Teshima of the Max-Planck-Institute for Physics (Munich, Germany)
and colleagues developed an HPD with a GaAsP photocathode 18 mm in diameter
for the Major Atmospheric Gamma-Imaging Cherenkov (MAGIC; La Palma, Canary
Islands) telescope.24 A photograph of the HPD is shown in Figure7.17. Gamma-ray
telescopes detect Cherenkov light, which is generated by the reaction of gamma
rays and the atmosphere. Since the number of photons generated by the gamma ray
is limited, the photodetector for this application required a large effective area and
the high quantum efficiency of a photocathode. The built-in silicon AD was 3 mm in
diameter, and the capacitance was about 20 pF; rise and fall times were about 1 ns
and 2.5 ns, respectively. This HPD had an advantage for photocathode high quantum
efficiency; it had about 50% efficiency at 500-nm wavelength and might find application in laser-scanning microscopes.
As another type of a single-pixel HPD, Hiroaki Aihara of Tokyo University
(Tokyo, Japan) and colleagues developed a 13-inch diameter HPD.25 A photograph
of the HPD is shown in Figure7.18. One of the applications for this HPD is the detection of neutrinos in the Hyper-Kamiokande experiment as water Cherenkov counter
the following Super-Kamiokande experiment. The photocathode was bialkali, and
167
Figure 7.17 The HPD for the MAGIC telescope.
Figure 7.18 The 13-inch HPD.
the AD was 5 mm in diameter. Since emitted electrons from the photocathode were
focused on an AD with high voltage (1020 kV), electron transit time from the photocathode to the AD was extremely fast (12 ns) compared to a conventional 13-inch
PMT (100 ns), which resulted in excellent timing resolution, a critical feature for the
accurate determination of the vertex point of neutrino interaction.
In addition to single-pixel devices mentioned, developments of multipixel HPDs
have been reported. One was an attempt by Toru Iijima of Nagoya University
(Nagoya, Japan) and colleagues to develop a 144-channel HPD with a bialkali photocathode.25 A photograph of the developed HPD is shown in Figure7.19. Considering
the requirements to cover an area of a few square meters with many HPDs and with
minimum dead area between them, a rectangular ceramic side wall was used. The
size of the HPD was 73 73 mm at the height of 30 mm for an effective area of
64 64 mm. This type of HPD has a proximity focusing structure and is able to
168
Figure 7.19 The 144-channel HPD for a RICH in the BELLE detector.
detect a single photon with high accuracy in a strong (1.5-T) magnetic field. One of
the possible applications of this device is in the BELLE detector used to measure
the decay of B meson, being upgraded in the near future at the collision point in
KEK (Tsukuba, Japan). The square HPD developed is likely to be installed in a ringimaging Cherenkov counter (RICH).
As another type of multipixel HPD, Suyama et al. developed a 64-pixel HPD, with
the size of one pixel 2 2 mm.25,26 The installed AD is the back-illumination type
with low loading capacitance. This HPD also has a proximity focusing structure, so
it can be operated in a strong magnetic field, and it has excellent response time.
7.4Summary
The HPD photodetector appeared first in the 1960s and has evolved to have high
performance for single-photon detection. The additional gain of an AD diode contributes to high-speed operation. Although HPD gain is still 1/10 compared to a
conventional PMT, it is relatively easy to detect a single photon with the help of
a remarkably efficient readout system.
The HPD was originally developed for high-energy physics experiments or astronomy in the 1990s, but now use of HPDs is spreading for bioapplications as a result of
its various advantages. The merits in performance are low afterpulse, long lifetime,
low cost, and high speed of operation due to the simple structure of the HPD. In the
near future, further improvements in avalanche gain can be expected. When that gain
increase is accomplished, the HPDs should be operable with lower voltage, the same
level as that for PMTs. In this case, photodetector applications would be expanded to
medical instruments, such as PET scanners and gamma cameras.
References

1. Hamamatsu Photonics, PMT Handbook, Hamamatsu, Japan, 2005.

2. Hamamatsu Photonics, Hybrid Photo Detector (HPD)R7110U Series. Technical manual, Hamamatsu, Japan, 2001.

169
3. M. Suyama, Development of a Multi-pixel Photon Sensor with Single-Photon Sensitiv

ity, doctoral thesis, Department of Particle and Nuclear Physics, School of Mathematical
and Physical Science, The Graduate University for Advanced Studies, 2002.
4. F. A. White and J. C. Sheffield, Reversed-biased P-N junctions in electron tubes, Proc.
IRE, 50(6), 15231524, 1962.
5. R. Kalibjian, A phototube using a diode as a multiplier element, IEEE Trans. Nucl. Sci.,
NS-12(4), 367369, 1965.
6. P. Chevalier, Photomultiplicatuer a haute resolution utilisant un multiplicateur semiconducteur, Nucl. Instr. Methods, 50, 346348, 1967.
7. E. A. Beaver and C. E. Mcllwain, A digital multichannel photometer, Rev. Sci. Instrum.,
42, 13211324, 1971.
8. E. A. Beaver and C. E. Mcllwain, Counting image tube photoelectrons with semiconductor diodes, Adv. Electron. Electron Phys., 33B, 863871, 1972.
9. R. G. Tull, J. P. Choisser, and E.H. Snow, Self-Scanned Digicon: A Digital Image Tube
for Astronomical Spectroscopy, Appl. Opt. 14, 1182, 1975.
10. R. Desalvo, Hybrid Photo Diode Tube, CLNS 87-92, Cornell University, Ithaca NY,
1987.
11. L. K. van Greest and K. W. J. Stoop, Hybrid phototube with Si target, Nucl. Instr.
Methods A, 310, 261266, 1991.
12. C. Datema, Hybrid photodiodes in scintillation counter application, Nucl. Instr. Methods
A, 387, 100103, 1997.
13. G. A. Johansen and C. B. Johnson, Operational characteristics of an electron-bombarded
silicon-diode photomultiplier tube, Nucl. Instr. Methods A, 326, 295298, 1993.
14. M. Suyama, Y. Kawai, S. Kimura, N. Asakura, K. Hirano, Y. Hasegawa, T. Saito, T.
Morita, M. Muramatsu, and K. Yamamoto, A compact hybrid photodetector, IEEE
Trans. Nucl. Sci., NS-44(3), 985989, 1997.
15. J. P. Choisser, Detecting photoelectron images with semiconductor arrays for multichannel photon counting, Opt. Eng., 16(3), 262266, 1977.
16. P. Cushman, A photomultiplier tube incorporating an avalanche photodiode, Nucl. Instr.
Methods A, 333, 381390, 1993.
17. R. Mirzoyan, An evaluation of the new compact hybrid photodiodes R7110U-07/40
from Hamamatsu in high-speed light detection mode, Nucl. Instr. Methods A, 422,
140145, 2000.
18. S. Matsui, M. Akatsu, Y. Enari, K. Fujimoto, Y. Higashino, M. Hirose, T. Hokuue,
K. Inami, A. Ishikawa, T. Matsumoto, K. Misono, T. Ohshima, A. Sugi, A. Sugiyama,
S. Suzuki, and M. Tomoto, HAPD time-resolution study under single-photon irradiation, Nucl. Instr. Methods A, 463, 220226, 2001.
19. A. Fukasawa, J. Haba, A. Kageyama, and M. Suyama, High speed HPD for photon
counting, IEEE Trans. Nucl. Sci., 55(2), 758762, April 2008.
20. X. Michalet, A. Cheng, J. Antelman, M. Suyama, K. Arisaka, and S. Weiss, Hybrid photodetector for single-molecule spectroscopy and microscopy, SPIE2008 proceedings,
2008.
21. V. Saveliev, The recent development and study of silicon photomultiplier, Nucl. Instr.
Methods A, 535, 528532, 2004.
22. V. Golovin and V. Saveliev, Novel type of avalanche photodetector with Geiger mode
operation, Nucl. Instr. Methods A, 518, 560564, 2004.
23. K. Kinoshita, M. Ito, and Y. Suzuki, Femtosecond streak tube, Rev. Sci. Instrum., 58(6),
932938, 1987.
24. M. Hayashida, R. Mirzoyan, and M. Teshima, Development of HPDs with an 18-mm-
diameter GaAsP photo cathode for the MAGIC-II, Nucl. Instr. Methods A, 567, 180183,
2006.
170
25. H. Nakayama, A. Kusaka, H. Kakuno, T. Abe, M. Iwasaki, H. Aihara, M. Shiozawa,

M. Tanaka, H. Kyushima, M. Suyama, and Y. Kawai, Development of a 13-in hybrid
photo-detector (HAPD) for a next generation water Cherenkov detector, Nucl. Instr.
Methods A, 567, 172175, 2006.
26. S. Nishida, I. Adachi, T. Iijima, H. Ikeda, S. Korpar, P. Krian, Y. Miyazawa, I. Nishizawa,
and T. Sumiyoshi, Development of an HAPD with 144 channels for the aerogel RICH of
the Belle upgrade, Nucl. Instrum. Methods A, 595, 150153, 2008.
27. M. Suyama, A. Fukasawa, J. Haba, T. Iijima, S. Iwata, M. Sakuda, T. Sumiyoshi,
F. Takasaki, M. Tanaka, T. Tsuboyama, and Y. Yamada, Development of a multi-pixel
photon sensor with single-photon sensitivity, Nucl. Instr. Methods A, 523, 147157,
2004.
28. M. Suyama, A. Fukasawa, J. Haba, T. Iijima, S. I wata, M. Sakuda, T. Sumiyoshi,
F. Takasaki, M. Tanaka, T. Tsuboyama, and Y. Yamada, Development of a multipixel
hybrid photodetector with high quantum efficiency and gain, IEEE Trans. Nucl. Sci.,
51(3), 10561059, June 2004.
8 High-Resolution
CdTe Detectors
and Application to
Gamma-Ray Imaging
Tadayuki Takahashi, Shin Watanabe, and
Shin-nosuke Ishikawa
Institute of Space and Astronautical Science (ISAS), JAXA

Kanagawa, Japan
Contents
8.1 Introduction................................................................................................... 171
8.2 High-Resolution CdTe Detector.................................................................... 173
8.3 CdTe Pixel Detector Module......................................................................... 176
8.4 Stacked CdTe Detector.................................................................................. 178
8.5 CdTe Double-Sided Strip Detector................................................................ 181
8.6 Si/CdTe and CdTe/CdTe Semiconductor Compton Camera.......................... 184
8.7 Conclusion..................................................................................................... 187
References............................................................................................................... 189
8.1Introduction
Efforts have long been made to develop room temperature semiconductors with high
atomic numbers and wide band gaps. These materials are useful not only in medical
and industrial imaging systems, but also in detectors for high-energy particle astrophysics and astrophysics in general. Among the range of semiconductor detectors
available for gamma-ray detection, CdTe and CdZnTe occupy a privileged position
due to their high density, the high atomic number of their components, and a wide
band gap. A large band gap energy (Egap = 1.44 eV) allows us to operate these detectors at room temperature.14
The high absorption efficiency of cadmium telluride (CdTe) or cadmium zinc telluride (CdZnTe)comparable with that of NaI and CsIis a very attractive feature.
171
172

1.00
CdTe
Si
0.50
Compton
(1/cm)
0.20
Pair
0.10
0.05
0.02
Photoelectric
0.01
0.1
0.5
1.0
5.0
10.0
50.0 100.0
Energy (MeV)
Figure 8.1 Linear attenuation coefficients in CdTe and silicon as a function of photon
energy. The intensity of photons can be expressed as I = I0 exp(x), where x denotes the path
length in centimeters.
As shown in Figure8.1, photoelectric absorption is the main process up to 300 keV

for these materials, as compared to 60 keV for silicon. Therefore, CdTe and CdZnTe
are expected to become more efficient once the gamma-ray energy exceeds a few
hundred kiloelectron volts. Despite long-term efforts to make improvements, however, high-resolution CdTe and CdZnTe detectors with energy resolution better than
a few kiloelectron volts (full width at half maximum or FWHM) have only recently
become available. In the 1990s, remarkable progress in the technology for producing
high-quality single crystals of CdTe by using the traveling heater method (THM)5
and CdZnTe by using the high-pressure Bridgman (HPB) technique68 dramatically
changed high-resolution room temperature detectors. The industry has further continued such efforts in the 2000s toward producing high-quality CdTe and CdZnTe
wafers.913 In addition to progress regarding the crystal, various technologies of
electrode configuration have been proposed and developed to overcome poor carrier
transport in the device, as described in this chapter.
To use CdTe and CdZnTe for detecting gamma rays with energy higher than a few
hundred kiloelectron volts, the bias voltage necessary to eliminate the low-energy
tail at a level to achieve resolution of a few kiloelectron volts must be higher than a
few hundred kiloelectron volts even with a thickness of a few millimeters of thickness. If the bias voltage is not high enough, only a fraction of the signal charge generated is induced at the detector electrode. The fraction and the resultant pulse height
depend on the depth of interaction. This positional dependency produces a shoulder
(tailing) in the peaks of gamma-ray lines toward the low-energy region. Increasing
the thickness of the CdTe/CdZnTe detector to improve the efficiency of detecting
high-energy gamma rays makes the effects of incomplete charge collection and the
High-Resolution CdTe Detectors
173
low-energy tail more significant. In some cases, the inactive region in the detector
volume increases along with the thickness, especially for the region near the anode.
Soon after the emergence of HPB-grown CdZnTe, new ideas based on the concept of single charge collection were proposed.1417 These ideas included the application of single-carrier detection techniques and apply these to semiconductors with
low values of (mobilitylifetime product) values for one type of carrier (typically holes). In this method, an attempt is basically made to change the detectors
charge induction property of the detector (i.e., its weighting potential) to improve
the charge induction efficiency (CIE) response.3 Another approach to improving the
spectral properties of CdTe detectors is the idea of forming a barrier electrode on
the tellurium-face of the p-type CdTe wafer as an anode.1821 A high Schottky barrier
between the electrode and CdTe makes the detector operate as a diode (CdTe diode).
The significant reduction in the CdTe diodes leakage current allows us to apply high
bias voltage for improved CIE without degrading the energy resolution.
Here, we review the achievements made in high-resolution CdTe diode detectors and their application to gamma-ray imaging. Recent progress made in CdTe
and CdZnTe detectors have been report by Takahashi and Watanabe2 and by Luke.3
Material properties of CdTe and CdZnTe have been reported by Owens and Peacock4
together with other compound semiconductor materials usable as radiation detectors.
Applications to nuclear medicine were reviewed in Barber; Scheiber; Darambara;
Verger et al.2326 The performance of CdTe and CdZnTe onboard astrophysical satellites were described by Limousin et al. for the IBIS instrument onboard INTEGRAL
and by Sato et al. for the BAT instrument used on the Swift gamma-ray burst
mission.27,28
8.2High-Resolution CdTe Detector

The technique for growing a large single CdTe crystal with good charge transport
properties has been established by the THM.5 After careful thermal treatment and
the selection of proper crystal orientation for the electrode system, the CdTe wafer
reportedly exhibits good charge transport properties for both electrons (ee = 1 2
10 3 cm2/V) and holes (hh = 1 10 4 cm2/V). Electrical resistivity of about 1
109 cm (p-type) is achieved by using chloride to compensate for the native defects.
Furthermore, the detector is free from stability-related problems in the typical electrode configuration with platinum that forms Ohmic contacts.5,28 The wafers uniform
charge transport properties are a very important aspect, not only for fabricating large
area strips or pixel detectors, but also for constructing a large-scale gamma-ray camera with many individual detectors. Single crystals of 100-mm diameter are now
commercially available for THM-CdTe with high uniformity.
Adopting a Schottky junction, in conjunction with the high quality of the THMCdTe manufactured by Acrorad, dramatically improves the energy resolution of
CdTe detectors.1821 Since the CdTe material grown by Acrorad has p-type resistivity,
a low work function metal, such as indium, can be used to form a Schottky barrier.
Moreover, the leakage current is significantly suppressed in the reverse bias operation of the In (anode)/CdTe/Pt (cathode) configuration. Thus, the leakage current of a
174
400
Counts
Counts
600
400
300
400
Fe K
100
200
500
E = 260 eV
200
0
600
Fe k
300
6
8
7
Energy (keV)
200
100
25
75
100
50
Energy (keV)
(a)
125
150
0
620
630
640 650 660

Energy (keV)
670
680
(b)
Figure 8.2 (a) 57Co spectrum obtained with the CdTe diode. Bias voltage of
800 V was applied. The detector size was 3 3 1 mm3 in size and operating temperature
was 40C. The energy resolution at 6.4 keV was 0.26 keV (FWHM). (Reproduced from
Takahashi, T. et al., Nucl. Instr. Methods, A541, 2005. With permission from Elsevier.)31
(b) High-energy spectrum of 137Cs above 620 keV with a 2 2 mm CdTe diode of thickness
0.5 mm. The applied bias voltage was 1,400 V and the operating temperature was 40C. The
energy resolution at 662 keV was 2.1 keV (FWHM).28
detector with area of 2 2 mm and thickness of 0.5 mm, the leakage current of the 2
2 0.5 mm3 detector is about 0.5 nA with bias voltage of 400 V at 20C. With this
high bias voltage, a CdTe diode with thickness of 0.5 to 1.0 mm becomes fully active,
and even holes generated near the anode face can be completely collected. Cooling
the detector further improves the energy resolution.
The theoretical energy resolution of CdTe can be calculated from statistical fluctuations in the number of electron hole pairs and Fano factor F.1 By using e = 4.5 eV and F =
0.15, the theoretical limit (FWHM) is 200 eV at 10 keV, 610 eV at 100 keV, and 1.5 keV
at 600 keV,2 provided that we could neglect electronic noise. As shown in Figure8.2, a
CdTe diode with area of 3 3 mm and thickness of 1 mm installed in the electronics
system manufactured by Amptek29 had energy resolution of 260 eV (FWHM) at 6.4 keV
when operating the detector at 40C.31,32 The reduction of the low-energy tail even in
the 662-keV line from 137Cs (Figure8.3) resulted in resolution of 2.1 keV (0.3%) obtained
with a 2 2 mm CdTe diode 0.5-mm thick at a bias voltage of 1,400 V.28 In the measurement, the charge signal is integrated in the Clear Pulse CP-5102 charge-sensitive preamplifier30 and shaped by an Ortec 571 amplifier. Intrinsic resolution by subtracting the
contribution of electronic noise was close to the prediction from e = 4.5 eV and F = 0.15,
suggesting that the resolution of the CdTe diode with a thickness of 0.5 to 1 mm almost
reaches the theoretical limit for a wide energy range from 10 to 700 keV.
To further reduce the leakage current of Schottky-type detectors, a CdTe diode
detector with a guard ring electrode was proposed and tested based on leakage current being proportional to the diode perimeter, not the area or volume.33 A guard ring
structure in the cathode electrode was consequently introduced. This improvement
resulted in reducing leakage current by another order of magnitude and enabled us
to operate the device at room temperature (i.e., up to 20C) with good performance.
175

10.00
Leakage Current (pA)
5.00
2.00
Al/CdTe/Pt
In/CdTe/Pt
1.00
0.50
0.20
0.10
0.05
100
200
300
Bias Voltage (V)
400
500 600
Figure8.3 Current voltage (I-V) characteristics at 20C of CdTe diode detectors with
In/CdTe/Pt and Al/CdTe/In electrode configurations, manufactured by Acrorad. Both detectors had an active area size of 2 2 mm2 0.5 mmt.
The typical current at 20C for this device with an active area of 2 2 mm2 and
thickness of 0.5 mm was 10 pA at a bias voltage of 400 V.33
Based on the idea of using a barrier (Schottkky) electrode, aluminum recently
emerged as an alternative electrode material to indium.3436 Because an Al/CdTe/Pt
electrode configuration also works as a Schottky diode for p-type CdTe, low leakage
current and good energy resolution comparable to those of In/CdTe/Pt detectors can
be achieved. Figure8.3 shows the current-voltage (I-V) characteristics of the of CdTe
diode detectors with the Al/CdTe/Pt and In/CdTe/Pt electrode configurations.39 As
shown in Figure8.3, the lower barrier height afforded by aluminum leads to higher
leakage current compared to that of indium. However, an aluminum electrode over
an indium electrode is that the aluminum electrode, which acts as an anode electrode, can be segmented into pixels or strips. Therefore, it is possible to fabricate
electron-collectingtype diode pixel detectors using aluminum in pixelated anodes
and platinum in the common cathode (Al-pixel/CdTe/Pt).36 Actually, monolithic Al/
CdTe/Pt detectors with guard-ring electrodes show spectral performance comparable with that obtained with In/CdTe/Pt detectors (see Figure8.4).
As described in our earlier articles (e.g., Takahashi et al.),22 the CdTe diode detector shows degraded spectrum resolution over time, similar to the so-called polarization effect in semiconductor devices. We noted that CdTe detectors with the anode
and cathode both made of platinum (Ohmic devices) do not exhibit the polarization
effect, and therefore a series of spectra taken over time does not change. However,
when Schottky contacts are introduced, we do see this polarization effect at room
temperatures and low bias voltages. We found that the polarization effect could be significantly suppressed if we make the detector thin (0.5 to 1 mm) and operating it with
high bias voltage at a low temperature of 20C. Under these operating conditions,
the stability of spectra and absence of polarization are clear for the operation as the at
176

+
In/CdTe/Pt
6,000
0.5 mm
5,000
59.5 keV
FWHM 1.0 keV
3,000
3,000
2,000
2,000
1,000
1,000
0
20
40
59.5 keV
FWHM 1.0 keV
4,000
Counts
Counts
0.5 mm
5,000
4,000
Al/CdTe/Pt
6,000
60
20
40
Energy (keV)
Energy (keV)
(a)
(b)
60
Figure 8.4 241Am spectrum obtained from CdTe detectors with (a) In/CdTe/Pt and
(b) Al/CdTe/Pt electrode configurations with a guard ring structure.
least a couple of days or more. We successfully reproduced the changes in spectra over
time quantitatively,37 by following the prescription given by Malm and Maritini.38 In
the model, a uniform distribution of deep acceptor levels in the In/CdTe/Pt device is
assumed for simplicity. A negative charge gradually accumulates over time through
the release of holes from the deep acceptor level (detrapping). When the bias voltage
is turned on, the number density of holes becomes very small, as does the recapture
(trapping) rate. The change in spectra can be explained by the time evolution electric
field distortion in the detector caused by accumulated negative space charge. The rate
of polarization is reduced when the operating detector is operated at low temperature,
due to the longer detrapping timescale at lower temperatures. The high electric field
either by high bias voltage or thin detector material makes the distortion of the electric
field negligible. When the bias is turned off, the polarization recovers by recapturing
holes. The timescale of recapturing is much shorter than that of the release of holes.
8.3CdTe Pixel Detector Module

For an imaging detector, the good energy resolution and the ability to fabricate
compact arrays expected from semiconductor detectors are very attractive features
compared with inorganic scintillation detectors coupled to either photodiodes or
photomultiplier tubes.40 By forming pixels on the electrode of a large CdTe device,
the small capacitance and low leakage current of the pixels help improve energy
resolution. Conversely, a large number of channels must be handled in the system,
requiring the implementation of very-low-noise analog circuits as a multichannel
application-specific integrated circuit (ASIC). U.C.E. energy resolution at around
1 to 2 keV is challenging if we have to rely on highly compact ASICs based complementary metal oxide semiconductor (CMOS). Recent advances made in ASIC have
led to developments of fine pitch pixel detector.4143 As an application of the CdTe
diode, we tried to make a fine-pitch CdTe diode imaging detector based on the ASIC
177
(a)
(b)
Figure8.5 (a) An 8 8 CdTe pixel detector module. The detector had 8 8 = 64 pixels
with a pixel size of 1.4 1.4 mm2. The detector was 0.5-mm thick. (b) A large-area CdTe
detector consisting of 4 4 = 16 CdTe pixel modules arranged on one plate. (Reproduced from
Watanabe, S. et al., Nucl. Instr. Methods A, 57, 9, 2010. With permission of Elsevier.)
developed by Bonn University44 and Caltech.45 In addition to signals from pixels

formed on the anode face, information from the cathode (common electrode) can be
used to obtain the depth of gamma-ray interaction. This interaction depth provides
a correction for the electron path length and the trapping, which are important for
thick detectors (~1 cm). He et al. performed extensive studies on this possibility.46,47
A large-area CdTe imaging detector with an area greater than a few tens of square
centimeters with an energy resolution of 1 to 2 keV (FWHM) is attractive for various
applications. We have used the modular structure approach in constructing a largearea imaging CdTe detector.48 Figure 8.5(a) shows a CdTe pixel detector module
consisting of an 8 8 CdTe pixel device, a fanout board, and a 64-channel readout
ASIC (VA64TA).49 The CdTe device is 13.35 13.35 mm in size and 0.5-mm thick.
This is a Schottky CdTe diode device that uses an indium anode and a platinum
cathode. The indium side is used as a common electrode; the platinum side is divided
into 8 8 (= 64) pixels. The pixel size is 1.35 1.35 mm, with a 50-m gap between
the pixel electrodes. To reduce leakage current, a guard ring electrode 1-mm wide is
attached at the outer edge of the detector. A thin layer of gold is evaporated on the
platinum side for ensuring good bump-bonding connectivity.
The substrate of the fanout board is made from a 96% Al2O3 ceramic material
with a thickness of 300 m. The fanout board consists of bump pads, through-holes,
and patterns that route signals from the bump pads on both surfaces of the ceramic
substrate. Stud bump-bonding technology developed specifically for CdTe detectors22 is employed to connect each pixel to the bump pads on the fanout board. The
studs are made with gold wire and a thin layer of indium printed on top of each stud.
The fanout board with the CdTe device was wire-bonded to an ASIC (VA64TA).50
Figure8.6 shows the gamma-ray spectra from 241Am and 57Co obtained with the
CdTe pixel detector module. Gain corrections were made for the 64 spectra from
all pixels, followed by the summing of all corrected spectra. Bias voltage of 600 V
was applied at an operating temperature of 5C. The energy resolutions were 1.2 and
1.5 keV (FWHM) at 59.5 and 122 keV, respectively.
178

241Am
3,000
57Co
59.54 keV
FWHM 1.2 keV
2,500
2,000
1,500
1,000
500
0
10
20
30
40
50
60
70
Energy (keV)
2,000
1,800
1,600
1,400
1,200
1,000
800
600
400
200
0
5C, 600 V
122 keV
FWHM 1.5 keV
14.4 keV
FWHM 1.1 keV
20
40
60
80 100
120
140
Energy (keV)
Figure 8.6 241Am and 57Co spectra obtained from a CdTe pixel detector module. Each
spectrum was constructed from spectra obtained from all 64 pixels. Bias voltage of 600 V
was applied at an operating temperature of 5C. The energy resolutions were 1.2 and 1.5 keV
(FWHM) at 59.5 and 122 keV, respectively. (Reproduced from Watanabe, S. et al., Nucl. Instr.
Methods A, 57, 9, 2, 2010. With permission of Elsevier.)
Once CdTe pixel detector modules have been established, a CdTe imager with
larger area can be produced by arranging the modules. Figure8.5(b) shows a CdTe
imager in which 4 4 (= 16) CdTe pixel detector modules are arranged on a plate.
The imager is 5.4 5.4 cm2 in total size. The imager works as a conventional gamma
camera with a collimator attached above the imager. Because the imager has high
energy resolution (~1%), various gamma-ray lines can be easily separated in the
spectra obtained. Therefore, the imager is capable of simultaneous multitracer imaging using different gamma-ray lines (see Figure8.7).
8.4Stacked CdTe Detector

In order to extend the application of CdTe diodes to the detection of megaelectron
volt gamma rays, an effective detector must be more than 1 cm thick. However, it is
difficult to obtain good energy resolution and good peak detection efficiency with
such a thick CdTe device due to incomplete charge collection. One approach to overcoming this difficulty is the idea of a stacked detector, in which several tens of thin
and fully active CdTe diode detectors are stacked together and operated as a single
detector.2,21,5153 Figure8.8 shows such a stacked detector consisting of 40 layers of
CdTe diodes each with an area of 21.5 21.5 mm2 and thickness of 0.5 mm.52 A gap
of only 0.7 mm has been achieved, with the detector housing using a ceramic sheet
0.5 mm thick. The stacked detector has a total volume of 9.2 cm3, thereby providing
efficiency of 20% at 500 keV and 7% at 1 MeV. In the stacked detector, the signal
from each layer is processed independently by using an individual analog chain.
The gamma-ray spectrum from the detector is obtained by summing the spectra
from all layers where pulse heights above a certain threshold are recorded. With this
179

140 keV
99mTc
6980 keV
20lTl
Figure8.7 Gamma-ray image taken from tobacco leaf that absorbed different kinds of
radioactive liquid (99mTc and 201Tl). The leaf was placed above the large-area CdTe detector
[(Figure8.5(b)] with a tungsten collimator. The different distribution between 99mTc and 201Tl
can be clearly seen.
Figure 8.8 The 40-layer CdTe stacked detector and a layer used in the detector.
(Reproduced from Watanabe, S. et al., Nucl. Instr. Methods A, 57, 9, 2010. With permission of
Elsevier.)52
approach, the energy resolution could be maintained at the same level as that of a
single layer. The energy resolution obtained with this stack detector was 1.6 to 1.7%
(FWHM) for the two peaks of 1.17 and 1.33 MeV in the energy spectrum of gamma
rays from 60Co.52
In addition to increased efficiency, the stack configuration and individual readouts
provide information on the depth of interaction. This depth information is useful for
reducing the background, since low-energy gamma rays can be expected to interact
in the upper layers; therefore, low-energy events detected in lower layers can be
rejected. Moreover, since the background rate is proportional to the detector volume,
180
Figure8.9 Photo of a CdTe stack detector with four layers. Each layer consists of 2 2 =
4 CdTe pixel modules. The pitch between each layer is 2 mm. (Reproduced from Watanabe, S.
et al., Nucl. Instr. Methods A, 57, 9, 2007. With permission of Elsevier.)
low-energy events collected from the first few layers in the stacked detector have a
high signal-to-background ratio compared with events obtained from a monolithic
detector with a thickness equal to the sum of all layers.
Figure8.10 shows the spectra of gamma-rays from 133Ba, obtained with the CdTe
stack detector consisting of CdTe pixel modules (Figure 8.9).48 Photoelectric absorption peaks are clearly seen in the spectra. The spectra from the first, second, third,
and fourth layers are shown in the figure. On the low-energy side, the first layer
detects most gamma rays. On the higher-energy side, the peak areas detected in all
layers are almost identical. This indicates that stacking detectors improve the detection efficiency for higher-energy gamma rays. As shown in Figure8.11, the energy
resolution for the 511-keV gamma ray is E(FWHM)/E ~ 0.9% at an operating temperature of 20C and under bias voltage of 600 V.
The stacked detector can also be applied to measure distance from a gamma ray
source. When the source at a distance of x emits monoenergetic gamma rays homogeneously, the photo peak counts detected in the ith layer Ni are given as

x

N i exp i 1 t
x + i 1 d
( ( ))
( )
(8.1)
where denotes the total photon cross section of detector material, t is the thickness of each layer, and d is the gap between layers. The first term corresponds to the
effect of blocking by the upper layers; the second term is introduced by the difference
in distance between each layer and the source. Owing to the second term, the ratio of
photo peak counts obtained in each layer changes with respect to distance from the
source. Therefore, once the ratio of each layer count is measured, the distance from
the source can be calculated by fitting with the function given in Equation 8.1.52
181

133Ba
Counts/bin (1/3 keV)
100,000
4,000
80,000
3,000
60,000
2,000
1,000
40,000
0
250
20,000
0
105
300
350
Energy (keV)
400
Sum of all layers

1st layer
356 keV peak

FWHM 2.4 keV
E/E = 0.67%
104
2nd layer
3rd layer
4th layer
103
102
81 keV peak
FWHM 1.9 keV
E/E = 2.3%
101
100
100
200
300
400
Energy (keV)
Figure8.10 133Ba spectra obtained with the CdTe stack detector. The spectrum from each
layer is shown in color; the summed spectrum of all layers is shown in black. The energy
resolutions (FWHM) achieved were 1.9 and 2.4 keV at 81 and 356 keV, respectively.
8.5CdTe Double-Sided Strip Detector

The approach of using a pixelated electrode design is widely adopted to obtain an
imaging device based on CdTe and CdZnTe. However, when a large imaging sensor
with dimensions greater than 10 cm2 and position resolution of a few hundreds of
micrometers is considered as the next target for a gamma-ray-sensitive imager,
pixel detectors can be very consuming in terms of electronic design.54 The number
of readout channels is N 2 for a N N pixel detector. To reduce the number of readout
channels, various methods have been proposed. These include a coplanar grid14 and
orthogonal strips.55 Orthogonal strip geometry was devised where the noncollective
strip contacts are made capacitive rather than conductive54 was made. With this configuration, only 2N readout channels are needed for the N N pixel detector.
The high-uniformity CdTe diode detector is attractive as a large-area imaging
detector. A thin CdTe diode detector, as we have demonstrated in a variety of detector
182

22Na
4,000
511 keV peak

FWHM 4.8 keV
E/E = 0.9%
3,000
2,000
1,000
100
200
300
400
500
Energy (keV)
Figure 8.11 The 511-keV gamma-ray spectrum obtained with the CdTe stack detector.
A 22Na radioisotope was used. FWHM for the 511-keV gamma-ray peak was 4.8 keV (E/E
up to 0.9%). (Reproduced from Watanabe, S. et al., Nucl. Instr. Methods A, 57, 9, 2007. With
permission of Elsevier.)
configurations, should be best suited since it has no inactive region in a device less
than a few millimeters thick. Although several attempts to make a strip electrode
on the indium side of the CdTe diode detector has resulted in poor performance, we
finally succeeded in making a double-sided strip detector (DSD) as an extension of
the Al/CdTe/Pt electrode technology.58
In our CdTe diode DSD, orthogonal strips on both sides of the detector provide
two-dimensional coordinate measurements for absorbed particles such as gammaray photons. Thanks to the properties of the aluminum electrode on the CdTe surface,
we could make stable contacts with a strip. As described in Section 8.2, aluminum
acts as a barrier electrode for p-type CdTe, and the resultant detector shows very
low leakage current and thus good energy resolution. To extract a signal from the
strips, we used gold stud bump bonding,22 rather than direct wire bonding to the aluminum and platinum electrodes. Figure8.12 shows one of our first prototype CdTe
DSDs with dimensions of 2.6 2.6 cm and a thickness of 500 m. The strip pitch is
400 m. In this detector, there are 64 strips for both the anode and cathode sides of
the detector. This detector can divide space into 64 64 (= 4,096) pixels for readout
from only 64 + 64 = 128 channels. Two ASICs (VA32TAs), used for our CdTe pixel
detectors, are used for the readout from strips.58 Based on our newly developed technology, we have already achieved a strip pitch down to 60 m.
Figure8.13 shows the gamma-ray images obtained with the CdTe DSD. These
are shadow images of brass nuts, a brass washer, and a soldering wire with gamma
rays from various radioisotopes: 241Am (60 keV), 133Ba (81 keV), and 57Co (122 keV).
183
Figure8.12 A CdTe double-sided strip detector. It was 2.6 2.6 cm2 and 500-m thick.
The strip pitch was 400 m with 64 strips formed on each side, and using two VA32TAs for
readout on each side. (Reproduced from Watanabe, S. et al., IEEE Trans. Nucl. Sci., 2009.
Copyright 2009 IEEE. With permission.)
The hole of a 2-mm nut and solder wire 0.6 mm in diameter can be clearly seen. It
can also be seen that a thin washer became transparent as the energy of gamma rays
increased. The DSD provides simultaneous imaging and spectroscopic information
with an energy resolution of 2.6 keV (FWHM) for 60-keV gamma rays.
Figure 8.14 demonstrates the spectral performance obtained with another prototype of CdTe DSDs with shorter strips.58 The size of the detector is 1.3 1.3 cm
and 500-m thick. The strip pitch is 400 m, with electrodes on both sides are
divided into 32 strips. For readout from these detectors, 64-channel analog ASICs
(VA64TA2s) were used for both the anode and cathode sides. The operating temperature was 20C at bias voltage of 500 V. The anode spectrum was obtained from
the signals of 30 out of 32 strips, while the cathode spectrum was generated from the
signals of all 32 strips. Energy resolution of 1.8 keV was obtained from both sides
of the electrode. Applying a high bias voltage such as 500 V for a thin device 0.5-mm
thick results in high charge collection efficiency with virtually no low-energy tail in
the spectra.
By collectively using the energy information on both sides, improved spectral
performance could be expected, provided that each noise component is independent.
When the energy resolution is similar for both the anodes and cathodes, spectral
performance could be improved by a factor of 2 by averaging the pulse heights of
strip electrodes from both the anode and cathode sides on a photon-by-photon basis.
Figure8.14(b) shows the averaged spectrum obtained from this equation. Improved
energy resolution of 1.5 keV (FWHM) was obtained by using this method.
184
60 keV
50
120
50
40
100
40
80
30
60
20
40
10
0
10
20
30 40
x (ch)
60
50
60
122 keV
80
60
40
20
10
20
0
10
20
30 40
x (ch)
50
60
nut (3 mm) Washer
80
40
60
30
Nut (2 mm)
solder
(0.6 mm)
40
20
20
10
0
100
81 keV
30
20
100
50
y (ch)
60
140
y (ch)
y (ch)
60
10
20
30 40
x (ch)
50
60
Solder
(1 mm)
Figure8.13 Shadow images obtained with the CdTe DSD prototype and a photo of the
target. Energies of the gamma rays were 60 (241Am), 81 (133Ba), and 122 (57Co) keV. The pixel
size of the images corresponds to a strip pitch of 400 m. (Reproduced from Watanabe, S.
et al., IEEE Trans. Nucl. Sci., 2009. Copyright 2009 IEEE. With permission.)
8.6Si/CdTe and CdTe/CdTe Semiconductor

Compton Camera
Compton cameras are used to reconstruct gamma-rayemitting source distributions
based on Compton scattering recorded in the detector. Compton scattering typically
plays a dominant role in the energy band from a few hundred kiloelectron volts to
10 MeV. It entails elastic scattering between the incident photons and electrons in the
scattering medium. In conventional Compton cameras, the incident gamma ray is
identified by successive interactions in the two detector layers. The ideal case would
be that a gamma-ray photon emitted from the source is Compton-scattered in the
first layer and then is photoabsorbed in the second layer. Once the locations and energies of both interactions are measured, Compton kinematics allows us to calculate
the energy and direction (as a cone in the sky) of the incident gamma ray by using the
following Compton equations:
185

400,000
400,000
0.5 mmt CdTe

60 keV
Anode
FWHM 1.8 keV
Cathode
FWHM 1.8 keV
200,000
300,000
200,000
100,000
0
60 keV
FWHM 1.5 keV
Counts
Counts
300,000
0.5 mmt CdTe

Reconstructed spectrum
100,000
20
40
60
20
40
Energy (keV)
Energy (keV)
(a)
(b)
60
Figure8.14 (a) 241Am spectra obtained using a CdTe DSD 0.5-mm thick. The operating
temperature was 20C at a bias voltage of 500 V. The energy resolution (FWHM) from
both anodes and cathodes was 1.8 keV for 60 keV. (b) Averaged spectra. (Reproduced from
Watanabe, S. et al., IEEE Trans. Nucl. Sci., 2009. Copyright 2009 IEEE. With permission.)
Ein = E1 + E2
(8.2)
1
1
cos = 1 mec 2

E2 E1 + E2
(8. 3)
where E1 denotes the energy of the recoil electron, E2, the energy of the scattered
gamma-ray photon, and , the scattering angle. For every event, a cone can be reconstructed as an opening angle of 2. The central axis of the cone is the scattering
direction of the gamma rays. The source is somewhere on the surface of the cone.
Compton cameras have the advantage that only a few photons are needed to recover
the position of sources without mechanical collimators in front of the camera. If the
direction of recoil energy can be measured, the Compton cone is reduced to a segment of the cone, whose length depends on the measurement accuracy of the recoil
electron.
Semiconductor imaging detectors are desired for Compton imaging. As expressed
in Equations 8.2 and 8.3, the energy and position resolution provided with semiconductors should improve the angular resolution and hence the sensitivity of the
Compton cameras. From this perspective, several semiconductor-based Compton
cameras have been proposed.59,60 However, so far, most of these cameras have been
developed based on combining a semiconductor, such as silicon, with scinitillators.
Since CdTe has large atomic numbers (48, 52) and high density (5.8 g/cm3), it has a
potential to replace scintillators and works nicely as an absorber detector. A semiconductor Compton camera based on HP-Ge has been reported elsewhere.61,62
186

Gamma-ray
Source
Compton Scattering
Si
CdTe
CdTe
Absorption
Figure8.15 Conceptual design of semiconductor Compton camera consisting of layers

of thin silicon and CdTe imaging layers. (Reproduced from Watanabe, S. et al., IEEE Trans.
Nucl. Sci., 2009. Copyright 2009 IEEE. With permission.)
The Si/CdTe Compton camera is a full-semiconductor Compton camera. It differs

from conventional Compton cameras in the sense that it uses many layers of silicon and
CdTe imaging detectors.63,64 In principle, each layer could act not only as a scattering
part but also as an absorber part. Figure8.15 shows the conceptual design of the Si/CdTe
semiconductor Compton camera. In order for the detector to cover a wide energy band
and the wide scattering angle of Compton interaction in the device, the energy threshold
of the detector must be low. Moreover, high Compton scattering probability is desirable
to obtain high efficiency as a Compton camera. Combining a low-Z material (Si) and a
high-Z material (CdTe) is suitable for obtaining high detection efficiency. In addition,
silicon works better than other semiconductor materials with larger atomic numbers
relative to the Doppler broadening effect. 65 For low-energy gamma rays (less than several hundred kiloelectron volts), the task of CdTe is to measure the energy of scattering
gamma rays through photoabsorption. When the energy becomes higher and the effect
of Doppler broadening in the angular resolution becomes smaller, Compton interactions
recorded in CdTe layers could be used to improve the detection efficiency.64
As gamma-ray energy increases to the megaelectron volt region, detection
becomes more difficult. One attractive idea to improve the detection efficiency is a
concept called the multiple Compton method,59,64,66 which uses the events with more
than one Compton interaction per event. A stack of many thin layers in the Si/CdTe
Compton camera can be used to record multiple Compton interactions in the camera.
The order of the interaction sequences, hence the correct energy and direction of the
incident photon, can be reconstructed by examination of the energy-momentum conservation for all possible sequences. Another important point of this method is that if
the incident gamma rays undergo at least the first three or four Compton scatterings,
187
Figure 8.16 Photo of the prototype Si/CdTe Compton camera. Four layers of CdTe
pixel detectors were placed underneath four layers of double-sided silicon strip detectors.
(Reproduced from Takeda, S. et al., IEEE Trans. Nucl. Sci., in press. Copyright IEEE. With
permission.)
which are sufficient to reconstruct the correct order, gamma rays do not need to be
fully absorbed by the detector and are allowed to escape.
Based on the development of high-resolution CdTe imaging detectors developed
in conjunction with the development of double-sided silicon strip detectors and lownoise readout ASICs, both imaging and spectral performances have been demonstrated for the Si/CdTe Compton camera.50,6770 Compton imaging for 60- to 662-keV
gamma-ray photons was successfully performed with a prototype, consisting of four
layers of CdTe pixel detectors placed underneath four layers of double-sided silicon strip detectors (Figure8.16). As shown in Figure8.17 (left), the angular resolution of Compton imaging was about 2 degrees for a point-like gamma-ray source
with 511-keV energy by a simple backprojection method. This value is consistent
with the theoretical limit due to the Doppler broadening effect. On the other hand,
by reconstructing events that scatter in a CdTe layer and are absorbed in another
CdTe layer, the angular resolution becomes about 10.67 Figure8.17 (right) shows the
background subtraction using the image. If we extract events in which the Compton
cone intersects with the bright spot in the image, most of the low-energy continuum
gamma rays disappear from the spectrum. Figure8.18 demonstrates the high performance of the Si/CdTe Compton camera in a reconstructed image of an extended
source obtained with the prototype. The target is soaked with a liquid radioisotope of
131I and located above the Compton camera at a distance of 3 cm. The C-like shape is
properly reconstructed, and the 3-mm gap is clearly resolved after image deconvolution. The prototype also achieved a 100 large field of view.70
8.7Conclusion
The recent advances made in CdTe detectors relative to diode configuration were
reviewed. After 10 years of research and development, CdTe diode detectors are now
188
300
Radius 2 deg
y (mm)
200
0.07
0.06
0.05
100
0.04
0.03
100
200
0.02
300
0.01
400
0
400300200100 0 100 200 300 400
x (mm)
300
511 keV
FWHM 8.0 keV
250
22
Na
200
Counts
400
150
100
50
0
100
200
300
400
Energy (keV)
500
600
Figure 8.17 (Left) Reconstructed image for a 511-keV point source with the Si/CdTe
Compton camera by a simple backprojection method. (Right) Spectra.69
Gap ~3 mm
Figure8.18 Reconstructed image with the Si/CdTe Compton camera. The 3-mm gap in the
C-shaped target is clearly resolved. The distance between the Compton camera and target was
3 cm. The 364-keV gamma-ray photons from 131I were used for Compton imaging. (Reproduced
from Takeda, S. et al., IEEE Trans. Nucl. Sci., in press. Copyright IEEE. With permission.)69
in the phase of real application. The fine energy and position resolution obtained
with the CdTe pixel detectors or CdTe DSDs are expected to lead to dramatically
improved performance in the area of gamma-ray imaging. In the fifth Japanese X-ray
satellite, ASTRO-H, CdTe diode detectors are used for both a hard X-ray imager
as a focal plane detector of the hard X-ray mirror and a soft gamma-ray detector
which is designed based on the concept of a narrow field-of-view Compton camera.71
Therefore, developing a CdTe imager and Si/CdTe Compton camera developed for
the ASTRO-H satellite would offer improved sensitivity of the gamma-ray detection
for various applications, including gamma-ray astronomy, nuclear medical imaging,
and nondestructive industrial imaging.
Acknowledgments
This research was conducted in collaboration with K. Nakazawa, G. Sato, T. Mitani,
K. Oonuki, K. Tamura, T. Kishishita, S. Takeda, M. Ushio, J. Katsuta (Institute of
189
Space and Astronautical Science [JAXA]), H. Tajima, T. Tanaka (Stanford Linear

Accelerator Center), M. Nomachi (Osaka University), Y. Fukazawa (Hiroshima
University), N. Kawachi, K. Arakawa (Japan Atomic Energy Agency [JAEA]), and
T. Nakano (Gunma University) and was funded by Japans Ministry of Education,
Science, Sports and Culture, through Grants-in-Aid in Japan (13304014, 14079207).
Images in Figure8.7 were taken in an experiment jointly conducted with the medical
department at Gunma University and JAEA.
References

1. G. F. Knoll, Radiation Detection and Measurement, 3rd ed., Wiley, 1999.

2. T. Takahashi and S. Watanabe, Recent progress on CdTe and CdZnTe detectors, IEEE
Trans. Nucl. Sci., 48, 950959, 2001.
3. P. N. Luke, Room-temperature replacement for Ge detectors? Are we there yet?, IEEE
Trans. Nucl. Sci., 54, 834842, 2008.
4. A. Owens and A. Peacock, Compound semiconductor radiation detectors, Nucl. Instr.
Methods, A531, 1837, 2004.
5. M. Funaki, T. Ozaki, K. Satoh, and R. Ohno, Growth and characterization of CdTe single
crystal for radiation detectors, Nucl. Instr. Methods, A436, 120126, 2000.
6. F. P. Doty, J. F. Butler, J. F. Schetzina, and K. A. Bowers, Properties of CdZnTe crystals
grown by a high pressure Bridgman method, J. Vac. Sci. Technol., B10, 14181422,
1992.
7. J. F. Butler, C. L. Lingren, and F. P. Doty, Cd1xZnxTe gamma-ray detectors, IEEE Trans.
Nucl. Sci., 39(4), 605609, 1992.
8. C. Szeles and M. C. Driver, Growth and properties of semi-insulating CdZnTe for radiation detector applications, Proc. SPIE, 3446, 18, 1998.
9. C. Szeles et al., Development of the high-pressure electro-dynamic gradient crystalgrowth technology for semi-insulating CdZnTe growth for radiation detector applications,
J. Elec. Mater., 33(6), 2004.
10. H. Chen et al., Characterization of large cadmium zinc telluride crystals grown by traveling
heater method, J.Appl. Phys., 103, 014903-1-5, 2008.
11. H. Chen et al., High-performance, large-volume THM CdZnTe detectors for medical
imaging and homeland security applications, IEEE Trans. Nucl. Sci., 6, 36293637,
2006.
12. H. Shiraki, M. Funaki, Y. Ando, A. Tachibana, S. Kominami, and R. Ohno, THM growth
and characterization of 100-mm diameter CdTe single crystals, IEEE Nucl. Sci. Symp.
Conf. Rec., 126132, 2008.
13. H. Shiraki, M. Funaki, Y. Ando, S. Kominami, K. Amemiya, and R. Ohno, Improvement
of the productivity in the growth of CdTe single crystal by THM for the new PET system, IEEE Nucl. Sci. Symp. Conf. Rec., 3, 17831787, 2007.
14. P. N. Luke, Single-polarity charge sensing in ionization detectors using coplanar electrodes, Appl. Phys. Lett., 65, 28842886, 1994.
15. H. H. Barrett, J. D. Eskin, and H. B. Barber, Charge Transport in arrays of semiconductor gamma-ray detector, Phys. Rev. Lett., 75, 156159, 1995.
16. H. B. Barber, D. G. Marks, B. A. Apotovsky, F. L. Augustine, H. H. Barret, J. F. Butler,
E. L. Dereniak, et al., Progress in developing focal-plane-multiplexer readout for
large CdZnTe arrays for nuclear medicine applications, Nucl. Instr. Methods, A380,
262265, 1996.
17. J. F. Butler, Novel electrode design for single-carrier charge collection in semiconductor
nuclear detectors, Nucl. Instr. Methods, A396, 427430, 1997.
190
18. T. Takahashi, B. Paul, K. Hirose, C. Matsumoto, R. Ohno, T. Ozaki, K. Mori, and

Y. Tomita, High-resolution Schottky CdTe diode for hard X-ray and gamma-ray astronomy, Nucl. Instr. Methods, A436, 111119, 1999.
19. T. Takahashi, K. Hirose, C. Matsumoto, K. Takizawa, R. Ohno, T. Ozaki, K. Mori, and
Y. Tomita, Performance of a new Schottky CdTe detector for hard X-ray spectroscopy,
Proc. SPIE, 3446, 2937, 1998.
20. C. Matsumoto, T. Takahashi, K. Takizawa, R. Ohno, T. Ozaki, and K. Mori, Performance
of a new Schottky CdTe detector for hard X-ray spectroscopy, IEEE Trans. Nucl. Sci.,
45, 428432, 1998.
21. T. Takahashi, B. Paul, K. Hirose, C. Matsumoto, R. Ohno, T. Ozaki, K. Mori, and
Y. Tomita, High-resolution Schottky CdTe detectors for hard X-ray and gamma-ray
observations, Nucl. Instr. Methods, A436, 111119, 2000.
22. T. Takahashi, S. Watanabe, M. Kouda, G. Sato, Y. Okada, S. Kubo, Y. Kuroda, M. Onishi,
and R. Ohno, High resolution CdTe detector and applications to imaging devices, IEEE
Trans. Nucl. Sci., 48, 287291, 2001.
23. H. B. Barber, Applications of semiconductor detectors to nuclear medicine, Nucl. Instr.
Methods, A436, 102111, 1999.
24. C. Scheiber, CdTe and CdZnTe gamma ray detectors in nuclear medicine, Nucl. Instr.
Methods, A448, 513524, 2000.
25. D. G. Darambara, State-of-the-art radiation detectors for medical imaging: demands and
trends, Nucl. Instr. Methods, A569, 153158, 2006.
26. L. Verger et al., Performance and perspectives of a gamma camera based on CdZnTe for
medical imaging, IEEE Trans. Nucl. Sci., 51, 31113117, 2004.
27. O. Limousin, C. Blondel, J. Cretolle, H. Dzitko, P. Laurent, F. Lebrun, J. P. Leray, et al.
The ISGRI CdTe gamma-ray camera first steps, Nucl. Instr. Methods, A442, 244249,
2000.
28. G. Sato, A. Parsons, D. Hullinger, M. Suzuki, T. Takahashi, M. Tashiro, K. Nakazawa,
et al. Development of a spectral model based on charge transport for the Swift/BAT 32K
CdZnTe detector array, Nucl. Instr. Methods, A541, 372384, 2005.
29. Amptek Inc., home page, http://www.amptek.com
30. Clear Pulse Company Ltd., home page, http://www2.clearpulse.co.jp/indexEng.html
31. T. Takahashi, T. Mitani, Y. Kobayashi, M. Kouda, G. Sato, S. Watanabe, K. Nakazawa,
Y. Okada, M. Funaki, and R. Ohno, High resolution Schottky CdTe diodes, IEEE Trans.
Nucl. Sci., 49(3), 12971303, 2002.
32. T. Takahashi, K. Nakazawa, S. Watanabe, G. Sato, T. Mitani, T. Tanaka, K. Oonuki,
K. Tamura, H. Tajima, T. Kamae, G. Madejski, M. Nomachi, Y. Fukazawa, K.
Makishima, M. Kokubun, Y. Terada, J. Kataoka, and M. Tashiro, Application of CdTe
for the NeXT mission, Nucl. Instr. Methods, A541, 332341, 2005.
33. K. Nakazawa, K. Oonuki, T. Tanaka, Y. Kobayashi, K. Tamura, T. Mitani, G. Sato,
S. Watanabe, T. Takahashi, R. Ohno, A. Kitajima, Y. Kuroda, and M. Onishi, Improvement
of the CdTe diode detectors using a guard-ring electrode, IEEE Trans. Nucl. Sci., 51(4),
18811885, 2004.
34. H. Toyama, A. Nishihira, M. Yamazato, A. Higa, T. Maehara, R. Ohno, and M. Toguchi,
Formation of aluminum Schottky contact on plasma-treated cadmium telluride surface,
J. Appl. Phys., 43, 63716375, 2004.
35. H. Toyama, M. Yamazato, A. Higa, T. Maehara, R. Ohno, and M. Toguchi, Effect of He
plasma treatment on the rectification properties of Al/CdTe Schottky contacts, J. Appl.
Phys., 44, 67426746, 2005.
36. S. Watanabe, S. Ishikawa, S. Takeda, H. Odaka, T. Tanaka, T. Takahashi, K. Nakazawa,
M. Yamazato, A. Higa, and S. Kaneku, New CdTe pixel gamma-ray detector with pixelated Al Schottky anodes, J. Appl. Phys., 46, 60436045, 2007.
191
37. G. Sato, Development and evaluation of CdTe/CdZnTe detectors for space applications,
masters thesis, University of Tokyo, 2002.
38. H. L. Malm and M. Martini, Polarization phenomena in CdTe nuclear radiation detector,
IEEE Trans. Nucl. Sci., 21, 322330, 1974.
39. S. Ishikawa, H. Aono, S. Watanabe, S. Takeda, K. Nakazawa, and T. Takahashi,
Performance measurements of Al/CdTe/Pt pixel diode detectors, Proc. SPIE, 6706,
67060M-67060M-10, 2007.
40. L. Rossi, P. Fischer, R. Tilman, and N. Wermes, Pixel Detectors: From Fundamentals to
Applications, from Fundamentals to Applications series, Springer-Verlag, Berlin, 2006.
41. M. Chmeissani et al., First experimental tests with a CdTe photon counting pixel detector hybridized with a Medipix2 readout chip, IEEE Trans. Nucl. Sci., 51, 23792385,
2004.
42. S. Basolo et al., A 20k pixels CdTe photon-counting imager using XPAD chip, Nucl.
Instr. Methods, A589, 268274, 2008.
43. F. Harrison, A. E. Bolotnikov, C. M. H. Chen, W. R. Cook, P. H. Mao, and S. M. Schindler,
Development of a high spectral resolution cadmium zinc telluride pixel detector for
astrophysical applications, Proc. SPIE, 4851, 823830, 2003.
44. M. Loecker, P. Fischer, S. Krimmel, H. Krueger, M. Lindner, K. Nakazawa,
T. Takahashi, and N. Wermes, Single photon counting X-ray imaging with Si and CdTe
single chip pixel detectors and multichip pixel modules, IEEE Trans. Nucl. Sci., 51,
17171723, 2004.
45. K. Oonuki, H. Inoue, K. Nakazawa, T. Mitani, T. Tanaka, T. Takahashi, H. C. M. Chen,
W. R. Cook, and F. A. Harrison, Development of uniform CdTe pixel detectors based on
Caltech ASIC, Proc. SPIE, 5501, 218228, 2004.
46. Z. He, W. Li, G. F. Knoll, D. K. Wehe, J. Berry, and C. M. Stahle, 3-D position sensitive
CdZnTe gamma-ray spectrometers, Nucl. Instr. Methods, A422, 173178, 1999.
47. F. Zhang, Z. He, G. F. Knoll, D. K. Wehe, and J. E. Berry, 3-D position sensitive CdZnTe
spectrometer performance using third-generation VAS/TAT readout electronics, IEEE
Trans. Nucl. Sci., 52, 20092016, 2005.
48. S. Watanabe, S. Takeda, S.N. Ishikawa, H. Odaka, M. Ushio, T. Tanaka, K. Nakazawa,
T. Takahashi, H. Tajima, Y. Fukazawa, Y. Kuroda, and M. Onishi, Development of semiconductor imaging detectors for a Si/CdTe Compton camera, Nucl. Instr. Methods,
A579, 871877, 2007.
49. H. Tajima, T. Nakamoto, T. Tanaka, S. Uno, T. Mitani, E. do Couto e Silva, Y. Fukazawa,
T. Kamae, G. Madejski, D. Marlow, K. Nakazawa, M. Nomachi, Y. Oakada, and
T. Takahashi, Performance of a low noise front-end ASIC for Si/CdTe detectors in
Compton gamma-ray telescope, IEEE Trans. Nucl. Sci., 51, 842847, 2004.
50. T. Tanaka, S. Watanabe, S. Takeda, K. Oonuki, T. Mitani, K. Nakazawa, T. Takashima,
T. Takahashi, H. Tajima, N. Sawamoto, Y. Fukazawa, and M. Nomachi, Recent results
from a Si/CdTe semiconductor Compton telescope, Nucl. Instr. Methods, A568, 375381,
2006.
51. S. Watanabe, T. Takahashi, Y. Okada, G. Sato, M. Kouda, T. Mitani, Y. Kobayashi,
K. Nakazawa, Y. Kuroda, and M. Onishi, CdTe stacked detectors for gamma-ray detection, IEEE Trans. Nucl. Sci., 49(3), 12921296, 2002.
52. S. Watanabe, T. Takahashi, K. Nakazawa, Y. Kobayashi, Y. Kuroda, K. Genba, M. Onishi,
and K. Otake, A stacked CdTe gamma-ray detector and its application as a range finder,
Nucl. Instr. Methods, A505, 118121, 2003.
53. R. Redus, A. Huber, J. Pantazis, T. Pantazis, T. Takahashi, and S. Woolf, Multielement
CdTe stack detectors for gamma-ray spectroscopy, IEEE Trans. Nucl. Sci., 51, 23862394,
2004.
192
54. G. Montemont, M.C. Gentet, O. Monnet, J. Rustique, and L. Verger, Simulation and
design of orthogonal capacitive strip CdZnTe detectors, IEEE Trans. Nucl. Sci., 54,
854859, 2007.
55. J. R. Macri, B. A. Apotovsky, J. F. Butler, M. L. Cherry, B. K. Dann, A. Drake, F. P.
Doty, T. G. Guzik, K. Larson, M. Mayer, M. L. Mayer, M. L. McConnell, and J. M.
Ryan, Development of an orthogonal-stripe CdZnTe gamma radiation imaging spectrometer, IEEE Trans. Nucl. Sci., 43, 14581462, 1996.
56. A. Shor, Y. Eisen, and I. Mardor, Optimum spectroscopic performance from CZT
gamma- and X-ray detectors with pad and strip segmentation, Nucl. Instr. Methods,
A428, 182192, 1999.
57. M. L. McConnel, J. R. Macri, J. M. Ryan, K. Larson, L.-A. Hamel, G. Bernard, et al.,
Three-dimensional imaging and detection efficiency performance of orthogonal coplanar CZT strip detectors, Proc. SPIE, 4141, 157167, 2000.
58. S. Watanabe, S. Ishikawa, H. Aono, S. Takeda, H. Odaka, M. Kokubun, T. Takahashi,
K. Nakazawa, H. Tajima, M. Onishi, and Y. Kuroda, High energy resolution hard X-ray
and gamma-ray imagers using CdTe diode devices, IEEE Trans. Nucl. Sci., in press,
2009.
59. T. Kamae, R. Enomoto, and N. Hanada, A new method to measure energy, direction,
and polarization of gamma rays, Nucl. Instr. Methods, A260, 254, 1987.
60. J. W. LeBlanc, N. H. Clinthorne, C. H., Hua, E. Nygard, W. L. Rogers, D. K. Wehe,
P. Weilhammer, and S. J. Wilderman, C-SPRINT: A prototype Compton camera system
for low energy gammaray imaging, IEEE Trans. Nucl. Sci., 45, 943949, 1998.
61. J. E. McKisson, P. S. Haskins, G. W. Phillips, S. E. King, R. A. August, R. B. Piercey, and
R. C. Mania, Demonstration of three-dimensional imaging with a germanium Compton
camera, IEEE Trans. Nucl. Sci., 41, 11821189, 1994.
62. E. A. Wulf, B. F. Philips, W. N. Johnson, R. A. Kroeger, J. D. Kurfess, and E. I. Novikova,
Germanium strip detector Compton telescope using three-dimensional readout, IEEE
Trans. Nucl. Sci., 50, 12211224, 2003.
63. T. Takahashi, K. Makishima, and T. Kamae, Future hard X-ray and gamma-ray observations, Astron. Soc. Pacific, 251, 210213, 2001.
64. T. Takahashi, K. Nakazawa, T. Kamae, H. Tajima, Y. Fukazawa, M. Nomachi, and
M. Kokubun, High resolution CdTe detectors for the next generation multi-Compton
gamma-ray telescope, Proc. SPIE, 4851, 12281235, 2002.
65. A. Zoglauer and G. Kanbach, Doppler broadening as a lower limit to the angular resolution of next-generation Compton telescopes, Proc. SPIE, 4851, 13021309, 2002.
66. R. A. Kroeger, W. N. Johnson, J. D. Kurfess, B. F. Phillips, and E. A. Wulf, ThreeCompton telescope: theory, simulations, and performance, IEEE Trans. Nucl. Sci., 49,
18871892, 2002.
67. K. Oonuki, T. Tanaka, S. Watanabe, S. Takeda, K. Nakazawa, T. Mitani, T. Takahashi,
H. Tajima, Y. Fukazawa, and M. Nomachi, Results of a Si/CdTe Compton telescope,
Proc. SPIE, 5922, 7888, 2005.
68. S. Watanabe, T. Tanaka, K. Nakazawa, T. Mitani, K. Oonuki, T. Takahashi, T. Takashima,
H. Tajima, Y. Fukazawa, M. Nomachi, S. Kubo, M. Onishi, and Y. Kuroda, A Si/CdTe
semiconductor Compton camera, IEEE Trans. Nucl. Sci., 52, 20452051, 2005.
69. S. Takeda, H. Aono, S. Okuyama, S. Ishikawa, H. Odaka, S. Watanabe, M. Kokubun,
T. Takahashi, K. Nakazawa, H. Tajima, and N. Kawachi, Experimental results of the
gamma-ray imaging capability with a Si/CdTe semiconductor Compton camera, IEEE
Trans. Nucl. Sci., in press.
70. S. Takeda, Experimental study of a Si/CdTe semiconductor Compton camera for the
next generation of gamma-ray astronomy, PhD thesis, University of Tokyo, 2009.
71. T. Takahashi et al., The NeXT Mission (ASTRO-H), Proc. SPIE, 7011, 70110O-1
70110O-14, 2008.
9 Caliste
Microcamera for Hard
X-Ray Astronomy
Olivier Limousin, Aline Meuris,
Olivier Gevin, and Francis Lugiez
CEA, Irfu
Contents
9.1 New Needs for Detection Systems in X- and Gamma-Ray Astronomy........ 193
9.1.1 Introduction....................................................................................... 193
9.1.2 Heritage of INTEGRAL..................................................................... 195
9.1.3 New Era of Direct Imaging in Hard X-Rays..................................... 196
9.2 Caliste Camera: Innovative Technology for Elementary Detection Units.... 199
9.2.1 Introduction....................................................................................... 199
9.2.2 Camera Description...........................................................................200
9.2.2.1 Detectors............................................................................. 201
9.2.2.2 Front-End Electronics.........................................................204
9.2.2.3 Hybrid Component.............................................................. 205
9.2.3 Performance.......................................................................................208
9.2.3.1 Setup...................................................................................208
9.2.3.2 Spectroscopy....................................................................... 210
9.2.3.3 Timing Issues...................................................................... 212
9.3 Perspectives................................................................................................... 214
9.3.1 Future Needs for Hard X-Ray Astronomy......................................... 215
9.3.2 Future Needs for Gamma-Ray Astronomy........................................ 215
Bibliography........................................................................................................... 216
9.1New Needs for Detection Systems

in X- and Gamma-Ray Astronomy
9.1.1Introduction
The discovery of the X-ray sky during the 1960s1980s opened a fundamental field of
research in astrophysics, recognized by the 2002 Nobel Prize in Physics to Riccardo
Giacconi, a pioneer. The increasing power of space observatories has revealed highenergy emission from objects of all scales in the universe, from compact sources
193
194
such as black holes and neutron stars to diffuse hot plasma pervading galaxies and
clusters of galaxies. The high-energy emission processes at work in these objects
are related to the two domains of accretion physics and particle acceleration mechanisms. Understanding the accretion process of matter onto black holes is fundamental in astrophysics since it has profound implications in physics (general relativity
testing in the black hole environment) and cosmology (interaction of matter with
supermassive black holes is suspected to play a role in the formation and evolution
of host galaxies). On the other hand, the origin of the highest-energy photons and
cosmic rays is still unknown, and firm evidence of hadron acceleration in suspected
astronomical sites is missing. Observing these processes at work in various acceleration sites is mandatory for resolving these issues.
The accretion process and acceleration mechanisms are best revealed by their
emissions in the X-ray and hard X-ray ranges, the latter probing the most energetic
and violent environments and their nonthermal population of energetic particles.
Below about 10 keV, astrophysics missions such as XMM-Newton and Chandra are
using X-ray mirrors based on grazing incidence reflection properties. This achieves
an extremely good angular resolution, down to 0.5 arcsec for Chandra, and a good
signal-to-noise ratio thanks to the focusing of the X-rays onto a small detector surface. This technique has been so far limited to energies below about 10 keV because
of the maximum focal length that can fit in a single rocket fairing. Hard X-ray and
gamma-ray imaging instruments are thus using a different imaging technique, that
of coded masks, as those onboard the International Gamma-Ray Astrophysics
Laboratory (INTEGRAL) or Swift missions. This nonfocusing technique has intrinsically a much lower signal-to-noise ratio than focusing telescopes and does not
allow angular resolutions better than a few arc minutes. In addition to the difference
in angular resolution, there is also roughly two orders of magnitude of difference in
point source sensitivity between X-ray and gamma-ray telescopes.
This transition of techniques unfortunately happens roughly at the energy above
which the identification of a nonthermal component is unambiguous with respect
to thermal emission. Considered from the low-energy side, this obviously strongly
limits the interpretation of the high-quality X-ray measurements, particularly those
related to the acceleration of particles. Considered from the high-energy side, this
prevents mapping the gamma-ray emission of extended sources to the required scales
to understand the emission mechanisms by comparing with lower-energy data. The
hard X-ray range is the energy domain at which fundamental problems of astrophysics have their essential signatures, either via nonthermal emissions characterizing
populations of particles accelerated to extreme energies or via thermal emissions
revealing the presence of very hot comptonizing plasmas such as those believed to
exist close to compact objects. In addition, whereas low-energy X-rays are stopped
by a relatively small amount of matter, hard X-rays are extremely penetrating and
can reveal sources that are otherwise left hidden. This has been shown in particular
by the measurements made by the INTEGRAL Soft Gamma-Ray Imager (ISGRI)
which has led to the discovery of a number of highly absorbed sources, mainly pulsars in supergiant systems.
A clear requirement for future high-energy astrophysics missions is thus to bridge
this gap of sensitivity by offering instrumentation in the hard X-ray range with a
Caliste
195
sensitivity and angular resolution similar to those of the current X-ray telescopes.
To take up this challenge, hard X-ray focusing optics is mandatory. Such optics can
readily be implemented by a simple extension of the current X-ray mirror technology
to long focal lengths.
9.1.2Heritage of INTEGRAL
The INTEGRAL satellite has been successfully operating since 2002. The mission is
devoted to gamma-ray sky observation between 15 keV and up to 10 MeV. The payload comprises two main instruments: IBIS (Imager onboard INTEGRAL satellite),
an imager with spectroscopic capabilities and spectrometer for INTEGRAL (SPI), a
cooled germanium spectrometer with imaging capabilities.
IBIS is based on a static passive coded mask aperture combined with a gamma
camera for the detector. With respect to previous telescopes of this kind, such as
SIGMA (Systme dImageria Gamma Masque Alatoire); onboard Granat satellite, IBIS uses two gamma cameras with discrete pixelsa CdTe array up to 1 MeV
and a CsI array above 1 MeVinstead of an Anger camera type. Pixel detector
technology enables drastic improvement of the sensitivity and significant extension
of the energy range down to 15 keV instead of 30 keV.
The angular resolution of a coded mask telescope is given by its mask element
size and by the mask-to-detector distance. The latter is only limited by the mission
resources (dimension, mass). The imaging performance of a coded mask telescope
is governed by the spatial resolution of the detector. Each pixel of the camera is an
individual spectrometric detector. Semiconductor detectors operating at room temperature with high Z and high density are most promising. Among them, CdTe is
the most mature and can be processed with any arbitrarily small size. Consequently,
limitation on the pixel size comes mainly from the number of electronic channels
one can operate simultaneously since one preamplifier per pixel is necessary. To
limit the volume, the electronics power consumption, and the related heat dissipation, dedicated integrated circuits (application-specific integrated circuits [ASICs])
are used. These considerations have driven the IBIS design. The detector of IBIS is
made up of two square detection planes of the same dimensions. Both layers are pixel
gamma cameras. The upper one, ISGRI (visible in Figure9.1), covers the low-energy
domain from 15 to about 200 keV, while the lower one pixelated imaging CsI telescope (PICsIT) takes care of the higher energies.
The ISGRI pixels are square CdTe detectors 4 4 mm and 2 mm thick. There
are 16,384 pixels, providing a useful area of 2,600 cm2. ISGRI is the first very large
CdTe gamma camera to be flown. Thanks to a mask-to-detector distance of 3.15,
12 arcmin angular resolution is achieved. ISGRI provides sky images with 5 arcmin
sampling, resulting in source location accuracy better than 1 arcmin for the brightest
sources.
The successful use of a very large amount of CdTe detectors in space with ISGRI
was a strong motivation to further develop a CdTe-based space gamma camera at
CEA (Commissariat lEnergie Atomique, France). We improved simultaneously
the energy resolution, the low-energy threshold, and the spatial resolution of the sensors toward imaging spectrometers with high pixel density.
196
Figure 9.1 View of the large ISGRI CdTe-based spectroimager inside the IBIS telescope
onboard the INTEGRAL satellite (ESA), flying since 2002. The camera is equipped with
16,384 CdTe crystals covering a 2,600-cm sensitive area. The ISGRI camera has been realized by CEA-Saclay with the support of CNES; this picture was taken at the LABEN plant in
Milan during the INTEGRAL/IBIS integration in summer 2001.
9.1.3New Era of Direct Imaging in Hard X-Rays

Thanks to current emerging space technologies, the design of orbital missions using
multiple spacecraft flying in a constrained formation is now possible. At the beginning of 2004, the French space agency (Centre National dtudes Spatales [CNES])
issued a call for a proposal for a scientific payload to be flown on a formation flight
demonstrator mission to be launched during the next decade. The Simbol-X mission has been proposed in this context by Dr. Ph. Ferrando and Dr. G. Pareschi,
astrophysicists at Laboratoire Astroparticules et Cosmologie (CEA/APC) and at
Observatorio Astronomico di Brera, respectively. Simbol-X is a hard X-ray pointed
deformable telescope and is based on a 20- to 30-m long focal length optics that
will extend the focusing techniques used in the X-ray domain to energies up to at
least about 80 keV. To do so, a set of two satellites is used: The first satellite carries
hard X-ray grazing incidence mirrors, and a second satellite houses a focal plane.
The telescope will offer an increase in sensitivity (see Figure9.2, which shows the
continuum sensitivity curve of Simbol-X vs. currently flying INTEGRAL/IBIS in the
same energy range, with the sensitivity defined as the minimum observable flux of a
source in a given exposure time at a given confidence level) and angular resolution by
two orders of magnitude compared to the current instruments in flight above 10 keV
(i.e., ISGRI onboard the INTEGRAL satellite).
The parameters that are driving the design of the focal plane assembly are (a) the
energy range; (b) the spectral resolution for the iron lines at 6.4 KeV and 44Ti lines
at 67.9 and 78.4 keV; (c) the 12-arcmin field of view of the optics; (d) the request
for oversampling the optics point-spread function of 20 arcsec half energy width v
197
Caliste
Exposure 1 Ms 3 detection E = E/2
105
Photons cm2 s1 keV1
Integral ISGRI
106
1 mCr
ab
107
108
1 C
rab
109
20
40
60
Energy (keV)
Simbol=X
80
100
Figure 9.2 Continuum sensitivity of Simbol-X for source detection at three sigmas for
a 106 exposure time and comparison with INTEGRAL/ISGRI in the same energy domain.
(Courtesy of Ph. Ferrando, CEA/APC.)
because of the need for a very low background. This has led to use of a combination
of a low- and a high-energy detector on top of each other, surrounded by a combination of active and passive shielding.
The focal plane of such a mission requires sensors in the range from 0.5 to 80 keV
at least. Below 10 keV, silicon sensors are well suited for performing high angular and high spectral resolution. Since charge-coupled devices (CCDs) are slow to
read out (in the millisecond region), new active pixel sensors based on monolithic
DEPFET macropixel arrays will be used in the program since they are much faster
(hundreds of microseconds). Assuming such technology, high spectral response is
achievable (<150 eV full width at half maximum [FWHM] at 6 keV) together with
fast readout.
Above 10 keV, the unprecedented focusing technique necessitates the development
of entirely new detector concepts. Owing to its well-known advantages (i.e., high
efficiency up to about 100 keV, room temperature or moderately cooled operations,
good spectrometric performance, and adequate maturity for space applications), we
chose to go ahead with Cd(Zn)Te semiconductor detectors. Moreover, this kind of
detector has been extensively and successfully used in space. Currently, CdTe-based
detectors are also taking advantage of mature electrode deposition techniques by a
photolithography process that enables high-quality pixelated contacts on monolithic
single crystals. The pixel size is currently only limited by the mounting technology
on hybrid modular components and by the power budget that constrains the number
of independent active channels into the camera.
Assuming a 20-m focal length with a 12-arcmin field of view having a 20-arcsec
angular resolution at 30 keV, we derive the camera sensitive area to be 64 cm 2. This
is much larger than any available Cd(Zn)Te single crystal. Such a camera is inescapably an assembly of subarrays, possibly a mosaic of elementary subarrays installed
198
Table9.1
Simbol-X Top Level Requirements Illustrating Detector Needs
Parameter
Simbol-X Requirement
Associated Detector Parameters
Energy band
~0.5 to > 80 keV
Spectral resolution
E/E = 40 at 6 keV
E/E = 50 at 68 keV
Field of view
(50% vignetting)
Angular resolution
>9 arcmin; goal 13 arcmin

(diameter)
<20 arcsec; goal 15 arcsec
Silicon pixel sensor < 10 keV

Cd(Zn)Te sensor up to 80 keV, 1- to
2-mm thick crystals
<10 electrons root mean square (rms) at
Si level
<100 electrons rms at Cd(Zn)Te level
64-cm sensitive area
Absolute pointing
reconstruction
On-axis effective area
~2 arcsec (radius 90%)
On-axis continuum
sensitivity
>600 cm2 at 8 keV

>450 cm2 at 2040 keV
>100 cm2 at 70 keV
<1 Crab (3 , 1 Ms) in 20- to
40-keV band
On-axis line sensitivity at

68 keV
<107 ph cm2 s1 (3 , 1 Ms)
Absolute timing accuracy
<100 s; goal 50 s
Mission duration
>2 years of scientific programs

(~3 years of mission)
625-m pixel pitch

16,384 channels
No constraint at detector level
Need for anticoincidence system,

requires high timing accuracy at
Cd(Zn)Te level < 300 ns
Need for anticoincidence system,
requires high timing accuracy at
Cd(Zn)Te level < 300 ns rms
No constraint from science at detector
level
Modular detection units, high reliability
and radiation hardness
next to each other. On the other hand, to perform significant oversampling of the
focal point, about 3 mm in diameter at half energy width in this configuration, a
750-m maximum pixel pitch is required to obtain sufficient image reconstruction
and accurate source localization. Simbol-X will use 625-m pixel pitch. Its 64-cm2
detector will house 16,384 independent spectrometric channels.
Combining small pixel detectors (small detector capacitance and low dark current) operated at moderately cool temperature (approximately 40 to 20C), new
development in low-noise and low-power microelectronics front ends and a suitable
hybridization process, it is possible to achieve a uniform spectral response much
better than 1.3-keV FWHM at 68 keV, specified by scientists for study of envelope
ejection speed in supernova remnants, for instance. Top-level requirements and associated detector parameters are summarized in Table9.1.
In hard X-ray astronomy, one has to observe very faint celestial sources, typically
a total of a few photons per minute detected from a source. Consequently, most of the
events recorded by a gamma-ray space instrument are due to cosmic ray background
and induced background (~one event per second per square centimeter). To optimize
Caliste
199
the background rejection onboard, we use an anticoincidence system that surrounds

the hard X-ray imager. This entails severe constraints at the hard X-ray detector level
in terms of dead time and timing accuracy. First, the dead time remains negligible
(~1%), having self-triggered mode readout. In this way, the dead out time is affected
only by the concerned pixel readout duration (~40 s), while other channels remain
open. On the other hand, high timing accuracy (~300 ns rms) is achievable even if
challenging when the pixels are read out independently (but no frame), in a photon
photon mode. CdTe has a reputation for being slow due to its charge carrier transport
properties. However, because small pixels diminish significantly the amount of dark
current in a chain, it is possible to set an internal electric field to the CdTe crystal
high enough to accelerate charge carriers and reach almost complete charge collection in a shorter time than the shaping time, avoiding ballistic deficit. Consequently,
the timing accuracy depends on the shaper transfer function, the low threshold value
(set as low as possible), and the signal shape at the discriminator level. Finally, the
timing accuracy also depends strongly on the energy deposit and the kind of particle
that interacts, basically protons or photons. For instance, a large energy deposit will
provoke faster triggering than an energy deposit close to the threshold level. This
induces time-tagging offsets, namely, time-walk, which may be calibrated on ground
and corrected afterward. The timing accuracy is the residual error after correction.
The time-tagging offsets are rather constant when energy deposits are above 100 keV
or so. Note that accurate timing needs some data processing to compute the real
event interaction date from the trigger date; consequently, it applies to digital anticoincidence systems. This calculation may be operated in real time onboard.
Finally, CdTe-based detectors for astrophysics must obviously comply with the
space manufacturing and reliability rules. For instance, from their conception,
devices must be tolerant to radiation (dose and single event latch-up or upsets) and
must survive the launch phase.
The new development of the Caliste microgamma camera is taking into account
all these requirements simultaneously.
9.2Caliste Camera: Innovative Technology

for Elementary Detection Units
9.2.1Introduction
Application of astrophysics in the hard X-ray domain entails extremely demanding and challenging performance for Cd(Zn)Te-based spectroimagers as discussed.
Obviously, high-quality single crystals in addition to specific low-noise analog integrated electronics are absolutely mandatory to reach this goal, but that is not enough.
First, CdTe pixel sensors and electronics parts have to be installed into hybrid components without degrading the intrinsic performances of both. On the other hand,
these hybrid detectors have to be assembled together to form a focal plane. Here,
direct imaging with a large field of view causes some additional constraints on the
focal plane architecture with respect to the dead zones, for example, that may exist
between modular devices.
200
Several approaches have been considered worldwide to assemble CdTe-based

spectroimagers with high pixel density: (a) direct flip-chip bonding of a pixelated
Cd(Zn)Te detector on top of a two-dimensional front-end ASIC, (b) double-sided
strip detectors, and (c) three-dimensional (3D) hybrids (as discussed here). For the
first approach, the main advantage is to operate an ultralow capacitance system that
enables extremely good spectral performance, close to the Fano limit. However,
the detector area is limited to the maximum size of an ASIC, which makes this
technology very challenging and limited in area. For the approach of double-sided
strip detectors, the main advantage is the very small pixel pitch achievable, down
to a few hundreds of microns, with low power consumption due to the low number
of active readout channels. Here, the limitation comes from the maximum achievable size of a single crystal. Since only four devices of this kind can be assembled
together, one may reach about 25 cm 2 in sensitive area. None of these techniques
matches a 64-cm 2 or larger surface for the moment. Our approach of 3D hybrids is
based on about a 1-cm 2 elementary detection unit that is juxtaposed to others on its
four sides to form an arbitrarily large detection surface. The advantages are related
to the modularity (and reliability), the response uniformity, and high production
yield. The limit comes from the minimum pixel pitch, which is currently about
580 m.
9.2.2Camera Description
According to the requirements described, we invented the Caliste device, a high-performance Cd(Zn)Te-based modular detection unit that can be butted on its four sides.
Caliste relies on space-qualified technologies, compliant with low noise, low power,
and radiation-hardened design. The concept of Caliste is based on its 3D architecture. Standard architectures (one dimension [1D]) of a multichannel analog front end
are stacked together and placed perpendicular to the sensor surface, inside the right
cylinder defined by the detector edges. All of the readout electronics are thus placed
below the sensitive material, leaving the sides free of any material. Two independent
modules may be positioned very close, with a few hundreds of microns space. By
stacking up to eight 32-channel ASICs, we succeed in equipping a 256-pixel detector in a square centimeter. However, this architecture is very flexible, and the same
ASICs stack may be used for larger pixel pitch if required by the user. We successively realized two versions of Caliste: Caliste 64 and Caliste 256. Before going
to the 580-m pitch version (Caliste 256), we decided first to design a 64-channel
hybrid (Caliste 64) with 1-mm pixel pitch inside a 1-cm2 surface to evaluate this new
technology. In this case, the hybrid is a stack of four 16-channel ASICs. This stack is
connected to a 64-pixel Cd(Zn)Te detector of any thickness.
Using a rather small single crystal of 1 cm2 is justified on one hand by the mismatch in coefficient of thermal expansion between the Cd(Zn)Te detector and the
hybrid materials and by the production yield on the other hand.
The elementary device Caliste 64, our first complete realization, is detailed next.
We describe the Cd(Zn)Te detector properties, the IDeF-X analog integrated circuits,
the hybrid architecture details, and the overall device performances.
201
Caliste
9.2.2.1Detectors
The detection materials used to design the Caliste 64 microcamera are based on CdTe
grown and processed by Acrorad (Japan) and Cd0.1Zn0.9Te grown by eV Products (United
States) and processed by Bruker Baltic (Latvia). These semiconductors have density and
charge transport properties well suited for X- and gamma-ray detection. Moreover, their
high resistivity enables these detectors to operate at room temperature or moderately
cooled and to apply high voltage to accelerate charges without inducing a too large dark
current that would harm the signal-to-noise ratio. Properties of these semiconductors
are summarized in Table9.2. CdTe has a resistivity lower than Cd0.1Zn0.9Te.
To obtain detectors with very low dark current, Acrorad processes CdTe crystals
with blocking contacts at the anode, from metals that have low work functions like
indium, aluminum, nickel, to increase the Schottky barrier height. Metals whose
work functions are close to those of CdTe and CdZnTe, such as platinum or gold,
are used to form quasi-ohmic contacts. Table9.3 contains work function values and
theoretical Schottky barrier heights for several families of Cd(Zn)Te detectors as well
as typical dark current levels obtained in the laboratory. Dark current was measured
for each pixel of the segmented anode with the setup illustrated in Figure9.3. The
64-pixel detector is mounted on a substrate, connected to four boards equipped with
relays. Current levels are measured with a precision of about 0.1 pA with Keithley
6517A ampere meters. Current maps and histograms, as shown in Figure9.4, emphasize good uniformity over the matrices.
Schottky CdTe detectors have the advantage of very low dark current at room
temperature. Moreover, the dark current is distributed in each individual pixel and
may reach values far below 100 pA. However, Schottky detectors are sensitive to
the polarization effect when they are biased. The effective Schottky barrier height
Table9.2
Properties of CdTe and CdZnTe for X- and Gamma-Ray Detection
Semiconductor
Cd0.9Zn0.1Te1
Atomic numbers
48, 30, 52
5.78
1.57
10.9
48, 52
5.85
1.51
10.3
5
3 1010
1,350
120
3 106
1 106
4.42
109
950
73
Density (g/cm3) (300 K)

Bandgap Eg (eV) (300 K)
Relative permittivity
Pair creation energy (eV/pair1) (300 K)
Resistivity ( cm) (300 K)
Electron mobility e (cm2/v.s)
Hole mobility h 2
Electron lifetime e (s) (293 K)
Hole lifetime h (s) (293 K)
Fano factor
Charge carrier concentration ni (cm3) (300 K)
1
2
0.14
From eV Products (USA) by high pressure Bridgman method (HPB).

From Acrorad (Japan) by travelling heater method (THM).
CdTe2
1.2 106
4.6 106
0.2
6.9 105
202
Table9.3
Classification of Several Families of Cd(Zn)Te Detectors, from Theoretical Schottky Barrier Values and Experimental Dark
Current Levels
Metal Work
Function
Theoretical Schottky
Barrier Height
(eV)
Typical Dark Current

at 0C, 200V/mm
Typical Dark Current

at 15C, 200 V/mm
Detector
Supplier
Type
Au-Ni-Au//
Cd0.9Zn0.1Te//Au
Pt//CdTe//Pt
In//CdTe//Pt
Al-Ti-Au//CdTe//Pt
eV-products
Bruker-Baltic
Acrorad
Acrorad
Acrorad
Ohmic n-type
4.4
5.1 eV (Au)
0.6
1.7 nA cm2
0.5 nA cm2
Ohmic p-type
Diode p-type
Diode p-type
5.79
5.79
5.79
5.43 eV (Pt)
3.97 eV (In)
4.18 eV (Al)
0.36
1.82
1.61
125 nA cm2
2.5 nA cm2
0.9 nA cm2
50 nA cm2
1.0 nA cm2
0.3 nA cm2
Note: Currents were determined from matrices with 64 pixels at 900 m (8.1 103 cm2).
Semiconductor
Work Function
(ev)
203
Caliste
Keithley 6517A
A
IGR
Ipixel
Ground
Anode
Pixels
Guard ring
Keithley
6517A A
CdZnTe
Ground
Cathode
Shielding
Ground
Figure 9.3 Setup for dark current measurement. Current is measured for each pixel individually thanks to a system of relays and Keithley 6517A ampere meters.
40
Number of Pixels
30
20
10
0.01 0.05 0.1 0.5 1
5 10 50 100 pA
4
6
Pixel Current (pA)
10
Figure 9.4 Current map and current histogram obtained with a 64-pixel Al-Ti-Au
Schottky CdTe detector cooled to 17C and biased at 200 V. Median current is 1.9 p.
decreases over time, causing a continuous increase of dark current. On spectroscopy,

the polarization effect leads to spectral resolution degradation, gain loss, then detection efficiency loss. The physical interpretation of this effect is a progressive detrapping of holes from the deep acceptor levels in CdTe, which create a space charge
region. The last affects the distribution of the internal electric field, which reaches
very low values close to the cathode, the entrance window of the sensor. The field
may become so low that charge carriers generated at the cathode side are not efficiently accelerated and do not contribute to the signal anymore. Polarization effect
204

10.0
0.12
0.20
0.08
0.5
0.9 mm 0.9 mm
0.58
10 mm
Guard
ring
10.0
1.0 mm 0.1 mm
10 mm
(a)
0.58
(b)
0.12
0.20
0.08
0.50
Figure 9.5 Patterns of the pixilated electrodes used for Caliste 64 and Caliste 256 cameras: (a) 64-pixel geometry; (b) 256-pixel geometry.
has an exponential trend with temperature (e.g., a moderate cooling temperature significantly stabilizes the detectors). Below 20C, we can operate them for several
days. The Caliste hard X-ray cameras were designed with thin Cd(Zn)Te detectors
(0.5-, 1-, and 2-mm thickness); their electrode patterns integrate 64 or 256 pixels and
a surrounding guard ring in 1 cm2 (cf. Figure9.5).
9.2.2.2Front-End Electronics
Dedicated front-end electronics, the imaging detector front end for X-ray (IDeF-X),
has been designed at CEA/Irfu (France) for spectroscopy with CdTe and CdZnTe, in
particular to be integrated in Caliste microcameras. These new ASICs are designed
to operate in a multichip configuration, well suited to our Caliste devices. IDeF-X is
a family of ASICs processed with the standard Austriamicrosystems (AMS) 0.35-m
complementary metal oxide semiconductor (CMOS) technology for the readout of
16 or 32 channels of Cd(Zn)Te detectors, with low input capacitance (2 to 5 pF)
and low dark current (1 pA to 1 nA). The goal of this ultralow-noise electronics is
to convert the amount of charges, created in the Cd(Zn)Te semiconductor detector
by the interaction of an impinging photon, into an amplified signal and to store the
amplitude of this signal, which is proportional to the integrated charge and hence to
the deposit energy. The analog channel of each ASIC generation is slightly the same,
as discussed next.
The first stage is a charge-sensitive preamplifier (CSA) based on a classical
folded cascode amplifier with a PMOS-type input transistor. The CSA is direct
current (DC) coupled to the input of the detector pixel; it means that the segmented
electrode of the detector is biased by the ASIC. The CSA integrates the incoming
charge on a feedback capacitor and converts it into voltage. The feedback capacitor is
discharged by a continuous reset system achieved with a PMOS transistor operating
in the subthreshold region that can source the leakage current of the detector. The
CSA has been designed to be preferably connected to the anode of a detector.
Caliste
205
The second stage of the chain is a pole zero cancellation (PZC) stage. Placed
after the CSA, it avoids long-duration undershoots at the output. It is also used to
amplify the signal, and it minimizes the influence of the leakage current on the
transfer function of the whole channel.
The two very front-end stages are followed by a filter stage. A fourth-order shaper
optimizes the signal-to-noise ratio; it is achieved by two second-order Sallen and Key
(S&K) low-pass filters with variable shaping times. This parameter allows setting the
best filter according to the input capacitance (hybrid and detector geometry) and
the real detector leakage current, depending on the operating temperature, detector
geometry, and applied voltage. Out of the filter stage, a discriminator compares the
shaped signal to an adjustable low-level threshold. In the IDeF-X V1.1 ASIC, the
discrimination threshold is common to all channels. In IDeF-X V2, each channel has
its own adjustable discrimination threshold thanks to an in-channel six-bit digitalto-analog converter (DAC). Finally, a stretcher realized with a peak detector (PkDet)
plus a storage capacitor stores the maximum of the shaped signal. To help either chip
characterization or functional tests at the hybrid level, an injection capacitor Cinj is
integrated in each channel to inject a calibrated charge to measure the gain of the
channel. A programmable input DC current source il simulates the dark current of
the detector when it is not present or turned off. In addition, the channel integrates
an inverting/noninverting voltage amplifier (A 2) to be able to connect the ASIC
to the cathode of a detector. Outside the channels, many configuration parameters
(shaping times, CSA current, test configurations, etc.) can be tuned by digital control
(programmable via a serial link in the case of IDeF-X V2).
A global trigger is built from the logical OR (G-OR) of the outputs of all the
individual discriminators (16 or 32). If at least one channel is hit, a trigger signal
is sent from the chip to initiate the readout sequence managed by FPGA: First, the
IDeF-X chip sends a serial digital pattern corresponding to the hit channels. Then,
the energy readout starts for the hit channels only or any desired channel according
to the readout mode; the outputs of the peak detectors are successively multiplexed
on the analog output via the common output buffer (Buff).
An overview of the described architecture is given in Figure 9.6. IDeF-X V2
includes a latch-up-hardened digital part. Characteristics of IDeF-X V1.1 and
IDeF-X V2 are summarized in Table9.4. Minimum achieved noise with these circuits, expressed as equivalent noise charge (ENC), is 33 electrons rms. In the complete detection system, this ENC value increases because of the dark current of the
detector in each pixel that contributes to parallel noise and because of the capac
itance of the ASIC board and the connections between ASIC inputs and detector
electrodes that contribute to series and 1/f noise. The dark current may vary according to operating conditions (voltage and temperature), while capacitances are fixed
once the design of the hybrid is done (track length, routing, and materials). This has
to be minimized in the hybrid component definition.
9.2.2.3Hybrid Component
Hybridization is the technological process that merges the sensor and the front-end
electronics dies to form a complete detection system into a unique part. Caliste 64
is the first prototype of our hybrid assembly, a 1 cm2 by 2 cm height elementary
206
Vtest
Vin(0)
Cinj
il
CSA
PZC
S&K
A +/2
S&K
PkDet Mux
Trigger
DAC
Discri
Vtest
Vin(15) Cinj
Vin(31)
il
CSA
G-Or
Energy
Bu
PZC
S&K
A +/2
S&K
PkDet Mux
IDeF-X V2 only
IDeF-X V1.1 & IDeF-X V2
DAC
Discri
Figure 9.6 Schematic of IDeF-X chips. An analog channel includes a DC-coupled continuous reset CSA, a pole zero cancellation (PZC) stage, a fourth-order Sallen and Key (S&K)
shaper, with an input signal polarity selector (A 2) for IDeF-X V1.1 and IDeF-X V2. These
last versions also include peak detectors (PkDet) and discriminators (Discri) for each channel. IDeF-X V2 integrates an in-channel six-bit DAC to achieve a low energy threshold.
Table9.4
Main Features of Front-End ASICs IDeF-X V1.1 and IDeF-X V2
Parameter
IDeF-X V1.1
IDeF-X V2
Chip size
Number of channel
Power supply
Power consumption
Gain
Dynamic range
Peaking times
Discrimination threshold
Radiation hardened design
SEL LET
Noise
3,000 m 4,000 m
16
3.3 V
2.8 mW channel1
200 mV fC1
60 ke
2,800 m 6,400 m
32
3.3 V
3 mW channel1
200 mV fC1
50 ke
0.99.6 s
0 to 5 ke (common to all channel)
No
12 MeV cm2 mg1
37 e + 7 e/pF
0.99.6 s
210 e to 4 ke (LSB = 65 e)
Yes
56 MeV cm2 mg1
33 e + 7 e/pF
Caliste
207
detection unit for large-area space applications. The Caliste design integrates 64
pixels read out by four 16-channel analog IDeF-X V1.1. The interface is optimized to
fit in a 7 7 pin grid array 1.27-mm pitch connector at the opposite side of the detector. Caliste 64 fulfills requirements for space devices related to power consumption,
radiation hardness, and environmental constraints (including thermal cycles, vibrations, and shocks).
The solution to integrate the ASIC chips below the detector is to put the chips
perpendicular to the detection surface. Each ASIC is responsible for reading out two
rows of pixels. The design and the fabrication of the Caliste 64 device is a collaborative work between CEA/Irfu and 3D Plus (France). The ASICs are mounted by
classical wire bonding on miniprinted circuit board (PCB) substrates. The last must
have good dielectric properties to avoid excess of noise due to dielectric losses. The
design is done in such a way that it limits parasitic capacitances and consequently
series and 1/f noise. The residual parasitic capacitance is estimated to be near 1 pF,
without the capacitance of the Cd(Zn)Te pixel, which depends on the detector pitch
and thickness but is clearly negligible with respect to other capacitances on the input
path. The four PCB substrates are stacked and molded inside an insulating epoxy
resin according to the 3D Plus technology. The resulting block is diced and metallized. The top surface of the block is prepared by laser ablation to receive a 1-cm2
matrix of 8 8 pixels. Each pixel pad is connected to an ASIC input. An interconnection stack drives signals from the main block to the bottom 7 7 pin grid array,
1.27-mm pitch. Finally, a 1-cm2 Cd(Zn)Te detector 0.5-, 1-, or 2-mm thick is fixed
using a polymer bump bonding technique at a moderate curing temperature to obtain
the complete Caliste 64 camera. Figure9.7 illustrates the design of Caliste 64.
The bottom interface is minimized to 49 pins, including analog and digital pins,
power supplies, and grounds. This design allows easy routing of an array of Caliste
64 units placed side by side. The reduction of output signals is made possible by sharing signals between ASICs. The common signals are routed together using the lateral
surfaces of Caliste 64. The routing is realized by a laser ablation process according to
the 3D Plus technology. Single-end and differential analog signals are routed according to specific rules to limit parasitic coupling with digital signals, for instance.
Figure9.8 shows a complete Caliste 64 camera with an Al Schottky CdTe detector.
Ten prototypes of cameras were realized with different kinds of detectors and different thicknesses. Spectroscopic characterizations are presented in Section 9.2.3.
After the validation of the hybridization concept with Caliste 64, a new generation of microcamera, the Caliste 256, was designed. This time, the device integrates four times more channels in the same volume as Caliste 64. It is based on
the same principle: Eight 32-channel IDeF-X V2 chips are stacked perpendicular
to the detector plane to read out two rows of pixels. The bottom interface is also
reduced to 49 connection pins. The fabrication process is basically the same. A new
aspect taken into account in the Caliste 256 design is heat dissipation because power
consumption is also four times higher in the same volume (816 instead of 188 mW).
Routing on PCB and on lateral faces has been optimized to reduce the thermal gradient between the bottom of the camera and the detector. However, the future versions
of Caliste to be operated in space will have drastically reduced power consumption
208
Planar cathode
Cd(Zn)Te detector
Pixelated anode
(64 pixels)
Main block
16-analog channel
IDeF-XV 1.1 ASIC
Interconnexion stack and

7 7 pin grid array
Figure 9.7 Schematic view of Caliste 64 camera. It is composed of a main block where
four IDeF-X V1.1 ASICs fixed on PCBs are integrated, an interconnection stack that connects
the signals from the main block to the pin grid array, and the detector is placed at the end of
the fabrication.
by at least a factor of about 5. This will bring the power consumption back to the
Caliste 64 level for 256 channels.
One of the most challenging fabrication steps for Caliste 256 with respect to
Caliste 64 is related to the detector mounting. Like Caliste 64, the interconnection
technique to provide electrical contacts between all detector electrodes and ASIC
entries is based on polymer bump bonding, but the detector pitch is 580 m instead
of 1,000 m. It obviously requires much more precise alignment of the electrodes
when the flip-chip process is performed. Such alignments are usual with indium
bump bonding on silicon chips, for instance, thanks to self-alignment of the pads
when indium reflows, but it becomes more difficult using conductive polymer contacts. Furthermore, the glue spots must be high enough to minimize the mechanical
strength due to the differential thermoelastic coefficient between the stiff crystal
and the electrical body in epoxy resin; otherwise, the detector could break when the
device is cooled. The spots are about 220 m high in the present design. The first
prototypes of Caliste 256 cameras are shown in Figure9.9.
9.2.3Performance
9.2.3.1Setup
The Caliste cameras (Caliste 64 or Caliste 256) are placed into a thermal enclosure for
electrical or spectroscopic performance tests. When at least one event is seen by the
camera, a trigger signal is sent to a board with an FPGA outside the vessel. The FPGA
Caliste
209
Figure 9.8 Caliste 64 microcamera equipped with a 2-mm thick Al Schottky CdTe detector from Acrorad. The electrical body integrates four IDeF-X V1.1 ASICs stacked perpendicular to the detector surface. Routing on two lateral surfaces connects signals common to
the four ASICs (power supply, etc.).
Figure 9.9 Caliste 256 microcamera equipped with a 1-mm thick Al Schottky CdTe
detector from Acrorad. The electrical body integrates eight IDeF-X V2 ASICs stacked perpendicular to the detector surface. The contact between detector and electrical body is only at
the glue spot level to minimize mechanical constraints between the two materials.
210
is in charge of the readout sequence, of the data encoding, and of the formatting of
data packets. As a result, for each event, a frame is generated containing the trigger
time, the position of the hit pixels, and the stored energies in these pixels. Hence,
Caliste 64 and Caliste 256 are spectroscopic imagers with time-tagging capability.
9.2.3.2Spectroscopy
In this section, we present spectroscopic results from the Caliste 64 device; 10 complete units were fabricated and tested with various types of detectors (Schottky CdTe,
CZT) and various thicknesses (0.5, 1, and 2 mm). An americium 241 radioactive
source was used for most spectroscopic characterizations. The source was installed
right above the detector in such a way that the entire array was almost uniformly lit.
Since the source was placed inside the vessel, no entrance window attenuated the
low-energy photons, which helped for evaluating the performances down to a few
kiloelectron volts. Good uniformity over all the matrices was observed. After individual energy calibration, the 64 spectra obtained were summed to obtain a sum
spectrum, even at room temperature. Of 10 tested matrices (i.e., 640 pixels), only
1 pixel had to be excluded from the sum of its matrix. The samples were cooled to
15C at the detector level to decrease the dark current, and therefore the electronic
noise, and to stabilize the Schottky CdTe detectors. This kind of detector can be
operated for several hours at 15C without spectral degradation, and no extra correction on the spectra is necessary to build the sum spectrum. Table9.5 summarizes
the excellent performance of the 10 cameras.
On one hand, Figure9.10 illustrates sum spectra obtained from single events only
with 2-mm thick Al Schottky CdTe and Cd0.1Zn0.9Te detectors. Peak-to-valley ratio
was computed as the ratio between the counts at the 59.54-keV line and the counts
at 57 keV. It is an indicator of the left tail of the 59.54-keV line, mainly caused by
Table9.5
Performance of 10 Caliste 64 Units
Sample
Type
Thickness
(mm)
E at 14 keV
(eV FWHM)
E at 60 keV
(eV FWHM)
1
3
4
5
6
7
8
11
12
14
Al-Ti-Au//CdTe//Pt
Al-Ti-Au//CdTe//Pt
Al-Ti-Au//CdTe//Pt
Al-Ti-Au//CdTe//Pt
Ni-Au//CdTe//Pt
Ni-Au//CdTe//Pt
Ni-Au//CdTe//Pt
Au-Ni-Au//CZT//Au
Au-Ni-Au//CZT//Au
Al-Ti-Au//CdTe//Pt
2
1
1
2
1
1
2
2
2
0.5
835
693
664
735
725
779
868
835
813
715
1,145
860
842
905
938
966
1,195
1,179
1,066
888
Note: Energy resolution is given for best sum spectra obtained from single events only
when the detector was cooled between 15 and 20C at the detector level.
211
Caliste
1,500
1,500
Counts
2,000
Counts
2,000
1,000
1,000
500
500
0
10
20
30 40 50
Energy (keV)
60
70
10
(a)
20
30 40 50
Energy (keV)
60
70
(b)
Figure 9.10 Sum spectrum from single events only with two Caliste 64 units at 15C.
(a) Caliste 64 unit equipped with a 2-mm thick Al Schottky CdTe detector biased at 800 V.
(b) Caliste 64 unit equipped with a 2-mm thick CZT detector biased at 1,000 V. Energy resolution at 59.54 keV was respectively 905 eV FWHM and 1,066 eV FWHM. Peak-to-valley
ratio between counts at 59.54 keV and counts at 57 keV was, respectively, 58 and 8.
charge loss in the bulk (due to the lifetime of the charge carriers). Peak-to-valley
ratio was equal to 58 for CdTe and 8 for CZT; charge loss in the bulk was significantly higher in CZT than in CdTe. On the other hand, temporal and spatial information with Caliste 64 enabled extraction of split events between two neighboring
pixels. After individual energy calibration of all the pixels, an americium spectrum
was built by summing the energies of the neighbors for all the couples of hit pixels.
Because of charge loss between the pixels (in the case of charge sharing), the main
line was not exactly centered on 59.54 keV but on an energy Edouble. An extra calibration was performed to compensate for this charge loss, an estimation of which can
be given by the formula

CLdouble =
59.54 Edouble ( keV)

59.54
(9.1)
Figure9.11 shows the americium spectra built with double events only from the
same acquisition as those used for Figure9.10. Energy resolution at 59.54 keV was
1.4 keV FWHM for Al Schottky CdTe and 2.9 keV FWHM for CZT. This indicates
that, again, charge loss between the pixels was higher in the CZT detector than in
the CdTe detector. According to Equation 10.1, charge loss for double events was
0.2% in CdTe and 2.4% in CZT. The study of split events underlines the fact that the
detectors with segmented electrodes are sensitive in interpixel zones; consequently,
inside the guard ring area, the sensor is not affected by any dead zone. Charge loss
between the pixels in the Schottky CdTe detectors was almost negligible in this
geometry (900-m pixel size, 100-m pixel gap); charge loss between the pixels can
be compensated and corrected in CZT detectors.
212
300
300
Counts
400
Counts
400
200
100
0
200
100
10
20
30
40
50
60
70
10
20
30
40
Energy (keV)
Energy (keV)
(a)
(b)
50
60
70
Figure 9.11 Sum spectrum from double events only with two Caliste 64 units at 15C.
(a) Caliste 64 unit equipped with a 2-mm thick Al Schottky CdTe detector biased at 800 V.
(b) Caliste 64 unit equipped with a 2-mm thick CZT detector biased at 1,000 V. Energy
resolution at 59.54 keV was, respectively, 1,390 eV FWHM and 2,890 eV FWHM. Charge
loss between the pixels was estimated as 0.2% and 2.4%, respectively.
9.2.3.3Timing Issues
In high-energy astrophysics, two key issues are related to timing. First, the accuracy
of the event time tagging is important for background rejection. Several high-energy
experiments integrate anticoincidence shielding. This detector surrounds the instrument observing sky. Background particles are mainly cosmic ray protons or secondary particles created by a proton interacting into the satellite materials. Coming
from every direction of the sky, they are detected by the instrument and the active
shielding at the same time (in a coincidence temporal window); hence, these events
are suppressed from the science data. The probability to reject two real events arriving by chance in the same coincidence window (false rejections) is all the lower since
this window is short, that is, the time-tagging accuracy of the instrument is good.
Second, dead time is the time during which the camera or part of the camera is
disabled, mostly due to signal shaping, readout procedures, and analog part reset.
Background rejection also induces dead time since data in the same coincidence
window are not taken into account in the science data. Since flux is low in the X- and
gamma-ray field, observation time can significantly increase with rather long dead
time for a required sensitivity.
Good time-tagging accuracy is not trivial to obtain with spectroscopic imagers
like Caliste cameras. First, the signal from the sensor is slow, and then the optimal signal-to-noise ratio entails long shaping times. With very low dark current,
the peaking time that minimizes ENC is on the order of 7 to 10 s in Caliste. As
a consequence, the time-walk, delay between the real arrival time of a photon and
the trigger of the analog channel, can last several microseconds. The trigger time
is not sufficient to determine precisely the event time. However, time-walk can be
213
Caliste
5,000
Time-walk (ns)
4,000
3,000
2,000
tPEAK = 9.6 s, ETH = 3 keV
1,000
0
50
100
150
Energy (keV)
Figure 9.12 Mean time-walk over 64 pixels of a Caliste 64 unit for two peaking times
(7.2 and 9.6 s) and two low-level thresholds (3 and 5 keV).
estimated and corrected by the energy measurement if the pulse shape and the lowlevel threshold are known. Figure9.12 illustrates time-walk calibration with Caliste
64. After time-walk correction (i.e., timing systematic error subtraction), some residual errors remain. They come mainly from three independent sources: electronic,
technological, or physical. First, the noise on the signal or the noise on the threshold
reference level may cause anticipated or late triggering of a channel. This phenomenon is responsible for statistical fluctuations on time-walk (jitter).
Then, technological mismatch in the microelectronics analog design of the microelectronics induces slight variations on the transfer functions between the channels
(gain, shaping time). As a result, a residual systematic error from channel to channel
exists after time-walk correction if a common law is used to perform corrections of
the entire camera.
Finally, different interaction depths in the Cd(Zn)Te detectors led to different
delays on the induced signals due to the difference of transport properties for holes
and electrons. According to the operating condition and the energy and nature of the
particles, the most probable interaction depth can be estimated, but some statistical
errors remain because of the uncertainty on the interaction depth for each event.
Figure 9.13 illustrates measurements of jitter and mismatch with Caliste 64 at
7.2-s peaking time. Time-tagging accuracy is better than 100 ns rms for energies
greater than 15 keV, taking into account the contribution of front-end electronics only.
To estimate dead time, we need to describe the readout procedure for Caliste.
When at least one pixel reaches the low-level threshold, the whole camera is locked
after latency time (so the pulse can develop and its amplitude height can be stored).
The readout sequence can then begin: The controller asks the number of the hit
214
Error on Time-walk (ns rms)
120
Noise
Mismatch
Quadratic sum
100
80
60
40
20
0
50
100
150
Energy (keV)
Figure 9.13 Contribution of front-end electronics to the time-tagging errors in Caliste 64.
Time-walk was measured many times on each channel at different energy levels for a 7.2-s
peaking and a 3-keV low threshold. Standard deviation on time-walk for a single channel
gives jitter. Standard deviation over the 64 channels of their mean time-walks gives a measurement of mismatch.
pixels and their corresponding amplitudes. In Caliste 64 as well as Caliste 256, there
is one output per ASIC. As a consequence, pixels of the same ASIC are read out successively by multiplexing. But pixels of different ASICs can be read in parallel if the
acquisition board houses one ADC per ASIC. Hence, the duration of the sequence
depends on the maximum number N of pixels per ASIC to read, with N inferior to 16
in Caliste 64 and inferior to 32 in Caliste 256. After the reading sequence, one extra
microsecond is needed to reset the peak detectors and reenable the Caliste camera.
Finally, the typical dead time is 20 s plus N.
9.3Perspectives
The current development of the Caliste detector demonstrates the feasibility of this
unprecedented configuration for a space hard X-ray elementary detection unit with
the ability to be butted on its four sides. The future steps of this work will be performed in the Simbol-X mission context, in which we have to realize a low-power
version of Caliste 256. This requires new efforts especially in microelectronics
design. A new IDeF-X version is currently under construction and will allow high
spectral performances with five times less power. Moreover, the design will help
provide a drastic simplification of the device interface and fewer connection pins for
easier routing when the devices are installed into a focal plane assembly. This new
device will be fully evaluated from a technological point of view to validate its reliability in space conditions.
Caliste
215
9.3.1Future Needs for Hard X-Ray Astronomy

In the hard X-ray domain, Caliste may be further studied to reach smaller pixel size
in the case of new mirrors with high angular resolution around 30 keV. This goal may
be reached by the 2020s and would increase sensitivity in the hard X-ray domain.
The challenge is to sample a point spread function of less than 10 arcsec mirror with
a 20-m focal length. In this situation, new efforts are necessary to obtain a pixel pitch
in the range from 200 to 300 m without sacrificing spectral performance. This also
means that the number of channels per square centimeter would rise by a factor of
about four from our current design and would require again a drastic reduction of
power consumption. It would also be beneficial to integrate much larger surface crystals, up to 4 cm2 or more, to minimize dead zones between two consecutive elementary detection units. The complexity of the device is more and more challenging, and
Caliste constitutes a first step toward this goal.
9.3.2Future Needs for Gamma-Ray Astronomy

Thanks to its very promising performance, Caliste could constitute a starting
point for gamma-ray spectroimagers at higher energy, in the megaelectron volt
region or above. This is the region where nuclear emission lines can be detected
and a region where some pulsars and active galaxies are strongly detected. In this
energy domain, the focusing technique is no longer applicable for direct imaging,
but the high pixel density and the excellent spectral response with accurate timing
capability may be applicable for the Compton telescope when installed below a set
of other pixel arrays (at least one layer) used as scattering medium, made preferably of a low-Z material (silicon, for instance). The Cd(Zn)Te detector is then used
as a segmented absorber and would preferably be thicker than currently available
versions (>5 mm). A Compton telescope allows high-energy source localization
using the Compton scattering process. In this physical process, the high-energy
photon interacts at least twice. In the top-level layer, the gamma-ray Compton
scatters off an electron, which loses its energy. The scattered photon then travels
down into a second level of Cd(Zn)Te material, which completely absorbs the scattered photon. The scattered photon direction is related to the energy deposit in the
first layer. Summing the two energy deposits gives the incident photon energy, and
from the line formed by the two interaction points, one can derive the primary photon incidence angle, but the azimuthal direction is still undetermined. The incident
angle gives a generator of a cone corresponding to all possible directions of the
primary photon. Projecting the cone back onto the sky gives the possible emission
sites of the gamma ray from a ring. Accumulating photons coming from the same
celestial source will determine several cones intercepting at the source position. It
is trivial to understand that the imaging performance of such a telescope is directly
based on the spectral response, spatial resolution, and coincidence capability (i.e.,
timing accuracy).
This technique is particularly exciting since it is extremely efficient for removing background. This technique has already been extensively used with scintillating
216
material-based instruments and would probably be much more accurate and sensitive
with the use of pixelated semiconductor sensors. Furthermore, the Compton scattering process is sensitive to gamma-ray polarization, and this would open the door to
new observable parameters from high-energy sources.
Acknowledgments
We wish to thank Philippe Ferrando from CEA (PI of the Simbol-X mission), CNES
(the French national space agency), and Marie-Ccile Vassal and Fabrice Soufflet
from 3D Plus for their support of this work.
Bibliography

1. P. Ferrando et al., Simbol-X: mission overview SIMBOL-X, Proc. SPIE, 6266, 62660,
2006.
2. F. Lebrun et al., ISGRI: the INTEGRAL soft gamma-ray imager, A&A, 411,
L141L148, 2003.
3. J. Treis et al., DEPFET based focal plane instrumentation for X-ray imaging spectroscopy in space, Proc. IEEE NSS-MIC Conf. Rec., N47-3, 22262231, 2007.
4. G. Lutz, Semiconductor Radiation Detectors, Springer-Verlag, Berlin, 1999.
5. B. P. F. Dirks, Study and modelling of the next generation Cd(Zn)Te X and gamma-ray
detectors for space applications, doctoral thesis, University of Paris, 2006.
6. H. Toyama et al., Formation of Schottky electrode for CdTe radiation detectors, Proc.
IEEE NSS-MIC Conf. Rec., N35N56, 13951398, 1999.
7. B. P. F. Dirks et al., Leakage current measurements on pixelated CdZnTe detectors,
Nucl. Instrum. Methods A, A567, 145149, 2006.
8. G. De Geronimo and P. OConnor, A CMOS fully compensated continuous reset system, Proc. IEEE NSS-MIC Conf. Rec., 584588, 1999.
9. Z. Y. Chang and W. M. C. Sansen, Low Noise Wide-Band Amplifiers in Bipolar and
CMOS Technologies, Kluwer Academic, Dordrecht, 160163, 1991.
10. F. Lugiez et al., IDeF-X V1.1: performances of a new CMOS 16 channels analogue readout ASIC for Cd(Zn)Te detectors, Proc. IEEE NSS-MIC Conf. Rec., 841844, 2006.
11. O. Gevin et al., IDeF-X ECLAIRs: an ultra low noise CMOS ASIC for the readout of
Cd(Zn)Te detectors, Proc. IEEE NSS-MIC Conf. Rec., 23512359, 2007.
Pixel Array
10 Hybrid
Detectors for
Photon Science
Heinz Graafsma
Deutsches Elektronen Synchrotron

Hamburg, Germany
Contents
10.1 Introduction................................................................................................... 217
10.2 Historical Overview....................................................................................... 218
10.3 Current Hybrid Pixel Detectors at Synchrotron Storage Ring Sources........ 220
10.3.1 The Medipix Family of Hybrid Pixel Detectors................................ 220
10.3.1.1 Medipix2 (MXR20)............................................................ 220
10.3.1.2 Timepix............................................................................... 222
10.3.1.3 Systems and Applications................................................... 222
10.3.1.4 Medipix3 Development....................................................... 223
10.3.2 The Pilatus-2 Hybrid Pixel Detector.................................................224
10.4 Cornell-Area Detectors Systems Corporation ADSC Pixel Detector
Project............................................................................................................ 226
10.5 Future Hybrid Pixel Detectors for Free-Electron Laser Sources.................. 227
10.6 The LCLS Detectors...................................................................................... 227
10.7 The European XFEL Detectors..................................................................... 228
10.7.1 The AGIPD Project............................................................................ 229
10.7.2 The LPD Project................................................................................ 232
10.7.3 The DSSC Project.............................................................................. 233
10.8 Summary and Outlook.................................................................................. 234
References............................................................................................................... 235
10.1Introduction
In X-ray photon science, the photon beam is used to probe the state of a sample by
spectroscopic, imaging, or scattering techniques. Even when we restrict ourselves to
the spectral range from a few kiloelectron volts to a few tens of kiloelectron volts,
in other words, soft-to-medium X-ray energies, as in this chapter, the diversity of the
applications is still large and ranges from fundamental research to materials science,
chemistry, and life sciences. Furthermore, the size of the sample under study, as well
217
218
as the spatial resolution, ranges from nanometers to centimeters. It is therefore clear

that a single detector, no matter how sophisticated, cannot cover all the fields, and that
many different detectors, tailored to the individual experiments, would be needed.
This large diversity of demands is one of the main challenges detector developers at
synchrotron sources face. A further difficulty is the exponential increase of source
brilliance since the 1980s, with a gain of approximately one order of magnitude every
2 years, putting high demands on continuous improvement of the detectors.
This chapter covers some of the two-dimensional hybrid pixel detectors specifically developed since the late 1990s for photon science applications in the few
kiloelectron volt to few tens of kiloelectron volt energy range. These hybrid pixel
detectors are all based on standard planar silicon sensors as the sensing element
bump bonded to application-specific integrated circuits (ASICs). The emphasis is
mainly on systems developed in Europe. The X-ray Pixel Array Detector (XPAD)
system developed by the Centre de Physique des Particules de Marseille (CPPM)
laboratory in Marseille (CNRS-IN2P3) is described in detail in a separate chapter. There are important systems in use at photon sources that are not covered in
this chapter, like charge-coupled devices (CCDs), pnCCDs, monolithic active pixel
detectors, and so on because of the need to focus, it does not constitute any judgment
regarding their usefulness or novelty.
I start with a short review of the systems developed in the past: the analog pipeline
HPAD developed at Cornell High Energy Synchrontron Source (CHESS) and the
Medipix1 or photon counting chip at European Organization for Nuclear Research
(CERN). The second part is dedicated to the Medipix suite of systems and the very
successful Pilatus-2 system. The next section delineates the requirements for the
two-dimensional X-ray detectors for the European X-ray free electron laser (FEL),
and three recently started projects (adaptive gain integrating pixel detector [AGIPD],
DEPMOS sensor with signal compression [DSSC], and large-pixel detector [LPD])
are presented. At the end of the chapter, a summary is given, as is an outlook for
future developments.
10.2Historical Overview
One of the first hybrid pixel detector systems specifically developed for photon science applications at synchrotron sources was the analog pipeline HPAD, developed
by the group of Sol Gruner at CHESS/Cornell.13 The system was designed for fast
time-resolved imaging experiments with microsecond framing times. To obtain statistically meaningful data within microseconds, photon counting is not an option,
and an integrating detection scheme is mandatory. To optimize the efficiency of the
experiments, eight consecutive images can be recorded and stored within the pixel
before reading out the detector. The readout chip (ROC) was designed and fabricated
in 1.2 m complementary metal oxide semiconductor (CMOS) technology with two
metal layers and linearized capacitors using the MOSIS services. A single chip has
92 100 pixels of 150 m square, giving a total area of 13.8 mm 15 mm. The ROC
was bump bonded to a 300-m thick silicon sensor.3 The schematic pixel layout of
the system is shown in Figure10.1 and consists of a standard charge-sensitive preamplifier followed by an eight-cell-deep analog pipeline.
219
Hybrid Pixel Array Detectors for Photon Science
Diode
+60 V
Input stage
IR
2 pf
SE
Storage stage
RE
Output stage
C1 C2 C3 C4 C5 C6 C7 C8
Vb
CB
C1C8: 130 fF
Figure 10.1 Pixel layout of the Cornell analog pipeline HPAD chip. (From Rossi et al.)3
This system was developed as a prototype and had a number of shortcomings, notably the limited radiation hardness of 30 krad, which prevented its use on a large scale.
Nevertheless, this prototype system has found use in a number of scientific applications for which the time evolution of a triggerable system is followed, like high-speed
radiography of diesel fuel injectors.4,5 This shows one of the advantages of hybrid pixel
detectors, for which larger areas are built from fully functional small modules. Owing
to the large variety of synchrotron experiments, large areas are not always needed,
and sometimes a single-chip module is sufficient. This very successful prototype also
showed the impact a detector custom designed for photon science can have.
In Europe, the Medipix collaboration centered at CERN was established in the
mid-1990s with the goal of disseminating the technology and know-how of hybrid
pixel detectors for charged particle detection developed for high-energy physics
(HEP) experiments to other fields of application, notably medical imaging (details
are available at http://medipix.web.cern.ch/MEDIPIX). In contrast to the integrating
HPAD developed by the Cornell group, which was targeting fast and high-flux experiments, the Medipix consortium focused on photon-counting or photon-processing
systems, giving very low background at the cost of maximum measurable flux. The
Medipix1 chip was designed in the SACMOS1 technology with two metal layers and
consisted of 64 64 pixels of 170 170 m. Each cell contained a charge-sensitive
amplifier, a lower-level threshold discriminator with a three-bit digital-to-analog
converter (DAC) for threshold fine-tuning, and a 15-bit shift register that could be
used as either a counter or a shift register. The chip had a total area of 1.7 cm2, of
which only 70% was sensitive since power was provided from three sides. Medipix1
could be read out in 384 s using a clock frequency of 10 MHz and had a front-end
noise of 250 e and a threshold dispersion of 100 e using the three-bit trim DACs.
The minimum detectable charge over the full chip was about 2,000 e. The Medipix1
chip was bump bonded to 300-m thick silicon and 200-mm thick GaAs sensors; the
latter is an important candidate for mammography applications at 17 keV.
220
This first system was successfully used by the members of the collaboration to
show the advantage in signal-to-noise performance of photon-counting systems and
low-energy fluorescence suppression over the more classical integrating systems in
X-ray imaging. This led to the rapid expansion of the number of interested institutes
and the establishment of the Medipix2 collaboration. The Medipix2 and Timepix
systems are described in detail in the next section.
Around the same time as the start of the Medipix project, another initiative, also
a spin-off of the work for the HEP experiments, started at the Paul Scherrer Institute
(PSI) in Switzerland: the Pilatus-1 system for the protein crystallography applications at the Swiss Light Source (SLS) at PSI. This system was also performing photon
counting to achieve the lowest possible noise performance. Pilatus-1 was designed in
the radiation-tolerant DMILL 0.8-m Bi-CMOS technology with two metal layers.
A chip consisted of 44 78 square pixels of 217 m and contained the usual photon-counting electronics.6 The DMILL process had the advantage of good radiation
resistance but suffered from very poor yield, which becomes a major obstacle with
large chips. Nevertheless, a full system was built with 1,120 967 pixels and used for
experiments at the SLS.7 The lessons learned from this exercise were invaluable for
the construction of the next-generation systems described in this chapter.
10.3Current Hybrid Pixel Detectors at Synchrotron

Storage Ring Sources
This section describes in some detail two hybrid pixel detector systems that have
found applications at synchrotron storage ring (SR) sources: the Medipix family and
the Pilatus system.
The SR sources are pulsed sources with bunch lengths on the order of 100 ps and
a bunch spacing ranging from a few nanoseconds to a few microseconds, depending
on the machine and the operation conditions. Despite this pulsed nature, the source
is mostly used as a continuous source in the experiments. Furthermore, in almost all
scattering as well as in most imaging experiments, the signal to be detected is monochromatic with a relatively low photon flux. This means that with small pixel sizes,
single-photon detection techniques are possible using either a simple lower-level or
window discrimination.
10.3.1The Medipix Family of Hybrid Pixel Detectors

As stated, the success of Medipix1 (photon counting chip) led to the establishment of
the much larger Medipix2 collaboration consisting of mainly European universities
and research institutes, but including some American members as well (information
available at http://medipix.web.cern.ch/MEDIPIX/Medipix2/indexMPIX2.html).
This increase of members also helped to finance the increased CMOS fabrication
costs associated with going to deeper submicron technologies.
10.3.1.1Medipix2 (MXR20)
Medipix2 was designed as the successor of Medipix1 (or photon counting chip) to
overcome many of the limitations in the first system. One of the main limitations was
221
the relatively large pixel size of 170 m, which would be acceptable for most photon science applications at synchrotron sources but was deemed too large for many
medical-imaging techniques, the original driving force behind the project and the
origin of the name. For instance, for mammography recognizing feature sizes down
to 50 m is one of the performance standards. Another major shortcoming of the
Medipix1 chip was the relatively low radiation hardness of less than 10 krad,8 a major
obstacle in both medical and synchrotron applications. Furthermore, building larger
systems by tiling was not possible due to the 500-m dead area at the borders. All
these issues made the Medipix1 chip a good prototype for evaluating the advantages
of photon counting compared to integrating in various X-ray imaging applications
but not suitable for building a functional imaging system for real applications.9 A
complete redesign was therefore undertaken by the CERN microelectronics group.
Medipix2 was designed in 0.25-m CMOS technology with six metal layers,
which permitted greatly enhanced functionality per pixel and smaller pixel sizes at
the same time (see Llopart10 and references therein). A single chip has dimensions of
16.120 14,111 m, contains 256 256 square pixels of 55 m, and can be butted
on three sides, with an insensitive region of approximately 2 mm at one edge needed
for bond pads and control DACs. The basic pixel functionality of the final version of
the chip, called Medipix2-MXR20, is given in Figure10.2 and the basic parameters
in Table10.1.
Each pixel contains a charge-sensitive preamplifier with leakage current compensation and a test capacitance; a lower- and upper-level discriminator, with three-bit
adjustment DACs; and a 13-bit counter with a maximum depth of 11,810 counts plus
overflow detection. Only the most sensitive transistors are designed with enclosed
gate structures for improved radiation tolerance compared to Medipix1. However,
the majority of the chip is designed with linear transistors to keep the pixel size
Previous pixel
Shutter
3 bit Thrs Adj
Polarity
Disc
L
Mask bit
Input
Disc
H
Ctest
Vth High
Test in
Mux
Clock out
Vth Low
Mux
Disc
logic
Conf
8 bits conf
13 bit
shift
regis.
3 bit Thrs Adj

Next pixel
Analogue
Figure 10.2 Pixel layout of the Medipix 2 MXR20 chip.
Digital
222
Table10.1
Medipix2 Family Parameters
Pixel size
Pixels per ROC
ROC size
Technology
Counter depth
Threshold
Threshold adjust
Readout clock
Quad (2 2) module area
Maxipix (5 1) module area
Sensor
55 55 m
256 256
16.12 m 14.11 mm
0.25-m CMOS
14 bit
Lower and upper level (MXR20)
3-bit DAC per pixel (+1 global Vth)
200 MHz
512 512 = 262,144 pixels
1,280 256 = 327,680 pixels
Silicon, GaAs, Cd(Zn)Te
as small as possible, resulting in an overall radiation tolerance on the order of few

hundreds of kilorads.10 After this significant threshold, shifts occur. This means that
the Medipix2 chip is one order of magnitude more radiation tolerant than its predecessor, but when hybridized with standard silicon sensors it is still not radiation
hard enough for a number of applications at synchrotron sources with photon energies above 10 keV. The measured equivalent noise charge (ENC) is approximately
110 e root mean square (rms), the threshold dispersion is 95 e rms, and the minimum detectable charge is 900 e. The chips can be read out either serially, giving
a readout time of 5 ms at a clock frequency of 200 MHz, or in parallel, giving a
readout time of less than 300 s at a clock frequency of 100 MHz.
10.3.1.2Timepix
After successful proof-of-principle studies using the Medipix2 chip in gas-filled
detectors together with electron-multiplying structures like GEMs and micromegas, for time projection chambers for the International Linear Collider (ILC),11,12
it was decided to build a modified version of the Medipix2-MXR20 chip, called
Timepix, that would give arrival time and total charge information as well as spark
protection circuitry. The arrival time can be determined with approximately 10-ns
resolution, and the total charge information is obtained using a time-over-threshold
principle. The added functionality in the pixel was obtained at the expense of the
upper threshold comparator.13 It is interesting to note that the Medipix family of
chips resulted as a spin-off from HEP; with the development of the Timepix system,
these developments are now giving spin-back to HEP.
10.3.1.3Systems and Applications
Various readout systems have been built for the Medipix2 chips: MUROS,14 DEMAS,15
USB,16 and Maxipix.17 The USB-lite, shown in Figure10.3a, is the smallest member
of USB-based systems and is an ideal system for evaluation purposes; these systems
are easy to install and can be operated from a notebook using Pixelman software18
developed by the same group of Pospisil at IEAP-CTU Prague. It uses the serial
223
(a)
(b)
Figure 10.3 (a) Single Medipix2 chip with USB-lite readout system. (b) The Maxipix 5
1 chip module with 1,280 256 pixels with fully parallel readout.
readout mode of Medipix and is limited to five frames per second for the USB1.0
interface and 200 frames per second for the USB2.0 interface, called RUIN, which is
the maximum speed of the Medipix2 chip in serial readout mode using a 200-MHz
clock. These USB interface systems have been an important component for the large
diversification of applications since they create a low-threshold route for new users
to a working system.
At the other end of the spectrum is the Maxipix system, which uses the parallel
readout mode of the Medipix2 chip.17 The system depicted in Figure10.3b contains
1,280 256 pixels and, since all five chips are read out in parallel, can be operated at a
1-kHz framing rate with less than 50% dead time. The modules can be butted on three
sides, which in principle allows for the construction of systems with (n 1,280) 512
pixels. This system is specifically developed for synchrotron applications for which
high fluxes are available and a fast framing time is desired for many time-resolved
experiments, for which changes in the sample occur on millisecond timescales.
The Medipix2 family of chips was used in many different configurations and
applications as well as in combination with different sensors or even without any sensor. A complete overview of all the fields in which the Medipix2 systems have been
applied would be outside the scope of this chapter and can be found on the Medipix
Web site (http://medipix.web.cern.ch/MEDIPIX/Medipix2/indexMPIX2.html).
Single-chip modules have been successfully used for the detection of visible light
photons, neutrons, electrons, and of course X-ray photons.
One of major successes of the Medipix2 project is the technology transfer to
PANalytical (http://www.panalytical.com), a company providing X-ray analytical
equipment. PANalytical commercializes the PIXcel detector, which contains a single Medipix2 chip hybridized to a silicon sensor for X-ray diffraction analysis, with
currently a few hundred systems operational worldwide.
10.3.1.4Medipix3 Development
One of the important issues in photon-counting or photon-processing hybrid pixel
detectors is charge sharing between neighboring pixels. The charge generated in a semiconductor by the absorption of an X-ray photon will spread laterally due to diffusion
224
and internal repulsion while drifting to the readout node. This spreading depends on
the absorbing material, the thickness of the material, and the applied bias voltage. A
10-keV X-ray photon absorbed at the entrance of a 350-m thick silicon detector biased
with 300 V typically will generate a 10- to 20-m diameter charge cloud at the readout
side of the detector. In certain cases, this effect can be used to advantage (i.e., to get
subpixel spatial resolution), but in most cases it is a disadvantage. In pure photoncounting mode, it can lead to either double counting or missed counts when photons
impact very close to the border of a pixel, even if the lower threshold is set to 50% of the
photon energy. When high-Z sensors, like Cd(Zn)Te, are used, the absorbed photons
can give rise to X-ray fluorescence, and the fluorescent photon can be reabsorbed either
in the same pixel as the primary event or in a different pixel, or it can even leave the
sensor undetected. Charge sharing and fluorescence will always lead to a low-energy
tail in the recorded energy spectrum, making spectral analyses of limited value. These
effects are more important for smaller pixel sizes, leading to a trade-off between spatial
and energy resolution as well as to fixed pattern noise. To overcome this limitation and
to develop a two-dimensional spectroscopic hybrid pixel detector, the Medipix3 project
was founded (http://medipix.web.cern.ch/MEDIPIX/Medipix3/homeMP3.htm).
The principle used to correct for charge sharing is by designing a matrix with
communicating pixels. Medipix3 is designed in eight-metal-layer 0.13-m CMOS
technology with the same pixel and chip dimensions as Medipix2 to keep backward
compatibility. The eight metal layers allow for connectivity between the pixels and
the 0.13-m technology for more functionality, or intelligence, in each pixel. Each
pixel contains a charge-sensitive amplifier, followed by a shaper. This shaper generates a current proportional to the integrated charge, which is copied to the four corners of the pixel. At these nodes, the currents coming from the four pixels sharing
the corner are summed. Subsequently, an arbitration circuit compares the summed
currents of all nodes of a four-pixel cluster and assigns the hit to the node with the
highest value. In this way, the shared charge is regrouped, and much improved spectroscopic information is obtained; also, the degradation due to fluorescence is partly
corrected. In addition, Medipix3 will employ a double counter per pixel, allowing for
simultaneous read-write operation, thus providing dead-time-free imaging.19
10.3.2The Pilatus-2 Hybrid Pixel Detector

The Pilatus-2 was developed as the successor to the Pilatus-1 system to correct for a
number of shortcomings7 and to profit from increased functionality per area thanks
to deeper submicron CMOS technology. Pilatus-2 was also custom developed for
protein crystallography at the SLS and is the largest system of its kind operational at
synchrotron sources at the moment. Besides protein crystallography, the system has
found many applications in other X-rayscattering fields.
The ROC is designed in a standard 0.25-m CMOS process using radiationtolerant layout techniques. Each pixel contains a charge-sensitive amplifier and an
AC-coupled shaper.20 The gain and the shaping time of the amplifier can be adjusted
by a single global voltage reference, allowing for fine-tuning of the analog performance depending on the experimental conditions. A higher gain setting allows
for the detection of lower-energy photons at the cost of maximum count rate since
225
an increased gain will also lead to wider pulses.21 For intermediate gain settings,
the standard mode of operation, a nonparalyzing dead time of 200 ns is obtained,
allowing for input count rates up to 4 mega counts per second (Mcps)/pixel.21 The
shaper is followed by a single-level comparator, allowing discrimination against lowenergy photons, which is particularly advantageous in experiments in which there is
a large fluorescence background. Since there is no upper-level comparator present,
the system does not discriminate against higher harmonic radiation, occasionally
present in synchrotron experiments. The lower-level discriminator can be set by a
global bias voltage and can be individually trimmed for each pixel by a six-bit DAC.
Careful threshold trimming of a 94,965 (100K) pixel module results in less than
50-eV threshold dispersion at 8 keV, whereas an untrimmed module has a threshold
dispersion of 343 eV (in high-gain mode setting). The overall energy resolution of a
trimmed Pilatus 100K module is around 1 keV at 8 keV.21
The lower-level comparator is followed by a 20-bit counter per pixel. The ROC
runs at 66.67 MHz, giving a readout time of 2.85 ms. Since a radiation-tolerant layout is used together with a 20-bit counter, the pixel size is 172 172 m, almost a
factor of 10 larger than the Medipix pixel area. The Pilatus ROC chip contains 60
97 pixels, and a standard 100K module consists of 8 2 ROCs bump bonded to a
single 320-m thick p-on-n silicon sensor (Hamamatsu) and forms the basic building block of all Pilatus detectors. Radiation damage tests at 12 keV have shown the
modules (sensor plus ROC) work properly up to 30 Mrad, above which threshold
shifts become significant. Retrimming allowed operation at doses above 50 Mrad.
The basic parameters of the Pilatus-2 ROC and 100K module are given in
Table10.2. The 100K module is depicted in Figure10.4a.
This modularity of the 100K module is used to build systems of any size, with the
largest one available currently 6 million pixels, as shown in Figure10.4b. The Pilatus
6M system has been successfully operated at the protein crystallography beamline at
the SLS at PSI, with unprecedented data quality due to the low-noise performance of
photon counting as compared to the integrating CCD-based systems conventionally
used for protein crystallography. The great success of the Pilatus systems has led to
Table10.2
Pilatus-2 Detector Parameters
Pixel size
Pixels per ROC
ROC size
Technology
Counter depth
Threshold
Threshold adjust
Readout clock
100K module area
100K module size
Sensor
172 172 m
60 97
17.54 10.45 mm
0.25-m CMOS (radiation hard)
20 bit
Lower level only
6-bit DAC per pixel (+1 global Vth)
66.67 MHz
487 195 = 94,965 pixels
83.75 33.56 mm
320 m p-on-n silicon (Hamamatsu)
226
(a)
(b)
Figure 10.4 (a) The Pilatus 100K module. (b) The Pilatus 6M detector for protein
crystallography.
the creation of a spin-off company, DECTRIS (http://www.dectris.com), which is

now commercializing not only the Pilatus detectors but also other solid-state detectors developed by PSI/SLS, like the Mythen silicon strip detector.
The detector group at PSI/SLS has started developing the Extreme Framing
System (XFS), which will become the follow-up of the Pilatus-2 system, at least
in a certain number of applications. This system will have smaller 75-m pixels as
well as a continuous read-write operation, allowing for true dead-timefree framing
operation, an important advantage in protein crystallography.
10.4Cornell-Area Detectors Systems Corporation

ADSC Pixel Detector Project
Area Detectors Systems Corporation (ADSC; Poway, CA) and Cornell University
are collaborating on the development of a pixel array detector for macromolecular
crystallography and applications requiring a fast frame rate and a wide dynamic
range. A simplified schematic of the ASIC is shown in Figure10.5.22
The basic operation of the ASIC is to integrate charge received from the X-ray
detecting diode as a voltage. When the voltage rises to a globally defined reference
level Vref, circuitry is engaged to simultaneously remove a fixed amount of charge
from the integrator and add a digital bit into an in-pixel 18-bit counter. Thus, the
detector counts units of charge, which may be adjusted from the equivalent of a few
to a few hundred X-rays. During readout, the residual analog level in the integrator
is digitized as a low-order digital word.
The basic tile, consisting of a 128 128 pixel ASIC hybridized to a silicon detector layer, has been developed and demonstrated.22 Attributes of this detector are a
very large dynamic range (per readout, 2.6 107 10-keV X-rays/pixel), the ability to
read out the entire ASIC continuously at 1 kHz, and a noise level of about a third of
227

VHV
Isig
Quantized charge
removal circuit
Charge removal trigger
Vref
NQ
18-bit counter
Digital readout
Mux.
Vout
Vth
S&H
Gate
Gated oscillator
Analog readout
mux.
Sample and hold control
Figure 10.5 A simplified schematic of the ADSC-Cornell pixel. (From Vernon et al.)22
a 10-keV X-ray/pixel/readout. ADSC is considering marketing large-area detectors

made of mosaics of this basic tile.
10.5Future Hybrid Pixel Detectors for Free-Electron

Laser Sources
The next generation of (X-ray) photon sources will be FELs based on single-pass linear accelerators, followed by long undulators where the process of self-amplification
by stimulated emission (SASE) creates fully coherent and extremely brilliant X-ray
beams. With 1014 X-ray photons per pulse of 100 fs, these sources will allow recording a full diffraction pattern of nanometer-size objects in a single shot. This, together
with the large coherence length, will open up new fields of science but at the same
time put unprecedented requirements on the detectors. As an example, for coherent
X-ray diffraction imaging (CXI), up to 104 X-ray photons per pixel per pulse need to
be recorded, whereas for other pixels in the same image single-photon sensitivity is
required. This means that integrating detectors with low noise, large dynamic range,
and high frame rate are needed. Since these detectors do not yet exist, the FEL projects contain a significant detector development program as part of their construction.
This is a change in attitude at photon sources toward detector developments and is
partly to the credit of the success of the mentioned solid-state detectors.
10.6The LCLS Detectors

The Linac Coherent Light Source (LCLS) at the Stanford Linear Accelerator was the
first hard X-ray FEL to become operational with its first beam in 2009. It has three
dedicated detector development projects.
228
The detector for the CXI experiments was developed by the Gruners detector
group at CHESS/Cornell. The requirements for this detector are the ability to collect
X-rays from femtosecond pulses (thereby precluding photon counting) with a signalto-noise ratio greater than three for single 8-keV photons, a dynamic range greater
than 1,000 X-rays, and continuous 120-Hz frame rate. The detector is a classical
hybrid PAD with a 500-m thick p-on-n silicon sensor bump bonded to the readout
ASIC. The ASIC was developed in the 0.25-m CMOS technology and has 185
194 square pixels of 110 m. The pixel analog front end consists of an integrating
charge amplifier, with two selectable gain settings, followed by an in-pixel digitization via a single-slope 14-bit ADC.23,24 The gain map is set at the start of the CXI
experiment, with the high-gain setting used for parts of the diffraction pattern that
have very low X-ray flux (less than a few X-rays/pixel/readout), and the low-gain setting is used for the parts of the diffraction pattern with more flux per readout. The
full 1500 1500 pixel system will be built from monolithic 2 1 chip modules and
have an adjustable central hole to let the direct beam pass through it.
The X-Ray Active Matrix Pixel Sensor (XAMPS) is the detector for the pumpprobe experiments and is developed and built by the detector group of the National
Synchrotron Light Source (NSLS) at Brookhaven National Laboratory (NBL). It
consists of monolithic sensors fabricated in high-resistivity silicon with an integrated
J-FET per 60- or 90-m square pixel for which the charge is integrated during exposure. The sensor area is read out in a column-parallel mode by ASICs, which are
developed in 0.25-m CMOS technology and employ a charge pump followed by a
14-bit ADC to cope with the large dynamic range required by the experiments.25,26
The same group is also developing a small-pixel detector for X-ray photon correlation spectroscopy experiments, for which the expected flux per pixel is greatly
reduced, but the pixel size should be as small as 35 m. More details about these
systems can be found in the given references.
It should be noted here that pnCCD detectors developed at the Max-Planck
Semiconductor Laboratory in Munich, Germany, are also being developed for
operation at LCLS and will most likely be the first two-dimensional detectors
in operation in the experiments at the end of 2009.
10.7The European XFEL Detectors

The European X-ray free electron laser (XFEL) to be built in Hamburg, Germany
poses an additional challenge for the detectors due to its nonuniform time structure,
depicted in Figure10.6. The accelerator uses superconducting technology originally
developed for the ILC, which delivers 0.6-ms bunch trains containing up to 3,000
pulses repeated 10 times per second. The 30,000 pulses per second is an advantage
over the 120 evenly spaced pulses produced at the LCLS. The disadvantage is that
the pulses are delivered 0.6% of the time; the other 99.4% of the time is needed
for the superconducting accelerator to evacuate the produced heat. This means that,
in addition to the requirements for the detectors for the low bunch repetition rate
machines like the LCLS and the Spring8 Compact SASE Source (SCSS), the detectors for the European XFEL have to store their images during the bunch trains inside
229

100 ms
100 ms
0.6 ms
99.4 ms
200 ns
X-ray photons
FEL
process
<100 fs
Figure 10.6 Time structure of the European X-ray free electron laser.
the detector head. An additional complication is that it is nearly impossible at this

stage to draft final detector requirements for some of the science experiments foreseen at the European XFEL, since many applications will be rather revolutionary
than evolutionary and as a consequence are still not completely defined.
The European XFEL has started three independent detector development projects
that attempt to solve the high burst mode frame rate and the required high dynamic
range in different ways. They each also target a different application or wavelength
range. The AGIPD project, coordinated by DESY in Hamburg, is described in more
detail in the following section. Briefly presented are the basic principles and functionalities of the LPD project, coordinated by STFC in the United Kingdom, and the DSSC
project coordinated by the Max-Planck Semiconductor Laboratory in Munich.
10.7.1The AGIPD Project

A consortium consisting of DESY, PSI, and the universities of Hamburg and Bonn is
developing the AGIPD to fulfill the requirements for single-particle imaging at hard
(12-keV) X-rays as well as X-ray photon correlation spectroscopy (XPCS). Singleparticle imaging will generate up to 104 X-ray photons per pixel per 100-fs pulse at
small scattering angles, while at high angles the pixel intensity will mostly be zero
photons. For correct classification of the images and reconstruction of the particle,
it is mandatory to cover the complete dynamic range, with single-photon sensitivity
at low intensities. The required angular resolution is 0.5 mrad, with a minimum of
1k 1k pixels but ideally 4k 4k pixels. To profit from the increased luminosity of
the European XFEL, a maximum number of images should be recorded during the
bunch train. The XPCS experiments require a much better angular resolution, ideally of 4 rad, but will have much lower maximum intensities. More details about the
requirements for the various experiments foreseen at the XFEL can be found in the
technical design report.27 One of the additional challenges is the radiation hardness
230
C1
Discr.
Analogue pipeline
Trim
DAC
Column bus
C2
Analogue
encoding
Control logic
C3
Readout amp.
Analogue pipeline
Leakage comp.
Filter/write amp.
Figure 10.7 Pixel layout of the AGIPD chip.
of the detectors, even though the total expected dose is difficult to calculate at this
stage of the project, since the science applications are not yet fully defined.
The AGIPD is being designed in technology that uses eight-metal-layer, 0.13-m
CMOS and uses dynamic gain switching to cover the large dynamic range and an
analogue pipeline to store recorded images during the 0.6-ms bunch train; the images
are subsequently read out and digitized during the 99.4-ms interval between bunch
trains. A schematic layout of the pixel cell is given in Figure10.7.
The first stage is a classical charge-sensitive amplifier, however, with three different gain settings that are dynamically switched to cover the full dynamic range
required. For instance, feedback capacitors Cf, of 100, 1,600, and 6,400 fF for the
high-, medium-, and low-gain settings would cover the ranges up to 256, 4,096, and
16,384 12-keV photons, respectively. One of the critical issues in this concept is
obviously the noise induced by the gain switching, but both simulations and first test
circuits demonstrated that the noise induced by the gain switching is well below the
Poisson noise of the incoming flux. If the global intensity distribution in the image is
known beforehand, which might frequently be the case in the experiments, the gain
settings can be preset and fixed. For every image recorded, the corresponding gain
setting is stored in a pipeline. After amplification, the signal is stored in an analog
pipeline, consisting of a series of storage capacitors, in concept similar to the original
AP-HPAD developed by the Cornell group3 and described in this chapter. However,
the AGIPD should have as many storage cells per pixel as possible, while keeping the
pixel size as small as possible. At the same time, the leakage during the storage time,
which can be as long as 100 ms, should have a negligible deteriorating effect, requiring low-leakage capacitors and special designs for the switches used for writing and
231
reading. Finally, the design should be sufficiently radiation hard to withstand the
expected high doses. The 0.13-m CMOS process offers low-leakage capacitors,
either as MIM-caps or DGN (dual oxide n-FET in n-well) caps. Test structures have
shown that both options give satisfactory performance, even after irradiation up to
100 MGy, the estimated maximum accumulated does after 3 years of operation.
The DGN caps are smaller, thus allowing more storage cells per pixel. A major
concern, however, are the switches, which should permit charging the capacitors
within the 200-ns interbunch spacing but prevent discharging due to leakage during the 100-ms intertrain spacing. Test structures have shown that this performance
is achievable with standard designs using dual-oxide DGPMOS FETs, which show
radiation hardness up to 1 MGy, whereas the less-standard dual-oxide DGNMOS
FETs show radiation hardness up to the maximum 100 MGy. The final technology
and design will be determined by a trade-off among pixel size, functionality per
pixel, and radiation hardness.
The design goals are 200 200 m pixels, containing more than 200 storage
capacitors, with 64 64 pixels per chip. The 2 8 chips will be bump bonded
to single silicon sensors constituting the basic module building block, similar to
the Pilatus modules. However, the silicon sensors and readout ASICs have to be
mounted in the experiment vacuum, while the interface electronics will be housed
at ambient pressure to facilitate access for maintenance. The detector will consist of
four quadrants and a central hole for the direct beam. The quadrants can be moved
with respect to each other to adjust the central hole size, similar to the CXI detector
developed for LCLS.23 The interface electronics are responsible for the digitization
during readout, for providing trigger signals during data taking to synchronize to the
photon bunches, and for the slow control signals. The communication with the back
end is via 10-Gb links.
The 200-m pixel size is most likely not compatible with many of the XPCS
applications, which require 4-rad angular resolution, which translates to 80-m
pixels at a maximum sample-to-detector distance of 20 m. This means that a dedicated design, for instance, with a single-gain stage and reduced number of storage
cells, may be needed. Another option, which might be acceptable for a number of
applications, is the use of a double grid masking structure in front of the pixels. With
100-m square holes, half the pixel size, any aperture between 0 and 100 m can be
selected by moving the grids relative to each other and to the underlying pixels. This
solution would reduce the effective pixel size, so that not more than one speckle per
pixel is recorded, without reducing the number of pixels. This solution would not be
completely the same as a detector with smaller pixels, since the sampling will be at
fixed a 200-m pitch, irrespective of the effective pixel size. This might cause problems in certain experiments in which the speckles to be measured are confined to a
small area in reciprocal space.
An important part of the AGIPD project is detailed simulations of the detector
performance and the influence of detector distortions and limitations on the science
to be extracted. Detector simulations are customary in HEP experiments but unusual
in photon science. To steer the developments and provide guidance in the decisionmaking process of various compromises, like noise performance and number of
232
Detector
geometry
Module
tiling
Special
pixels at
ASICs
border
Photon
absorption
Electron
creation
Electron
drift
Thickness
material
Fano
factor
Charge
spreading
Parallax
Electron
collection
Charge
sharing
Dark
current
Charge
explosion
Amplication
Electron
storage
Amplier
noise
Leakage
Readout
ADC
Gain
switching
Implementation: IDL
Figure 10.8 Schematic modular layout of the detector and science simulation program
HORUS.
storage cells versus pixel size, a modular program, HORUS, has been developed.28
The schematic layout of the program is given in Figure10.8.
The program takes an expected scattering image as input and calculates the distorted image that will be recorded. The program permits modifying or disabling any
effect, which allows for detailed study of the various contributions to the image distortion. The recorded image can then be analyzed by the application scientist to determine whether the image distortions are acceptable or detrimental to their science.
10.7.2The LPD Project

The second two-dimensional detector development launched by the European X-ray
FEL is the LPD project managed by the detector group of the Rutherford Appleton
Laboratory in conjunction with the University of Glasgow. The LPD project attempts
to achieve the large dynamic range required by employing three different gain settings in parallel, each followed by its own analog pipeline. The principle is depicted
in Figure10.9. To implement three analog pipelines in parallel per pixel with 500
storage cells each, the pixel size cannot be less than 500 m, which is relatively large
for some photon science applications, hence its name.
20
5
1
Sensor
Preamplifier Multiple gains
Figure 10.9 Pixel layout of the LPD chip.
Pipeline
233
The LPD front-end module will include an interposer between the silicon sensor
and the ROC, which gives the flexibility to have different pixel sizes and layouts
between the silicon sensor and the ROC. It also provides space between the sensor
and the ROC for hidden wire bonds as well as extra radiation shielding, relaxing the
required radiation hardness of the ASIC. Each chip will contain 16 32 pixels, and
8 1 chips will be bump bonded to a monolithic silicon sensor giving 128 32 pixel
tiles. Of these tiles, 2 8 will be used to construct so-called supermodules with
256 256 pixels. These supermodules can then be used to construct detectors of
any size. The LPD, due to its large size, is very well matched to the requirements
of pump-probe experiments with liquid samples, for which large total angular coverage with limited angular resolution is required. An interesting option for the LPD
is to string together the three different analog pipelines, if the required gain setting is
known beforehand, which is normally the case for liquid-scattering experiments, for
which the scattering pattern is very reproducible. This will allow recording up to
1,500 images per bunch train and thus a near optimal use of the high luminosity of
the European XFEL.
10.7.3The DSSC Project

The previous two projects attempt to solve the dynamic range challenge by using
multiple gains or dynamic gain switching, and both store the images in an analog
pipeline. The DSSC uses a nonlinear response of the active sensor pixels to cope
with the dynamic range and digital memory to store images inside the pixels.29 The
DSSC project is developed by a consortium consisting of the MPI-Semiconductor
Lab in Munich, which is also the coordinator, the Politecnico di Milano, DESYHamburg, and the Universities of Heidelberg, Siegen, and Bergamo. The principle of
a single DEPFET of the DSSC sensor is given in Figure10.10. The electrons generated by absorbed photons are stored in potential minima underneath the gate of the
FET and thus under the conductive channel, as in a standard DEPFET.30 However,
in the DSSC DEPFET design, the potential minimum, also called the internal gate,
extends beyond the gate region into the source region. The result is that the first electrons generated will be stored in a potential minimum right under the gate, giving a
Gate
Source
Drain
Internal gate
Fully depleted silicon
Back contact
Figure 10.10 Principle of the DSSC sensor with a nonlinear gain response.
234
large effect on the source drain current, whereas subsequent electrons will be stored
only partly under the gate and increasingly under the source region, thus having a
reduced effect on the source drain current.
This nonlinear response of the FET provides the so-called signal compression
and results in a large dynamic range. The actual size of a DSSC pixel is much larger
than the FET area. The transistor is surrounded by a small drift volume to collect
the charge generated by the incoming radiation into the internal gate. An additional
advantage of the DEPFET is the low-noise performance, which makes this detector well suited for experiments using lower-energy X-rays, down to a few hundred
electron volts, whereas the AGIPD and LPD projects are focused on the harder X-ray
experiments around 12 keV. The nonlinear amplification and the intrinsic DEPFET
low noise allow achieving a dynamic range of about 6,000 1-keV photons per pixel
and single-photon resolution simultaneously. The DSSC design foresees hexagonal
pixels, which give a more homogeneous drift field and a faster charge collection
than square pixels, and, with 136-m long sides, results in a 200-m bump-bond
pitch, the minimum provided by the 0.13-m IBM CMOS process used for the ASIC
design. The DEPFET sensor will be bump bonded to read out ASICs with an amplifier, shaper, and ADC per pixel as well as digital storage memories. The advantage
of digital storage over analog is the absence of signal leakage; thus, images can be
stored for indefinite times. The added challenge, however, is the high speed on chip
digitization, which has to be done well within the 200-ns minimum bunch spacing.
The baseline design foresees the possibility of storing more than 500 images per
bunch train. The readout ASIC will have 64 64 pixels, and 2 4 ASICs will be
bump bonded to a monolithic DEPFET sensor. Two of these monolithic units will be
used to build a ladder with 128 512 pixels, which can then be stitched together to
form the final 1k 1k detector.
10.8Summary and Outlook

Since the late 1990s there has been a significant change toward detector developments at synchrotron sources. Traditionally, detectors used at synchrotron sources
were adopted and adapted from other science applications or were commercially
available but not explicitly developed for specific experiments. In this chapter, I illustrated some of the hybrid pixel detectors developed for photon science applications.
The success of this approach is probably best illustrated by the Pilatus detectors
developed at PSI for the SLS, which are now finding wide-ranging applications at
synchrotron sources throughout the world. With the X-ray FELs in the United States
and Europe, for the first time a significant budget for detector development was available at the beginning of the construction of the facilities, which is resulting in stateof-the-art, custom-designed systems pushing the frontiers of detector developments.
It is reasonable to assume that this trend will continue in the future.
There is still a lot of room for improvement of the systems currently used or even
the ones under development. On the side of the sensors, the developments of edgeless, or better edge-sensitive, sensors will reduce dead areas in the focal plane detectors, which is highly desirable for real space-imaging experiments. Also higher-Z
sensors are needed since the X-ray spectrum at synchrotron sources extends well
235
beyond 15 keV, at which silicon sensors become inefficient. The development of

GaAs, Cd(Zn)Te, and germanium sensors are of particular interest. On the ASIC
side, deep submicron CMOS allows for ever-increasing digital sophistication per
area, providing the possibility to create intelligence in the pixel, as demonstrated in
the Medipix3 project. However, the analog circuits do not benefit from this trend;
on the contrary, they suffer from the reduced voltage swings allowed. Therefore,
the development of three-dimensional ASIC integration technology with throughchip vias is of extreme interest for photon science, since it will allow even further
increased functionality per pixel area, as well as the use of different optimized technologies for the analog and the digital parts of the circuitry. Photon science applications are particularly well placed to profit from this new technology, since they do
not suffer from material budget restrictions like HEP applications. Consequently,
photon science might well be the driving force in solid-state detector developments
for years to come.
Acknowledgments
I gratefully acknowledge the help and input from C. Brnnimann, M. French, S.
Gruner, M. Porro, A. Schwarz, and C. Youngman.
References

1. S. L. Barna, J. A. Shepherd, M. W. Tate, R. L. Wixted, E. F. Eikenberry, and S. M.

Gruner. IEEE Trans. Nucl. Sci., 44, 950956, 1997.
2. E. F. Eikenberry, S. L. Barna, M. W. Tate, G. Rossi, R. L. Wixted, P. J. Sellin, and S. M.
Gruner, J. Synchrotron Radiat., 5, 252255, 1998.
3. G. Rossi, M. J. Renzi, E. F. Eikenberry, M. W. Tate, D. Bilderback, E. Fontes, R. Wixted,
S. Barna, and S. M. Gruner, J. Synchrotron Radiat., 6, 10961105, 1999.
4. A. G. MacPhee, M. W. Tate, C. F. Powel, Y. Yue, M. J. Renzi, A. Ercan, S. Narayanan,
E. Fontes, J. Walther, J. Schaller, S. M. Gruner, and J. Wang, Science, 295, 12611263,
2002.
5. W. Cai, C. F. Powel, Y. Yue, S. Narayanan, J. Wang, M. W. Tate, M. J. Renzi, A. Ercan,
E. Fontes, and S. M. Gruner, Appl. Phys. Lett., 83, 16711673, 2003.
6. Ch. Brnnimann, R. Baur, E. F. Eikenberry, S. Kohout, M. Lindner, B. Schmitt, and R.
Horisberger, Nucl. Instrum. Methods A, 465, 235239, 2001.
7. Ch. Brnnimann, E. F. Eikenberry, B. Henrich, R. Horisberger, G. Heulsen, E. Pohl, B.
Schmitt, C. Schulze-Briese, M. Suzuki, T. Tomizaki, H. Toyokawa, and A. J. Wagner,
Synchrotron Radiat., 13, 120130, 2006.
8. C. Ponchut, J. L. Visschers, A. Fornaini, H. Graafsma, M. Maiorino, G. Mettivier, and
D. Calvet, Nucl. Instrum. Methods A, 484, 396406, 2002.
9. C. Ponchut, and F. Zontone, Nucl. Instrum. Methods A, 510, 2934, 2003.
10. X. Llopart, Doctoral thesis, Mid Sweden University, 2007.
11. P. Colas, A. P. Colijn, A. Fornaini, Y. Giomataris, H. van der Graaf, E. H. M. Heijne, X.
Llopart, J. Schmitz, J. Timmermans, and J. L. Visschers, Nucl. Instrum. Methods A, 535,
506510, 2004.
12. M. Campbell, M. Chefdeville, P. Colas, A. P. Colijn, A. Fornaini, Y. Giomataris, H.
van der Graaf, E. H. M. Heijne, P. Kluit, X. Llopart, J. Schmitz, J. Timmermans, and
J. L. Visschers, Nucl. Instrum. Methods A, 540, 295304, 2005.
236
13. X. Llopart, R. Ballabriga, M. Campbell, L. Tlustos, and W. Wong, Nucl. Instrum.

Methods A, 581, 14851494, 2007b.
14. D. San Segundo Bello, M. van Beuzekom, P. Jansweijer, H. Verkooyen, and J. L.
Visschers, Nucl. Instrum. Methods A, 509, 164170, 2003.
15. M. Maiorino, R. Martinez, G. Pellegrini, G. Blanchot, M. Chmeissani, J. Garcia, M.
Lozano, C. Puigdengoles, and M. Ullan, Nucl. Instrum. Methods A, 563-1, 9699, 2006.
16. Z. Vykydal, J. Jakubek, and S. Pospisil, Nucl. Instrum. Methods A, 563-1, 112115, 2006.
17. C. Ponchut, J. Clment, J.-M. Rigal, E. Papillon, J. Vallerga, D. LaMarra, and B.
Mikulec, Nucl. Instrum. Methods A, 576, 109112, 2007.
18. T. Holy, J. Jakubek, S. Pospisil, J. Uher, D. Vavrik, and Z. Vykydal, Nucl. Instrum.
Methods A, 563, 254258, 2006.
19. R. Ballabriga, M. Campbell, E. H. M. Heijne, X. Llopart, and L. Tlustos, Nucl. Sci.
Symp. Conf. Rec. IEEE, 6, 35573561, 2006.
20. P. Kraft, A. Bergamaschi, Ch. Brnnimann, R. Dinapoli, E. F. Eikenberry, B. Henrich, I.
Johnson, A. Mozzanica, C. M. Schleptz, P. R. Willmott, and B. J. Schmitt, Synchrotron
Radiat., 16, 368375, 2009.
21. P. Kraft, A. Bergamaschi, Ch. Brnnimann, R. Dinapoli, E. F. Eikenberry, H. Graafsma,
B. Henrich, I. Johnson, M. Kobas, A. Mozzanica, C. M. Schleptz, and B. Schmitt,
IEEE Trans. Nucl. Sci., 56, 3, 758764, 2009.
22. W. Vernon, M. Allin, R. Hamlin, T. Hontz, D. Nguyen, F. Augustine, S. M. Gruner,
Ng. H. Xuong, D. R. Schuette, M. W. Tate, and L. J. Koerner, Proc. SPIE 6703, paper
6703-22, 2007.
23. L. J. Koerner, H. T. Philipp, M. S. Hromalik, M. W. Tate, and S. M. Gruner, Proc. PIXEL
2008 Int. Workshop, Jinst 4, P03001, 2009.
24. H. T. Philipp, L. J. Koerner, M. S. Hromalik, M. W. Tate, and S. M. Gruner, IEEE Nucl.
Sci. Symp. Conf. Rec., 15671571, 2008.
25. A. Dragone, J. F. Pratte, P. Rehak, G. A. Carini, R. Herbst, P. OConnor, and D. P.
Siddons, IEEE Nucl. Sci. Symp. Conf. Rec., 29702975, 2008.
26. G. A. Carini, W. Chen, A. Dragone, J. Fried, J. Jakoncic, A. Kuczweski, Z. Li, J. Mead,
R. Michta, J.-F. Pratte, P. Rehak, and D. P. Siddons, IEEE Nucl. Sci. Symp. Conf. Rec.,
15721577, 2008.
27. M. Altarelli, et al. European X-ray Free Electron Laser Technical Design Report, 2006.
28. G. Potdevin, U. Trunk, and H. Graafsma, Nucl. Instrum. Methods A, 607, 5154, 2009.
29. M. Porro, L. Andricek, A. Castoldi, C. Fiorini, P. Fischer, H. Graafsma, K. Hansen, A.
Kugel, G. Lutz, U. Pietsch, V. Re, and L. Struder, IEEE Nucl. Sci. Symp. Conf. Rec.,
15781586, 2008.
30. M. Porro, G. Ferrari, P. Fischer, O. Haelker, M. Harter, S. Herrmann, N. Hoernel, R.
Kohrs, H. Krueger, P. Lechner, G. Lutz, I. Peric, R. H. Richter, L. Strueder, J. Treis, M.
Trimpl, and N. Wermes, IEEE Trans. Nucl. Sci., 53, 401408, 2006.
a Photon11 XPAD,
Counting Imager for
X-Ray Applications
Patrick Pangaud and Pierre Delpierre
Universt de la Mditerrane
Marseille, France
Jean-Franois Brar
Institute Neel, UJF-GNRS

Grenoble, France
Contents
11.1 Introduction................................................................................................... 237
11.2 From High-Energy Physics to X-Ray Imagers.............................................. 238
11.2.1 Silicon Detector Developments for Microvertex Detectors in
High-Energy Experiments................................................................. 238
11.2.2 From High-Energy Particle Physics Detectors to X-Ray
Cameras............................................................................................. 241
11.3 Hybrid Pixel Detectors.................................................................................. 242
11.3.1 Sensors............................................................................................... 242
11.3.2 Front-End Electronics........................................................................ 243
11.3.3 The Bump-Bonding Technique......................................................... 243
11.4 XPAD, a Hybrid Pixel Detector..................................................................... 243
11.4.1 The XPAD1 and XPAD2 Chips........................................................244
11.4.2 The XPAD3 Chip.............................................................................. 247
11.5 Applications of X-Ray Imagers..................................................................... 253
11.5.1 Crystallography Application in Material Sciences............................ 253
11.5.2 Biomedical Application: Computer Tomography.............................. 257
11.6 From 2D to 3D Approach: A New Hybrid Trend.......................................... 259
References...............................................................................................................260
11.1Introduction
Photon-counting imagers have appeared recently in high-energy physics experiments. In these experiments, localization of ionizing particles has been done for several years by multiwire proportional chambers (MWPCs) with many variations (drift
237
238
chambers, TPC, etc.). For vertex applications, these detectors have been progressively replaced by silicon sensors, first arranged in microstrip detectors, and more
recently (1997) in hybrid pixel detectors. These detectors are extensively used in the
big detectors of the Large Hadron Collider (LHC) at the European Organization for
Nuclear Research (CERN, Geneva, Switzerland). Some of these detectors, adapted
to photon detection, have appeared more recently to cover other areas, such as medical imaging or crystallography experiments.
This chapter relates the story of one of these hybrid pixel detectors, the X-ray
pixel chip with adaptable dynamics (XPAD), intended for X-ray applications. At the
beginning of the XPAD project, it was well established that the increase in experimental possibilities associated with the new synchrotron sources are often limited by
the quality of available detectors. To increase the functionality of cameras with goniometers at the beamline D2AM at the European Synchrotron Research Facilities
(ESRF, Grenoble, France), a new two-dimensional (2D) detector was developed
using hybrid pixel technology. Its aims were noise reduction, large dynamical range,
and high counting rate as required by material science studies by X-ray scattering.
Increasing the readout speed was also requested. The project started with the Centre
de Physique des Particules de Marseille (CPPM, France) laboratory in 1999, with
the first prototype, XPAD1, tested in 20002001. Then, a second version (XPAD2,
20022003) followed that corrected a few problems. The third-generation, XPAD3,
with a new architecture, arrived in 2006. At the same time, hybrid pixel technology
was starting to be considered in computed tomographic (CT) scanners to improve
image contrast.
Hybrid pixel detectors integrated sensors (silicon diode) with a complementary
metal oxide semiconductor (CMOS) chip. Each sensor pixel is coupled via a microscopic metallic ball (bump-bonding technology) to a monolithic electronic counting
device in a dedicated circuit. By placing XPAD circuits side by side, a module 7 by
1.5 cm could be assembled in tiles to form a large 7 by 12 cm detector. This is an
appropriate surface for crystallography applications or small-animal imaging. The
last version of the XPAD detector is linear up to 107 photons per second per square
millimeter, has a half million square pixels on a 130-m side, and can read out data
up to 500 frames per second.
11.2From High-Energy Physics to X-Ray Imagers

11.2.1Silicon Detector Developments for Microvertex
Detectors in High-Energy Experiments
In high-energy physics experiments, one of the detector goals is to localize the trajectories of the charged particles. In the past, this task was usually accomplished
by bubble chambers. The invention of the MWPC by George Charpak in 1968 was
a revolution that led to so-called electronic experiments. In the 1970s, the bubble
chambers disappeared and were replaced by wire chamber trackers in all the experiments. However, even if the spatial resolution (hundreds of micrometers) was enough
for the core of the experiments, it was not for short lived particle vertexes around
the target. The first silicon microvertex detectors appeared in the beginning of the
XPAD, a Photon-Counting Imager for X-Ray Applications
239
Figure 11.1 Mounting of the second forward pixel layer on the DELPHI vertex detector
(down).
1980s in fixed target experiments in which the beam from the accelerator hits a
target. The NA1 experiment (19821984) at CERN was the first with a target made
of layers of silicon strip detectors that measured the ionization created in each layer
by charged particles going through it. But, the microvertex detectors became crucial
with the new particle colliders. These accelerators produce two beams (electrons
or protons) running in reverse direction that collide at the center of the experiment.
The detector has a cylindrical geometry, and it was mandatory to localize with high
spatial resolution (few micrometers) the particles with a short lifetime (only few
hundreds of micrometers of travel before disintegration) emitted from the collision
point. The success of the NA1 silicon strip detectors led to the choice of this type
of tracker for collider experiments. For example, all the large electron project (LEP,
CERN) experiments (ALEPH, DELPHI, L3, and OPAL) had a silicon strip detector working for all the LEP lifetime (19892000). At the same time, a silicon vertex
detector (the SVX) was installed on the CDF experiment at FNAL (Chicago, IL).
Figure11.1 shows the mounting of the DELPHI vertex detector with the cylindrical
strip layers.
In the silicon strip detectors, the charges generated in the crystals are collected by
long and thin (a few tenths of a micrometer) strip diodes, and charge amplifiers are
situated at the end of the strips. They have the requested resolution but only in one
direction. It is possible to get the second direction information by strip structures on
the backplane (double-sided strip detectors), but for high density of particles one has
to face many ambiguities.
By the end of the 1980s, different groups started to think about two-dimensional
(2D) silicon detectors. In this case, the sensor is now divided in more or less square
240
pixels, each connected to one pixel cell of a dedicated application-specific integrated

circuit (ASIC). The problem of these surface connections was already under study
by electronic companies to save space by assembling different layers of ASICs. The
microelectronic group at CERN was aware of this technique and started to solve the
problems of reducing the size of the contact pads and designing a compact electronic
chain to fit with a sensor pixel size of about 100 m. The CPPM group was trying to
solve this problem in an economic way using a plastic tape filled by microballs that
make a one-directional contact by surface compression. This technique was working
for contact pads of at least 100-m diameter.
At that time, there was discussion at CERN on how to build detectors for LHC,
and a large research and development effort was undertaken. In particular, at the
time there was no solution available for the vertex detector regarding the high density
of particles and high radiation damage immunity required. The need for a precise
2D detector with a high signal-over-noise ratio led to the hybrid pixels, and in 1989
an international group (RD19) was created by the LHC/CERN committee to develop
hybrid pixel detectors for LHC.
In 1991, the first demonstrator was built1 for the WA97 experiment (heavy ion
fixed-target experiment at the SPS/CERN accelerator). Two prototypes of the electronic chip have been built, and one turned out to be suitable for this experiment.
Each pixel included an amplifier, a discriminator, a delay, a coincidence circuit, and
a latch. All that circuitry was placed under the sensor pad surface, which was 75
500 m. The bump bonding was done by solder bumps of 30 30 m using a technique developed at GEC Marconi (Caswell, UK). Matrixes of 16 64 pixels have
been built and were bump bonded on detector ladders 5 cm long. There were six
ladders per detection plane and two planes 5 cm 5 cm.
The RD19 team split into the Medipix collaboration (CERN) and the CPPM pixel
group. The CPPM group with a team from the DELPHI experiment at LEP built a
hybrid pixel detector to increase the tracking capability in the forward direction. In
1996, two layers of pixels on both sides of the DELPHI vertex detector (1.2 million
pixels) were installed.2 Each layer was split in two half cones that included 19 modules each. The modules used 6.4 1.5 cm silicon sensors on which 16 ASICs were
bump bonded. One strong requirement was to reduce the material, then the module support is a small piece of ceramic, and the connection to the external world is
done by thin flexible printed circuits as shown in figure. The mounting of the second
forward pixel layer on the cylindrical strip detector is shown in Figure 11.1. This
was the first large pixel detector (1.2 million pixels) installed on a high-energy collider. With these pixel layers, the track reconstruction in the forward direction was
increased by a factor of two.3 This pixel detector was working until the end of LEP
operation in 2000.
From the success of this detector, all the LHC experiments (ALICE, ATLAS,
CMS, LHCb) included a pixel vertex detector. For example, for the ATLAS experiment, the pixel vertex detector included three cylindrical layers 0.8 m long followed
by three disks on each side for about 100 million pixels (Figure11.2).4,5 This detector can localize particle tracks with accuracy better than 10 m in space and 20 ns
in time.
241
Semi-filled shell of its staves with modules
Figure 11.2 View of the inner part of the ATLAS pixel vertex detector.
11.2.2From High-Energy Particle Physics Detectors to X-Ray Cameras

The CPPM took a large part in these developments, in particular for DELPHI and
ATLAS, and the pixel team was contacted by physicists working at the ESRF synchrotron. This machine of the third-generation X-ray light sources has made major
progress in material sciences. However, the performances of the usual detectors were
not sufficient to take the full benefit of the intense monochromatic X-ray beam provided by these sources. It became obvious that the hybrid pixel detectors would
provide a big step forward in satisfying the needs of this field of scientific research
thanks to the possibility that they may reach a very high (almost infinite) dynamic
range. Most of the X-ray cameras in use work by integration of the collected charge,
which is then digitized by analog-to-digital converters. The converter capacity is
limited, and the noise cannot be subtracted, leading to a restricted dynamic range.
The photon-counting functionality of the hybrid pixel detectors corrects this defect
and makes increasing the image frame rate by a large factor possible.
These attractive features led the CPPM group to propose hybrid pixel detectors
for synchrotron X-ray light sources.6 In collaboration with a CNRS team working
at ESRF, a dedicated chip was designed (the XPAD1), followed by a second version
(the XPAD2). This second chip was good enough to build a large-size X-ray camera
(7 12 cm).7,8 This camera was extensively tested on the D2AM/ESRF X-ray beamline, and the good results have motivated the SOLEIL synchrotron X-ray light source
to collaborate for the development of a new chip (the XPAD3). For this circuit, submicronic technology has been chosen to reduce the pixel size to 130 m. CdTe sensors have also been investigated. Also, for the Synchrotron Light Source (SLS) at the
Paul Scherrer Institut (PSI, Villigen, Switzerland), the detector group put much effort
into the development of a large hybrid pixel detector. This group, with the benefit of
the investments and electronic engineers coming from the CMS vertex detector, succeeded in building large-size hybrid pixel detectors for X-ray. Some people from this
group have created a dedicated firm (Dectris, Switzerland); some engineers from the
CPPM pixel group are going to do the same (imXPAD, France).
242
Other domains in which hybrid pixel detectors should bring substantial progress
are small-animal imaging and medical imaging. To test the interest in this field, in
2005 CPPM started to build a CT scanner for small animals (the PIXSCAN)9 with
the XPAD2 as the X-ray detector. The Institut de Biologie du Dveloppement de
Marseille Luminy (IBDML, France) biologists joined the CPPM group to validate
this new instrument. Now, the last generation of the XPAD detector, the XPAD3, is
being installed inside a small-animal micropositron emission tomograph (PET),
allowing simultaneous PET and CT images.
11.3Hybrid Pixel Detectors

Progress in microelectronics in the 1990s encouraged compactness of the data processing. The development of hybrid pixels has followed by including a whole dataprocessing sequence for detection of particles in a few square millimeters with a low
power budget.10 Hybrid pixel detectors can be seen as microstrip detectors that have
been cut up and that offer individual detection zones that are smart enough to work
separately. The pixels are called hybrids because of the decoupling of the in-sensor
detection and the data-processing electronics, and various types of sensors can be
used. This means that sensor and electronics are fabricated on different substrates
and finally joined. The sensor transforms the particles into electronic charges and
feeds electronic devices through a small connection. This connection between the
sensor and electronics must be carried out with the most reliable possible contact to
secure the signal, which should not be degraded. After using anisotropic conductive
tape at the beginning, the hybrid pixels rapidly adopted the microball connection.
11.3.1Sensors
The sensors are semiconductor diodes that transform the energy losses of an incoming
particle into electric charges. The basic material can be silicon, germanium, CdTe, or
GaAs. The amount of energy needed to create an electronhole pair E e/h in silicon is
about 3.6 eV, but its efficiency is limited to photon energies below 20 keV according
to the Z and the density of this substrate. The CdTe material has an obviously higher
absorption but suffers bad yield for a large surface and is very delicate. Its Ee/h is
about 4.4 eV with efficiency close to 100% until 60 keV. The GaAs material, another
high-Z material, is also a difficult material for realizing large surface sensors with
low leakage current. All these materials are not very tolerant of ionizing radiation.
When divided into small portions, the pixel diode exhibits a capacitance that
is much smaller than strips but that cannot be ignored. The depleted capacitance
is about 5 fF for a silicon pixel 100 m square with a thickness of 300 m. It is
necessary to add to it the interpixel coupling capacitance, which is an order of magnitude higher. Finally, by adding the capacitance of the microball, we can observe
a total capacitance of about 200 fF. The reverse-biased diode also has a leakage
current of about 10 pA that grows with the effects of irradiation. We must take into
account these two parameters because their contributions to the total noise are not
negligible and degrade the resolution of the chain.
243
11.3.2Front-End Electronics
The electronic charges coming from the sensor must be transformed, sorted, analyzed, and eventually stored. The amount of charges follows a Poissonian law distribution according to the physics of its creation. These charges will be transformed into
either tension or current signal, and this signal will be amplified in turn. For that, we
use a preamplifier, whose role is to amplify and preserve the signal without damage.
This one must be compact, with a low power budget and low noise, be linear, and be
adapted to the experiment. A feedback capacitor or resistor transforms the charges or
equivalent electrical current into voltage. A feedback switch cuts off the signal and
must be used to prevent accidental pile up when new charges are coming. A second
stage of amplification may exist to increase the signal-to-noise ratio. Finally, the
signal often feeds a comparator since only the presence of the particle is exploited,
whatever its energy loss. The signal is now digital and can be memorized and processed. Various architectures exist according to the experiments requirements: massive storage of the data, processing speed, readout of the pixels and reading speed,
correlation with an external trigger. The digital data become easy to transfer, are not
modifiable, and can be copied. Another parameter is the homogeneity of the pixels
in the same matrix. Microelectronic devices are process dependent and suffer some
physical parameter dispersions, which in turn give fluctuations in pixel performance.
To fix these fluctuations, pixels often include an individual and adjustable stage.
It is obvious that the complexity of these electronics comes especially from the
compactness of the pixel. In such a restricted place, it is necessary to find a compromise between the space and the number of functions. Moreover, every pixel must be
autonomous and not influenced by its neighbors. With more than 10,000 pixels in no
more than 1.5 cm2, the power budget also becomes a large challenge.
11.3.3The Bump-Bonding Technique

The pixel part of the sensor is connected directly to its electronic part by a single
bond, and a pixel detector corresponds to a multitude of individual amplifier chains,
all independent. The connection of the sensor to its electronics has required much
research and testing. The conducting tape used at the beginning offered good connections, but its dielectric constant was too high and influenced the resulting signal.
From the existing ball-grid array (BGA) for the flip-chip devices, some companies
like IZM in Germany or Alenia Microsystems (AMS) in Italy succeeded in creating
microballs by the bump-bonding technique with balls a few tens of micrometer in
diameter. These balls are made of solder (SnPb) or indium (depending on the process)
and are deposited on under-bump metallization, since the top metal layer (aluminum)
of the chips is not directly wettable. Techniques involved in such a process are derived
from the ones used in microelectronic circuit fabrication and involve full wafers.
11.4XPAD, a Hybrid Pixel Detector

Development of XPAD chips started with the reuse of the know-how coming from the
development of hybrid pixels for high-energy physics. The idea was to design a chip
244
for X-ray applications in parallel with the ATLAS vertex subdetector. Electronics
choices were driven at that time by the intensive use and knowledge of the AMS
technology and by the availability of spare sensors from a previous DELPHI development. Collaboration with the D2AM beamline at ESRF allowed determination
of the specifications. With this third-generation synchrotron radiation source, diffraction, macromolecular crystallography, and small-angle X-ray scattering (SAXS)
studies require simultaneous measurements of very intense spots and weak diffuse
scattering that is often only just above background level. The detectors therefore have
to offer very low noise and wide dynamic range over more than six decades; furthermore, they should have fast readout to allow the study of fast kinetics. Conventional
experimental settings do not fulfill these requirements. Slits associated with scintillators and photomultipliers allow only point detection; many scattered photons
are lost, and the acquisition time is long. Large-area charge-coupled device (CCD)
detectors with a phosphor screen and optical demagnification with fibers have a low
dynamic range, slow readout, and spread function with large tails. These deficiencies
justify the interest of the hybrid pixel-counting detectors. In the same way, X-ray
imagers are intensively used for medical imaging, and hybrid pixel detectors could
improve some performance in dose reduction and contrast enhancement.
11.4.1The XPAD1 and XPAD2 Chips

The first version of circuit XPAD, manufactured in 2000 in the AMS 0.8-m CMOS
process, contained 600 pixels distributed in 24 columns of 25 elements. The choice
of 330-m square pixel size has been driven by the reuse of the silicon sensor of
CERNs DELPHI detector. The XPAD1 global architecture has been built on one
of the front-end chips dedicated to the development of the ATLAS experiment. Some
new approaches have been implemented for X-ray diffraction experiments.11
The architecture of the XPAD imager had to be customized for X-ray experiments. In X-ray applications, more than 50% of the pixels are hit, and high-rate
photon counting and a large dynamic range are required. This requires a periodic
data readout of the pixel counters to avoid data overflow. The XPAD1 chip had a
size of 11.9 by 8.2 mm. Its main characteristics include a photon-counting rate from
1 to 107 photons s1 pixel1, adjustable and pixel-to-pixel tunable threshold, and an
on-the-fly readout.
Each pixel consists of an analog part followed by a digital counter and configuration registers. A charge-sensitive amplifier using an automatic pole-zero cancel
lation, based on a folded cascade approach, supplies a differential-in differential-out
amplifier for a high power supply rejection ratio (PSRR). This last stage feeds in turn
a differential discriminator. A fine threshold adjustment has been implemented in
each pixel by a four-bit internal digital-to-analog converter (DAC). The LSB range
is globally set so that the trim resolution can be balanced with the trim range. A
16-bit overflow adjustable counter and a readout system constitute the digital part of
the pixel. After a selected acquisition time, the scanner-based readout system carries data to the bottom of the chip. The technical requirements were a power budget
below 100 W per pixel to avoid any extra cooling system in the final detector, an
incoming X-ray energy range from 5 to 50 keV, a threshold spread reduced to a few
245
15
14
Counter
1
0
Mux
Data
Select
Scanner
Latch
Pixel
4-bit counter
Pre-charge subtract
From register
Scan
manager
Dynamic
delay
From data
acquisition
Scanner
delay
out
From last pixel
hit
Overflow select
Reset
Serial
out
Figure 11.3 Diagram of the readout part of XPAD1.
10 e rms for the whole chip after calibration, and positive and negative polarity
input charge acceptance for the preamplifier. Therefore, CdTe sensors were expected
to connect to the XPAD1 circuit.
The pixel array formed a large serpentine shape, with each pixel serially connected to its neighbor. During the pixel readout time, a four-bit counter memorizes
the position of the overflow pointer set into each pixel. The first 600-bit output data
stream out, clocked externally by a 33-MHz master clock (PCI standard) provided
by the data acquisition (DAQ). The mechanism used reads the nth bit of all pixel
counters to form 600-bit output data, then decrements the global four-bit counter
and starts a new cycle on the (n 1)th bit. This mechanism authorizes the charge
acquisition and the readout phase simultaneously. This mechanism also allows a real
infinite dynamic range. By extension, if one wants to fasten data acquisition, it is
possible to read only the 16th bit of the 16-bit counter. Successive values of this bit
are added in an external 32-bit memory present in the DAQ board, and at the end of
the acquisition, all 16-bit pixel counters are read out and added to the same memory
as shown in Figure11.3.
A main register is used for chip configuration along with several global registers
for setting purposes. These registers configure the XPAD1 chip into a given state by
control of the polarizations in each stage of the pixel. Several eight-bit DACs officiate
as bias generators to polarize all analog stages of pixel and to fix the global threshold
value. In the same way, a control part of a pixel containing five latches locally stores
injection, mask, and threshold fine-tuning information. All pixel control parts are
246
Figure 11.4 XPAD1 bump bonded with the DELPHI sensor.
chained from pixel to pixel and can be addressed by the main register. The injection
function is useful for testing the pixel functionality by injecting a calibrated charge
in a selected pixel. An analog chopper strobes a DAC-generated voltage level when a
dedicated digital signal is applied. The resulting analog step is converted into charge
by an input capacitor present in the selected preamplifier.
The first results gave a foreseeable dispersion of the offsets at the input of the discriminators. The root mean square (rms) of the distribution of these offsets represents
the global dispersion of the circuit. This value of the standard deviation measurement was more than 3,500 e before correction. But, by applying the local four-bit
threshold correction, the rms value scaled down to 1,100 e. This dispersion, higher
than required, was far from Gaussian, and long tails of untunable pixels were present. Power consumption of the circuit was around 600 mW. The chip was produced
in 4 wafers, and several sets of 10 chips were bump bonded on DELPHI sensors to
form X-ray detector modules. These modules were intensively tested in the D2AM
beamline at ESRF and demonstrated that the signal was reasonably separated from
the noise for energy above 10 keV (Figure11.4).
The XPAD2 chip was manufactured during 2002. The goal of this second version was to correct a small bug that hindered the reading of all the depth of the pixel
counters and to fix the too-large dispersion of the local offsets. For this last point, the
DIDO amplifier architecture was modified, and the local DAC was expanded from
four bits to six bits. But, with the exception of these modifications, all other functions
of the chip were kept to conserve the compatibility with the DAQ system. For this
version, we developed a new sensor geometry that allowed tiling of several XPAD2
modules, avoiding the dead zone present at the edges of the previous sensors. Each
sensor can receive eight chips. This 500-m thick silicon sensor was realized by
247
Figure 11.5 Detector XPAD2, 8 12 cm nude (left) and mounted on the D2AM/ESRF
beamline (right).
Canberra (Belgium) and forms a 4,800-pixel array. Each hybrid was fixed in a small
printed circuit board (PCB), and eight PCBs were abutted to cover a 68 68 mm
total surface. DAQ parallelism allowed readout of the whole detector within 2 ms
thanks to an Altera Nios acquisition card in which all modules were plugged. An
onboard memory allows storing 423 images, each with an exposure time as small
as 10 ms.
Finally, data were transferred to the acquisition personal computer (PC) via a
100-Mb Ethernet link. First images were dedicated to powder diffraction and kinetics measurements recorded on the D2AM/ESRF beamline (Figure11.5). Using the
same detector, we also investigated the small-animal imaging capabilities of this
device. A X-ray CT scanner called the PIXSCAN was built with the XPAD2 detector
to evaluate the advantage of photon-counting detection in in vivo experiments.
This new version corrected the threshold dispersion. The value of the standard
deviation measured was 1,500 e before correction and 50 e after correction.
However, the XPAD2 chip still had a non-Gaussian pixel offset distribution, with
many pixels remaining far from the adjustment value. The real rms value of the dispersion, taking into account these pixels, was 700 e after correction. These untunable pixels are quite annoying for good image quality. Nevertheless, we produced
twelve 4-inch wafers to allow the construction of two large detectors. The last one
has been working since 2004.
11.4.2The XPAD3 Chip

Two main motivations induced moving the XPAD2 to a new version: the obsolescence of the AMS 0.8-m CMOS (the process concluded at the end of 2004) and
the not-so-good qualities of the X-ray imager. The XPAD2 exhibits intrinsic wide
threshold dispersion, which limits its usage to photon energy above 12 keV. Below
this value, it was impossible to set the thresholds for the whole detector, which leads
in turn to a high fraction of useless pixels. This problem was mainly due to the sensitivity of the design to the fluctuations of the CMOS process. So, we decided to restart
the project from scratch with a new technology, the IBM 0.25-m radio-frequency
(RF) CMOS. In fact, we decided to rebuild all the components, like the chip, the
sensor, and the DAQ system, and all test functions made previously but obsolete.
SOLEIL synchrotrons team joined the collaboration at this point (in addition to
the D2AM team of the ESRF) and participated in fixing specifications. The biology
248
requirements were not forgotten, and some of them have been introduced into the
XPAD3 development. But, with the universal chip design a nightmare, we decided
to split the project into two developments. The first one, XPAD3-S (S standing for
silicon), accepts positive charges (holes) at its input and offers an energy range from
0 to 35 keV with a single threshold. The second one, XPAD3-C (C standing for CdTe),
accepts electrons and has an energy range of 60 keV with a windowed energy selection. The C version is very useful for experiments for which good contrast is needed.
These two versions were manufactured in 2006 with twelve 8-inch wafers.12
The choice of the IBM deep-submicron technology was motivated because this
technology was extensively used in the LHC experiments (in which our group
is also involved) because it is radiation tolerant. Sustainability of this technology
was also thought to be sufficient. Global considerations showed that the dispersion
control was the principal issue. Like any hybrid pixel circuits, the XPAD3 chips
combine a global threshold setting whose step is programmable with a local dispersion correction setting. This combination should allow us to adjust all the pixels at
the same threshold value. All subparts were studied from this perspective.
Images from X-ray detectors need to be of very good quality. Spatial resolution is
a quality element, and the pixel size must be kept at a minimum. But, the detection
efficiency and the noise rejection must also be considered. As a general principle,
hybrid pixels tend to minimize noise counts by fixing the threshold above electronics noise pulses. Nevertheless, in such matrices, electronic noise by itself is generally
less important than perturbations on the signal induced by pickup from the substrate
and by power supply distribution. Signal integrity is high-level. For instance, the discriminator uses a current mode approach. This mode allows a good PSRR and low
power consumption and can be implemented in a small area. Moreover, this device
needs a direct current at its input to fix the threshold. Thus, the local offset correction and the global threshold can be made by an easy current mode ladder DAC. To
provide good X-ray image quality, the analog front-end part must be linear as far
as possible for the best threshold accuracy. For the XPAD3-S chip, we required a
threshold precision of 140 e, which gives in turn about 500-eV resolution (silicon).
The preamplifier has been designed to have 100-e rms noise and to have half this
equivalent noise charge (ENC) value for the LSB of the threshold. By minimizing
the electronic noise, we expected to reach threshold value under 5 keV. XPAD3-S
and XPAD3-C differ only by the in-pixel functions, but they have the same footprint.
Additional requirements of the XPAD3-C pixel must fit within the same area. To
fulfill this last requirement, we released the signal accuracy to fix the noise value of
the preamplifier to 130-e rms.
The reliability is also an important consideration. A novel architecture
replaced the serpentine matrix. This previous architecture was retained from its
twin ATLAS project but suffered from bad yield when a dead pixel broke the
serial link. So, we decided to connect independently each of the 80 columns (of
120 rows) contained in the chip to a very fast conveyor that serializes data to the
LVDS output interface. In the worst case, one column may be out of order when
a pixel dies.
From the previous experience, we noticed that the power budget was a big problem, especially because of cooling difficulties. The XPAD detector must be put into
249
Table11.1
XPAD Chip Specifications
Version
XPAD3-S
XPAD3-C
Number of pixels
Pixel size
Full image readout time
9600
9600
130 130 m
2 ms
130 130 m
2 ms
Counting rate
Pixel counter depth
On-the-fly readout
Power
Input polarity
Selectivity mode
Nonlinearity
Electronic noise (rms)
Threshold adjustment resolution
Up to 106 photons/pixel/second
12 bits + 1 bit overflow
Yes
Up to 106 photons/pixel/second
12 bits + 1 bit overflow
Yes
<70 W/pixel
Hole collection
Single threshold
<10% over 35 keV
<140 e
50 e typically
<70 W/pixel
Electron collection
Double threshold
<10% over 60 keV
<140 e
50 e typically
a closed box. To help keep a low power budget, we decided to decrease the nominal
operating voltage recommended by the 0.25-m technology from 2.5 to 2 V and
to monitor the temperature by a thermal sensor designed into the chip. The maximum global power budget for both chips was fixed at the same value as for XPAD2
(600 mW) but with 16 times more pixels. The power budget for each pixel was scaled
down to a maximal value of 70 W.
Finally, a frame rate of 500 images/second was fixed to be able to read out all
12-bit counters twice as fast as required by the maximum counting rate. The ingenious overflow readout option was also kept (see Table11.1).
Both developments of XPAD3 followed the same scheme concerning the global
considerations. A signal integrity approach has been included in every step of the
design process. We included decoupling capacitors in each pixel and at the end of each
column. To keep threshold dispersion at minimum, we used intensive process dispersion simulation to fix the dimension of the local pixel correction setting. Finally, to
reach the readout speed requirement of 2 ms/frame, we designed low-noise readout
electronics. Usually, the operation of the analog part together with significant digital
activity poses problems because of the interference due to the transience of commutation generated by the digital part. The control signals addressed to the numerous
pixels of the matrix have been distributed in time and space to decrease the impact
of digital commutations. This method allows writing and reading operations to and
from a set of registers present in each pixel, while the circuit is kept operational and
able to count incoming events. The reading dead time is null.
A new chip means new pixel definition. From the requirements, the choice of
pixel size fluctuated from 100 to 150 m. We fixed the 130-m square size by a layout implantation arrangement to respect all the specifications. To avoid any digital
cell fault risk, we used standard cells from the IBM library. These cells are bigger
than custom ones, so we have been forced to allow a bigger space for digital functions by dividing the area of pixels in equal parts, 65 m for the front-end analog
250

Guard rings + shielding strategy
Readout
logic
Conguration
registers
12-bit
counter
Current mode
DAC
Bump pad
CSA
&
OTA
Decoupling
capacitors
Threshold;
Current mode
comparator(s)
Figure 11.6 Pixel implantation of XPAD3.
part and 65 m for the digital data-processing part. To secure the analog front-end
blocks and especially the preamplifier, because analog functions do not like digital
activity, we adopted a guard ring strategy. A p+ diffusion guard ring encircles the
digital part, and a p+ surrounded by an n+ diffusion encircles the analog part. Last
but not least, a grounded shield covers the digital part to prevent induction of any
parasitic signals on the sensor. The pixel is composed of a charge preamplifier, an
operational transconductance amplifier (OTA), and a current mode comparator. The
preamplifier is a classical low-noise, folded cascoded one that transforms charge
information into a voltage signal. The charge from the sensor is integrated into a
10-fF feedback capacitor and is reset by a classical MOS transistor polarized in the
ohmic zone for the XPAD3-S and by a feedback OTA for the XPAD3-C. The use of
a feedback OTA was intended to linearize the preamplifier when electrons are collected. The second stage amplifies and converts the signal from voltage to current to
feed the current mode comparator (Figure11.6).
This comparators compact architecture received the current signal from the OTA
and several compensation currents at its unique input node. The particularity of the
XPAD3-C is to select an energy located within a selected window. Two thresholds
with two comparators make it possible. Concerning the digital part, the physical
requirement was to have a photon-counting rate of 2.106 photons/pixel/second. We
also fixed the pixel readout rate to 2 ms. The depth of the counter was determined
251
12-bit counter + Overflow (Read only)

lift: a 12-bit transfer
9-bit register (Read/Write)
Hit
OVP
C 11
From/to pixel i+1
D 11
Pixel i+1
M 11
Cf 8
MUX
OVF/
HIT
D0
C0
Hit (Test)
XOR
M0
Cf 0
From/to pixel i1
Pixel i+1
Local DAC value

(6 bits for XPAD3-S)
(5+2 bits for XPAD3-C)
+
Analogue mask
Digital mask
Injection test enable
Figure 11.7 Pixel architecture of XPAD3.
by 12 bits plus a 13th bit for the overflow mode. To have an on-the-fly readout, we
adopted a local memory made by a set of registers with the same depth as the one of
the counter. This memory, called lift, allows write and read operations into the
pixel (Figure11.7). Finally, a nine-bit configuration register is used to fix the local
threshold adjustment DAC (six bits), to mask the analog and digital functions of the
pixel, and to monitor the injection of the calibration pulse into the pixel. To keep the
same nine-bit register, the two local adjustment DACs needed for XPAD3-C are only
five bits for the lower one and two bits for the upper one, with two mask and injection
bits used instead of three.
The global write-in and readout part placed at the bottom side of the matrix feeds
all columns with data configuration and receives data from the lift of each pixel. This
four-bit depth readout conveyor is able to download a frame in 2 ms by several sets
of four steps of the lift readout clock. Each pixel countershifts data to lift register
while the circuit is counting new events. A latency time of 4 ms is allowed to let the
conveyor work. When the conveyor is full, 80 steps of the master clock flush the first
data out to the DAQ memory through a four-bit LVDS interface. The one-bit depth
write-in conveyor is working at the beginning of the chip calibration, particularly to
fix a homogeneous threshold adjustment for all pixels. To perform a very quick readand-write operation without noise effect, the master clock is delayed spatially from
the bottom left to the top right of the matrix.13
A global configuration memory is inserted to program the local configuration of
pixels and to program a set of control currents provided by the eight-bit DAC. These
bias controls dispatch bias signals to each pixel. They fix not only the preamplifier
and feedback currents but also the global threshold values needed for the discriminator. This global threshold adjustment also needs a last control current that monitors
the minimum step value (LSB). This minimum step value, which can be set as low as
50 e, is the same for the threshold adjustment placed inside the pixel. The matrix is
252
large, so we had to adopt a bias distribution strategy to be sure that all pixels receive
the same bias whatever their position. The bias control delivers currents that are
divided by 80 before supplying each column. At the bottom of each column, a bias
column cell replicates this information by transforming it into voltage to avoid the
effect of voltage drops on distributed currents. Concerning the charge injector, an
eight-bit DAC has been chosen to cover the whole dynamic range necessary to calibrate XPAD3 circuits without physical signal. Finally, we placed a temperature sensor inside XPAD3 to monitor any unexpected variation of the operating conditions.
The first tests demonstrated very good performance, but some problems appeared.
The injector for calibration has a slow rising time, which modifies the signal shape of
the preamplifier. A ballistic deficit exists, making it difficult to calibrate chips without physical signal. This problem is due to the impedance of the track, which drives
injector current signal to the injector capacitors in each pixel. The second problem
is the bad power supply track distribution on the bottom of the matrix, leading to a
left-right effect in bias values, harmful especially to the bias column cell. Bias signals are defective for the whole matrix in the left-right direction, and the calibration
process had to be reexamined to be sure to have all pixels in a working state. For
that, we increased the minimum step of the local adjustment value. The XPAD3-S
chips have less accuracy, but all pixels can be tuned. We explored these problems
in two ways. A focus ion beam (FIB) surgery cut and replaced power supply tracks
on the right position and made a fine analysis of the defect injector track. Because
the power budget of the XPAD3-C is higher, one can guess that biasing problems
will have more serious consequences. In fact, the chip cannot be operated properly.
Finally, the command gate that enables pixel counting has a rise time incompatible
with experiments that require fast starting and stopping.
The first characterization results gave nonlinearity lower than 4% for the
XPAD3-S circuit and lower than 2.5% for the XPAD3-C circuit. The XPAD3-S chip
has a global electronic noise of 127 e rms and a threshold adjustment dispersion of
57 e with a threshold adjustment resolution of about 150 e. The power budget was
under the expected value, with 40 W per pixel for XPAD3-S. The on-the-fly readout
was working with the expected readout time of 2 ms/frame, and the counting rate
reached a value of 106 photons/pixel/second.14
Seven XPAD3-S chips have been bump bonded to a 500-m silicon sensor to
form a 7.5 1 cm module of 67,200 pixels. Each module contains its own DAQ made
by an ATLERA FPGA and 4Go SRAM. By tiling eight modules, the large-surface
pixel detector is operational, and ESRF and SOLEIL started working with it. A
double-chip XPAD3-S detector bump bonded with a 700-m p-type CdTe by AJAT
(Finland) is also operational and has given good results for several energies at
ESRF.15
This version of XPAD promises good experimental results, particularly for
high-contrast biomedical applications. Despite the problems explained, we found a
maximum of three defective pixels for a module, and pixels have uniform behavior
without suffering any cross-talk signals from neighbors.
A new version of the XPAD3 chip is under development, not only to remove the
problems mentioned but also to add some improvements, like widening of the linear
253
range to 60 keV and adding a six-bit LVDS readout bus instead of the four-bit one.
An additional objective was to make the chip work with both polarities of the input
signal. This requirements was resolved by designing two different chips: XPAD3-2S
and XPAD3-2C. Last but not least, a radiation hardness design was adopted to
improve reliability against ionization radiation.
11.5Applications of X-Ray Imagers

11.5.1Crystallography Application in Material Sciences
An important challenge in material science is to reach better knowledge of a material at the nanometer and atomic scales. For this purpose, two methods are commonly used. The first one is based on electron microscopy, which gives images of
selected zones, showing, for example, defects in the atomic arrangement. The second
uses the X-ray diffusion properties and leads to interference images associated with
mean properties of the sample. Data obtained using this method are very sensitive to
atomic distance and strain in materials.
Since the X-ray wavelength is close to the distance between atoms in solids, their
diffusion interferes constructively in crystalline samples, which gives rise to Bragg
peaks. Between these peaks, the X-ray diffusion reveals the imperfection of the
crystalline network. These properties have been used since the beginning of the
20th century to reveal the structure of materials. Moreover, it has been possible to
crystallize proteins and then to access their atomic structure. The availability of new
X-ray sources produced by third-generation synchrotrons like ESRF or SOLEIL has
improved the crystallographic method applications, which are now able to obtain
information coming from a quantum dot made of few atoms at the surface of semiconductors. However, such experiments are very demanding, and their weak point
is often detector problems, which frequently appear in crossed slits and scintillation
counters or fiberoptic coupled CCDs.
The XPAD detector was built to overcome the limitations of such detectors; its
design was based on the typical needs in material science. Indeed, a high-quality
crystal, like those used in electronics processing, can reflect a significant part of the
incoming X-ray intensity within the Bragg peaks, with the diffuse level in the best
case 106 times smaller, but often 109 times smaller. As a first example of the need for
a detector with a wide dynamic range, we look at new materials called quasi-crystals.
The first ones were obtained by quenching some liquid metal alloys in the 1980s; in
2007, the first real structure of one of them was published.16 They differ from classic
crystal; their symmetry allows rotation axis (about fivefold) that are not compatible
with any lattice translation. This led to strange properties and infinities of diffraction
peaks. As an example, Figure11.8 shows the diffraction peaks that appear when the
quasi-crystal AlPdMn is oriented near the [0 3.1495 0] direction. Around the most
intense peak, for which the maximum intensity was not recorded, a few weaker peaks
appear, but by looking more carefully, one can distinguish a few other families of
weaker peaks. The intensity of these peaks is directly connected to the atomic structure. Of these very sharp peaks, Figure11.8 is characterized by a diffuse scattering
254
300
??
250
??
200
??
150
??
100
??
50
??
0
100
1000
10,000
Intensity
1.E5
1.E6
Figure 11.8 Diffusion around [0 3.1495 0] of AlPdMn.
6,000
6,000
XPAD
Intensity
2,000
2,000
1,000
1,000
600
600
400
CCD
4,000
Intensity
4,000
50
100
150
Pixels
200
250
400
100
200
300
Pixels
400
500
Figure 11.9 Cross section, along black line shown in Figure 11.8, for the measure with the
XPAD3 detector and with a CCD camera.
almost centered on the peaks. The intensity of this diffuse scattering is at least three
orders weaker than the peak one, but it reveals the intrinsic disorder of such a structure, such as the phason associated with atoms jumping between equilibrium sites.
To provide evidence of the advantage of using a direct counter such as XPAD3-S,
we compared the recorded intensity profile along the black line in Figure11.8 to the
same profile extracted from CCD data. The comparison of both slices (Figure11.9)
showed that the dynamic of the pixel detector allows better characterization of this
diffuse scattering (the signal-to-noise ratio of the CCD pattern was low because the
accumulation time was reduced so the detector was not saturated).
255
Another example of the interest in pixel detectors is given by powders used as

catalysts. Before the optimization of the catalyse processing, some questions have
to be answered: What happens in the structure during the reaction? Where are the
incoming molecules located? Answering such queries often requires locating and
differentiating minor dopants in the zeolite phase, which can be achieved using
the anomalous effect: the small modification of diffracting power of the atoms near
their X-ray absorption edges. Such experiments have been carried out on zeolite
at the D2AM/ESRF beamline; the collection of each data set has required numerous hours using the classical setup. A new data collection has been attempted using
the XPAD2 detector (330-m pixel size) on the same beamline (Figure11.10). The
use of XPAD allowed reduction of collection time to less than 25 minutes for each
data set with similar quality,17 which is a decrease by more than a factor of 12 of the
previous collecting time, but statistical analysis of the data showed that the collection
time may have been reduced by more than 100 without damaging the needed quality for performing such structural refinement. An advantage of using a 2D detector
rather than a one-dimensional (1D) one is that it allows measuring a wider angular
aperture without decreasing the resolution since the collected diffraction pattern has
a nonnegligible curvature at small angles.
Small-angle scattering experiments also require 2D detectors to integrate all the
diffuse intensity and eventually to detect some anisotropy of the sample. In soft
condensed matter, most of the properties are not directly concerned by the atomic
arrangement but by clusters at an intermediate scale between nanometers and
micrometers. A part of this scale can be reached by light scattering, but for absorbing
samples or at lower scale, the diffusion of X-rays (or neutrons) is performed: X-rays
are diffused at very low angles around the direct beam; the shape of this diffusion allows characterization of the clusters and their organization. The experimental
drawback is that numerous orders of magnitude have to be measured on a signal that
vanishes with q2 for the wider ones to q4 for quicker ones. If we make a comparison with the use of a CCD camera, the sharp point-spread function of the XPAD3
detector, which has no Lorentzian tail, appears to be an advantage. The measured
signal reaches the background more quickly and allows consequent characterization
of weaker scatterers, such as biological solutions, as indicated in Figure11.11.
Finally, pixel detectors are well suited for demanding grazing incidence synchrotron experiments. A surface diffraction experiment was attempted at SOLEIL
(DIFABS beamline) using an energy of 10 keV and a single-chip XPAD3-S Si detector. Surface diffraction has strong experimental requirements: measuring the contribution of a few atomic layers at the surface of a perfect crystal, which means detecting
a very weak signal close to intensities that are more than 109 times stronger. In this
experiment, the reciprocal space intensity map around the (111) diffraction peak of
the substrate was reconstructed (Figure11.12). This allows characterization of the
distortions of a few GaInAs layers epitaxially grown in a GaAs single crystal; these
strains modify the electronic properties. It must be noted that such experiments realized with XPAD3 cannot be performed with CCDs. They are commonly done with
crossed slits and scintillation counters scanning point by point, with a corresponding
waste of time and photons.
256
200
200
400
600
800
1,000
1,200
200
100
100
200
10
1,400
100
1,600
1,000
1,800
10,000
2,000
1.E5
2,200
1.E6
2,400
1.E7
2,600
200
100
0
200
100
2,800
3,000
3,200
3,400
3,600
0
4,000
200
3,800
100
100
0
0
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Intensity
1.4
1.8
2.0
*10E5
XPAD values
Fitted values
20,000
Counts/s
1.6
15,000
10,000
5,000
20
22
24
26
Diraction Angle (20)
Figure 11.10 Debye Scherrer reconstructed film from XPAD2 images. Rietveld fit of
XPAD2 experimental data of the CaSrX-zeolite at 16.097 keV; this zoom allows comparison of the experimental data (open circle) to the fitted data (line). The analysis of the results
showed that if the data processing had been fully enhanced, the collection time could be
reduced by more than a factor of 100, opening the way to real in situ measurement with gas
flowing though the reactor.
257
120
50
100
150
200
250
300
120
100
100
80
80
60
60
40
40
20
20
50
10000
100
1.E5
150
200
1.E6
Intensity
250
1.E7
300
1.E8
Distribution radiale du Lupolen
Intensite
XPAD (100s)
CCD (10s)
1000
10 adu/pixel
100
10
le5 photons/pixel
0
0.5
1
1.5
2 theta (14 keV)
Figure 11.11 Two-dimensional small-angle X-ray scattering from Lupolen as measured

with the XPAD detector. Radial distribution extracted from the 2D pattern for XPAD3 and
the CCD.
11.5.2Biomedical Application: Computer Tomography

At the beginning of the development of the XPAD chip, we focused on the benefits
of the X-ray hybrid pixels to build a micro-CT scanner dedicated to the study of
small animals. By counting and processing each single photon, instead of performing integration of a flux, hybrid pixel detectors for X-ray medical imaging ensure
catching better images with optimized contrast resolution compared to CCD-based
detectors. The electronic noise below the threshold low limit, the linear response,
and the infinite counting dynamic are advantages that make photon-counting detectors extremely attractive for medical applications.
The imXgam group of the CPPM laboratory, in collaboration with the IBDML,
is interested in the development and the production of the CT scanner and in the
optimization and validation of the scanner for the study of noninvasive visualization
258

Log10(l)
1.14
0.100
0.892
1.68
2.47
3.27
4.06
4.85
5.64
0.866
1.13
0.867
0.823
Figure 11.12 Reciprocal space map around the (111) peak of the GaAs substrate.
X-ray source
XPAD2
Mouse
Figure 11.13 Small-animal CT scanner PIXSCAN with XPAD2 chip.
of developmental abnormalities of transgenic mice. The PIXSCAN CT scanner is an

all-in-one micro-CT scanner that includes an X-ray source, a large surface XPAD
detector, and a rotating gantry between them. With a very large dynamic of photon
counting at very low detector noise, the first prototype of PIXSCAN with the XPAD2
detector has enabled study of 3D tomographic reconstruction thanks to the step-bystep rotating holder (Figure11.13). The body region imaged using an X-ray source
with a wide conic beam aperture (60 keV, 30 W, 0.5 mA) allows half of the mouse
to be imaged at once. The PIXSCAN system is installed inside a shielded box, and
most mechanical actions are remotely controlled.
The first use of the PIXSCAN scanner for animal pathology study was the detection of the nonprogression of lung metastasis in a murine model of cancer with non
259
invasive monitors.18 There were 360 projection images of mice acquired every degree
in 15 min. With a pixel size of 330-m square, a geometrical zoom factor of 2, and a
total of 36,800 pixel area, we reconstructed 200 200 200 voxels of 165 m. Image
reconstructions were performed using FDK algorithms. These images showed quality similar to the CT scanners based on charge integration detectors, but to underline
lung tumor masses we were compelled to increase the dose, upper to the low dose
rate required. By using smaller pixels, faster readout data, and better contrast detection, it will be possible to limit the dose rate and to improve spatial resolution to
detect micrometric tumors.
The next version of PIXSCAN with a XPAD3-S detector is under development
and will replace the XPAD2 to have a bigger detection surface 12 7.5 cm with eight
modules of seven chips each to make a full-body scan. With 130-m pixel square,
precise threshold adjustment, and very fast frame readout, we expect to enhance the
contrast of regions using iodine contrast agent. The XPAD3-C chip, with its energy
windowing function, would be a good candidate to fit the iodine photoelectric absorption jump, but because of the calibration defect, the XPAD3-S chip will replace it by
using fast threshold adjustment steps. We expect to get both faster images around the
jump and to enhance the contrast by offline image subtraction. In parallel, we will
build a PIXSCAN dedicated to higher energies by assembling a CdTe sensor with
XPAD3-S. Owing to the lack of large-size CdTe wafers, only modules of two chips
are expected to form a final 6 6 cm detector with 3 4 module arrangement.
Another exciting project is to combine the CT scanner with a PET scanner. While
an X-ray CT scanner allows imaging the mass density of living tissues using an
external X-ray tube, a PET scanner images gamma rays directly emitted from the
tissues. The PET/CET imaging system for small animals in a common gantry could
avoid the juxtaposition of both modalities, which is quite difficult due to animal
position variations between the two exams. Both detection systems merged together
with an X-ray tube in a fully integrated PET/CT device make possible the acquisition of simultaneous emission and transmission scans for mice. The imXgam group
obtained the ClearPET scanner from the Crystal Clear collaboration (another international group originally named RD18 created by CERN). The ClearPET/XPAD
scanner will use the actual ClearPET scanner in addition to the next version of the
PIXCAN with the XPAD3-S detector. It will consist of an X-ray tube from RTW
with 50-m focal spot size and the 118 76 mm XPAD3 X-ray imager with 130-m
square pixels positioned within the PET detector ring. The CT scanner modality has
an axial field of view (FOV) of 59 mm and a transverse FOV of 38 mm with a 2
magnification. The X-ray cone beam passes through the shared FOV without irradiating the PET detectors directly.
11.6From 2D to 3D Approach: A New Hybrid Trend

A novel approach of micropackaging called 3D Integration Technology (3D-IT) was
developed to extend the reach of the Moore law of microelectronics. The idea relies
on assembly of several devices to transform planar geometry into 3D devices instead
of shrinking 2D technology. This technique is already used in the integrated circuit (IC) industry to stack memory chips on top of a central processing unit (CPU).
260
Imager arrays like hybrid pixel detectors are good candidates to experiment on using
this new 3D approach, which would enable new capabilities at the chip, device, or
wafer level.
There are several ways to create stacked structures, and for our application, it is
necessary to retain the possibility of assembling small-size pixels, the ones on top of
the others. Thus, connections passing through the substrate make it possible to interconnect the functions between themselves, such as the analogical detection part on the
digital data-processing part. With the existing sensor already on top, that leads to three
layers at least. In addition to compactness and heterogeneous device integration, we can
obtain interconnection impedance reduction and thus an increase of signal bandwidth.
Two approaches exist via first and via last. In the via-first approach, the throughsilicon-vias (TSVs) are made in a foundry before or after the front-end-of-the-line
(FEOL, which are the first steps of IC fabrication during which the individual devices
are patterned in the semiconductor prior to metal layer deposition). It allows smaller
vias that are closer to the devices and in great number. Alternatively, in the via-last
approach, via are added in postprocessing, after the back-end-of-the-line (BEOL,
which is the step at which individual devices are interconnected with wiring on the
wafer). This makes it possible to reuse the old circuits not envisaged at the origin
for 3D integration. On the other hand, vias can be drilled only in the free area of the
circuit, they are larger, and the overall process results are more expensive.
Another aspect is the way of positioning the tiers between themselves. Stacking
is made on either the wafer level (wafer to wafer) or the chip level (chip to wafer). On
the wafer-to-wafer level, we obtain better precision of contact alignment to the detriment of the yield because all the chips present on the wafer are not working. In the
chip-to-wafer approach, handling is more tedious as a result of the number of dies.
But only good dies are deposited, so the yield is better.
The last aspect of the technology consists of wafer thinning, which is necessary
to contact the vias on both sides of the wafers. The aspect ratio of a TSV is no more
than 5 to 10, which means that a 2-m via will have a maximal total depth of 10 to
20 m. In via first, thin vias in turn give wafers about 10 m thick. In via last,
because the vias are larger, they will have a thickness of about 50 m.
References

1. E. H. M. Heijne et al., Development of silicon micropattern pixel detectors, Nucl. Instr.

Methods A, 348, 399408, 1994.
2. P. Delpierre et al., Large scale pixel detectors for DELPHI at LEP200 and ATLAS at
LHC, Nucl. Instr. Methods A, 342, 233239, 1994.
3. K. H. Becks et al., Commissioning of the DELPHI pixel detector, Nucl. Instr. Methods A,
418, 1521, 1998.
4. G. Aad et al., The ATLAS Experiment at the CERN Large Hadron Collider, J. Instrum.,
3, Article S08003, 2008.
5. G. Aad, ATLAS pixel detector electronics and sensors, J. Instrum., 3, Article P07007,
2008.
6. P. Delpierre, Pixels detectors and silicon x-rays detectors, J. Phys., IV 4, C9-11C9-18,
1994.

261
7. P. Delpierre et al., X-ray pixel detector for crystallography, IEEE Trans. Nucl. Sci.,
48(4), 987991, 2001.
8. J. C. Clemens et al., Large surface X-ray pixel detector, IEEE Trans. Nucl. Sci., 49(4),
17091711, 2002.
9. P. Delpierre et al., PIXSCAN: Pixel detector CT-scanner for small animal imaging,
Nucl. Instr. Methods A, 571, 425428, 2007.
10. E. H. M. Heijne, Semiconductor micropattern pixel detectors: a review of the beginnings, Nucl. Instr. Methods A, 465, 126, 2001.
11. L. Blanquart et al., XPAD, a new read-out pixel chip for X-ray counting, IEEE Nucl. Sci.
Sympo. Conf. 1520, Oct., 9297, 2000.
12. P. Pangaud et al., XPAD3: a new photon counting whip for X-ray CT-scanner, Nucl.
Instr. Methods A, 571, 321324, 2007.
13. Matrice de Cellules Electroniques Patents INPI: No. FR2007/054974 and No.
FR2008/0855480.
14. P. Pangaud et al., XPAD3: first results of XPAD3, a new photon counting chip for X-ray
CT-scanner with energy discrimination, IEEE Nucl. Sci. Sympo. Conf., Nov., 1418,
2007.
15. S. Basolo et al., A 20kpixels CdTe photon-counting imager using XPAD chip, Nucl.
Instr. Methods A, 589, 268274, 2008.
16. H. Takakura et al., Atomic structure of the binary icosahedral Yb-Cd quasicrystal, Nat.
Mat., 6, 5863, 2007.
17. S. Basolo et al., Application of hybrid pixel detector to powder diffraction, J. Sync.
Radiat., 14, 151157, 2007.
18. F. Debarbieux et al., Repeated imaging of lung cancer development using PIXSCAN,
a low dos micro-CT scanner based on XPAD hybrid pixel detectors, IEEE Trans. Nucl.
Sci., in press.
Electron
12 Ultrafast
Beam Tomography
Uwe Hampel
Forschungszentrum Dresden-Rossendorf
Dresden, Germany
Contents
12.1 Motivation and State of the Art..................................................................... 263
12.2 ROFEX: An Ultrafast Electron Beam CT Scanner.......................................266
12.3 Scanning Procedure and Data Processing..................................................... 274
12.4 Application Examples.................................................................................... 275
12.5 Conclusions and Perspectives........................................................................ 278
Acknowledgment.................................................................................................... 279
References............................................................................................................... 279
12.1Motivation and State of the Art

Since its invention in the 1970s, computed tomography (CT) has become an indispensable diagnostic tool in medicine. CT scanning using X-rays was the first clinical tomographic imaging technique and was later complemented by tomographic
imaging modalities based on other physical principles, such as magnetic resonance
imaging (MRI), positron emission tomography (PET), and single-photon emission
computed tomography (SPECT). Today, medical radiology is still the primary application field for CT, but this imaging technology has diverted also into other fields of
science and engineering, for instance, nondestructive testing, flow measurement, and
small-animal imaging.
The basic principle of tomography consists of the reconstruction of a crosssectional map of a physical quantity from a set of integral measurements. In X-ray
CT, an attenuation map is recovered from a set of radiographic projections, that is,
a number of attenuation line integrals obtained from different angular views with a
rotating compound of an X-ray tube and an X-ray detector. Todays helical CT scanners use a high-power X-ray tube with a power dissipation of up to 100 kW and a
focal spot roughly between 0.5 and 1.5 mm together with a multielement scintillation
X-ray detector arc. Such machines achieve around 0.5-mm spatial image resolution
and scan rates of up to two frames per second. For most medical questions, this is
a sufficient acquisition speed. However, faster scanning is constantly of interest in
medicine and other fields. In X-ray CT scanning, speed is clearly limited by the
inertia of the rotating system components, while X-ray power and detector response
would in principle allow for faster acquisition. The scanning speed of other medical
263
264
tomography modalities is generally slower, in the range of seconds to some minutes.

In SPECT and PET, the acquisition rate is typically limited by admissible activity of
the radiotracer, and in MRI it is limited by the signal-to-noise ratio and the magnetization time constants of the tissue.
A closer look into the existing tomographic modalities shows that X-ray techniques are the most obvious candidates for fast imaging. X-ray sources can be made
sufficiently powerful to yield good photon statistics at small detectors in short time
intervals and at reasonable distance from the source. One application that has in
the past fostered the development of faster CT scanners is cardiac diagnostics. The
human heart, with a regular beating frequency of 50 to 100 beats per minute at rest,
cannot be caught in motion with classical helical scanning machines. This problem
was addressed by the electron beam cardio CT introduced in the 1980s.1 Unlike CT
scanners of the third generation with their rotating sourcedetector, compound cardio CT uses a scanned electron beam, which is produced by an electron gun, shaped
and deflected by electromagnetic beam optics, and scanned across a large metal
target that surrounds the patient. In this concept, the detector is fixed and aligned
coplanar with the target, covering a sufficient angular scanning range. Cardio CT
achieves acquisition rates of up to 50 frames per second. Faster scanning is hampered by the large sourcedetector distance of almost 1 m and physical limits in the
detector response time.
There are fields other than medicine in which fast tomography is of great interest but speed demands also are more stringent. Small-animal imaging, for instance,
has become an essential tool in preclinical diagnostics today. Many pharmaceutical
and scientific laboratories use small-animal PET, MRI, and CT scanners for drug
development, tumor research, and other targets. As with their medical counterparts,
todays small-animal scanners are not able to visualize dynamic and nonperiodic life
functions and are therefore of limited use for live animal imaging. Electron beam
tomography as a fast technique has so far not found application in this field, although
there were proposals for that.2 Compared to humans, small animals require much
faster scanning to visualize structures in the thorax region. Thus, the heartbeat of
a mouse is typically in the range of 500 to 650 beats per minute, and breathing is
about 200 breaths per minute at rest. These numbers indicate that a CT scanner for
live small animals should achieve frame rates of about 500 frames per second to capture the beating heart. Therefore, the radiation dose must be kept low enough, while
spatial resolution should be in the range of 100 m. A way to meet these conflicting
requirements would be the introduction of an electron beamcone beam CT with twodimensional detectors. Therefore, small-animal cardio CT most probably will not be
just a downsizing of its human antetype but instead require novel scanner designs.
Multiphase flow is another important potential application of fast tomography.
Multiphase flows are to be found in many industrial areas, such as in chemical and
nuclear facilities, hydrodynamic machinery, and mineral oil processing. Prominent
representatives are water-steam mixtures in cooling circuits or liquid-gas-particle
flows in chemical reactors. The nature of multiphase flow is intricate. The flow constituents, with their different physical properties, produce a merely unpredictable flow
mechanical behavior that may include complex heat and mass transfer mechanisms.
Ultrafast Electron Beam Tomography
265
Understanding, modeling, and simulation of multiphase flow phenomena present a

challenging problem for scientists and engineers. Many investigations of multiphase
flows are being conducted at small-scale experimental facilities, where the availability of adequate measurement technology is a key issue. Imaging techniques that
can noninvasively disclose the distribution of certain physical quantities of flowing
mixtures, such as gas holdup, interfacial area density, species concentration, temperature, phase velocity, and pressure, are highly desired. A real breakthrough in flow
imaging was achieved with the introduction of high-speed video imaging. Modern
high-speed cameras enable flow imaging at frame rates that may soon approach 1 million frames per second. However, since multiphase flows are commonly opaque, this
imaging technology is in most practical cases not applicable.
Since tomography techniques are commonly considered the only applicable imaging technology for multiphase flows, most of the existing medical imaging modalities have been in some way made available for this purpose in the past. The longest
history belongs to the application of X-ray and gamma-ray tomography for imaging
of stationary flows or time-averaged phase distribution measurement. Application
examples range from the measurement of holdup distributions in pipes or reactor
vessels3,4 to the imaging of liquid distributions in periodically working machines.5
The use of MRI for multiphase flow imaging was first reported by Mantle et al.6 Their
machine was specially designed for flow measurement and can achieve frame rates
of up to 8 Hz and a spatial resolution below 1 mm. Special features of this technique
are chemical species imaging via chemical magnetic resonance (MR) shift and flow
velocity measurement via spin transport. However, since the use of magnetic materials is forbidden, MRI is not universally applicable to flow measurement problems.
Although PET itself is not a candidate for fast tomography, a special technique
called positron emission particle tracking (PEPT) has been developed and is frequently used in flow studies.7 PEPT is able to track the motion of positron-emitting
particles in mixtures and flows and has real high-speed capability. After activation
in a cyclotron, the particles are brought into the system of investigation. While in
motion, they frequently emit positrons, which produce annihilation radiation when
they encounter electrons. This radiation is registered by a PET camera. Because only
a few registered annihilation events are needed to uniquely recover position information, this method is very fast and can track particle trajectories with a time resolution
down to 1 ms.
Electrical tomography has become especially popular for flow imaging,8,9 and for
a long time it was the only available fast imaging technique for multiphase flows.
Electrical tomography is based on the measurement of integral electrical parameters
of an object, such as electrical resistance, capacitance, or impedance, from a set of
electrodes located at the surface of that object. Electrical tomography is capable
of high-speed imaging in the range of 1,000 images per second and even higher.
However, since electrical fields cannot be constrained to a straight and narrow path
like X-rays, electrical tomography by virtue of the underlying physics cannot produce images of high resolution and contrast.
Tomographic flow scanners based on high-energy radiation, which achieve sufficiently high frame rates of at least 100 frames per second, are so far rare. Johansen
266
et al. introduced an isotopic scanner with five stationary americium 241 sources that
scans at more than 100 frames per second.10 The restriction to five projections results
in an inverse problem with limited data, which compromises spatial resolution and
requires an algebraic reconstruction technique. Fast scanner concepts based on X-ray
sources can be generally divided into those with multiple gated X-ray sources and
those with use of an electron beam. Tomographic systems based on multiple gated
X-ray tubes have been introduced, for instance, by Morton et al.,11 Hori et al.,12 and
Misawa et al.13 Morton et al. proposed a system with 156 X-ray sources, grouped in
12 tubes with 13 switchable focal spots per tube. The tubes were designed for up to
50-kV tube voltage. Their scanner can achieve a frame rate of up to 50 frames per
second. Misawa developed a system with 18 pulsed X-ray tubes having a maximum
160-kV tube voltage and 10-mA tube current and reaching 263 frames per second.
Horis system is made of 60 X-ray tubes with 120-kV tube voltage and maximum
10-mA tube current and achieves 2,000 frames per second. So far, the presented
setups still have rather low X-ray power, which may be attributed to their prototype
design. A critical factor of the concept may be reliability and availability for longterm use, since X-ray sources usually have a limited lifetime.
Our group has proposed and implemented an ultrafast electron beam CT
approach,5 which is a straightforward adaption of the known medical electron beam
CT. We started with simple limited-angle-type tomography to demonstrate the applicability of the method. With the experimental setup, a frame rate of up to 10 kHz was
achieved.14 This is an order of magnitude above that of existing systems. Meanwhile,
the technology has been implemented in a dedicated flow tomography scanner with
full-angle tomography, which can be operated at up to a 7-kHz frame rate.15 This
system is described further next.
12.2ROFEX: An Ultrafast Electron Beam CT Scanner

Essentially, an electron beam X-ray tomographic scanner consists of an electron
gun, beam optics, a metal target, a fast X-ray detector with a data acquisition unit,
and supplemental components, such as high-voltage generator, vacuum system, coil
current drivers, beam-monitoring components, and a computer for data acquisition,
device control, and image reconstruction. The principle of electron beam tomography is shown in Figure12.1, here for a flow measurement application. The electron
gun produces an electron beam by electrostatic acceleration of electrons emitted
from a heated tungsten cathode. The electron beam passes an electromagnetic lens
system for beam focusing and centering and further an electromagnetic deflection
unit for two-dimensional beam scanning. The electron beam is focused onto a circular metal target that is at some distance from the gun. The distance is determined by
achievable deflection angle of the beam deflection system, the required scan diameter (target size), and some other geometric constraints. When impinging on the
target, the electrons are stopped, and their energy is converted to heat and radiation.
Less than 1% of radiation is X-rays. Most of the X-rays are bremsstrahlung, and only
a smaller portion is characteristic radiation from the target metal. During scanning,
the electromagnetic beam deflection system sweeps the focal spot rapidly across the
target surface on a circular path. This way, a moving X-ray source is created.
267
Electron gun
Focusing coils
Deflection coils
Electron beam
Target and
X-ray detector
Test section
Figure 12.1 Principle of electron beam tomography.
Thermal load on the target is a common problem that has to be looked after with
X-raygenerating systems. Similar to rotating anode tubes, the thermal power is
distributed on a ring when the beam is circularly scanned across the target. Figure12.2
shows a typical temperature distribution over time for a single point on the focal
spot path.
The prompt temperature rise in the focal spot is given by

8P
2 f R c d3
(12.1)
for a beam with a circular profile, focal spot diameter d, scanning frequency f, target
radius R, target heat conductivity , target material density , target heat capacity c,
and beam power P.16 With a 1-kHz scan with a beam of 10-kW power and 1-mm focal
spot diameter, the prompt temperature rise would be theoretically about 820 K and
somewhat less in reality since the actual focal spot is larger due to target inclination.
With respect to melting and sublimation, the tungsten target can well withstand this
temperature rise. However, the temperature gradients produce considerable thermal
stress. To account for this effect, the target is commonly made of several segments
that are separately mounted on a copper support. The ring temperature follows a
relationship given by
268
Temperature
critical
ring
Time
Figure 12.2 Temperature as a function of time in a single focal path point.
ring = k
d (n + 1)

2 R
(12.2)
where n is the number of circulations, and k is a coefficient that accounts for the
targets heat storage capacity and heat removal. Active cooling by a water-cooling
circuit as well as fast scanning helps to prevent damage to the target.
Radiation detection is performed by an X-ray detector that consists of multiple
detector elements arranged contiguously in a circle inside the target opening. The
detector needs to be very fast. During one circulation of the electron beam, a multiplicity of radiographic projections must be acquired, usually between 100 and 500,
depending on the required spatial resolution. For very fast scanning with frame
rates of 1,000 frames per second or more, this demands either very fast scintillation
converters with low afterglow or fast room temperature semiconductor detectors.
Further, the detector ring needs to be aligned in a coplanar way to the focal spot
path. This is not exactly possible, since the detector would then obstruct the X-rays.
Therefore, the detector is placed at a small axial distance relative to the focal spot
path. Practically, this leads to some axial uncertainty in image reconstruction since
the X-rays are no longer running in a plane. Usually, this error has no severe influence on image quality.
Inside the scanner, a vacuum has to be maintained. The gas pressure requirements
are moderate, typically 0.1 mPa in the cathode chamber and 1 mPa in the beam tube
and the scanner head. Although the moderate gas pressure implicates some electron loss, the presence of positively charged ions created by the beam itself helps to
counteract the space-charge effect in the beam. Gas pressure increases in the long
term as a result of vacuum leaks and during operation by sublimation processes at
the cathode, target, and components hit by scattered electrons. Therefore, frequent
evacuation is necessary. Usually, a turbomolecular pump with some prepump is sufficient equipment to secure stable gas pressure during operation.
269
A notable difference between an electron beam tomograph and an X-ray tube is

the unipolar operation of the first. In the electron beam tomograph, the cathode is on
full high-voltage potential up to 150 kV, while the anode and the parts behind the
anode hole (beam tube and target) are grounded. In X-ray tubes, cathode and anode
are charged with half of the absolute potential but at different polarity. While giving
the same acceleration voltage, this reduces insulation efforts enormously and makes
high-power, high-voltage supplies much more economic. Forming and steering an
electron beam, however, requires a field-free space behind the anode to avoid uncontrolled beam deflection. Further, only this design makes the insulation issue of the
scanner components practicable. As a drawback, higher insulation effort and more
expensive high-voltage supply devices are required.
The electron beam tomography scanner Rossendorf Fast Electron Beam X-ray
CT (ROFEX), built and commissioned at Forschungszentrum Dresden-Rossendorf
in 2007, has primarily been designed for multiphase flow measurement applications.
A computer-aided design (CAD) drawing of the scanner along with a photograph
are shown in Figures12.3 and 12.4, respectively, and a component view is shown in
Figure12.5. The scanner differs from the medical electron beam tomograph in some
notable ways. As stated, multiphase flow imaging requires frame rates of at least
1,000 frames per second. Therefore, the sourcedetector distance is kept relatively
small to achieve high photon flux at the detector. Monte Carlo simulations were run
X-ray
detector
Target
Beam tube
Electron
beam
generator
Figure 12.3 CAD drawing of the ROFEX scanner operated at Forschungszentrum

Dresden-Rossendorf.
270
Figure 12.4 Photograph of the ROFEX scanner mounted at a vertical pipe of a thermal
hydraulic test facility.
to simulate X-ray propagation within a scanner model. As a result, it was found that
sourcedetector distance should not exceed 200 mm to secure a photon flux of 104
photons per microsecond and square millimeter with no material in the X-ray fan.
Therefore, size and arrangement of target and detector have been chosen appropriately. Another different feature is the use of Cadmium-Zinc-Telluride (CZT) room
temperature semiconductor detectors instead of scintillation detectors.
The electron beam gun is a triode system operated with a small-size tungsten
bolt cathode that is indirectly heated by electron bombardment. A Wehnelt electrode
and anode form a Rogowski transducer, which guarantees maximal independence
of focal spot size on the electron beam current. The Wehnelt electrode is used to
control the beam current by applying a maximum reverse voltage of 3 kV relative
to cathode potential. Gas pressure in the cathode chamber is kept at about 0.1 mPa by
a scroll prepump and a turbomolecular main pump operated in series. After passage
of the anode, the accelerated electrons traverse the beam-forming section, which
contains the centering, focusing, and deflection coils. The quadrupole centering coil
is used to align the beam exactly with the optical axis of the focusing and deflection
271

Measurement
and control PC
HV
generator
Vacuum
system
Beam
focusing
and
deflection
system
Detector
electronics
Figure 12.5 Components of the ROFEX scanner.
system. This is necessary to compensate the effects of geometric misalignments in

the electron optics, which would otherwise cause beam astigmatism and aberration.
The focusing lens provides beam concentration into a 1-mm focal spot at the target
in 1-m distance. The maximum beam deflection angle is 17.5 in both x and y directions. The cutoff frequency of the deflection system is 10 kHz.
The far end of the electron beam column has been designed in a horseshoe shape.
The opening in the scanner head enables accommodation of the scanner to pipes,
vessels, and other extended objects. A metal target with a 240 opening angle and
256-mm outer diameter is mounted inside the scanner head. The optical axis of
the electron beam system runs at an angle of 30 relative to the axis of the object.
The beam cone is therefore an oblique cone, and circular scanning across the target
requires an elliptical beam deflection pattern. The target is a compound of a massive copper body with tungsten alloy plates on the top surface. Such a combination
is typical for heavy-duty X-ray targets. The tungsten has a high melting point and
enables the highest dissipation of beam energy, whereas the copper has superior
heat conduction and allows efficient heat removal. Active cooling is provided with a
water-cooling circuit and channels that run directly underneath the tungsten surface
of the target. The target is surrounded by a set of electron catcher sheets for beam
monitoring and control. The X-rays produced on the target enter the scanning section via a thin aluminum window, which provides appropriate beam hardening. The
scanning geometry is schematically shown in Figure12.6.
The X-ray detector is arranged in a circular way along the inner wall of the horseshoe enclosure. Its diameter is 132 mm. This gives a free space of about 120-mm
diameter for accommodation of objects. The detector ring has a modular design and
can easily be separated into parts in order to dismantle and mount the scanner head.
272

Electron
beam
Target
X-ray
Imaging
plane
Detector
Figure 12.6 Axial scanning geometry (not to scale).
The detector is comprised of 240 CZT room temperature semiconductor pixels, each
1.5 1.5 1.5 mm. The axial offset between the active detector plane and the focal
spot path is 5 mm. The detector pixels are operated in current mode. Choice of suitable detector technology is a crucial issue. An ideal detector should have good linear
current output for photon fluxes in the range of 103 to 105 photons per square millimeter and microsecond, further fast response time less than 1 s, and low afterglow.
Some scintillation crystals, for instance, lutetium orthosilicate (LSO) or LaCl3(Ce),
are sufficiently fast but have been excluded because of too low overall conversion
efficiency. CZT and CdTe room temperature converters are a more appropriate
choice. Direct conversion is very efficient, and pixellated detectors are readily available. A general problem is device polarization at high photon flux, which is inherent
to this detector type. Since device polarization is a long-term effect, it can be tolerable to some degree in fast-scanning applications, provided suitable reference measurements can be performed. It should also be noted that, owing to different methods
in crystal growth, preparation, and electrical contact technology, commercial CZT
and CdTe detectors do exhibit quite different response functions. For the ROFEX,
detectors have been selected with fabrication technology that results in rather stable
long-term behavior, fast response time, and low afterglow. The time constant of the
detectors is typically 500 ns, and afterglow is lower than 5%.
A block diagram of signal-processing hardware is shown in Figure 12.7. Each
detector pixel is connected to a multistage amplifier circuit providing transimpedance conversion and dynamic range adaption. An integrated gain selection stage
allows us to choose an optimum operating speed and integration time constant for
273

Ub
Ub
Selectable gain
amplier
512
CS
Data
Clock
12 bit
ADC
Select 0
Select 1
Power o
CZT
detector
pixel
4 GByte data RAM module
12 bit
ADC
Data multiplexer
Selectable gain
amplier
Micro controller
USB 2.0 controller
Measurement PC
Figure 12.7 Block diagram of signal-processing hardware. (Reproduced from Fisher, F.

et al., Meas. Sci. Technol., 19, 094002, 2008. With permission.)15
the given imaging problem. The amplifier output voltage signals are converted by
a 12-bit analog-to-digital converter (ADC) with 500-kHz bandwidth limitation. A
temporary data random-access memory (RAM) provides storage capacity for the
vast amount of data acquired during scans. The data acquisition system is designed
for fully synchronous sampling of 512 detector channels. Control of amplifiers,
ADCs and RAM modules, and data transfer to the personal computer (PC) is performed by a single microcontroller. The maximum sampling rate of the detectors is
1 MHz and channel.
The scanner is complemented by some auxiliary components that are necessary
for its operation. Beam monitoring is one important feature. Electron catcher sheets
arranged around the target are used to assist control of deflection patterns and measurement of effective beam diameter. In addition, a small observation charge-coupled
device (CCD) camera is mounted inside the scanner casing to view the focal spot
path during adjustment operations. Between scanner head and anode, a protective
slit mask is mounted that is made of copper and actively cooled by a water circuit.
This mask restricts the electron beam cone to the target surface area and blanks out
the beam when circulating across the area of the target opening. It protects the scanner head from thermal destruction in the case of a malfunction of the beam-steering
system or a wrong manual steering operation. Furthermore, this mask can be used as
a beam dump to park the electron beam at full power. All scanner parts are controlled from the 19-inch rack, which contains the control components, such as coil
current amplifiers, vacuum pump controller, and control PC. All measurement and
274
control software is implemented in LabVIEW, except the detector data acquisition

and image reconstruction software, which is implemented in C++.
12.3Scanning Procedure and Data Processing

A tomographic scan is essentially performed by simultaneous switching of the
Wehnelt voltage and deflection pattern generator and triggering of the detector
data acquisition. For synchronization of beam sweep and detector data acquisition,
the coil current drivers deflection signals are fed into two additional channels of the
detector data acquisition electronics. With the given RAM capacity, a scan as long as
10 s at full 1-MHz sampling rate can be made. After each scan, the data RAM can
be read out via the universal serial bus (USB) 2.0 interface.
In the scanning mode, the electron beam is swept across the target in a circular
way. The beam deflection figure is an ellipse, given by
x (t ) = x 0 + a cos t
y (t ) = b sin t
(12.3)
where x and y denote the deflection angles produced by the x- and y-deflection
coils, respectively; t is the time; and is the angular frequency. The ellipse has a
constant offset x0 in one direction since the optical axis does not coincide with the
target ring center. The elongations of the ellipse are defined by a and b. The three last
parameters are determined by geometrical relations (i.e., target ring diameter, distance between focus of deflection cone and target, and target plane inclination relative to the optical axis of the scanner). The focal spot path geometrically represents a
cut through a skewed elliptical cone, which makes its mathematical description quite
complex. The scanning pattern is stored as a digital data sequence in the control PC
and can be output via a two-channel DAC card at a user-defined rate.
A CT scan results in a set of temporal data from each of the ND detectors. These
data are grouped into NF subsets of size ND NT, with NT being the number of temporal points for a complete electron beam revolution and NF the number of frames in
the scan. The recorded deflection signals are used to find the reference points of the
electron beam revolution. In the next step, the raw data of each frame are mapped
from the temporal domain into the angular domain of the target. That is, a projection
data matrix of size ND NS is computed where NS is the number of equidistantly
distributed source positions on the target. The underlying transformation is comparatively complex and nonlinear because the source trajectory on the target is a cut
of a skewed elliptical cone. The data can now easily be resorted into a fan beam data
set such as for a conventional CT scanner. For these fan beam data, the line integral
X-ray attenuation values are calculated according to

Em ,n = log
I m ,n I m( d )

I m( 0,n) I m( d )
(12.4)
275
Here, I denotes the X-ray intensity encoded in the detector readings; m and n are the
indices of detector and projection, respectively; superscript (d) denotes a previously
acquired dark reference; and superscript (0) indicates a previously acquired reference measurement with no object in the tomographic cross section. Eventually, these
extinction data are reconstructed by a classical fan beam reconstruction algorithm to
images on a given pixel grid, which is typically 128 128 pixels.
12.4Application Examples
Time
Figures12.8 to 12.10 illustrate some selected application examples for multiphase

flow problems. The first is a bubble column, which is often used in the chemical
industry to run large-scale chemical reactions involving a gas and a liquid. Gas is
injected into a column of stagnant liquid, and gas bubbles consequently move upward
in the liquid, driven by buoyancy. A design objective for such devices is to create a
finely dispersed bubbly flow with small bubbles to achieve a high interfacial area
density for mass transfer in the column. However, if the gas flow rate chosen is too
high, bubbles will grow due to coalescence and may form unfavorably large and fastrising plugs with less interfacial area and consequently an inefficient reaction rate.
As shown in Figure12.8, the ROFEX scanner allows analyzing gas fraction distributions in such a bubble column. The column used here was 60 mm in diameter and
made of acrylic glass. If cross-sectional images of a full scan are stacked above each
other, a pseudothree-dimensional (3D) data set results, which gives a good idea of
Scanning
plane
Gas
inlet
Figure 12.8 Scan of a 60-mm diameter bubble column. Left: Sketch of the column. Right:
Pseudothree-dimensional view of the gas distribution for different gas flow rates computed
from the reconstructed image sequence.
276

0 ms
1000 ms
Figure 12.9 Axial scan and slice images of large gas bubbles (dark) in a fluidized bed.
The column has 60-mm diameter and contains nascent polypropylene particles that are
100 m to 1 mm (white).
the gas distribution in the flow. Axial cuts of these 3D data can be directly processed
and displayed. Bubble columns are a particular example for which high-speed video
cameras would fail to quantitatively disclose the gas distribution at gas fractions
above 5% because of the restricted view through the gas bubble swarm.
Another example is imaging the interior of particle flows, something that was not
possible previously. One industrially relevant application is fluidized beds, in which
particles are fluidized in a gas flow. Examples are production of polyolefin or pharmaceuticals, coal firing, or coffee bean roasting. Gas flow in a fluidized bed is not
completely understood and is difficult to model. Figure12.9 shows exemplary crosssectional images of a fluidized particle bed in a 60-mm diameter vessel. The vessel
is filled with polyethylene powder with particles 100m to 1mm in size, and gas is
injected into the column from the bottom. The gas forms larger bubbles, which pass
through the bed. Although the spatial resolution of the ROFEX is not sufficient to
resolve structures down to the single particle, the passage of the gas through the bed can
be reconstructed. The scanning rate in this example was 1,000 frames per second.
Figure 12.10 shows a sequence of cross-sectional images of gas bubbles rising
from an injection needle. Gas injection systems are important components of chemical reactors and are often subjects of optimization. In this example, the flow was not
277
Figure 12.10 Gas bubbles (bright) rising from a three-hole injector needle into liquid.
The liquid-filled vessel was scanned as a reference; therefore, all invariant material is not
shown, and the disturbed gasliquid boundary at the top of the vessel appears with negative
contrast.
scanned horizontally but instead vertically. The injector was placed in a small vessel
with a 10-cm diameter and 10-cm height and filled with water. The image sequence
hence shows a real vertical cross section and no longer a temporal axial cut as for
the bubble column.
As stated, small-animal scanning is an interesting subject for fast CT scanners
but certainly requires some new scanner design for optimal use. To demonstrate the
capability of small-animal scanning, the ROFEX scanner was used. The scanner was
placed such that the targetdetector setup was in an upright position (Figure12.11).
The animal was a male Wistar rat anesthetized with desflurane (30% O2/N2O). It was
placed in a specifically designed closed animal bed for PET, CT, MRI with temperature control. The animal was transversally scanned as shown in Figure12.11. For
slice selection, 0.1-s overview scans at 150 kV and 2 mA were performed. Diagnostic
scans were done at 150 kV and 10 mA for no longer than 15 s.
Figure12.11 (bottom) shows snapshots of an image taken with 1000 frames per
second at the center of the animals heart. The resolution is sufficient to disclose the
anatomical structures of lung, spine, and heart. The heart contraction is rather complex, and the images show an overlay of lateral and axial motion, the latter leading
to an almost complete disappearance of the heart in some parts of the sequence. The
applied radiation dose was measured as 90 mGy/s. Comparison with a commercial
high-resolution animal CT scanner showed that 20-s ultrafast scanning gives comparable exposure as a standard 8-min high-resolution scan. In summary, the results
demonstrated that the heart motion can be fully tracked, and ventricular volume may
be determined.
278
Figure 12.11 Cardiac scan of a rat made with 1,000 frames per second.
12.5Conclusions and Perspectives

Fast tomography is of great interest not only in medicine but also in small-animal
diagnostics, flow measurement, and nondestructive testing. Demands on fast scanners vary with the application, from very fast imaging rates in flow measurement
and nondestructive evaluation of dynamic processes in materials and components
to good penetrability and low doses in biomedicine. In fact, there will always be
a compromise between demands and technological capabilities. Ultrafast electron
beam CT has recently been qualified to achieve a very fast imaging speed, which is
particularly needed not only in flow measurement but also in small-animal imaging.
So far, this technology can scan only moderately attenuating objects, such as biological matter or flows confined in moderately absorbing vessels, made, for example, of
plastics, aluminum, or titanium. Further, there is currently no real 3D imaging capability. However, electron beam CT has a great technological potential that exceeds by
far the potential of multitube fast X-ray CT or other nonX-ray techniques. Electron
beams can be steered and guided in a rather complex way. This gives rise to new
279
concepts of ultrafast multiplane and helical electron beam scanning as it is already

partly realized in cardio CT. Further, the use of two-dimensional detectors may one
day allow cone beamelectron beam tomography. This idea is especially appealing when it comes to small-animal imaging, since this way the dose can be greatly
reduced. However, the required detector technology is, with respect to the development state of planar two-dimensional fast X-ray to electron converters as well as with
respect to powerful data acquisition and processing electronics, far from mature.
Regarding the problem of penetrability, for instance, to scan flows confined in heavy
pressurized vessels, there is the choice of using high-power electron beam accelerators in the future. Thus, ultrafast electron beam CT presents itself today as an imaging method with remarkable prospects.
Acknowledgment
I wish to thank all my coworkers in the FZDs X-ray tomography group, particularly
my PhD student, Frank Fischer, for their continuing efforts in mastering this exciting
new technology.
References

1. D. P. Boyd, M. J. Lipton, Cardiac computed tomography, Proc. IEEE, 71, 298307, 1983.
2. G. Wang, Y. Liu, Y. Ye, S. Zhao, J. Hsieh, S. Ge, Top-level design and preliminary physical analysis for the first electron-beam micro-CT scanner, J. X-Ray Sci. Technol., 12,
251260, 2004.
3. N. Reinicke, G. Petritsch, D. Schmitz, D. Mewes, Tomographic measurement techniquesvisualization of multiphase flows, Chem. Eng. Technol., 21, 718, 1998.
4. M. P. Dudukovic, Opaque multiphase flows: experiments and modeling, Exp. Therm.
Fluid. Sci., 26, 747761, 2002.
5. U. Hampel, M. Speck, D. Koch, H.-J. Menz, H.-G. Mayer, J. Fietz, D. Hoppe, E.
Schleicher, C. Zippe, H.-M. Prasser, Ultrafast X-ray computed tomography with a linearly scanned electron beam source, Flow Meas. Instrum., 16, 6572, 2005.
6. M. D. Mantle, A. J. Sederman, L. F. Gladden, S. Raymahasay, J. M. Winterbottom, E. H.
Stitt, Dynamic MRI visualization of two-phase flow in a ceramic monolith, AIChE J.,
48, 909912, 2002.
7. D. J. Parker, R. N. Forster, P. Fowles, P. S. Takhar, Positron emission particle tracking
using the new Birmingham positron camera, Nucl. Instrum. Methods A, 477, 540545,
2002.
8. R. A. Williams, M. S. Beck, Process Tomography: Principles, Techniques and Applica
tions, Butterworth-Heinemann, Oxford, UK, 1995.
9. T. York, Status of electrical tomography in industrial applications, J. Electron. Imaging,
10, 608619, 2001.
10. G. A. Johansen, T. Frystein, B. T. Hjertaker, . Olsen, A dual sensor flow imaging
tomographic system, Meas. Sci. Technol., 7, 297307, 1996.
11. E. J. Morton, R. D. Luggar, M. J. Key, A. Kundu, L. M. N. Tvora, W. B. Gilboy,
Development of a high speed X-ray tomography system for multiphase flow imaging,
IEEE Trans. Nucl. Sci., 46, 380, 1999.
12. K. Hori, T. Fujimoto, K. Kawanishi, Development of ultra-fast X-ray computed tomography scanner system, IEEE Trans. Nucl. Sci., 45, 2089, 1998.
280
13. M. Misawa, I. Tiseanu, H.-M. Prasser, N. Ichikawa, M. Akai, Ultra-fast x-ray tomography for multi-phase flow interface dynamic studies, Kerntechnik, 68, 8590, 2003.
14. M. Bieberle, F. Fischer, E. Schleicher, D. Koch, K. S. D. C. Aktay, H.-J. Menz, H.-G.
Mayer, U. Hampel, Ultra fast limited-angle type X-ray tomography, Appl. Phys. Lett.,
91, 123516, 2007.
15. F. Fischer, D. Hoppe, E. Schleicher, G. Mattausch, H. Flaske, R. Bartel, U. Hampel, An
ultra fast electron beam X-ray tomography scanner, Meas. Sci. Technol., 19, 094002,
2008.
16. H. Morneburg, ed., Bildgebende Systeme fr die medizinische Diagnostik, 3rd ed.,
Publicis MCD Verlag, Munich, Germany, 1995.
Imaging
13 Compton
Principles and Practice
Carolyn E. Seifert
Pacific Northwest National Laboratory

Richland, Washington
Contents
13.1 Motivation...................................................................................................... 282
13.1.1 Medical Radiotagging and Diagnostics............................................. 282
13.1.2 Emission Imaging Methods............................................................... 283
13.2 Physics of Compton Scatter Imaging.............................................................284
13.2.1 Ideal Conditions.................................................................................284
13.2.2 Effect of Binding Energy and Electron Motion................................. 286
13.2.3 Event Sequencing............................................................................... 287
13.3 Compton Image Reconstruction.................................................................... 291
13.3.1 Backprojection................................................................................... 291
13.3.1.1 Simple Backprojection........................................................ 291
13.3.1.2 Filtered Backprojection....................................................... 293
13.3.2 Iterative Reconstruction Methods...................................................... 294
13.3.2.1 Maximum Likelihood Estimation...................................... 294
13.3.2.2 Expectation Maximization................................................. 296
13.3.2.3 System Response Function P(A|j)...................................... 298
13.3.2.4 Weighted MLEM................................................................ 299
13.3.2.5 Convergence and Early Termination.................................. 301
13.4 Compton Imaging Systems............................................................................302
13.4.1 Compton Imaging Metrics.................................................................302
13.4.1.1 Compton Imaging Efficiency..............................................302
13.4.1.2 Angular Resolution............................................................. 303
13.4.1.3 Field of View.......................................................................304
13.4.2 Detector Requirements, Composition, and Design...........................304
13.4.2.1 Scintillators......................................................................... 305
13.4.2.2 Si Drift Detectors................................................................ 305
13.4.2.3 Double-Sided Silicon or Germanium Strip Detectors........307
13.4.2.4 Segmented Coaxial Germanium.........................................307
13.4.2.5 Pixellated CdZnTe..............................................................308
13.4.2.6 Gas and Liquid Time Projection Chambers.......................308
13.5 Limitations of Compton Imagers and New Opportunities............................309
References............................................................................................................... 310
281
282
13.1Motivation
This chapter explores the physics behind and implementation of Compton imagers
for medical diagnostics. Beginning with a brief introduction of traditional nuclear
medicine technologies, this chapter develops the physics framework for Compton
scatter kinematics, introduces various methods of image reconstruction, presents
several practical implementations of Compton imagers, and finally discusses opportunities for future development.
Compton imaging is the use of Compton gamma-ray scatter kinematics to reconstruct an image of a gamma-ray source distribution. Knowing both the energy and
position of gamma-ray interactions in a detector enables reconstruction of the initial
gamma-ray direction to within a conical surface of probability. Measuring many
such Compton cones enables reconstruction of the distribution of radioactivity present in the patient. Unlike other medical imaging systems, Compton imagers do not
require mechanical collimators or superb timing resolution to function; only position
and energy sensitivity are needed.
13.1.1Medical Radiotagging and Diagnostics

Gamma-ray imaging systems reveal two- and three-dimensional information regarding spatial distribution of radionuclides for a variety of applications including medical diagnostics, astrophysics, and national security. In nuclear medical diagnostics
gamma-ray imaging provides information to help diagnose functional performance
of various systems in the body. A radioactive isotope binds to a tracer molecule,
which is then injected or ingested into the body. The tracer molecule travels through
the body, preferentially collecting in specific tissues or bone due to the chemical
nature of the tracer. As the radionuclides decay, they emit gamma rays, which are
detected by the gamma-ray camera. The resulting image provides the distribution
of tracer molecules in the body, which relates to a specific function in the body. For
example, radioactive iodine gathers in the thyroid and thus produces high-resolution
images to aid in diagnosis of thyroid abnormalities. Also, using gated emission computed tomography, it is possible to image the motion of the heart in 4-D: three spatial
dimensions plus time.
Typical medical diagnostics are performed with 99mTc as the gamma emitter.
This nuclide emits a gamma ray at 140 keV. Other radiotracers include 129I, 18F,
and 139Lu. Fluorine-18 is an isotope used for positron emission tomography (PET).
Medical radionuclides are selected based on both emitted gamma-ray energies and
on ability to bind to a given tracer molecule that targets specific areas within the
body. Low-energy gamma-ray emitters are preferred, because a higher likelihood
of capture in the gamma-ray imager allows for a lower quantity of radiotracers, and
thus a lower dose given to the patient However, the goal of high intrinsic detection
efficiency must be accompanied by a high escape fraction from the body, which
is accomplished using higher-energy gamma rays. In practice, typical gamma-ray
energies range from 140 to 511 keV.
Compton Imaging
283
13.1.2Emission Imaging Methods

A typical imager consists of a large planar scintillator that emits visible and nearultraviolet (UV) photons when excited by gamma radiation. These photons are then
detected in an array of position-sensitive photomultiplier tubes, which provide twodimensional spatial information on the location of the gamma-ray interactions in the
scintillator. This is called an Anger camera. Alternatively, pixellated semiconductor
detectors can also be used; in these systems, the gamma rays are converted into
electronhole pairs that drift under bias and generate electronic signals on collecting
electrodes. Each electrode is read out independently, providing the two-dimensional
spatial information for this type of detector.
For either detector system, a collimator composed of high-density, highatomicnumber material such as lead or tungsten is placed in front of the detector to allow
only gamma rays from preferred directions to enter. Parallel, divergent, or convergent holes in the collimator attenuate gamma rays incident from oblique angles and
provide a correlation between the gamma-ray interaction locations in the detector
and the originating directions of the gamma rays, resulting in a two-dimensional
image of the radiation emitted from the patient. When such images are taken at
regular projection angles around the body, a three-dimensional reconstruction of the
distribution of radioactivity in the patient is possible. This imaging method is called
single-photon emission computed tomography (SPECT).
Positron emission tomography is a similar technique, except that two detectors located on opposite sides of the patient are used to detect the simultaneously
emitted 511-keV photons from the positrons-emitted radioisotope delivered to the
patient. Because the two photons are emitted in nearly opposite directions, the
possible origin of the two photons can be narrowed to a line between the gammaray interaction locations in the two detectors. In addition, measurement of the
difference in the time each gamma ray was detected can further narrow the origin
to a segment of that line. Three-dimensional distributions of the radioactivity
in the patient are reconstructed from many such detections. The two detectors
can be rotated around the patientas with typical SPECT systemsor the PET
detectors can be oriented in a continuous ring around the patient to achieve much
higher efficiency.
In any nuclear diagnostic imaging application, minimizing patient dose is critical. The imaging system sensitivity is inversely linked to the radiation dose required.
Maximizing sensor efficiency is the first step to increasing the sensitivity. The Anger
cameras typically used in SPECT and other collimator systems by nature attenuate a
large fraction (over half) of the radiation emitted by a patient. Increasing imaging efficiency is a key goal for Compton imagers, which require no collimators for imaging.
An imaging system capable of simultaneously providing diagnostic information
on both the functional and the structural aspects of the body would be beneficial
from operational and medical standpoints by reducing patient preparation and scanning time and enabling high-fidelity corrections for patient positioning. Owing to
the nature of the collimators typically employed, SPECT and Anger camera systems
284
cannot be used for other diagnostic imaging methods, such as X-ray computed
tomography (CT) or PET. However, Compton imagers may provide similar emission
imaging capabilities as Anger cameras while also functioning as multimodal detector systems.
13.2Physics of Compton Scatter Imaging

13.2.1Ideal Conditions
In Compton scatter, a gamma ray interacts with an atomic electron, providing enough
energy to overcome the binding energy and eject it from the atom. The gamma ray,
which transfers energy to the electron in the form of kinetic energy, is scattered at an
angle with respect to its initial direction. The gamma ray may then scatter again or
undergo other types of interactions in the material. Of particular interest in Compton
imaging is photoelectric absorption, in which the total scattered gamma-ray energy
is transferred to an atom, which then ejects an electron. After any interaction, the
energy carried by the electron is then deposited in the surrounding material through
electron scattering, which generates electronion pairs in gases or electronhole
pairs in scintillators and semiconductor detectors. For initial gamma-ray energy E 0,
deposited energy E1, and scattered gamma-ray energy E, the Compton scatter angle
can be calculated via Equation 13.1:

cos = 1
mec 2 E1

E0 E
(13.1)
where mec2 is the rest mass energy of an electron (511 keV). Equation 13.1 is commonly called the Compton scatter formula and is derived under the assumptions that
the electron is unbound and at rest.
E1 is measured in the detector for each detected gamma ray. For a known E0, the
scattered gamma-ray energy E can be determined mathematically (E = E 0 E1) or
by measuring its energy directly by summing the energies deposited in subsequent
interactions in the detector. The latter method is preferred for several reasons. First,
any scatter occurring before gamma rays reach the detector will reduce the incident
gamma-ray energy and result in an incorrect value of E 0 used in the Compton scatter calculation. Second, the direction of the scattered gamma ray must be known to
reconstruct the gamma-ray incident direction (as discussed in the next paragraph),
and thus the scattered gamma-ray must interact a second time in the detector anyway. Measuring the energy deposited in the second gamma-ray interaction is generally a trivial addition to the measurement of its location. E0, when not known a
priori, is generally calculated by summing the energies of all interactions. By means
of measured or calculated values for E0, E1, and E, the scatter angle of the gamma
ray can be determined.
With the determined Compton scatter angle, it is possible to determine the incident
direction of the gamma ray to within the surface of a cone. The opening half angle of
285
Compton Imaging
Y
Z
Figure 13.1 Illustration of Compton scatter imaging. The line between the first and second interaction locations determines the axis of a cone whose half angle is given by the
Compton scatter formula. The cone is projected onto the source plane. Summing cones from
many events gives an estimate (image) of the source distribution. (Courtesy of E. Fuller.)
the Compton cone is given by the Compton scatter angle, as illustrated in Figure13.1.
The axis of this cone is defined by the antiparallel vector to the scattered gamma-ray
direction, which is determined by the line vector from the first to the second gammaray interaction location. Thus, in any Compton imager, two gamma-ray events are
required to reconstruct a cone of probability of the incident gamma-ray direction.
The Compton cone is projected onto an image surface, which is typically a plane
at some known distance from the detector face, resulting in ellipsoid rings representing possible emission locations. The image surface could also be spherical,
cuboid, or any other shape convenient for reconstruction or data interpretation. Each
detected gamma ray results in at most one Compton cone, and summing cones over
many gamma rays results in an estimate of the distribution of radioactivity in the
patient. (A more detailed description of Compton image reconstruction is given in
Section 13.3.)
Ideally, a gamma ray will scatter once and then be absorbed in the detector such
that only two interactions occur. It is possibleand even likelythat a gamma ray
will scatter twice and escape, depositing less than its full energy in two interactions,
or scatter multiple times before being absorbed in the detector, depositing its full
energy in three or more interactions. When E 0 is known or suspected (as is the case
in medical imaging), it is possible to determine whether full energy was deposited
in the detector or some energy was lost due to scatter or escape outside the detector.
A simple comparison of the total energy deposited against the known gamma-ray
energy enables discrimination of partial-energy events, if desired. When the full
gamma-ray energy is deposited in three or more interactions, only E0, E1, and the
positions of the first two interactions are needed to reconstruct the Compton cone.
286
13.2.2Effect of Binding Energy and Electron Motion

Equation 13.1 assumes that electrons are both unbound and at rest. The assumption
of free electrons has few consequences for Compton imaging. The effect of including
the binding energy is to restrict Compton scatter to only those electrons whose binding energy can be overcome by the energy deposited in the scatter event. A reduction
in the electron shell cross section results, and the electron kinetic energy is reduced
by the binding energy of the shell. A deexcitation X-ray is emitted from the atom
with an energy equal to the binding energy. For typical position resolution on the
order of 1 mm, very little X-ray escape is predicted, and the total energy deposited
in the detector at the interaction location is the same as the total energy lost by the
scattered gamma ray, as calculated using the Compton scatter formula.
The momentum of the electron before the scatter, on the other hand, has a much
more significant effect on Compton imaging. The Compton scatter formula predicts
a one-to-one relationship between the scatter angle and the scattered gamma-ray
energy. If the finite electron momentum is included, for a fixed scatter angle the
scattered gamma-ray energy will deviate from that predicted by Equation 13.1. The
relationship between the initial gamma-ray energy E 0, scattered gamma-ray energy
E, the scatter angle , and the electron momentum in the scatter plane pz is given in
Equation 13.2.1,2

pz = mc
E0 E E0 E(1 cos ) /mc 2

E02 + E2 2 E0 E cos
(13.2)
If the initial electron momentum were known, then Equation 13.2 could be used
in place of the Compton scatter formula (Equation 13.1) to calculate the precise scatter angle. However, the initial electron momentum cannot be known a priori. It may
be possible in some detector designs to measure the electron momentum after the
scatter and deduce the initial momentum from the measured energies (see Section
13.3.4), but this is not typically possible. As a result, there is no recourse except to
use the Compton scatter formula in Equation 13.1 and account for the additional
uncertainty due to the motion of the electron in the calculated scatter angle.
The electron momentum vector can be oriented in any direction. In some scatter
events, it will have an additive effect, increasing the scattered gamma-ray energy; in
others, it will reduce the scattered gamma-ray energy. The result is a distribution of
possible energies centered about the value predicted by the Compton scatter formula.
Similarly, for a fixed deposited energy the actual Compton scatter angle will also
vary about the mean predicted value. This is called Doppler broadening.
The effect of Doppler broadening is to degrade the angular resolution of the Compton
imager. Doppler broadening is most severe for high electron momentum, which occurs
in materials with high atomic numbers. Figure13.2 shows the scattered gamma-ray
energy distribution in several detection materials for a 90 scatter of 511-keV photons.
These distributions are calculated analytically using Equation 13.2 and the HartreeFock Compton profiles,3 using the method described by Ribberfors and Berggren.2
287
Compton Imaging
Dierential Distribution
(relative probability per keV)
0.035
0.03
Si
Ge
CZT
Hgl2
511 keV gamma rays

90-degree scatter
0.025
0.02
0.015
0.01
0.005
0
200
210
220
230
240
250
260
270
280
290
300
Scattered Gamma-Ray Energy (keV)
Figure 13.2 The effect of electron momentum in several detection materials on the calculation of the scattered gamma-ray energy assuming a 90 scatter of 511-keV gamma rays.
Note that the shape of the distribution for each material is dependent on the filled
electron levels in the atom, and that Doppler broadening is not a Gaussian effect.
Although the Compton profile of each electron shell is Gaussian, the total atomic
distribution is the weighted sum of the profiles and not the convolution of them. As
such, using a full width at half maximum (FWHM) value to estimate the uncertainty
due to Doppler broadening will greatly underestimate the effect. In addition, because
the profiles differ for each element, the FWHM will result in a variable measure
of the Doppler broadening effect. (Angular resolution metrics are discussed further
in Section 13.3.)
13.2.3Event Sequencing
As described, the axis of a Compton reconstruction cone is determined by the vector
from the second to the first gamma-ray interaction. Determining the Compton cone,
then, relies on the ability to determine the order of gamma-ray events in the detector.
The sequence of interactions in some types of Compton imagers is clearly defined.
NASAs Compton Gamma-Ray Observatory, for example, had two sets of detectors
separated by a sufficient distance such that the timing between interactions in the
first and second detector arrays could be measured.4 Only gamma rays that first
scattered in the plastic scintillators and then were absorbed in the NaI(Tl) backplane
scintillators were accepted for imaging. In other designs, the backplane detector is
shielded so that it does not receive direct irradiation. The design of a sufficiently
large two-detector imager can then require the first detected event to occur in the
front-plane detector.
For Compton imagers in which the time between gamma-ray events cannot be
measured (because of small form factor or poor timing resolution), all events for a
given gamma ray appear to occur simultaneously. Neither shielding nor timing can
288
help determine the sequence of events, and alternative strategies must be employed for
determining the sequence of events. The only information available on a sequenceby-sequence basis is the number of events, the positions and energies deposited for
each event, and the expected initial gamma-ray energy. The kinematics of Compton
scattering provide some guidance for choosing the proper event order.
In the following discussion, event or interaction describes a single occurrence of
a detectable photon interaction such as photoelectric absorption, Compton scatter,
or pair production. A detectable interaction is one in which the energy deposited
exceeds a given threshold (determined by the electronic noise of the spectrometer). A
sequence or track is the series of interactions from a single gamma ray in the detector.
Thus, a sequence consists of one or more events. Finally, sequence reconstruction,
sometimes referred to as gamma-ray tracking in the literature, is the determination
of the order of interactions in the detector.
Sequence reconstruction techniques differ according to the number of interactions observed. Tracks that consist of only one event cannot be used for Compton
imaging, as previously discussed. At least two interactions must occur in the detector
to determine both the scatter angle and the cone axis.
For tracks with two events, determining the order of interactions is for the most
part educated guessing, with the method of guessing depending on the total observed
energy. For any initial gamma-ray energy below 256 keV, the energy distributions
of the deposited and scattered gamma-ray energies are distinct. This indicates that
for a full-energy sequence in which the gamma ray is scattered and then absorbed in
two interactions, the sequence order is absolutely distinguishable. The first interaction always deposits less energy than the second. Figure13.3a shows the deposited
energy distributions for the scatter and absorption events, which were calculated
using the KleinNishina differential Compton cross section5 for 140-keV gamma
rays. For gamma-ray energies above 256 keV the scatter and absorption distributions overlap, as shown in Figure13.3b for 511-keV gamma rays. For these gamma
rays, the knowledge of the deposited energies is insufficient to determine the order
of events absolutely.
From Figure13.3, it is clear that there are some energies that are kinematically
impossible to deposit via scatter processes given the incident gamma-ray energy. The
Compton edge, corresponding to a backscatter event, is the highest energy that can
be deposited in scatter by a gamma ray and is determined using the Compton scatter
formula as in Equation 13.3. If the sequence order is known, then a sequence can be
rejected if the first event deposits an energy higher than the Compton edge. This is
called the Compton edge test, which can be used to reject some sequences in which
the full gamma-ray energy has not been deposited.

Eedge =
E0
2
mc
1 + e 1 cos
E0
E0

2mec 2
1+
E0
(13.3)
In any two-event sequence in which more than 256 keV is deposited, at least one
of the two potential sequence orders is kinematically possible, and one must attempt
to determine the correct sequence. As a result, the Compton edge test cannot be
289
Compton Imaging
140
Deposited Energy (keV)
120
Absorption (Second Interaction)
100
80
60
Scatter (First Interaction)
40
20
0
20
40
60
100
120
80
Scatter Angle (degrees)
140
160
180
140
160
180
600
Absorption (Second Interaction)
500
400
300
200
Scatter (First Interaction)
100
0
20
40
60
80
100
120
Figure 13.3 Deposited energies for two-event scatter-absorption sequences as a function

of scatter angle for initial gamma-ray energies of (a) 140 keV and (b) 511 keV.
used when the sequence order is unknown unless more information is available.
Fortunately, as incident gamma-ray energy increases, the probability of the first
event depositing more energy also increases. This would appear to contradict the
KleinNishina differential cross-section relationship in which higher-energy gamma
rays are more likely to be forward scattered, resulting in lower energies deposited
on average in scatter events. However, owing to the finite size of the detector system
used for Compton imaging, the gamma ray must scatter at a large angle in the first
event to sufficiently reduce its remaining energy such that the probability for capture is high. Thus, because Compton imaging selectively filters sequences ending
in absorption (full-energy deposition), large-angle scattersand thus high-energy
depositionsare more likely to occur in the first event of a two-event sequence. To
illustrate this concept, Figure13.4 shows the distribution of energies deposited in
the first and second events of full-energy two-event sequences in a large germanium
290
Relative Probability per keV
250
200
Absorption event
(second interaction)
Scatter event
(first interaction)
150
100
50
0
50
100 150 200 250 300 350 400 450 500 550
Figure 13.4 The distribution of energies deposited in the first and second interaction of
full-energy sequences at 511 keV in a large germanium detector. In this example, the scatter
event deposits more energy than the absorption event in 60% of the full-energy two-event
sequences.
detector. The high probability of backscatter in the first interaction is evident in

the figure.
This suggests a simple way to determine which event in the sequence actually
came first. If the total energy deposited is greater than a determined threshold value,
the higher energy was most likely deposited first. If the total energy observed is less
than the threshold, the opposite assumption is made: The first interaction is the one
that deposited less energy. The actual threshold value depends on the detector material and geometry and can be determined using radiation transport simulations. For
a small CdZnTe detector, a threshold of 400 keV was determined to provide the best
sequencing efficiency for two events.6
Now that there is a simple sequence reconstruction technique for total deposited
energies above 256 keV, the Compton edge test can be applied. If the chosen sequence
is not kinematically possible despite full-energy deposition, then the sequence order
was incorrect and should be reversed, or the sequence can be discarded. Using this
method, the probability of correctly sequencing two events in the detector ranges
from 50% (at 400 keV) to 100%, depending on the energy of the initial gamma ray.
Traditional Compton imagers did not use with more than 3 interactions for imaging, but several methods have been developed for sequencing multiple events. For
three detected events, there are six possible event sequences. Choosing between
possible tracks requires a figure of merit (FOM) to be assigned to each track. The
sequence with the best FOM is then chosen. Two FOMs have been proposed by other
researchers: a chi-squared statistic relating the difference in the second scatter angle
calculated using both deposited energies and known interaction positions710 and a
likelihood function based on the KleinNishina scatter probability (probabilistic
method).11,12 In general, the likelihood of correctly reconstructing high multiplicity
sequences ranges from 60% to 80%, depending on the detector design and sequence
reconstruction method used.
Compton Imaging
291
13.3Compton Image Reconstruction

Image reconstruction techniques typically fall into two categories: direct (Fourier)
and iterative methods. Direct reconstruction uses approximate deterministic expressions to reconstruct an object from its forward projections. Iterative methods use
statistical models and can more accurately treat the physics of gamma-ray detection
and account for stochastic processes, although the iterative nature of the reconstruction necessitates longer computation times than the direct methods.
The two most popular methods for reconstructing images from Compton scatter
data are backprojection and maximum likelihood (ML) estimation. Backprojection
is a direct reconstruction method that relies on fast Fourier transforms and has been
used extensively for 3D tomographic imaging from X-ray projection data. It usually
requires first a backprojection step to obtain a blurred estimate of the image and
then a filtering step to remove the blurring. (Sometimes, the filter is applied before
the backprojection step.) Because the projections from Compton scatter data are not
straight lines as in X-ray projections but rather conic sections, backprojection is not a
precise image reconstruction technique for Compton imaging. As a result, ML estimationan iterative reconstruction methodhas been used for Compton imaging
since the 1980s with much success.
13.3.1Backprojection
13.3.1.1Simple Backprojection
Backprojection gained popularity in early Compton imagers because of its simplicity. The mathematics of backprojection were well understood in terms of X-ray
tomography, for which a 3D image is reconstructed from its 2D straight-line forward
projections. In Compton imagers, the forward projection is the observed data: deposited energies and interaction locations. Reconstructing images requires summing the
backprojected cones on a surface, as previously illustrated. Usually, backprojection
is performed in the source plane parallel to the front face of the detector. The source
distance must be known a priori; in medical imaging applications, this can be the
average distance to the patient, or the image can be reconstructed at a series of distances, bringing features into or out of focus at each depth.
Backprojection is performed in the following manner: The axis of the Compton
cone is determined by drawing a ray from the second interaction through the first
interaction location. The cone vertex is the first interaction point, and the cone opens
in the direction of the axis ray with a half angle equal to the first Compton scatter
angle. The cone has a finite width due to the uncertainty in both the axis direction
and the calculated Compton angle. The cone shell is projected onto the imaging
surface. The imaging surface is mathematically discretized into pixels. Wilderman
et al.13 described a process for determining the proper backprojections with a planar
image surface. Rohe et al.14 described the calculation for a spherical image volume
inside a shell of detectors. Each image pixel that intersects the backprojection cone is
then given a value of 1 for that gamma-ray sequence. All other pixels are assigned a
value of 0. The image pixel values are sometimes normalized such that the sum of
292
the intensities for each pixel in the ring equals 1. The process is repeated for each set
of measurements, and the backprojections are summed.
The width of the backprojection cone is given by the estimated angular uncertainty. There are two independent components to the angular uncertainty: error in
the Compton scatter angle, calculated using deposited energies, and error in the cone
axis direction, calculated using interaction positions. The angular uncertainties due
to energy and position can then be added in quadrature to yield the total angular
uncertainty and hence the width of the backprojection cone.
The uncertainty in angle calculated using energies de is calculated as in Equation
13.4, given the uncertainty dE1 in the energy E1.
de =
mec 2
dE1
( E0 E1 )2 sin
(13.4)
This relation assumes that the initial gamma-ray energy E 0 is known absolutely
and is not determined by the sum of deposited energies, each of which has an
uncertainty associated with its measurement. In the case of medical imaging, this
assumption amounts to requiring that the total deposited energy equal the expected
gamma-ray energy and then using the absolute known gamma-ray energy (rather
than the sum of deposited energies) in Equation 13.1 to calculate the Compton scatter angle.
The angular uncertainty due to energy resolution becomes very large for both
small and large scatter angles, as shown in Figure13.5 for several values of energy
resolution. For small scatter angles, the first event deposits very little energy. In this
case, the error in the energy can be a large fraction of the deposited energy, and
thus the angular uncertainty is also very large. For backscattered gamma rays, the
Angular Uncertainty (degrees)
50
0.5%
1%
3%
7%
12%
45
40
35
30
25
20
15
10
5
0
20
40
60
80
100
120
140
160
180
Figure 13.5 The energy-resolution-induced angular uncertainty as a function of Compton

scatter angle for several values of energy resolution reported at 662 keV. The energy uncertainty is assumed to scale with the square root of deposited energy and includes a constant
noise term.
293
Compton Imaging
Figure 13.6 The method for calculating angular uncertainty due to position resolution for
detectors with discrete position resolution.
deposited energy is not sensitive to large changes in scatter angles. As expected, the
effect of energy resolution on the angular uncertainty is most profound for lowerenergy gamma rays.
Angular uncertainties based on interaction positions are estimated using the
uncertainties in the x, y, and z coordinates of the gamma-ray interactions. Ordonez
et al. provided a formalism for expressing the angular uncertainty. For isotropic
spatial resolution dr1, the angular uncertainty for any given sequence is given by
Equation 13.5, where R1 and R2 are the distance from the source to the first and second gamma-ray interaction locations, respectively.

2
2
2
dr = (dr2 ) + (dr1 ) 1 + R1 + 2 R1 cos

2
2
2
R2
R2
R1 R2
(13.5)
Other methods for determining the geometric contribution to angular uncertainty

are also possible. Lehner provided an empirical method for calculating geometric
uncertainty in a system with discrete position resolution.6 In that work, the average
angular deviation relative to a line connecting the two voxel centroids is calculated
for randomly selected points R1 and R2 as shown in Figure13.6.
Typically in Compton imagers, the position resolution dominates the angular
uncertainty unless the system is sufficiently large such that the first and second
interactions are separated by a distance much larger than the position resolution.
Two-plane Compton imagers like CGRO meet this requirement, and the angular
uncertainty is dominated by the energy resolution in the front detector plane.
13.3.1.2Filtered Backprojection
In the simple backprojection method presented, the backprojection cones are simply
summed on the image surface without any image filtering. However, as well known
in X-ray CT, filtering out the blur introduced in the reconstruction process will significantly improve image resolution and enable better medical diagnostics. Several
methods for filtered backprojection have been tested. As with CT, the filtering step
can be performed before or after the summation of cones. In this section, both of these
methods are referred to as filtered backprojection, although the latter is sometimes
294
referred to in the literature as backprojection filtering to indicate the sequence of

operations. One recent example of filtered backprojection was published by Xu and
He in 2006.16 Unlike previous attempts at direct reconstruction of Compton data, the
method developed by Xu and He accounts for the fact that the measured data set cannot be represented by a KleinNishina distribution for Compton scatter angles due
to both finite detector size and minimum energy thresholds in the detector. Although
image resolution was not measured directly, it can be inferred from the data presented in the manuscript that imaging resolution using filtered backprojection is better than 7 FWHM for a system in which simple backprojection cannot completely
separate sources that are separated by 65. Also, because the filtered backprojection
can be performed on a cone-by-cone basis, it is possible to apply this method in realtime to measured data using fast Fourier methods.
13.3.2Iterative Reconstruction Methods

Iterative reconstruction methods have an advantage in that the physics of Compton
scatter and radiation detection can be incorporated into the reconstruction process,
reducing much of the blur. The most widely used iterative reconstruction method for
Compton image data is maximum likelihood estimation. This method maximizes the
likelihood that a given (unknown) source distribution produces the data observed.
Because the source distribution is not known and the data can be considered incomplete, the reconstruction process involves iterating over increasingly likely estimates
of the source distribution. Iterations proceed until convergence of the source distribution estimate, or until the reconstruction is terminated. This section describes the
mathematics behind maximum likelihood estimation and provides several variations
on the method.
13.3.2.1Maximum Likelihood Estimation
The source estimate (image) is discretized into M pixels, and p(A|j) denotes the
system model, that is, the probability of observing the measurement set A given a
gamma ray originating from source pixel j.
Conventionally, the measurements A are also discretized. A given set of observed
interaction locations and deposited energies for a measurement corresponds to one
bin in p(A|j). Many bins are required to account for all possible combinations of
locations and energies; for k attributes in a given measurement with precision b measured in bits, at least 2kb elements are required.15 One research group estimated that
over 2 1010 elements were required in the system model p(A|j) per source pixel for
their system.16 At even 1 byte per integer value in the bin (counting up to 255 events
in a single bin), the model would require nearly 20 GB of memory, which can be difficult to obtain and is certainly an inefficient way to store the data.
In list mode, the measurements are stored in a list during acquisition. Each measured parameter is considered as a point in a continuous measurement space, and
no binning of data is required. As a result, information is retained throughout the
reconstruction that would otherwise be lost in the binning of positions and energies
in conventional ML. The total number of memory elements required in list mode is
295
Compton Imaging
only Nk. Thus, for large values of k, 2kb >> Nk, and list mode reconstruction becomes
much more efficient. There are at least eight parameters per measurement: the 3D
positions of at least two interactions plus the energies deposited at those locations.
The list mode ML equations were derived by Parra and Barrett.17 The methodology is briefly reviewed here, and the Parra and Barrett notations and conventions
are adopted here. Consider a source distribution f = [f1, f 2, , fj , , f M], where fj
is the expected number of photons from source pixel j. Let sj be the probability that
a gamma ray from source pixel j is detected anywhere. Then, the probability of a
detected gamma ray originating from source pixel j given the distribution f is given
by Equation 13.6.
P( j|f ) =
f js j
M
f s
(13.6)
n n
n=1
The probability p(A|f) of observing a measurement A due to a gamma ray that originated from anywhere in the distribution f is given in Equation 13.7, where p(A|j) is
the probability of observing a measurement A given that the gamma ray originated
in source pixel j, also known as the system function or model.
M
p( A|f ) =
p( A|j)P( j|f )
(13.7)
j =1
The goal of ML reconstruction is to maximize the likelihood that the source estimate generated a given set of observed measurements. The log-likelihood of observing the measurements given the source distribution L(A1, ,AN|f) can be expressed
as in Equation 13.8.
N
L ( A1 ,..., AN |f ) =
ln

i =1
j =1
p( Ai |j) f j s j N ln
f s
j j
(13.8)
j =1
When the data are collected, typically the measurement time T is fixed, and the
number of detected sequences N becomes a random variable that is drawn from a
Poisson distribution as in Equations 13.9 and 13.10.

P( N |T , f ) =
(T ) N exp(T )

N!
(13.9)
s f
j j
j =1
(13.10)
296
Then, the likelihood function is calculated via the product of the independent likelihoods of observing A and N, as in Equation 13.11.

L ( A1 ,..., AN , N |T , f ) = ln p( A1 ,..., AN |f ) P( N |T , f )
(13.11)
The likelihood function is then maximized with respect to the source distribution
f. Since f is unknown, an iterative process must be used to determine the estimate of
f that maximizes the likelihood of observing the given data. Various maximization
procedures can be used, but the expectation-maximum (EM) technique is used most
often with ML reconstruction. Together, they are known as maximum likelihoodexpectation maximum or MLEM.
13.3.2.2Expectation Maximization
The key principle for the EM algorithm is the assumption that the observed data
are incomplete and are drawn from a set of (unknown) complete data. Two assumptions are required: The source distribution is a priori independent of the parameters of the missing data process, and the missing data are missing at random.18
The log-likelihood function of the complete data then provides all information necessary to exactly reconstruct the source distribution. However, the complete data
set is unknown, and it cannot be sampled. Instead, the expectation value of the loglikelihood function is calculated based on the current estimate of the source distribution and the observed data. The source distribution that maximizes the expected
value of the log-likelihood then becomes the current estimate, and the process is
repeated until the log-likelihood function converges or the process is artificially
stopped. Thus, the two steps in the EM algorithm are first a calculation of the
expectation of the log-likelihood function based on the current source estimate
and then a calculation of the source estimate that maximizes the calculated loglikelihood function.
Suppose the observed data are expanded by the unobserved variables zij, such that
zij = 1 if event i originated in pixel j and zij = 0 otherwise. In other words, assume
there are some gamma rays emitted from the source that were not detected. It is clear
that zij has only one nonzero entry in each row because a gamma ray can originate
from only one source pixel.
The probability of observing measurement zi is then determined by Equation 13.12.
M
P(zi |f ) =
z P( j|f )
p(zi |j) P( j|f ) =
j =1
(13.12)
ij
j =1
The log-likelihood function then reduces to the expression in Equation 13.13.

L ( A1 , z1 ,..., A N , z N , N |T , f ) =

z (ln P( j|f ) + ln p(A |j)) + ln P(N |T , f )
ij
i =1
j =1
(13.13)
297
Compton Imaging
As previously stated, the unknown data zij cannot be sampled. Thus, in the expectation step of the EM algorithm, the expected value Q(f|f (t)) of the log-likelihood
function (defined in Equation 13.14) is calculated based on the measured data
A1, , AN and the (fixed) current estimate f (t) of the source distribution f, where t
indicates the iteration number. The calculation involves simply replacing zij with its
expected value, given in Equation 13.15.

Q( f |f (t ) ) = E L ( A1 , z1 ,..., AN , z N , N |T , f )|A1 ,..., AN , N , f (t )

P( j|f (t ) ) p( Ai |j)
zij ( f (t ) ) = P( j|Ai , f (t ) ) =
P(k|f
(t )
(13.14)
(13.15)
) p( Ai |k )
k =1
After the expectation step is complete, the derivative of the expected value of the
likelihood is set to zero, while f (t) is fixed, and Equations 13.16 and 13.17 are obtained
for the updated algorithm.
N
f j(t+1) =
i =1
f j(t+1) =
f j(t )
T
zij ( f (t ) )

Ts j
i=1
(13.16)
p( Ai |j)
p( A |k ) f
i
(13.17)
(t )
k
k
k =1
Thus, given any initial estimate f (0), a new source estimate can be calculated
from the system model and the sensitivity of the device. In practice, only the
updated expression is required, and the expectation and maximization steps are
performed simultaneously. The initial estimate f (0) is typically the initial unfiltered
backprojection image of the data, although a uniform field or other image estimate
could be used.
Some authors16,19,20 have performed the expectation step on the complete data
rather than the likelihood function. Shepp and Vardi19 have shown that this method
also converges to a ML point. However, it leads to a slightly different result for the
iteration expression,16,20 as given in Equation 13.18.

f j(t+1) =
f j(t )
Ts j
i=1
p( Ai |j)
p( A |k ) f
i
(13.18)
(t )
k
k =1
The sensitivities sj are moved outside the summations. (The original MLEM
algorithm by Dempster, Laird, and Rubin18 uses the expectation of the likelihood
298
function, except in the special case of exponential families, for which the vector of
sufficient statistics is used.) It is believed that there may be some small differences in
the two methods, especially for systems in which the sensitivity widely varies from
pixel to pixel.
The EM algorithm has several advantages over other maximization procedures.
Dempster and colleagues18 showed that the likelihood after each step is nondecreasing,
meaning that successive iterations lead to a source estimate that is at least as likely as
the previous estimate to have produced the observed data. Any positive initial source
estimate automatically results in nonnegative subsequent estimates.20 Furthermore,
under some reasonable conditions (such as N M: the number of observed sequences
is greater than or equal to the number of image pixels), the log-likelihood function
is strictly convex and necessarily has a single global maximum.17 In this case, it has
been proved that successive iterations of the expectation and maximization steps will
lead to global convergence.20 Convergence is achieved when the global maximum of
the likelihood function has been reached. Lange and Carson proved that the image
obtained after convergence is independent of the choice of initial estimate f (0).20 In
practice, image reconstruction is rarely computed through all the iterations required
to achieve convergence, and stopping rules and other justifications for early termination of the algorithm are often applied (see Section 13.3.2.5).
13.3.2.3System Response Function P(A|j)
Wilderman et al.21 proposed an analytical system model for list mode ML of Compton
scatter images. A value for P(A|j) that can be calculated on the fly is advantageous
and preferable to a system matrix that bins the data (resulting in lost information)
and requires dedicated memory or a table lookup (resulting in lengthy reconstruction
times). The probability of observing a given measurement A = [E0, E, E, , r01, r12,
r 23, ] given a gamma ray incident from pixel j is then given by Equation 13.19.

P( A|j) = exp t ( E0 )r01
) dd exp ( (E)r )
C
12
(13.19)
where t(E) is the total absorption cross section at energy E; E 0 and E are the initial
and scattered gamma-ray energies, respectively; r01 is the attenuation distance between
the source pixel and the first interaction; r12 is the attenuation distance between the
first and second interactions; and dC/d is the differential Compton cross section,
which is approximated by the KleinNishina cross section divided by r12.2 Thus, the
system model is the product of the probabilities of survival of the initial gamma ray
to the first interaction point, scatter at the observed angle , and survival of the scattered gamma ray to the second interaction location. The KleinNishina cross section
is given in Equation 13.20,5 where = E 0/mec2 is the ratio of the initial gamma-ray
energy to the rest mass energy of an electron.

1
d C
d 1 + (1 cos )
1 + cos2
2 (1 cos )2
1 +
(13.20)
2
2
(1 + cos )[1 + (1 cos )]
299
Compton Imaging
In this calculation, is the angle of scatter that would be observed if the gamma ray
were incident from pixel j and interacted at the measured locations r1 and r 2.
In 2007, Xu and He introduced a different system response function, as in Equation
13.21 (for two-event sequences ending in absorption).
P( A|j) =

1 E0 d1 d c ( E0 )
1 E d2
e
e
sin
d E ( E2 )2
1
p ( E2 )
2 dE12 + dE22
( E0 E1 E2 )2 / 2 ( dE12 +d
dE22 )
(13.21)
In this function, Ei is the attenuation coefficient for energy Ei, di is the distance traveled in the detector before the ith interaction, and p(E) is the photoelectric cross
section at energy E. Xu and He termed this type of reconstruction energy-imaging
integrated deconvolution. In this method, both the initial gamma-ray energy and the
incident gamma-ray location are reconstructed together in the deconvolution step,
resulting in reconstruction of both the spatial and energy distributions of the radioactive source. The method was shown to preserve the correct gamma-ray emission ratios
for a 133Ba source, despite the different system response at each gamma-ray energy.
13.3.2.4Weighted MLEM
It is clear from the discussion that higher probabilities are assigned to measurements
in which the interaction distances are small. Furthermore, image pixels that lead to
smaller apparent Compton angles, where the KleinNishina formula is peaked over
the energy range of interest for Compton imagers, are also assigned a higher probability. Thus, the product fj P(A|j) is the forward projection and should produce the
measurements most likely to be observed. However, the most likely measurements
are also those with the most uncertainty. Small distances between interactions result
in very large geometric uncertainties. Forward-scattered gamma rays lose a small
fraction of their energies, leading to large energy uncertainties as well. Thus, the
reconstructed angular uncertainties for the most likely sequences will be large.
Ideally, the sequences leading to high angular uncertainty should be weighted
less than those that lead to low uncertainty. A more accurate knowledge of the source
location should be possible through such weighting. Using angular uncertainty to
weight sequences appropriately seems an obvious choice. Defining Yi as the inverse
of the estimated angular uncertainties di , which are calculated as the quadratic sum
of the energy and position components discussed previously, the weighted MLEM
algorithm then uses Equation 13.22 in place of Equation 13.18.

f j(t+1) =
f j(t )
Ts j
i=1
Yi p( Ai |j)
p( A |k ) f
i
(13.22)
(t )
k
k =1
The image must then be normalized to preserve absolute source intensity information, as in Equation 13.23.
300
( t +1)
j
Nf j(t )
Ts j
(d ) p( A |j)
p( A |k ) f
(d )
i
i=1
1 i=1
(13.23)
(t )
k
k =1
Equations 13.22 and 13.23 represent the weighted ML method,6 which demonstrated higher image resolution for a compact CdZnTe-based Compton imaging system. Shown in Figure13.7 are images generated using simple backprojection, filtered
backprojection, MLEM and weighted MLEM for the same set of 511-keV gammaray source data.
Iterative reconstruction calculations are time consuming. Block iterative or
ordered subset methods, when combined with traditional iterative image recon
struction algorithms, have shown accelerated convergence in SPECT image
reconstruction,22,23 but have not been traditionally applied to Compton imaging.
These methods differ from traditional MLEM in that only portions of the full data
are used at once. The resulting image from one subset is used as the starting value
for the next. As a result, fewer iterations (defined as a cycle through all the subsets)
are needed to achieve convergence. The result is nearly an order of magnitude gain in
a)
c)
b)
d)
Figure 13.7 Compton images generated using (a) simple backprojection, (b) filtered backprojection, (c) MLEM, and (d) weighted MLEM for the same set of 511-keV gamma-ray
source data.
Compton Imaging
301
computational speed in some implementations.22 Ordered subset methods have been

proved to converge for noiseless data but result in a repeating cycle of nonconverging
images for real data.
13.3.2.5Convergence and Early Termination
In practice, iterations of the MLEM algorithm are usually terminated before convergence is reached. Images resulting from convergence of the likelihood function are
noisy because the data creating them are noisy.24 Veklerov, Llacer, and Hoffman stated
that because of the inherent statistical processes involved in radiation emission and
measurement, a source distribution identical to the recovered image could not have
possibly generated the data.25 When the image reconstruction is iterated until the ML
solution is reached, the resulting image becomes unacceptably noisy. Typically, the
reconstruction process is stopped while the image is still relatively smooth.
One possible way to determine the proper number of iterations is to monitor the
progress of the reconstructed image with increasing iterations. When the image
begins to degrade rather than improve, the user could stop the reconstruction process. This is a dissatisfying method for early termination because it is user dependent. It is preferable to use a statistical or at least disciplined method for determining
the stopping point. One simple method involves simply terminating the reconstruction after a given number of iterations. However, the rate of convergence depends on
both the number of observed events in the detector and the spatial distribution of the
source.26 Therefore, using the same number of iterations for all source distributions
will yield widely variable results. It is necessary to determine an independent, casespecific stopping criterion. Several such methods have been proposed.
In 1988, Veklerov, Llacer, and Hoffman proposed a method based on the Poisson
nature of radiation measurements.25 After each iteration, the hypothesis that the observed
data are a Poisson sample drawn from the source distribution given in the image is tested.
If the hypothesis cannot be rejected, then the iterations are stopped. For a small number
of iterations, the image does not sufficiently fit the observed data; for a large number of
iterations, the image becomes too noisy to have generated the data. In this way, the reconstructed image passes through a region of feasible images.27 For computer-generated
Poisson random data, the authors observed that the iterative process consistently passes
through such a region. Choosing any of the images in the region is acceptable, although
with the test outlined, the first image to satisfy the criterion is chosen.
The authors observed for real PET imaging data that a feasible image was never
found. With the computer-generated data, the same system function is used to perform the generation and reconstruction of the data; this is not necessarily the case for
real imaging data. The system function used in the ML estimation has some uncertainty associated with it. This observation led to a weaker stopping criterion based
on the variance of the observed data and the reconstructed image.27 This weaker
stopping criterion was found to be more robust and led to acceptable images from
real PET data.
Another possible stopping criterion is to choose a solution near the ML solution.
Vardi, Shepp, and Kaufman28 suggested calculating the change in the likelihood
value between iterations and terminating the reconstruction when the difference is
302
small (i.e., convergence is near). Thus, a sort of diminishing returns approach can
be used.
Kontaxakis and Tzanakos proposed a different stopping criterion based on the
factor that is multiplied by fj(i) in Equation 13.23.29 This factor approaches 1.0 as
the image approaches the ML estimate. The authors stated that a value of 0.8 produces images that are close to the optimal image, independently of image shape, the
number of counts in the image, and the system configuration. The optimal image is
defined as having the minimum root mean square (rms) error and minimizing a 2
distribution similar to that used by Veklerov.
A reconstruction method that maximizes the likelihood and then terminates the
iterations before that ML solution is obtained can be conceptually difficult to accept,
regardless of the way in which the stopping point is chosen.24 The ML image does
not depend on the choice of initial source estimate, but if the reconstruction is halted
early, the independence will no longer hold. As a result, there is still much debate
about the use of ML estimators for reconstructing Compton imaging, emission
tomography, and PET imaging data. Using a stopping criterion with MLEM may not
be the theoretically optimal reconstruction method, but it has been shown via simulations and experiments to produce low-noise, high-resolution images.
13.4Compton Imaging Systems

13.4.1Compton Imaging Metrics
Compton imagers are typically described by three metrics: imaging efficiency, angular resolution, and field of view. Each requires some attention. Metrics are intended
to provide a quantitative assessment of the performance of a given Compton imager
under given conditions. Unfortunately, no standards exist for reporting Compton
imaging performance, and both the experimental conditions and the metrics themselves vary widely in the literature.
13.4.1.1Compton Imaging Efficiency
The Compton imaging efficiency can be defined in terms of the various processes
that lead to rejection of gamma-ray sequences. First, the gamma-ray must interact
in at least two locations within the detector. Some gamma rays will pass through the
detector unscathed, and any single events are discarded; obviously, these gamma
rays are not useful for Compton imaging. For two or more measured events, the
interactions must be properly sequenced. The full gamma-ray energy is also usually required to be deposited in the detector. (Another efficiency loss arises when
the image reconstruction surface does not fully encompass the source object, but
in medical imaging [or 4 field-of-view systems, for that matter], this situation is
never encountered.) Thus, the actual Compton imaging efficiency is given by a combination of detection efficiency, multiple-site efficiency, full-energy efficiency, and
sequence reconstruction efficiency. None of these efficiencies is independent of the
others, and all are dependent on the incident gamma-ray energy.
Detection efficiency can be measured using standard methods for medical imaging systems. Also, the fraction of detected gamma rays resulting in multisite or
Compton Imaging
303
full-energy sequences can be determined experimentally. However, it is impossible

to measure sequence reconstruction efficiency empirically. The correct sequence cannot be determined, and thus the fraction correctly sequenced is unknown for any
given measurement; this value must be determined through radiation transport simulations. In practice, the Compton imaging efficiency is wholly determined through
simulations. Imaging efficiencies vary widely depending on detector system design.
A close-packed pixellated CdZnTe detector array can image between 10% to 20% of
detected gamma rays between 100 and 3000 keV33, resulting in an expected intrinsic
imaging efficiency of 5% to 8% for low energies. COMPTEL, by comparison, used
two detector arrays spaced 1.5 m apart to achieve high angular resolution, which
limited the imaging efficiency to less than 0.01%.
To maximize intrinsic imaging efficiency, a Compton imager should have closely
packed detector arrays with a high probability of initial scatter and final absorption.
To achieve a high likelihood of scatter at energies as low as 140 keV, it is necessary
to select a detector material with low atomic number (Z). However, low-Z materials
are also poor absorbers. To compromise between the need for both gamma-ray scatter and absorption, one design uses a Si detector front plane detector with a Ge back
plan.34 According to simulations, the Si-Ge hybrid imager should achieve a maximum combined efficiency of 6% detection, multi-site interaction, and full-energy
deposition at 140 keV. No data were provided on sequencing efficiencies, but a value
of 60% to 80% is reasonable at all energies, which would result in an intrinsic imaging efficiency of 4% to 5%.
13.4.1.2Angular Resolution
There are several metrics used to measure angular resolution of Compton imaging
systems. In this chapter, the uncertainty (as a one-sigma value) was used to estimate
the angular uncertainty for a given Compton cone. However, in light of the discussion of Doppler broadening and the non-Gaussian nature of the angular uncertainty,
a one-sigma metric is insufficient to fully describe the angular resolution performance of Compton imaging systems.
Angular resolution is often quoted as a FWHM of the point-source response function (PSF). In a Gaussian distribution, 68% of the distribution is contained within the
FWHM. For the non-Gaussian distributions from Doppler broadening, a 68% containment value can also be reported.30 For the simple backprojection image shown in
Figure13.7, the 68% containment value is nearly twice the value of the FWHM. The
choice of uncertainty metric is somewhat arbitrary, but it is clear from the Doppler
broadening discussion in Section 13.2 that the FWHM will underestimate the blur
in the image.
Image resolution is generally described in terms of the PSF. To the extent that
image reconstruction is a linear process, metrics drawn from the PSF are a good
indication of the performance of Compton imagers for extended source objects, as in
medical imaging. However, in iterative image reconstruction methods, the assumption of linearity is no longer valid, and superposition of PSFs will not approximate the
image obtained from extended sources.
To achieve the best angular resolution performance, a Compton imaging system
must have both excellent energy resolution and good position resolution relative to
304
average distance between interactions. The hybrid Si-Ge Compton imaging system
described above is one such system and has an expected angular resolution of 34
FWHM. By comparison, CGRO had a very low <delta>r/r ratio (indicating high
position resolution) but relatively poor energy resolution.
None of the resolution metrics described in this section can be used to evaluate
utility for medical imaging applications. Rather, the metrics are used to comparatively assess different imaging detector designs. To quantitatively evaluate imaging
performance for medical imaging, one must use traditional image quality metrics,
such as modulation transfer function, detective quantum efficiency, and noise power
spectrum. The reader is referred to image quality texts for more information on this
topic.
13.4.1.3Field of View
Compton imagers can have an unlimited field of view. In these systems, called 4
imagers, the image surface is typically a sphere (with the detector system at its
center), and the image is reconstructed as an angular map of the sky, as in astronomy
applications. Usually, there is some variation in sensitivity for sources at different
orientations due to the inevitable anisotropy of the detector system.
The field of view for Compton imagers may be limited by either the design of
the system or the choice of the operator. Medical systems are a good example of the
latter: The direction to the patient is always known, and any gamma rays appearing
to originate from other directions can be ignored. Two-plane Compton imagers are
often limited to a forward field of view within 60 of the normal to the face of the
front plane. At more oblique angles, the probability of a scattered gamma ray being
absorbed in the second detector decreases significantly, and there is low likelihood
of observing full-energy 2+-event sequences from these gamma rays.
13.4.2Detector Requirements, Composition, and Design

The first working prototype Compton imager was reported in 1983 by Singh and
Doria from the University of Southern California.37 Three-dimensional tomographic
and 2-D planar images were reconstructed using a cylindrical 6 mm diameter
6 mm thick high-purity germanium (HPGe) front-plane detector separated by 5 cm
from an uncollimated NaI(Tl) scintillation detector in the back plane. The initial
gamma-ray energy was known and used to discard partial energy deposition events.
A 9 full-width at half-maximum (FWHM) angular resolution can be inferred based
on data published in the manuscript. In 1988, Brechner and Singh used simulations
to compare the above Compton imaging (electronically collimated) system with
mechanically collimated systems.38 They reported a simulated factor of 20 gain in
sensitivity compared with a cone-beam collimated gamma camera (which has better sensitivity than a traditional parallel-hole collimated camera).
Since 1983, advances in position-sensitive gamma-ray spectroscopy systems have
enabled development of a wide variety of Compton imager designs and applications.
The choice in detector materials and system design is influenced by overall cost,
size and efficiency requirements, tolerable electronics complexity, and desired energy
Compton Imaging
305
and angular resolution over the gamma-ray energy range of interest. In this section,
the design tradeoffs are discussed for various gamma-ray detection technologies.
13.4.2.1Scintillators
The key challenge in using scintillators for Compton cameras is obtaining sufficient position resolution within the scintillation material to confidently determine
the gamma-ray interaction location. COMPTEL used large monolithic detectors in
which the interaction position within each scintillator could not be known. More
recent advances in scintillator technology enable determination of the interaction
position through electrical and optical means. In the former case, position-sensitive
photomultiplier tubes (PMTs) can use Anger logic or multiple PMT anodes to determine the position on the photocathode at which most of the light arrives. This position corresponds to the gamma-ray event location in two dimensions. Determining
the third dimension (depth in the crystal) is the focus of much research in medical
imaging systems.3941 It is also possible to use optical means to determine event
position. For example, scintillation detectors can be grown in narrow columns with
minimal optical coupling between columns within a single detector. This geometry
minimizes the spreading of scintillation light in the lateral direction and, coupled
with a position-sensitive photodetector, provides a more accurate determination of
interaction position. Typically, these systems exhibit poorer energy resolution than
monolithic scintillator crystals due to the photon losses at the column surfaces.
Position resolution in scintillators is typically limited to several millimeters or
more. This resolution is poor compared to achievable resolution in semiconductor
detectors. Also, scintillators exhibit poorer energy resolution than semiconductors in
general. Thus, angular resolution performance of scintillator-based Compton cameras is generally poor, especially for low-energy gamma rays.
COMPTEL is a primary example of a Compton imaging system using monolithic scintillator detectors. This Compton telescope was selected to fly on NASAs
Gamma Ray Observatory (later referred to as the Arthur Holly Compton Gamma
Ray Observatory, or simply the CGRO).42 Schonfelder, who helped develop the
first (non-imaging) Compton telescope, led the international team of scientists that
designed COMPTEL, shown in Figure 13.9. The front-plane detector was an array
of seven liquid scintillator cells coupled to photomultiplier tubes (PMTs). The back
plane detector consisted of 14 NaI(Tl) crystals also coupled to PMTs. The observed
energy resolution was 12% FWHM at 511 keV and 6% FWHM at 2.75 MeV. Due to
the large separation distance between detectors, the angular resolution of the system was at or below 4.7 FWHM for gamma-ray energies between 1 and 10 MeV.
COMPTEL was not intended for imaging low-energy gamma rays relevant to medical imaging.
13.4.2.2Si Drift Detectors
As semiconductor detectors improved throughout the 1980s and 1990s, more
researchers looked to silicon, germanium, and CdZnTe detectors as possibilities for
Compton imaging.
Thin silicon strips can be used to track the recoil electrons from Compton scatter
interactions, as shown in Figure13.8.
306

Compton
Event
Circle
Event
Arc
Incident
Gamma
Rays
Pair
Event
Silicon
Strip
Detector
Converter/
Tracker
Scattered
Electron
Scattered
GammaCsI
Calorimeter
Figure 13.8 Electron tracking with silicon strip detectors. (Reproduced from Bhattacharya,
D. et al., IEEE Trans. Nucl. Sci., 42, 4, 1995. With permission from Elsevier.)50
COMPTEL
GAMMA RAY
Gamma Ray scattered;
light emitted
Light recorded
Gamma Ray absorbed,

light pulse emitted
and recorded
Figure 13.9 A schematic of the COMPTEL telescope flown on the Compton GammaRay Observatory. (Courtesy of the COMPTEL collaboration; available at http://heasarc.gsfc.
nasa.gov/docs/cgro/images/epo/gallery/cgro/index.html.)
Compton Imaging
307
13.4.2.3Double-Sided Silicon or Germanium Strip Detectors

Double sided strip detectorstypically silicon or germanium semiconductorsuse
orthogonal strips on two flat sides of the crystal to determine the lateral coordinates
of gamma-ray interactions. The depth of interaction can then be determined by differences in the drift times to the two sets of electrodes. Strip detectors enable 3-D
determination of events to within a line pitch (or better43) using a relatively small
number of readout channels. For a detector area A and number of strips per side
N, a position resolution of A/N2 requires only 2N readout channels. The primary
drawback of double-sided strip detectors is the relatively small detector thickness
(typically on the order of 1 cm) compared with scintillation detectors, a limitation
that is typically overcome by stacking detectors. Also, double sided strip detectors
are often used as a front plane or scatter detector; a high-efficiency scintillator is
then selected as the back plane or absorption detector to achieve a higher imaging
efficiency than possible with only Si or Ge strip detectors.
The Tracking and Imaging Gamma-Ray Experiment (TIGRE), developed by the
University of California at Riverside in 1995,44 used seven 3.2 cm 3.2 cm 300 mm
Si strip detectors stacked to provide 2 mm of scatter plane thickness and 36 CsI(Tl)
scintillators with dimensions 1 cm 1 cm 1.7 cm for the rear absorber layer. The
device was designed for astronomical imaging in the range of 0.3 to 100 MeV. The
group proposed to track the recoil electron through the Si detector layers. Knowing
the Compton scatter angle and the recoil electron direction eliminates the azimuthal
uncertainty in the backprojected image, resulting not in a cone but in a ray. Using
up versus down electron momentum determination, they measured a 10 FWHM
angular resolution at 900 keV. At 511 keV, the measured resolution was approximately 11 FWHM.
In early 2004, Zhang, Rogers, and Clinthorne from the University of Michigan
proposed a Compton imager for scintimammography.45 The camera is a dual-head
imager consisting of two planes of 1-mm thick Si arrays 10 cm 10 cm placed
directly above and below the breast followed by two 40 cm 40 cm 2 cm thick
NaI(Tl) detectors. Using a realistic simulated anthropomorphic phantom, they estimate an absolute efficiency of several percent and a spatial resolution of several
millimeters for gamma-ray energies up to 511 keV, depending on the depth of the
tumor in the breast. They demonstrate better confidence in detecting 5 mm tumors
using Compton imaging compared with traditional scintimammography.
13.4.2.4Segmented Coaxial Germanium
Segmented coaxial germanium detectors can also be used for Compton imaging.46
Coaxial germanium detectors provide good efficiency and high energy resolution for
high-energy gamma rays. However, at energies of interest to medical imaging, the
position resolution in these detectors is on the order of centimeters, rather than milli
meters, resulting in few sequences with multiple detected interaction sites as well as
relatively high angular uncertainties. Using segmented coaxial germanium detectors for imaging low-energy gamma rays results in low-efficiency and low-resolution
Compton images.
308
13.4.2.5Pixellated CdZnTe
Pixellated semiconductor detectors can provide excellent position resolution. In these
systems, the anode is divided into square pads typically ~1 mm in size, which is large
enough to contain the average electron charge cloud of gamma-ray interactions up
to 1 MeV. In some systems, the anode pixels are separated by a steering grid electrode that is biased to a negative voltage in order to steer electrons away from the
gap between pixels and toward the anode pixels themselves. Pixellated detectors
are typically read out using multi-channel application-specific integrated circuits.
The readout of many parallel channels adds complexity to the system, but such technologies have been well established in high-energy physics applications.
Pixellated CdZnTe has an advantage over germanium-based spectrometers in that
it can achieve 1% FWHM energy resolution at 662 keV without cryogenic cooling.47
It also has higher stopping power due to its higher average atomic number and density. However, a higher stopping power also translates to lower scatter probabilities
at low gamma-ray energies, such as those used in medical imaging.
Du et al. developed the first Compton imager prototype composed of two 1-cm3
CdZnTe detectors with three-dimensional position sensitivity in 2001.48 Each detector had an array of 121 pixel anodes with a steering grid electrode. The detectors
were separated by 5 cm, and the field-of-view was limited to scatter angles between
20 and 80. Only gamma rays that scattered in the first detector and were absorbed
in the second detector were used for imaging. The measured intrinsic efficiency at
662 keV was only 1.5 10 4, as a result of the small back plane detector size compared
with the distance between detectors. The imager had a measured angular resolution
of about 5 FWHM at that energy.49 By 2004, Compton imaging had been demonstrated in a single pixellated CdZnTe detector, and by 2008 this technology was
extended to arrays of detectors acting in concert, as shown in Figure 13.10. The
CdZnTe array systems can achieve better than 10 FWHM angular resolution and an
intrinsic imaging efficiency between 56% at 662 keV.33
13.4.2.6Gas and Liquid Time Projection Chambers
A time projection chamber (TPC) is a gas- or liquid-filled device with charge readout. An electric field is established in the chamber to drift electrons and ions resulting
from gamma-ray interactions. Often a magnetic field is also applied to limit diffusion of the electrons. Multi-wire readout enables determination of the event location
in two dimensions, and the third coordinate is determined from the electron drift
time. Gas-electron multipliers, pixel electrodes, or other structures can also provide
three-dimensional readout. The electronics used to read out multi-wire TPCs is similar to that of double-sided strip detectors. Each wire has its own readout channel. The
primary advantage of time projection chambers is the achievable instrument size;
high-energy physics experiments have used TPCs tens of meters in size.
In 2003, Orito et al. proposed using a TPC filled with gaseous xenon as the scatter detector and surrounding scintillators for detecting the gamma-ray absorption.50
The TPC has pixel electrodes only 400 mm in size, and thus it is possible to track
the recoil electron from Compton scatter processes, similar to the UC Riverside
silicon strip detector design. They estimate between 1% and 0.1% intrinsic efficiency
Compton Imaging
309
Figure 13.10 Pixellated CdZnTe Compton imaging array.
between 100 keV and 2 MeV and an angular resolution on the order of 15, although
the performance has not been measured. The major drawback of gaseous TPCs is
the low interaction efficiency. Xenon pressurized to 50 atm has a density of only
0.5 g/cm3, an order of magnitude lower than semiconductor detector materials.
To achieve higher gamma-ray sensitivity, Aprile et al. proposed the first liquid
xenon Compton imager, composed of a 20 cm 20 cm 7 cm (active volume)
time-projection chamber enclosed in a cylindrical pressure chamber.51 This liquid
xenon system achieved a density of 3 g/cm3. Another proposed design contained
two large area (2500 cm2) time-projection chambers in a single pressure vessel.52
With a 1s energy resolution of 2.5% at 1 MeV and 1s position resolution of 1 mm,
the expected 1s angular resolution varied from 0.4 to 2.4 over the 120 field-of-view.
High-resolution imaging is possible with TPCs, but efficiency suffers greatly compared with semiconductor and scintillator detectors.
13.5Limitations of Compton Imagers

and New Opportunities
Despite the recent successes in laboratory demonstrations, Compton imaging systems have not been widely deployed for medical imaging. Some technological progress is needed to make Compton imaging a viable imaging modality for medical
diagnostics. First, the requirement for high-resolution position sensitivity results
in a large number of readout channels in Compton imaging detectors. Advances in
multichannel integrated circuits are needed to achieve low-noise, high-throughput
readout of thousands of channels simultaneously. Fortunately, this development is
310
paralleled in the high-energy physics community, which commonly instruments

thousands of readout channels for large collider experiments. Several of the systems discussed in Section 13.4 are advancing the state of the art in multichannel
readout systems, and further progress in this technology is expected to continue.
Second, the traditionally meager yield obtained in the fabrication of highquality, position-sensitive detectors, coupled with extensive testing and characterization of these detectors, serves to increase the sensor cost exorbitantly for
any large-area imager. Crystal growth methods that produce high-quality detector
material and fabrication processes that yield low-noise contacts and low-leakage
surfaces are needed to drive down the materials costs inherent in any Compton
imaging system.
Third, image reconstruction that takes advantage of parallel processing environments may result in significantly faster image reconstruction algorithms. The
iterative image reconstruction methods described require a significant amount of
memory and processing power. To the extent that the reconstruction of each pixel
value is independent of other pixels in the image, the calculations can be distributed
among a network of processors all acting independently.
Finally, Compton imaging systems that can also operate in Anger mode or enable
PET reconstruction will find a wider audience. Significant research is already focusing on combined PET-CT systems that can perform both transmission and emission
tomography using the same imaging detectors. Critical advancements in high-rate
spectroscopic detector technology have enabled the development of prototype PET-CT
systems. The same advancements applied to multimodal Compton imagers could generate a new crop of medical imaging systems with higher diagnostic capabilities.
References

1. R. Cesareo, A. L. Hanson, G. E. Gigante, L. J. Pedraza, S. Q. G. Mahtaboally, Interaction

of keV photons with matter and new applications. Phys. Rep., 213(3), 117178, 1992.
2. R. Ribberfors, K.-F. Berggren, Incoherent x-ray scattering functions and cross sections
by means of a pocket calculator. Phys. Rev. A, 26(6), 33253333, 1982.
3. F. Biggs, L. B. Mendelsohn, J. B. Mann, Hartree-Fock Compton profiles for the elements. Atomic Data Nucl. Data Tables, 16, 201309, 1975.
4. V. Schonfelder, A. Hirner, K. Schneider, A telescope for soft gamma ray astronomy.
Nucl. Instrum. Methods, 107, 385394, 1973.
5. G. F. Knoll, Radiation Detection and Measurement. 3rd ed. New York: Wiley, 2000.
6. C. E. Lehner [Seifert], Z. He, F. Zhang, 4-Pi Compton imaging using a 3-D positionsensitive CdZnTe detector via weighted list-mode maximum likelihood. IEEE Trans.
Nucl. Sci., 51(4), 16181624, 2004.
7. G. J. Schmid, I. Y. Lee, M. A. Deleplanque, P. Hendriks, F. S. Stephens, K. Vetter, S. J.
Asztalow, R. M. Clark, R. M. Diamond, P. Fallon, R. Kruecken, A. O. Macchiavelli,
R. W. MacLeod, Gamma-ray cluster identification in a spherical shell of highly segmented germanium detectors. IEEE Trans. Nucl. Sci., 44(3), 975978, 1997.
8. I. Y. Lee, Gamma-ray tracking detectors. Nucl. Instrum. Methods Phys. Res., A422,
195200, 1999.
9. U. G. Oberlack, E. Aprile, A. Curioni, V. Egorov, K. L. Giboni, Compton scattering
sequence reconstruction algorithm for the liquid xenon gamma-ray imaging telescope
(LXeGRIT). SPIE Proc., 4141, 168177, 2000.
Compton Imaging
311
10. S. E. Boggs, P. Jean, Event reconstruction in high resolution Compton telescopes.

Astron. Astrophys. Suppl. Series, 145, 311321, 2000.
11. R. A. Kroeger, W. N. Johnson, J. D. Kurfess, B. F. Phlips, E. A. Wulf, Three-Compton
telescope: theory, simulations, and performance. IEEE Trans. Nucl. Sci., 49(4),
18871892, 2002.
12. C. J. Pearson, J. J. V. Dobon, P. H. Regan, P. J. Sellin, E. Morton, P. J. Nolan, A. Boston,
M. Descovich, J. Thornhill, J. Cresswell, I. Lazarus, J. Simpson, Digital gamma-ray
tracking algorithms in segmented germanium detectors. IEEE Trans. Nucl. Sci., 49(3),
12091215, 2002.
13. S. J. Wilderman, W. L. Rogers, G. F. Knoll, J. C. Engdahl, Fast algorithm for list mode
back-projection of Compton scatter camera data. IEEE Trans. Nucl. Sci., 45(3), 957962,
1998.
14. R. C. Rohe, M. M. Sharfi, K. A. Kecevar, J. D. Valentine, C. Bonnerave, The spatiallyvariant backprojection point kernel function of an energy-subtraction Compton scatter
camera for medical imaging. IEEE Trans. Nucl. Sci., 44(6), 24772482, 1997.
15. H. H. Barrett, T. White, L. C. Parra, List-mode likelihood. J. Opt. Soc. Am., A14(11),
29142923, 1997.
16. S. J. Wilderman, N. H. Clinthorne, J. A. Fessler, W. L. Rogers, List-mode likelihood
reconstruction of Compton scatter camera images in nuclear medicine. IEEE Nucl. Sci.
Symp. Conf. Record, 1999, 17161720, 1999.
17. L. Parra, H. H. Barrett, List-mode likelihood: EM algorithm and image quality estimation demonstrated on 2-D PET. IEEE Trans. Nucl. Sci., 17(2), 228235, 1998.
18. A. P. Dempster, N. M. Laird, D. B. Rubin, Maximum likelihood from incomplete data
via the EM algorithm. J. R. Stat. Soc., B39(1), 138, 1977.
19. L. A. Shepp, Y. Vardi, Maximum likelihood reconstruction for emission tomography.
20. K. R. Lange, E. M. Carson, Reconstruction algorithms for emission and transmission
tomography. J. Comp. Assisted Tomog., 8(2), 306316, 1984.
21. S. J. Wilderman, J. A. Fessler, N. H. Clinthorne, J. W. LeBlanc, W. L. Rogers, Improved
modeling of system response in list mode EM reconstruction of Compton scatter camera
images. IEEE Trans. Nucl. Sci., 48(1), 111115, 2001.
22. H. M. Hudson, R. S. Larkin, Accelerated image reconstruction using ordered subsets of
projection data. IEEE Trans. Med. Imaging, 13(4), 601609, 1994.
23. C. L. Byrne, Accelerating the EMML algorithm and related iterative algorithms by rescaled block-iterative methods. IEEE Trans. Image Proc., 7(1), 100109, 1998.
24. L. Shepp, From convolution algorithms to maximum likelihood. IEEE Nucl. Sci. Symp.
Conf. Record, 14411444, 1990.
25. E. Veklerov, J. Llacer, E. J. Hoffman, MLE reconstruction of a brain phantom using a
Monte Carlo transition matrix and a statistical stopping rule. IEEE Trans. Nucl. Sci.,
35(1), 603607, 1988.
26. T. Hebert, R. Leahy, M. Singh, Fast MLE for SPECT using an intermediate polar representation and a stopping criterion. IEEE Trans. Nucl. Sci., 35(1), 615619, 1988.
27. J. Llacer, E. Veklerov, Feasible images and practical stopping rules for iterative algorithms in emission tomography. IEEE Trans. Nucl. Sci., 8(2), 186193, 1989.
28. Y. Vardi, L. A. Shepp, L. Kaufman, A statistical model for positron emission tomography (with discussion). J. Am. Stat. Assoc., 80, 8:37, 1985.
29. G. N. Kontaxakis, G. S. Tzanakos, Practical application of a new stopping criterion for the
EM-MLE image reconstruction for PET. Proceedings of the 18th Annual International
Conference of the IEEE Engineering in Medicine and Biology Society, 848849, 1996.
30. E. Novikova, Testing LECS: Doppler broadening extension of Geant4. Unpublished,
pp. 129, 2003.
312
31. Y. Du, Z. He, G. F. Knoll, D. K. Wehe, W. Li, Evaluation of a Compton scattering camera using 3-D position sensitive CdZnTe detectors. Nucl. Instrum. Methods Phys. Res.,
A457, 203211, 2001.
32. Y. Du, Development of a prototype Compton scattering camera using 3-D position sensitive CZT detectors, in Nuclear Engineering and Radiological Sciences. Ann Arbor:
University of Michigan, p. 137, 2001.
33. C. E. Seifert, M. J. Myjak, and D. V. Jordan, Simulated performance of the
GammaTracker CdZnTe handheld radioisotope identifier, IEEE Nucl. Sci. Symp. Conf.
Record, 940044, 2007.
34. K. Vetter, M. Burks, C. Cork, M. Cunningham, D. Chivers, E. Hull, T. Krings, H. Manini,
L. Mihailescu, K. Nelson, D. Protic, J. Valentine, D. Wright, High-sensitivity Compton
imaging with position-sensitive Si and Ge detectors, Nucl. Instrum. Methods Phys. Res.,
A579, 363366, 2007.
35. E. Aprile, V. Egorov, K.-L. Giboni, S. M. Kahn, T. Kozu, U. Oberlack, S. Centro, S.
Ventura, T. Doke, J. Kikuchi, E. L. Chupp, P. P. Dunphy, D. H. Hartmann, M. D. Leising,
H. Bloeman, XENAa liquid xenon Compton telescope for gamma-ray astrophysics in
the MeV regime, SPIE Proc., 3446, 8899, 1998.
36. J. Bernabeu, N. H. Clinthorne, Y. Dewaraja, C. Lacasta, G. Llosa, M. Mikuz, S. Roe, W.
L. Rogers, A. Studen, P. Weilhammer, L. Zhang, D. Zontar, Development of a high efficiency and high resolution Compton probe for prostate imaging, Nucl. Instrum. Methods
Phys. Res., A527, 5861, 2004.
37. M. Singh, D. Doria, An electronically collimated gamma camera for single photon emission computed tomography. Part II: Image reconstruction and preliminary experimental
measurements, Medical Physics, 10(4), 428435, 1983.
38. R. R. Brechner, M. Singh, Comparison of an electronically collimated system and a
mechanical cone-beam system for imaging single photons, IEEE Trans. Nucl. Sci.,
33(1), 649653, 1988.
39. R. S. Miyaoka, T. K. Lewellen, H. Yu, D. L. McDaniel, Design of a depth of interaction
(DOI) PET detector module, IEEE Trans. Nucl. Sci., 45(3), 10691073, 1998.
40. A. Saoudi, C. M. Pepin, F. Dion, M. Bentourkia, R. Lecomte, M Andreaco, M. Casey, R.
Nutt, H. Dautet, Investigation of depth-of-interaction by pulse shape discrimination in
multicrystal detectors read out by avalanche photodiodes, IEEE Trans. Nucl. Sci., 46(3),
462467, 1999.
41. K. C. Burr, A. Ivan, D. E. Castleberry, J. W. LeBlanc, K. S. Shah, R. Farrel, Evaluation
of a prototype small-animal PET detector with depth-of-interaction encoding. IEEE
Trans. Nucl. Sci., 51(4), 17911798, 2004.
42. V. Schonfelder, R. Diehl, G. G. Lichti, H. Steinle, B. N. Swanenburg, A. J. M. Deerenberg,
H. Aarts, J. Lockwood, W. Webber, J. Macri, J. Ryan, G. Simpson, B. G. Taylor, K.
Bennett, M. Snelling, The imaging Compton telescope on the Gamma Ray Observatory,
IEEE Trans. Nucl. Sci., NS-31(1), 766770, 1984.
43. M. Burks, E. Jordan, E. Hull, L. Mihailescu, K. Vetter, Signal interpolation in germanium
detectors for improved 3-D position resolution, IEEE Nucl. Sci. Symp. Conf. Record, 2,
11141118, 2004.
44. T. J. ONeill, D. Bhattacharya, S. Blair, G. Case, O. T. Tumer, R. S. White, A. D. Zych,
The TIGRE desktop prototype results for 511 and 900 keV gamma rays, IEEE Trans.
Nucl. Sci., 42(4), 933939, 1995.
45. L. Zhang, W. L. Rogers, N. H. Clinthorne, Potential of a Compton camera for high performance scintimammography, Physics in Medicine and Biology, 49, 617638, 2004.
46. G. J. Schmid, D. A. Beckedahl, J. E. Kammeraad, J. J. Blair, K. Vetter, A Kuhn, Gammaray Compton camera imaging with a segmented HPGe, Nucl. Instrum. Methods Phys.
Res., A459, 565576, 2001.
Compton Imaging
313
47. F. Zhang, Z. He, G. F. Knoll, D. K. Wehe, J. E. Berry, and D. Xu, Improved resolution
for 3-D position-sensitive CdZnTe spectrometers, IEEE Trans. Nucl. Sci., in publication,
2004.
48. Y. Du, Z. He, G. F. Knoll, D. K. Wehe, W. Li, Evaluation of a Compton scattering camera using 3-D position sensitive CdZnTe detectors, Nucl. Instrum. Methods Phys. Res.,
A457, 203211, 2001.
49. Y. Du, Development of a prototype Compton scattering camera using 3-D position sensitive CZT detectors, in Nuclear Engineering and Radiological Sciences, University of
Michigan: Ann Arbor, 137, 2001.
50. R. Orito, H. Kubo, K. Miuchi, T. Nagayoshi, A. Takada, T. Tanimori, M. Ueno, A
novel design of the MeV gamma-ray imaging detector with Micro-TPC, Nucl. Instrum.
Methods Phys. Res., A513, 408412, 2003.
51. E. Aprile, V. Egorov, K.-L. Giboni, T. Kozu, F. Xu, T. Doke, J. Kikuchi, T. Kashiwagi,
G. J. Fishman, R. Rehage, D. Trice, The electronics readout and data acquisition system
for a liquid xenon time projection chamber as a balloon-borne Compton telescope, Nucl.
Instrum. Methods Phys. Res., A412, 425436, 1998.
52. E. Aprile, V. Egorov, K.-L. Giboni, S. M. Kahn, T. Kozu, U. Oberlack, S. Centro, S.
Ventura, T. Doke, J. Kikuchi, E. L. Chupp, P. P. Dunphy, D. H. Hartmann, M. D. Leising,
H. Bloeman, XENAA liquid xenon Compton telescope for gamma-ray astrophysics
in the MeV regime, SPIE Proc., 3446, 8899, 1998.
53. C. E. Ordonez, W. Chang, A. Bolozdynya, Angular uncertainties due to geometry and
spatial resolution in Compton cameras, IEEE Trans. Nucl. Sci., 46, 11421147, 1999.
14 Multimodality
Imaging with MR/PET
and MR/SPECT
Troy Farncombe
Hamilton Health Services

Hamilton, Canada
Contents
14.1 Multimodality Imaging Systems................................................................... 316
14.1.1 Radionuclide Transmission Imaging................................................. 317
14.1.2 CT-Based Coregistration................................................................... 317
14.1.3 Image Coregistration......................................................................... 318
14.1.4 Image Display.................................................................................... 318
14.1.5 CT Attenuation Correction................................................................ 319
14.1.6 Clinical Applications of PET/CT and SPECT/CT............................. 320
14.1.7 Problems with CT-Based Coregistration........................................... 321
14.1.8 Requirements for PET/MR and SPECT/MR..................................... 322
14.2 Basics of PET and SPECT Imaging.............................................................. 323
14.2.1 Scintillation Detectors....................................................................... 323
14.2.2 Photodetectors................................................................................... 324
14.2.3 Acquisition Electronics...................................................................... 327
14.3 Imaging with MRI......................................................................................... 332
14.3.1 Basics of MRI.................................................................................... 332
14.4 Alternative MR Configurations..................................................................... 333
14.4.1 Low-Field MR................................................................................... 333
14.4.2 Field-Cycled MR............................................................................... 333
14.5 Alternative Radiation Detectors.................................................................... 334
14.5.1 Scintillation Detectors....................................................................... 334
14.5.2 Fiberoptic Coupling........................................................................... 334
14.5.3 Photodiodes....................................................................................... 335
14.5.3.1 Avalanche Photodiodes....................................................... 336
14.5.3.2 Multipixel Geiger-Mode Avalanche Photodiodes............... 342
14.5.4 Solid-State Detectors......................................................................... 343
14.5.4.1 Si(Li) and Ge(Li) Detectors................................................344
14.5.4.2 CdTe and CdZnTe Detectors...............................................344
315
316
14.6 Other Considerations.....................................................................................346

14.6.1 Magnetic Compatibility.....................................................................346
14.6.2 System Design...................................................................................348
14.6.3 MR Attenuation Correction............................................................... 349
14.7 Conclusions.................................................................................................... 350
References............................................................................................................... 352
14.1Multimodality Imaging Systems

Since its inception, nuclear medicine imaging has always been known as a functional
imaging technique rather than an anatomical imaging method. Three-dimensional
(3D) images obtained from single-photon emission computed tomographic
(SPECT) or positron emission tomographic (PET) devices depict the accumulation
of radiopharmaceutical by specific cells, thus representing cellular function (see
Figure 14.1a). As a result, these images lack the spatial information and instant
recognition of other medical imaging techniques, such as X-ray computed tomography (CT) or magnetic resonance imaging (MRI). The underlying information
conveyed to an educated reader, however, is substantial, since the distribution of
radiopharmaceutical within the body provides physiological information about the
body rather than strictly anatomical information. Nuclear medicine images often
provide an early glimpse into disease progression, since physiological changes
occur prior to structural changes appearing. Even so, nuclear medicine has long
Figure 14.1 Typical whole-body PET scan showing the distribution of 18F. fluorodeoxyglucose (left). A radionuclide transmission scan using Cs-137 (right). Note the absence of
contrast between soft tissues.
Multimodality Imaging with MR/PET and MR/SPECT
317
been described as unclear medicine because images typically lack any readily
identifiable landmark information, such as bones or lungs. In fact, an ideal nuclear
medicine image should contain no anatomical information whatsoever but rather
depict the highly specific uptake of radiotracer in certain cell types. Admittedly, the
lack of structural information may introduce some difficulties in localizing disease
or framing the extent of disease. To provide some measure of anatomical structure and to correlate the radiopharmaceutical distribution with anatomy, numerous
methods have been used.
14.1.1Radionuclide Transmission Imaging

Among the first methods used to obtain anatomical information along with functional information was radionuclide transmission measurements.14 This type of
acquisition places a radioactive source opposite the detector and rotates it around the
body with the radiation detector measuring the resultant transmitted radiation. By
also measuring a blank transmission scan (i.e., no patient) and using the relation

dl
I = I 0e i
it is possible to reconstruct the 3D distribution of linear attenuation coefficients in the

patient. Because the detectors operate in photon-counting mode rather than in current
mode, images produced using transmission-based imaging suffer from poor spatial resolution and poor tissue contrast due to the low number of detected photons. In essence,
images made using radionuclide transmission can distinguish soft tissue from air but
little else (see Figure14.1b). This is reasonably acceptable for performing attenuation
compensation but generally is not of much value for accurately localizing disease.
The shortcomings of radionuclide transmission-based imaging has led to it being
relegated to perform attenuation correction only. In fact, until the early 2000s, most
PET scanners still used a Germanium-68 or Cesium-137 source for attenuation
correction, and, even now, several SPECT machines offer radionuclide transmission sources (typically Gadolinium-153) for attenuation correction.
14.1.2CT-Based Coregistration
Because of the limitations of radionuclide transmission imaging, many investigators turned to combining X-ray CT images with nuclear medicine PET or SPECT
images.5,6 Typically, PET or SPECT imaging would be performed on one system,
followed by CT imaging on another system, often in a different physical location.
Care has to be taken to image the patient in the same orientation on both systems, and often external fiducial markers are used to aid the coregistration process.
Once acquired, PET or SPECT images would be matched to CT images by shifting
the radionuclide images in three dimensions until they aligned with the CT. Image
coregistration such as this often requires substantial operator interaction and a keen
eye, since image alignment is usually subjective.
318
In the late 1990s, the value of incorporating a high-resolution X-ray CT into a

PET gantry was realized, and the first PET/CT imaging system was developed.7,8 By
colocating the PET and CT in the same room and sharing a common patient bed, it
is possible to obtain accurately coregistered PET and CT images in much reduced
time. The benefits of PET/CT soon became apparent,9 and combined PET/CT was
commercialized in the early 2000s. Sales of PET/CT have increased to the point at
which currently no major equipment manufacturer offers a PET-only device for routine clinical imaging. Similarly, the combination of SPECT and CT was seen to be
clinically useful, so the development of SPECT/CT devices was initated,10 and these
are available commercially through all major vendors.
14.1.3Image Coregistration
When performing radionuclide (PET or SPECT) and anatomical (CT or MRI) imaging with different physical systems, the accuracy of coregistration becomes paramount. In the simplest case, it is assumed that the patient can be represented as a
rigid body and is free to move about only six degrees of freedom from one scan to
the next. Thus, a given coordinate in one scan corresponds to a transformed coordinate in the other scan via the transformation
x cos cos cos sin sin sin cos cos sin cos + sin sin x x

y = cos sin cos sin sin + cos cos sin sin cos sin cos y + y
z sin
z z
sin cos
cos cos

where , , and x, y, z represent rotation and translations, respectively, about
the three orthogonal axes.
Often in rigid body transformations, it is up to the user to manipulate the transformation parameters to arrive at a suitably coregistered image. However, since this is a
subjective assessment and dependent on user interactions, it is often not reproducible
from user to user. As a result, a number of methods11 have been used to quantify the
degree of coregistration, principal among them the use of mutual information.12,13
14.1.4Image Display
Once anatomical (MR or CT) and functional (PET or SPECT) information has been
gathered and coregistered, it must be presented to the interpreting physician in the
clearest and most concise manner possible. Often, even though images are coregistered, images are viewed separately in split windows. When the interpreter selects
a location on either the anatomical or functional image, the viewer is usually shown
three orthogonal views centered at the selected location.
As an alternative, it is common to present coregistered images simultaneously in
a fused display using alpha blending.14 In this type of display, each image is shown
superposed on the other but in a different color scale. For example, a PET and CT
319
Figure 14.2 Alpha blending of PET/CT images shown (from left to right) = 0 (CT only),
= 0.5 (equal PET and CT), = 1.0 (PET only).
image can be displayed as a fused image by combining the red, green, and blue
(RGB) color channels of each image appropriately. Given the PET and CT images
as 24-bit color images (8 bits in each channel), a fused image can be created with
altered RGB color channels through the transformation
R
R
R
G = G + 1 G
B
B
B
where is given as an opacity value in the range 0 to 1 controlling for mixing of the
two fused images (see Figure14.2).
14.1.5CT Attenuation Correction

Along with providing accurate spatial localization when overlaid with PET or SPECT
images, coregistered X-ray CT images can also be used to compensate for the attenuation of photons during the PET or SPECT data acquisition process.2,4,8,15 Since
X-ray CT is acquired using a polyenergetic spectrum of X-rays, resultant images do
not represent a linear attenuation coefficient but rather are represented in Hounsfield
units (HUs), with air defined as 1,000 HU and water as 0 HU, with all other tissue
types distributed accordingly along this scale. To perform accurate photon attenuation correction, CT images must first be converted into units of linear attenuation
(i.e., cm1) specific to the PET or SPECT isotope energy. Various schemes exist for
this,1619 but in the simplest implementation, CT images are rescaled to linear attenuation coefficients via the expression

HU
+ 1
PET = PET
H 2O
1, 000
320
At low tissue densities, the conversion of CT number to linear attenuation coefficient

follows this relation, but for CT numbers greater than soft tissue, this conversion is
no longer linear with the same slope but rather becomes
(
(
)
)
CT PET PET
H 2O
H2O bone
PET = PET
+
HU
H 2O
1, 000 CT CT
bone
H 2O
To implement this type of conversion, calibrations must first be performed to

CT
evaluate CT
bone and H 2O for each CT gantry and X-ray tube.
14.1.6Clinical Applications of PET/CT and SPECT/CT

Probably the most common application of PET/CT imaging is 18F-fluorodeoxyglucose
(FDG) imaging for oncology.7 With standard FDG PET imaging, the distribution
of FDG follows that of glucose metabolism (i.e., areas of high glucose metabolism
result in large accumulations of FDG). Since many malignant tumors have a high
affinity for glucose, it follows that these tumors also would have a high affinity for
FDG. Because FDG depicts cellular glucose metabolism, the resulting PET images
lack any inherent anatomical information and typically have spatial resolutions on
the order of 69 mm, thus making it difficult for interpreting physicians to pinpoint
the exact location of a malignant tumor with certainty.
The introduction of PET/CT, however, has been able to change this. By acquiring
a CT image of the same area of the body and overlaying the PET and CT on the same
spatially coregistered image, it is possible to determine the location of FDG-avid
tumors with a high degree of accuracy. In the case of head and neck cancers that
typically spread to the neck lymph nodes, PET/CT has made it possible to correlate
the uptake of FDG to specific lymph nodes, thus making surgical excision more
precise.19,20
Imaging 111In-labeled white blood cells (WBCs) with SPECT has become an
important procedure for determining sites of infection following orthopedic replacement surgery or in patients with fever of unknown origin. The immune system
recruits leukocytes to fight bacteria at areas of infection, so labeling these cells permits physicians to localize injury. An ideal WBC SPECT image would lack any
anatomical information, since the distribution of WBCs would be specific only to
the site of infection. Thus, the resultant image would depict a diffuse spot of activity with no anatomic context. As a means to provide some additional localization
information, it is common to take advantage of the multi-isotope imaging capabilities of SPECT to perform simultaneous 99mTc-methylene diphosphonate (MDP) bone
imaging. The uptake of 99mTc-MDP is confined to the skeletal system, providing
a rough anatomical context to place the resultant WBC information. While dualisotope imaging provides some measure of anatomical context, it fails to provide
the spatial resolution needed to accurately localize infection. With the introduction
of SPECT/CT, anatomical imaging is made much simpler and more precise, since
WBC distributions can now be correlated exactly with patient morphology.2125
321
Myocardial perfusion imaging with SPECT is one of the most commonly performed procedures in nuclear medicine and is used to detect perfusion abnormalities
in the heart. The distribution of 99mTc-labeled compounds such as sestamibi or tetra
fosmin is related to the coronary arterial blood supply. Since the blood supply to the
heart is affected by the diameter of the coronary arteries, any reduction in perfusion
is typically the result of a narrowing in the arteries, usually the result of either calcification of the lumen or buildup of atheromatous plaques (artherosclerosis) within the
arteries. While useful for determining the extent of perfusion abnormalities, SPECT
is incapable of determining the exact site of narrowing. However, with SPECT/CT
imaging, it is possible to obtain anatomical information and correlate it with perfusion imaging to determine where the perfusion defects originate. In addition, CT
imaging is able to provide quantitative measures of calcification (calcium scoring) or
3D angiography (computed tomographic angiography, CTA) to depict the 3D coronary artery structure. Complementary information such as this in a single SPECT/CT
exam enables patients to be treated more effectively and in a more timely fashion.26
14.1.7Problems with CT-Based Coregistration

While multimodality imaging with X-ray CT has proved itself valuable for a number
of applications, it is not immune to problems. Because of the polyenergetic nature of
X-ray production, a significant amount of low-energy X-rays exist in most spectra.
Because of the energy dependence on photon attenuation, low-energy X-rays are preferentially attenuated as they pass through the body. The result is a progressive increase
in the average X-ray energy, commonly referred to as beam hardening, the result of
which produces, sometimes significant, streaking artifacts in resultant CT images
(Figure14.3). Materials such as aluminum or copper are often used to reduce these
low-energy X-rays as they pass from the X-ray tube, but usually this filtration does
not remove all of this component. Images with beam-hardening artifacts may lead to
misinterpretation and incorrect attenuation compensation to the nuclear image data.
Figure 14.3 Examples of artifacts seen in X-ray CT scans. (Left) Image showing streak
artifacts resulting from beam hardening through dental fillings. (Right) CT image of the
thorax showing mild respiratory motion artifact. Note the stair-step effect in the dome of
the liver at the point of the arrow.
322
Because of the different timescales of data acquisition involved in nuclear medicine imaging and X-ray CT (minutes vs. seconds), it is possible that physiological
processes occurring over short (or long) timescales can also interfere with the imaging method. For example, in the case of lung cancer imaging, FDG-PET scans may
be acquired over the course of several minutes; thus, motion artifacts as the result
of normal tidal breathing essentially get averaged out over the course of imaging
to obtain a single PET image. Contrast this to CT imaging, for which the entire lung
volume may be imaged in a single breath-hold in just a few seconds. Since the PET
image represents an average motion, CT and PET images are inherently misregistered,27, 28 resulting in the possibility of incorrect localization.
While the addition of X-ray CT imaging has had a profound impact on nuclear
medicine, there are some applications in which CT is not the optimal imaging modality. Because X-ray CT characterizes materials based on electron density (X-ray
attenuation is primarily dependent on photoelectric effect and Compton scattering),
it follows that materials or tissues with similar electron densities may not produce
much variation in X-ray attenuation. This is particularly true in imaging areas such
as the brain, where white and gray matter have very similar properties and cannot
typically be distinguished with X-ray imaging.
Because of the shortcomings of X-ray CT imaging, there is interest in alternative
approaches for performing anatomical/functional correlative imaging. MRI is one
technique that is capable of high-resolution anatomical imaging that, at the same
time, produces superior soft tissue contrast compared to X-ray CT. In addition, in
contrast to X-ray CT, which produces high-resolution images of anatomy, MRI is
also capable of bridging the gap between anatomical and functional imaging by
being capable of performing studies such as those for brain activation, chemical
metabolism, or perfusion. As a result of these advantages, there is increasing interest
in combining MRI with molecular imaging using PET or SPECT.
14.1.8Requirements for PET/MR and SPECT/MR

The MRI machines use high-strength magnetic fields and radio-frequency (RF)
generators to manipulate the magnetic moments of certain atoms. Typical magnetic
fields for clinical MRI machines range from 0.1 (open field) to 4.7 T (functional MRI,
fMRI), with 3-T machines quickly becoming the norm. The high field used in MR
requires that any material placed within the MRI field have both negligible effect on
the field homogeneity and normal operation under such a large magnetic field.
The design of most MRI systems also imposes a constraint on the size of the
incorporated PET or SPECT device, since high-field MRI systems use a cylindrical
gantry with a fixed bore size. The ideal PET or SPECT imaging system that would be
incorporated into an MRI device would need to have the following characteristics:
Must have the ability to operate in a high magnetic field with negligible
effect on imaging performance.
Must be made of magnetically compatible materials and produce negligible
effect on magnetic field homogeneity.
323
Must be small and relatively lightweight.

Must produce image quality comparable to stand-alone systems.
As can be expected, the logistics of incorporating PET or SPECT radiation detectors into a typical MR gantry are complex.
14.2Basics of PET and SPECT Imaging

To investigate the various approaches to combining MR with PET or SPECT, it is first
important to describe the basics of radionuclide imaging and the imaging systems.
14.2.1Scintillation Detectors
Overwhelmingly, scintillation detectors make up the majority of radiation detection
methods used for PET or SPECT imaging. Briefly, these detectors usually consist of
inorganic crystalline materials that fluoresce in the presence of ionizing radiation.
Since many materials scintillate when exposed to ionizing radiation, the appropriate scintillator material must be selected for the chosen application. For example, in
PET imaging, the requirement of the scintillator is to stop 511-keV photons and to
distinguish two detections based on very short timescales. Thus, PET scintillators
must have high stopping power (effectively a high density) and a very fast response
time (i.e., short phosphorescence time).29 Since most imaging used in SPECT uses
lower-energy radionuclides, stopping power is less of a concern, but the ability to
distinguish different photon energies is important. Thus, scintillators with high light
output and good energy resolution are important. In the case of combined MR/PET
or MR/SPECT, an appropriate scintillator must be chosen that meets these criteria but is also compatible with the high magnetic field present in MR.30 Table14.1
Table14.1
Properties of Some Common Scintillators Used in Nuclear Medicine
Imaging Equipment
Material
Density
(g/cm3)
Effective
Atomic
Number (Zeff)
Emission
Wavelength
(nm)
Decay
Time (s)
Hygroscopic
NaI(Tl)
BGO
CsI(Tl)
CsI(Na)
LSO
LYSO
GSO
LGSO
3.67
7.13
4.51
4.51
7.40
7.19
6.71
6.5
51
73
54
54
65
64
58
59
415
480
540
420
420
420
440
415
0.230
0.300
0.68
0.63
0.040
0.050
0.060
0.065
Yes
No
Slight
Slight
No
No
No
Yes
Paramagnetic
Source: Lecomte, R., Eur. J. Nucl. Med. Mol. Imag., 36, 569585, 2009. With permission.
No
No
No
No
No
No
Yes
Yes
324
presents some of the properties of the most common scintillators used in PET and
SPECT imaging. As expected, the Gadolinium-based scintillators, Gd2SiO5 (GSO)
and Lu0.4Gd.6SiO5 (LGSO) have very high magnetic susceptibility, thus making them
unsuitable for PET/MR or SPECT/MR.30
Regardless of the material used, any scintillator used for radionuclide imaging
must be capable of absorbing the high-energy gamma rays originating from the
radioactive decay and converting this energy into optical photons. When a highenergy gamma ray is incident on a scintillator, three different interaction types are
possible: photoelectric absorption, Compton scattering, or pair production. In typical
nuclear medicine imaging, only the first two effects are relevant, since pair production requires incoming gamma-ray energies in excess of 1.022 MeV.
In the photoelectric effect, high-energy gamma rays interact with inner shell electrons of the scintillator. Since the energy of the high-energy gamma ray is higher
than the binding energy of the electron, an electron will be ejected from the atom
with a kinetic energy equal to the difference between the original gamma-ray energy
and the binding energy. In the Compton effect, the incoming gamma ray is initially
absorbed by the scintillator atom, thus producing an electron ionization but also
emitting a secondary photon. The energy of the ejected electron and the secondary
photon is equal to the original gamma ray. The secondary photon is then free to interact via additional photoelectric or Compton interactions. In both cases, the ionization
electrons that are produced give rise to the fluorescent properties of scintillators.
14.2.2Photodetectors
For the radiation to be detected, the light produced in the scintillator material must be
detected by a photodetector and converted to a measurable electrical signal. Various
approaches exist for this conversion, but the most common is through the use of the
photomultiplier tube (PMT).
Essentially, a PMT consists of an evacuated glass housing with numerous components inside it. These components consist of (a) photocathode, (b) multiplying region,
and (c) collector. The photocathode is on the front surface of the PM tube and is
responsible for the conversion of incident light into electrons. This conversion takes
place via the transfer of energy from the incoming photon to the material electrons.
For most photocathode materials, this requires approximately 24 eV of energy for
each liberation. Once liberated, these electrons must now transit the photocathode
material, be emitted from the photocathode, and be accelerated to the first detection
stage. To be emitted, the electrons require at least enough kinetic energy to escape
the potential barrier holding the electrons to the material (i.e., the work function,
typically a few electron volts).31
As a result of the energy required to liberate electrons (the work function), most
photomultipliers are more sensitive to higher-energy light photons (i.e., shorter wavelengths) and in fact are optimal for scintillators like NaI(Tl) or lutetium orthosilicate
(LSO), which fluoresce in the blue/green region of the visible light spectrum. After
the electron is liberated, it must travel to the cathode surface to be ejected. Typically,
electrons will only travel a very short distance (several nanometers) before being
reabsorbed, so it is imperative that the photocathode be kept as thin as possible.
325
However, in keeping the cathode thin, it becomes less efficient in absorbing the photons from the scintillator in the first place. The efficiency in converting scintillation
light to photoelectrons is deemed the quantum efficiency (QE) and is typically in the
range 2030% for modern PMTs.31
At the heart of a PMT is the electron multiplication stage. An electron emitted
from the surface of the photocathode will typically have very low kinetic energy
(on the order of a few electron volts). If an electrode (the dynode) placed near
the photocathode is held at a high positive potential relative to the photocathode
(say, 100 V), then the emitted electron will be accelerated across the gap, thereby
acquiring a kinetic energy of 100 eV en route. This electron will then collide with
the dynode, resulting in more electrons being liberated. Since the creation of each
electron requires sufficient energy to overcome the band gap (23 eV), it is possible
that the original 100-eV electron will ionize several electrons within the dynode
material. However, not all of these electrons will have enough energy to traverse
the dynode material and subsequently make it to the surface. As a result, only a
small fraction of the electrons liberated will retain enough energy to be ejected
from the dynode surface. When another dynode is placed in close proximity to
the first dynode and again held at a positive potential relative to the first dynode,
then the ejected electrons will be accelerated across this gap and will impact the
next dynode, after which this process repeats itself. A typical PMT may have 612
such dynode stages, with each dynode stage increasing the number of electrons
produced. At each dynode stage, the number of low-energy liberated electrons
can be represented by . Thus, after N stages, the overall photomultiplier gain is
given by

Gain = N
where is the overall multiplier tube efficiency (i.e., the number of detected photoelectrons per number of emitted photoelectrons). It is not uncommon for modern PM
tubes to achieve gains on the order of 106 to 107. It is also worth noting that the PMT
gain is a function of the applied electrical potential. Since a higher voltage is applied
between the dynodes, more secondary electrons are produced when the incoming
electron ionizes the dynode material, and more of these secondary electrons are
accelerated toward the next dynode. In addition, each electron acquires more kinetic
energy en route. So, while more electrons are incident on the dynode, they penetrate
the dynode material to a greater depth, thus making it more difficult for the secondary electrons to escape the material. In addition, at typical room temperatures, there
is a finite probability of an electron being spontaneously emitted from the photocathode even though a scintillation event did not occur. When operated with high
operating bias, this single electron may result in a large photocurrent at the anode
that may be mistaken for an event.
The final dynode of a photomultiplier is called the collector anode and has the
highest potential difference between all the dynodes and the photocathode. As such,
it represents the end of the line for secondary electron emissions. Between the anode
326
Cc
RL
Rd
Rd
Rd
Rd
Rd
Cs
Cs
Cs
Cs
Rd
Rd
Rd
Rd
Rd
Out
RL
+HV
Out
RL
HV
Rd
Rd
Rd
Rd
Rd
Cs
Cs
Cs
Cs
Rd
Rd
Rd
Rd
Rd
Figure 14.4 The layout of a conventional PMT. The PMT is shown in the positive bias
(top) and negative bias (bottom) configurations. The photocathode is shown on the left in both
cases. (Adapted from Knoll.)31
and the high voltage, a load resistor is used to drain the photocurrent generated
through the PMT.
As described, the successive PMT dynodes must maintain a positive potential difference between them so that secondary electrons get accelerated from the current
dynode to the next dynode.
These voltages are typically configured in one of two ways:

1. Positive bias: In this case, the photocathode is held at zero potential, and
each successive dynode has a corresponding higher positive bias applied,
with the full bias applied to the anode (see Figure14.4). This is most usually applied via a resistive divider network that separates each dynode with
a resistor. Because the anode voltage is high, the resultant photocurrent
pulse will ride atop a large direct current (DC) bias. To remove this bias,
the anode output is then alternating current (AC) coupled by means of a
capacitor placed in parallel with the load resistor. AC coupling ensures that
only the photocurrent passes through the capacitor and not the underlying
high-voltage bias.
2. Negative bias: To eliminate the need for AC coupling, it is instead possible
to negatively bias the PMT (see Figure14.4). In this configuration, instead
of the anode being held at high positive bias, the photocathode is held at
327
a high negative voltage. Again, each successive dynode is held at a higher

positive voltage with the anode held at ground. In this way, the resulting
photocurrent pulse is riding atop an inherent 0-V bias, thus negating any
need for AC coupling.
With either bias configuration, it is important that the PMT current along the
resistive network be kept large enough to supply the dynodes with enough bias to
keep them at constant voltage even in the event of large photocurrents. This is particularly true for later-stage dynodes, which may be producing a substantial photocurrent prior to the anode. Thus, to keep this bias stable, it is common for the last
few dynodes also to use a stabilizing capacitor across them to supply the current lost
from the dynodes during photocurrent pulses.31
When secondary electrons are liberated from each dynode, they typically have
very low energy and are accelerated via the applied potential difference to the next
dynode. This acceleration requires a finite amount of time for each electron, since
each electron will have a different amount of kinetic energy when ejected. Over the
course of all electrons progressing through the dynodes, there will be a spread in
the time it takes for electrons to transit from the photocathode to the anode (i.e., the
transit time). The most crucial step is the initial transit of primary photons from the
photocathode to the first dynode and is primarily due to the different distances that
electrons must travel from one location on the photocathode to the dynode.31 In an
effort to reduce the transit time spread, photocathodes are often curved to equalize,
as much as possible, the photocathodedynode distance as well as incorporating
focusing electrodes to guide the electrons to the dynode.
Since PMTs use vacuum tubes and streams of low-energy accelerated electrons,
they are inherently susceptible to magnetic fields. When placed in the vicinity of
even low-strength magnetic fields, the electrons traversing from dynode to dynode
are subject to magnetic forces that may affect the electron trajectories. Such effects
have been seen even for small magnetic fields on the order of 10 Gauss,32 thus making PMTs impractical to use for photodetectors in conventional MRI devices. In an
effort to reduce the effect of small magnetic fields, PMTs are usually wrapped in a
thin magnetic shield such as Mumetal.31
14.2.3Acquisition Electronics
The electrical signal output from a PMT, although amplified through the electronmultiplying process, is still rather small and must be further amplified and shaped.31
A typical PMT output consists of a sharp drop in voltage (several nanoseconds) when
the scintillator initially fluoresces followed by a longer tail, usually lasting several
hundred nanoseconds. The amount of photocurrent produced by a photomultiplier
is proportional to the initial photon energy, and even though amplified through the
PMT, the signal is still generally too small to be measured reliably. Thus, the output
from the PMT is passed through a preamplifier to produce an output signal that
is proportional to but greater than the input signal. This is usually accomplished
through the use of either a voltage-sensitive preamplifier (VSA) or a charge-sensitive
328

R2
R1
Vin
Vout
(a)
Rf
Cf
Vin
Vout
C1
(b)
Figure 14.5 (a) (Top) A voltage-sensitive preamplifier. The output Vout is proportional to
the input signal Vin multiplied by the ratio R2 /R1 but is dependent on the detector capacitance
(not shown). (Bottom) A typical charge-sensitive preamplifier. The output signal Vout is now
independent of the detector capacitance and decays with time constant Rf Cf . (b) The effect
of CR-RCn pulse shaping on preamplifier signals. With more RC stages, the output pulse
becomes more symmetric but with a longer rise and decay time.
preamplifier (CSA). To overcome the capacitive loading of the connecting wires, it is

usually located as close as possible to the photodetector.

Vmax =
Q
C
A typical VSA is shown schematically in Figure 14.5a and consists of a basic

inverting amplifier. Given an AC-biased input, the maximum voltage across the bias
capacitor on the input side is given by

Vmax =
Q
C
After passing through the amplifier, the input voltage is inverted and scaled by the
ratio of the resistors such that
329

4
CR-RC
Output (arbitrary units)
3.5
3
CR-RC2
2.5
CR-RC3
CR-RC4
1.5
1
0.5
0
20
40
60
80
100
120
Time (s)
140
160
180
200
(c)
Figure 14.5 (continued)
Vout =
R2
Vin
R1
Such a preamplifier design works fine for detectors whose capacitance does not
change, such as PMTs, but for semiconductor detectors whose capacitance may
change under different operating parameters, the ever-changing relationship between
Q and Vmax makes consistency difficult to achieve.
In contrast to the VSA, a CSA is shown schematically in Figure14.5b. In this
configuration, the output from the photodetector is first coupled through an input
capacitor Ci before being passed through the amplifier. An input charge builds up
on the feedback capacitor Cf and is dissipated through the feedback resistor Rf . The
subsequent output voltage through the CSA is thus

Vout =
Q
Cf
In this scenario, the resultant output voltage is only a function of the integrated
charge Q produced in the photodetector. Under realistic circumstances, the output
pulse Vout will rise very quickly to a maximum that is proportional to the detected
charge, followed by a slow decay described by the time constant Rf Cf . It is common
for the initial rise time to be several nanoseconds long, but the ensuing decay proceeds for several hundred microseconds.
As a result of these long tails, it is possible that in the presence of high count rates
each ensuing output pulse is added to some residual of the previous pulse. Thus, the
detected absolute output voltage may be severely overestimated. If the count rate is
330
high enough, it is possible that the output voltage will be above the saturation level of
the amplifier, thus making each detection appear with the same magnitude.
Since it is only the height of the initial rise that is important, it is common to
employ additional circuitry to provide pulse shaping to extract only this portion.
Examples of such shapers include CR-RC circuits or Gaussian shapers. If the output
pulse from a preamplifier (essentially a step function) is fed through a simple CR
circuit, the output is a differentiated signal equal to

Vout = Vinet /
where represents the time constant RC. Similarly, if an input is directed through
an RC circuit with the same time constant , the output is an integration of the input
equal to,

Vout = Vin 1 et
If the preamp pulse is fed through the CR-RC circuit in series, the output signal
is equal to

Vout = Vin
t t /
e
By combining multiple RC stages, the net result is a CR-(RC)n pulse shaper of

which the output is described by

Vout =
Vin
n!
t t /
e
This type of shaper reduces the output tail further and achieves a Gaussian-shaped
output function after the addition of only about four RC stages. While the output
shape becomes more Gaussian in shape (as shown in Figure 14.5c) and the tails
decay to baseline faster than with a standard CR-RC shaper, the initial rise time does
become longer, thus limiting the peak counting rate.
In PET and SPECT imaging, to measure high count rates, it is common to clip the
incoming PMT pulse while the scintillator is still fluorescing. When clipped, the preamplifier will only integrate over a short time period and accumulate only a portion
of the total charge possible. While the use of pulse clipping can improve the overall
count rate capability of the detector, the trade-off is a larger noise component in the
amplified signal due to accumulation of less-than-complete charge.
In current PET and SPECT designs, it is common for each shaped signal to then
be digitized via an analog-to-digital converter (ADC). It is normally sufficient to
use 8-bit or 10-bit flash ADCs for this task. To distinguish actual events from background noise, the use of a summing-and-triggering circuit is used. Signals from all
PMTs are passed through a summing circuit to determine the total signal produced
331
during a scintillation event. The total signal is then compared to a preset threshold
(situated well above background noise levels), and if the event signal is above this
threshold, a trigger signal is sent to each PMT channel, thereby initiating the ADC
conversion. Within each channel, a peak detect-and-hold circuit latches on to the
peak of the shaped pulse and holds it until sampled by the ADC. Once sampled, the
peak detect circuit is reset to baseline to await the next pulse.
For each scintillation event, the location of the event must now be determined.
SPECT imaging typically uses a large, monolithic scintillator coupled to an array of
many photodetectors. For each scintillation event, multiple photodetectors will sense
the light produced, and by using all the signals, the (X,Y) location of the scintillation
can be determined. This is typically performed using an Anger logic-type algorithm
or some other variant by which the scintillation location is determined via a weighted
averaging in both the X and Y locations depending on the detector position, such as
N
X=
i i
i =0
and
N
Y=
i i
i =0
where and are weighting factors for the ith digitized signal.
PET scanners typically consist of multiple modular detector blocks with each
block comprised of a 2 2 array of PMTs coupled to a scintillator. For each scintillation, the light is shared among only four PMTs, thus making the localization
somewhat simpler. Using the four digital PMT outputs, the (X,Y) location of the
scintillation can be determined via the equations

X=
( A + B) (C + D)
( A + B + C + D)
Y=
( A + C ) ( B + D)
( A + B + C + D)
and

where A, B, C, and D correspond to the four PMT outputs, respectively.

The total gamma ray deposited energy E is simply determined by summing all
the PMT signals. Once the (X,Y) locations are determined, it is then possible to display a histogram of detected positions on a computer monitor. In the case of SPECT,
the detector then needs to rotate around the patient to acquire angular projections
from which the 3D distribution of radioactivity can be determined. In the case of
PET, an entire ring of detectors is positioned around the patient and acquires all possible projection angles simultaneously without any detector motion.
332
14.3Imaging with MRI

14.3.1Basics of MRI
To align the proton magnetization vector within the object, MRI uses a high, static
magnetic field B0. Usually, this magnetic field is on the order of 1.53 T and is produced using a superconducting magnet. This magnet is usually cylindrical with an
open bore of approximately 70-cm diameter and 2-m long, although some low-field,
open-gantry systems consist of a flat permanent magnet.
Within the bore is a set of three additional magnetic fieldproducing coils. These
coils are the gradient coils Gx, Gy, and Gz and are used in conjunction with the main
magnetic field to alter the local magnetic field within the bore. Also within the bore
is a set of RF coils. These coils are used to transmit an RF pulse to the protons of
the object, causing the magnetization vector to change as a result of the applied magnetic field. Since the proton magnetization decays back to its normal state following
RF excitation, a subsequent RF pulse is emitted from the protons that is detected by
the coil. By using a series of RF pulses in conjunction with altering the magnetic
gradients within the bore, images can be obtained depicting localized properties of
the magnetization vector. The RF coils may transmit and receive, transmit only, or
receive only. In the case of transmit-only coils, a receiver set of coils must also be
used to detect the emitted RF signal from the protons. It is important to note that RF
coils should be placed as close as possible to the object being imaged to obtain the
highest signal-to-noise ratio (SNR). This factor will become important when considering where PET or SPECT inserts can be placed within the bore.
MRI is able to distinguish tissues based on their proton density or magnetization
properties, such as their spin-spin (T2) or spin-lattice (T1) relaxation times. Since
these properties vary in the proximity of different tissues, MR images can depict
subtle changes in tissues even though the tissue density does not vary much. Given
the improvements in soft tissue contrast in MRI compared to CT and the complementary functional information obtained, it seems natural to combine MRI with
nuclear medicine imaging. In fact, combining PET or SPECT with MR images has
been performed rather routinely using separate imaging systems and software coregistration methods. However, given the success of combined PET/CT and SPECT/CT,
the combination of PET/MR or SPECT/MR in a single imaging gantry would seem
to offer several advantages over separate imaging systems, not least of which is the
improved diagnostic value of perfectly coregistered images. As we will discuss,
however, the development of PET/MR or SPECT/MR is not necessarily as straightforward as the development of PET/CT or SPECT/CT.
The biggest hurdle with simply grafting a PET or SPECT machine onto an MR
gantry is that the radionuclide imaging system must operate in the vicinity of a high
magnetic field strength (typically 1.53 T, but possibly upward of 914 T for animals). The high magnetic field creates problems for conventional nuclear instrumentation, so alternative techniques must be used. Some attempts have focused on using
alternative MR configurations, while other attempts have investigated alternative
nuclear detectors. We investigate some of these approaches in the following section.
333
14.4Alternative MR Configurations
To achieve higher spatial resolutions with higher signal to noise, there has been a
push in MRI for higher and higher magnetic field strengths. To achieve these high
magnetic fields, most MR machines use cryogenic superconducting magnets, making it impossible to cycle MR magnets on and off to provide sequential imaging with
PET or SPECT. However, by using alternative approaches to generate the magnetic
fields needed, it becomes possible to perform coregistered sequential nuclear/MRI.
14.4.1Low-Field MR
As mentioned,32 even small magnetic fields can have a significant effect on the performance of PMTs. As such, conventional PMT-based radiation detectors cannot be
placed in the vicinity of high-field MR scanners. Goetz et al.,33 however, have taken
the approach of using a nonsuperconducting, low-field MR (0.1 T) in conjunction
with a single-detector NaI(Tl) SPECT camera to conduct sequential SPECT/MR
imaging in small animals. While resulting in lower spatial resolution and lower SNR
compared to high-field MR, the advantage of using a low-field MR lies in the fact
that the 5-Gauss line is situated only 15 cm from the edge of the main magnet,
thereby making it possible to place a conventional PMT-based SPECT camera at this
location. In this configuration, the animal is first imaged using the pinhole SPECT
camera, then manually moved the 15 cm or so to the MR system. Since the animal
remains in the same imaging cell for both scans, any deviation in positioning can
be remedied with a relatively simple rigid-body transformation. In fact, coregistration parameters between the MR and SPECT images were first determined using a
three-tube geometric phantom, and for all successive imaging experiments using
animals, the SPECT scans were coregistered to the MR images using the same set
of registration parameters.
While the low-field MR was not capable of providing high-resolution anatomical
imaging or spectroscopic information, the spatial resolution and soft tissue contrast
was deemed of sufficient quality to be used for anatomical coregistration with SPECT.
In addition, it was pointed out that if the only requirement of the MR is for SPECT
colocalization, then low-field MR is more than adequate for this purpose.
14.4.2Field-Cycled MR
As another alternative to cryogenically maintained static magnetic fields, research
has focused on using field-cycled MR.3436 In conventional MR, the main static magnetic field is used both to induce the magnetization within the object and to create
the magnetic environment for the transverse magnetization to precess. Thus, the field
strength affects the amount of magnetization generated in the sample and must be
kept as uniform as possible.
In contrast, the field-cycled MR consists of two nonsuperconducting magnets, one
large field for generating the polarizing magnetization and a smaller homogeneous
magnet for readout. The large-field magnet is first applied to create a net magnetization in the object. After a short time (on the order of 1 s), the polarizing magnet is
334
turned off, and the readout magnet is turned on, again for a short time. When the
readout magnet is turned, the net magnetization produced by the polarizing magnet
will precess at the Larmor frequency while it decays. During this time, RF pulses and
gradient fields are applied as in conventional MRI to gather the resonance signal.
Since the field-cycled MR does not use a static magnetic field, it is possible to
incorporate conventional PMT-based radiation detectors into the gantry.36 These
detectors would not be operated simultaneously with the MR, but rather acquisitions
would be interleaved between the MR pulse sequences when the main and readout
magnetic fields are cycled off.
14.5Alternative Radiation Detectors

Developments in radiation detection have made it possible to consider alternatives to
the conventional scintillator/PMT arrangement.
14.5.1 Scintillation Detectors

In many respects, scintillation detectors still offer many advantages over other radi
ation detection methods. Reasonable cost, good energy resolution, and high detection
efficiency are just some of the advantages scintillation detectors offer. In addition,
if appropriate scintillators are chosen, their relative insensitivity to high magnetic
fields enables them to function properly in the presence of high magnetic fields. By
the same token, the high electrical resistance of most scintillators reduces the effect
of eddy currents within the detector material, thus reducing the influence of the scintillator on the magnetic field. For these reasons, scintillating detectors are still a good
option for combined radionuclide/MRI. However, care must be taken either to place
the photomultipliers a sufficient distance from the main magnet or to use alternative,
magnetic-compatible photodetectors.
14.5.2Fiberoptic Coupling
The magnetic field at the center of most clinical MRI systems is typically between
1.5 and 3 T; however, this decreases rapidly with distance such that by about 3 m
from the center, most magnetic fields are less than 10 mT.37 Thus, if the sensitive
PMTs and electronics of a PET or SPECT device can be placed in this lower-field
area, then it would permit simultaneous imaging. This was the approach taken in the
development of the first simultaneous PET/MR detectors.3739
The McPET I was a single, 38-mm diameter ring consisting of forty-eight 2 2
10 mm LSO scintillators with each crystal coupled to an individual pixel of a multi
channel photomultiplier. This detector was integrated between the poles of a 0.2-T
vertical field MRI. So that the magnetic field exhibited negligible effect on the photo
multipliers, the PMTs were placed about 3 m from the magnet and connected to the
scintillators via a 4-m long and 2-mm diameter optical fiber. At this distance, the
magnetic field strength was measured to be less than 0.1 mT.
Since some light will inevitably be lost during transit through the fiberoptic device,
the significantly greater light output of LSO compared to bismuth germanate (BGO)
335
lends itself well as the scintillator of choice even though the detection efficiency is
slightly less. Since the crystals were oriented with their long axis parallel to the axial
direction, detection efficiency was somewhat reduced, as only a 14% detection efficiency was realized, thus contributing to an energy resolution of 41%. Timing resolution for this system was also somewhat high at 20 ns. Nevertheless, this system was
able to show that simultaneous PET and MRI was indeed possible using separated
scintillators and photodetectors and that no noticeable distortion effects were seen in
either the PET or MR images.
This system was further developed into the McPET II, which used seventy-two
2 2 5 mm crystals oriented radially in a single slice and coupled to the same
photodetection subsystem. By reorienting the crystals, an improvement in detection
efficiency to 34% was obtained with similar energy resolution (45%) and timing
resolution (26 ns) to the McPET I. This system was tested in magnetic fields up to
9.4 T without showing any noticeable effects on the PET images.
While the use of fiberoptic coupling showed the feasibility of simultaneous PET and MRI, a number of limitations are present with this approach. The
attenuation of scintillator light output along the long optical fibers reduces the
energy resolution and coincidence timing. As a result of the issues with using
PMTs for photodetection, there has been increased interest in alternative types
of photodetectors.40
14.5.3Photodiodes
Photodiodes are semiconductor photodetectors that convert incident light into electrical current proportional to the intensity of the incoming light. As such, they have
replaced PMTs in some applications.41,42 Photodiodes consist either of PN-type (pand n-type semiconductor layers) or PIN-type materials (p- and n-type layers with
a depleted i-type region between them). Most photodiodes used in conjunction with
scintillators are of the PIN type due to their superior sensitivity and responsiveness. Compared to PMTs, photodiodes offer higher intrinsic quantum efficiency, low
power consumption, low operating bias, and insensitivity to magnetic fields.
As light photons enter the p-layer, electronhole pairs are produced and collected
on the boundary layers of the i-type region, driven there by an applied reverse bias
across the PN contacts. Once at the anode, the electrons contribute to a small amount
of current. Through the collection of many light photons, several electronhole pairs
are liberated, and a small, but detectable, photocurrent is produced.
The inherent QE of typical PIN photodiodes is 5070%, several times higher
than for PMTs.31 However, photodiodes typically have a different response curve
as a function of wavelength compared to photomultipliers. Most PMTs have a peak
QE around 400 nm, thus making them suitable for use with common scintillators
such as NaI(Tl), LSO, or BGO. However, photodiodes typically have responses
peaked more toward the red end of the visible light spectrum, making them more
efficient with scintillators such as CsI(Tl).
It is worth noting that for each visible light photon incident on the photodiode, a
maximum of only one electronhole pair will be generated (theoretical QE of 100%),
so the conventional photodiode is a unity gain device. Since most scintillation events
336
only generate a few thousand scintillation photons, the resultant charge buildup is
quite small for a single scintillation. Even so, with a low-noise CSA, it is often possible to detect the resultant charge buildup from a single event over the background
detector noise.
With the small level of signal produced in photodiodes when operating in pulse
mode, they are extremely sensitive to electronic noise. The two most important factors contributing to noise are the photodiode capacitance and the leakage current.
Because of the PN junction, a photodiode has an inherent capacitance that increases
with area but decreases with thickness. It is typical that photodiodes have capacitances on the order of 2050 pF/cm2, although some may be as high as 300 pF/cm2.
Because photodiodes are semiconductors, they will always show a small amount of
conductivity as a result of thermal ionizations. This leakage current increases with
temperature as well as with increasing thickness. Small fluctuations in the leakage current give rise to jitter that may obscure the small signal from a legitimate
photoevent. Since the leakage current decreases with temperature, it is common for
photodiodes to be actively cooled to reduce the dark current enough to be able to
detect photoevents.
To limit the device capacitance and dark current, photodiodes are typically only
available in small sizes (typically less than 1 cm2). Thus, it is tempting to consider
using arrays of photodiodes coupled to large-area scintillators to replace conventional PMTs. The problem with this, as pointed out by Groom,31,43 is that as the light
collection increases with photodiode area, so does the inherent noise; as a result,
very little gain in signal to noise is actually obtained. Thus, most applications of
photodiodes as radiation detectors use individual scintillation crystals coupled to
individual photodiodes.44 With this design, most of the light produced in the scintillator interacts within a single photodiode, thus maximizing the SNR. Because of the
relatively small size of photodiodes compared to the effective area of most imaging
devices, designs that incorporate photodiodes as photodetectors typically require the
use of multichannel digital electronics and parallel processing. However, as a result,
these devices do not suffer from the same dead time considerations as conventional
monolithic scintillators and Anger logic detectors.
The Digirad 2020Tc is the first commercially available photodiode-based gamma
camera. This camera uses a 64 64 array of CsI(Tl) detectors (3 3 mm) with each
detector pixel coupled to a PIN diode. Since there is no sharing of the scintillation
light across multiple diodes, this camera does not suffer from excessive noise on each
detector channel, although the system is also actively cooled to reduce background
noise. It is noted that although this system uses PIN diodes, it has not been designed
to operate in the vicinity of an MR imager.
14.5.3.1Avalanche Photodiodes
As mentioned, photodiodes are unity gain devices and suffer from poor SNRs
when detecting the small signals produced in scintillation events. To mitigate
this problem and achieve some level of signal amplification, the avalanche photo
diode (APD) was developed. A typical APD device is shown in Figure14.6a. In
essence, an APD consists of a PN semiconductor with a drift region between the
337

Scintillation
light
p+
Electric eld
Depth
p
n+
(a)
3,000
Current (nA)
APD A(Id)
APD A(Ip)
APD B(Id)
APD B(Ip)
2,500
2,000
1,500
1,000
500
0
100
200
300
400
500
600
Bias voltage (V)
(b)
Figure 14.6 (a) Representation of a standard reach-through avalanche photodiode.
Incoming light photons liberate electronhole pairs in the semiconductor material, which then
are accelerated through the multiplying region, thus liberating more electronhole pairs. The
subsequent collection of electrons at the anode is amplified through the avalanche process. (b)
Plot of gain versus bias for two silicon avalanche photodiodes (Hamamatsu S8550-1010). Note
the breakdown voltage is different for both diodes even though the model is the same. (c) Plot
of gain versus operating bias for a silicon avalanche photodiode (Hamamatsu S8550-55). In
addition to an increase in gain at lower operating bias when cooled, the thermal noise content
is significantly lowered when cooled, thus increasing the signal to noise. (Reproduced from
Kataoka, J. et al., Nucl. Instrum. Methods A, 564, 2006. With permission from Elsevier.)45
p and n junctions. When a high reverse bias is applied across the junctions, it
sets up a nonlinear electric field through the multiplying region. When exposed
to light, electronhole pairs are created in the drift region, which, because of
higher electron mobility, results in the electrons preferentially being pulled
toward the anode. Because of the high electric field residing in this region, the
electrons are accelerated and collide within the semiconductor matrix, thereby
resulting in more ionizations and more electronhole pairs. The secondary electrons are then free to be accelerated and undergo further ionizations and so forth.
Eventually, the freed electrons will reach the anode and result in an amplified
electrical signal.
338

200
from 20 to +20 deg
(left to right)
t = 5 deg step
100
Avalanche gain
50
20
10
5
2
1
0
100
200
High voltage (V)
300
400
(c)
Figure 14.6 (continued).
As the applied electrical potential increases, the number of impact ionizations

in the multiplying region also increases, thereby resulting in increased signal gain.
However, at high bias, the normally immobile holes will be attracted toward the
cathode and produce impact ionizations en route. The additional electrons get pulled
easily toward the anode, producing an additional, undesired, cascade effect. The
point at which this occurs is the breakdown voltage Vbr and is temperature dependent. When operated above breakdown voltage, the number of electronhole pairs
produced via the original interaction and subsequent multiplying process is no longer
proportional to the intensity of incoming light. For this reason, APDs are generally
operated at voltages somewhat below the breakdown voltage at a given operating
temperature.
It is important to note that while the gain of an APD is greater than standard PIN
diodes, the gain is dependent on the operating voltage and the operating temperature.45 Figure14.6b depicts the system gain for a standard 10 10 mm APD exposed
to light from a red light-emitting diode (LED) over a range of operating voltages.
As the bias is increased, the overall gain of the APD is also increased. However, in
addition to increasing the APD signal, the noise increases with bias. There comes
a point prior to breakdown voltage at which the noise is increasing faster than the
useful signal; thus, the SNR starts to decrease. It is at this point, when the SNR is
maximal, that it is usually optimal to operate the APD.
Figure14.6c depicts the overall APD gain as a function of operating temperature.
It is seen that by cooling the APD at a given operating bias, the gain will be higher
when at lower temperatures. In addition, the breakdown voltage decreases when
cooled, so that for a given gain, the APD can be operated with a lower operating
339
bias.46,47 Furthermore, the APD background noise also decreases when cooled, thus
improving the SNR. This temperature dependence on gain can become a problem,
however, since it means that the APD output varies as the device warms up or cools
down. For most APDs, this variation in gain as a function of temperature is on the
order of 23%/C, thereby making it rather challenging to operate an array of APDs
without some form of active gain compensation.45
One of the first APD-based radiation detectors for use with PET was the smallanimal PET developed by Lecomte et al. at the University of Sherbrooke.4852 This
device used two rings of 256 3 5 20 mm BGO scintillators, each coupled to an
individual APD. This device has since been commercialized as the LabPET,53 and
although the device has not been shown as yet to operate in the vicinity of magnetic
fields, these devices were the first to show that a stable imaging system can be developed using APDs.
Pichler et al.5456 replaced the PMT photodetectors on a clinical PET detector module with an array of APD detectors for use in simultaneous PET/MR. This detector
uses a 3 3 array of 5 5 mm APDs coupled to LSO scintillators. The output of each
APD is then fed into a CSA and associated electronics similar to PMT detectors.
In an effort both to improve the spatial resolution of large-area APDs and to reduce
the complexity of multichannel acquisition electronics when used with individual
APD elements, position-sensitive APDs (PSAPDs) have been developed.57 In essence,
PSAPDs are large-area APDs (up to 14 14 mm) with a resistive layer on the back
face on which multiple contacts are placed (see Figure14.7a). The usual configuration is to divide the back face into quadrants with separate outputs for each quadrant.
Because the total charge produced in the APD is shared among four anodes, PSAPDs
require high gain to overcome the inherent noise associated with each anode.
When coupled to an appropriate scintillator [usually pixelated CsI(Tl) or LSO], it
is possible to acquire the four quadrant signals for each scintillation event and subsequently to determine the appropriate scintillator detector that gave rise to these signals via the same equations used for PET imaging (p. 327). Given the relatively low
SNR on each channel, PSAPDs also use a fifth detector channel on the top layer to
obtain a summed signal over the entire array. This output is useful for obtaining lownoise information for subsequent energy discrimination. These devices have been
investigated for use in preclinical imaging owing to the possibility of high resolution
(300 m58) for both PET59,60 and SPECT.61
Using these PSAPDs, Cherry et al.6265 developed a preclinical PET insert for
hybrid PET/MRI (Figure14.7b). This device uses a ring of LSO detector modules
fiberoptically coupled to PSAPDs. The entire ring consists of 16 detector modules,
with each detector module comprised of an 8 8 array of LSO scintillator crystals
(1.43 1.43 6 mm). A short 6 6 fiberoptic bundle (1.95 1.95 mm per fiber) is used
on each module to carry the light information from the scintillators to the 14 14 mm
PSAPD. This bundle incorporated a 90 bend to minimize the radial extent of the
insert. Each PSAPD used five outputs (four back layers for position information
plus the top layer for timing and energy discrimination), with each channel passing through a low-noise CSA. As the APD and preamps were placed inside the MR
field, they were shielded with a cylinder of high-frequency laminate to minimize RF
interference. Outputs from the preamps were then fed to standard PET acquisition
340
A
B
C
D
8 8 array of
LSO crystals
14 14 mm2
PSAPD
6 6 optical
fiber bundle
Charge-sensitive
preamplifiers
PCB populated with

nonmagnetic components
15 cm
Figure 14.7 (a) A position-sensitive avalanche photodiode. Note that the anode is divided
into quadrants, thus resulting in a charge division across the four anodes. The individual
signals are combined to determine the (X,Y) location of the original scintillation event.
(b) Preclinical PET/MR ring produced at the University of California at Davis. The detector
ring consists of a series of LSO detectors fiberoptically coupled to PSAPDs and chargesensitive preamplifiers. The entire ring and electronics are enclosed within a copper can.
(From Catona, C. et al., J. Nucl. Med., 47, 2006. With permission.)62
341
electronics located at a safe distance from the MR magnet. To reduce dark current,
the PSAPDs were cooled to approximately 5C by blowing cold nitrogen gas over
them. Of note is the fact that a relatively large 40-ns coincidence timing window
was used to discriminate coincidence events. This was explained by the fact that the
top-layer PSAPD signal has a position-dependent delay, since it was recognized that
signals originating from the edge of the PSAPD would typically be measured sooner
than signals originating from the center. Because the purpose of this system is preclinical imaging, it was noted that the large coincidence window would not represent
a significant problem, since anticipated random events will be quite low because of
the relatively low radioactive doses administered to the animals.
As can be imagined, as the bias voltage is continually increased, additional impact
ionizations occur within the APD at the same time that some electrons are collected
at the anode. However, there will come a point when more electrons are produced
than can be collected at the anode. This is referred to as the breakdown voltage, and
at this point, the ensuing runaway production of electronhole pairs leads to a large
photocurrent that is no longer representative of the original light intensity. When
operated in this way, the APD is said to be operating in Geiger mode, since the
detection of a single light photon in the APD results in a large photocurrent. Because
the output is no longer proportional to the incident light intensity, APDs used as
alternatives to PMTs for imaging are typically operated well below breakdown voltage. However, as the APD bias is increased above breakdown, the generation of
electronhole pairs increases exponentially. Since the semiconductor material has
some series resistance, more of the voltage is dropped across the resistance as the
photocurrent grows, thus limiting the avalanche effect until, at some point, the voltage across the high-field region is reduced to breakdown when the generation of
electrons is matched by the extraction of electrons at the anode. So, in the continued
presence of applied bias, this large photocurrent will continue to flow indefinitely,
thereby rendering the detector unresponsive to further photodetections. To further
detect photons, the APD must be quenched.
Two means of quenching an APD are possible: passive quenching or active
quenching. In the former case, the bias is simply removed from the APD once it is
in the steady state. Once removed, the inherent resistance and capacitance result in
a gradual decline in photocurrent until the avalanche process no longer occurs, thus
resulting in dead time while the APD discharges. In active quenching, the APD bias
is again removed after the avalanche process is initiated, but once the discharging
starts, a quenching circuit steps in to shunt the APD and quickly discharge the device,
thereby reducing the APD dead time. Once discharged, the bias can then be reapplied through the use of a switch. For simplicity, most applications of Geiger-mode
APDs use a large series resistor Rs between the power supply and the APD, thus
creating a virtual open circuit that continually charges the APD with time constant
RsC while allowing the APD to discharge with time constant RC.
For all diode-based photodetectors, every effort is made to reduce the number of
high-energy gamma rays from interacting within the photodiode by using thick scintillators and making the diode as thin as possible. Even so, it is still possible that a gamma
ray will penetrate the scintillator and interact within the diode material itself. When
this occurs, electronhole pairs are created the same as when low-energy light photons
342
interact, thus giving rise to electrical signals. This is problematic, since the signal from
a direct ionization is several times larger than that which would be detected from the
scintillator, thus making it possible that spurious events may have a large effect on subsequent processing.
14.5.3.2Multipixel Geiger-Mode Avalanche Photodiodes
Since the output signal of a single-channel Geiger-mode APD is not proportional to
the deposited gamma-ray energy, APDs operated above Vbr are not useful for imaging
applications. However, since they are extremely sensitive, it is possible to measure
single optical photons with these devices. If the APDs were miniaturized sufficiently
and an array of such APDs configured, then for a given scintillation event in a crystal,
each APD element would be capable of detecting a single optical scintillation photon
and outputting a large photocurrent. Since each Geiger-mode APD would output the
same voltage when it detects a photon, simply summing the total signal over all elements would give an indication of the total number of optical photons detected. This
is the basis of the multipixel Geiger-mode APD, sometimes referred to as solid-state
or silicon photomultipliers (SSPMs or SiPMs, respectively),6668 depicted schematically in Figure14.8a.
Most SiPM designs use an array of several thousand APD detector cells, all
operating in Geiger mode. Each cell is typically between 20- and 50-m2; thus,
the entire array is usually around 1 to 3 mm2. When a scintillation event occurs,
the light photons will spread out through the scintillator and will be incident on the
SiPM, which then initiates an avalanche process that creates a large current through
the cell, thereby leading to an infinite gain if not for a quenching resistor integrated
into each cell. Because many light photons will be generated during a scintillation,
many cells will undergo the avalanche cascade, thereby leading to an overall current
proportional to the number of cells that undergo avalanche. Since each cell can only
react to a single light photon and output a given current, it is important that there are
more cells than light photons expected and that the time required to reset each cell
be as short as possible.
The overall gain of silicon photomultipliers approaches that of PMTs (around
106),69 yet they do not suffer from the same sensitivity to magnetic fields as PMTs.
In addition, they offer superior gain to either PIN diodes or APDs but with increased
robustness, reduced noise, and the capacity to operate at much reduced bias (~30 V).
Because of these benefits, many research groups are currently investigating the use
of SiPMs for PET or SPECT imaging.7074 Commercial SiPMs are available through
suppliers such as Hamamatsu, SensL Limited, and Zecotek, among others. Further
development of SiPMs has led to arrays of SiPMs such as the large-area SensL
SPMArray. This device consists of a 4 4 array of 3 3 mm SiPMs, with each
SiPM comprised of 3,640 microcells (see schematic in Figure14.8b). This array can
be operated in single-channel mode, in which all SiPMs in the array are summed
together, or in position-sensitive mode, in which each of the 16 elements outputs
an independent signal. With this configuration, the SiPM array would normally be
coupled to pixelated scintillators and the 16 outputs used to determine the site of
photon interaction.
343
V>Vbr
(a)
14 mm
3 mm
(b)
Figure 14.8 (a) Schematic of a multipixel Gieger-mode avalanche photodiode. Each APD
element is typically between 25 and 50 m, with the entire array usually about 1 1 mm.
(b) SiPM array using 4 4 SiPMs. The output of each SiPM can be independent (i.e., 16 channels per array) or summed for a single output. When summed, the array behaves like a conventional PM tube, and when operated in multichannel mode, the array becomes position
sensitive for high-resolution applications.
14.5.4Solid-State Detectors
There has been renewed interest in replacing the scintillator/photodetector combination with solid-state radiation detectors for PET and SPECT. The inherent robustness
of semiconductor detectors in strong magnetic fields make them an attractive alternative to scintillator/photodetectors; however, their relatively high cost due to low
manufacturing yields has limited their use, until recently, to research only.
In essence, semiconductor radiation detectors behave similarly to PIN diodes as
described in this chapter. When a high-energy gamma ray interacts within the semiconductor, a number of electronhole pairs are produced. The migration of these
electrons and holes across an applied electric field gives rise to a measurable electric
charge at the electrode. The electronhole cloud of charge that is created has been
seen to remain relatively intact (less than 250 m) as the charges migrate through the
344
electric field, limiting the charge spread to less than 0.5 mm from the original inter
action location,31 thus making them ideal for high-resolution PET or SPECT imaging.
14.5.4.1Si(Li) and Ge(Li) Detectors
Lithium-doped silicon- and germanium-based semiconductors have been used extensively for spectroscopy applications because of their high-energy resolution (23%
at 662 keV). While capable of extremely high energy resolution, Si(Li) and Ge(Li)
semiconductors suffer from an inherently high thermal noise component, thus making it necessary to cryogenically cool the material. In addition, the relatively low
stopping power of silicon and germanium materials limit their usefulness to primarily spectroscopy applications, although some use has been made in SPECT or PET
applications.75
14.5.4.2CdTe and CdZnTe Detectors
The most commonly used semiconductor radiation detectors are cadmium telluride (CdTe) and cadmium zinc telluride (CdZnTe). While CdTe and CdZnTe have a
density sufficient to stop 511-keV photons for PET, most crystals of CZT can only
be grown of sufficient thickness for low-energy applications (typically 3- to 5-mm
thick). Thus, CZT arrays are primarily used for SPECT applications, although some
groups have investigated them for use with PET.76
The CZT detectors are typically either pixelated into discrete detector elements
or used as a monolithic block. In the former case, each pixel requires its own anode
and cathode, thus resulting in an inherent improvement in spatial resolution as
pixels are made smaller and smaller; however, this comes at the cost of increased
readout complexity in the requirement for thousands of individual readout channels. In the case of large single blocks of CZT, there is a single common cathode
for the entire block, but individual anodes are distributed in an array on the back
side of the block. Because of the small amount of diffusion of the charge carrier
cloud, there is minimal spread of the carriers as they traverse the detector block.
GammaMedica-Ideas produced a prototype CZT detector insert for preclinical
SPECT/MRI,77,78 shown in Figure14.9. This insert consists of 24 CZT detector modules arranged in three octagonal rings. Each module uses a single block of 25.4
25.4 5 mm CZT with a single cathode and with a 16 16 array of pixelated anodes.
A 500-V potential bias is applied to the cathode, thus creating the large electric
field required to draw electrons to the anodes to produce a measurable signal.
Each anode of the detector block is connected to a single channel of a multiplechannel application-specific integrated circuit (ASIC).79 In total, each ASIC has
128 input channels, with each channel incorporating a CSA, pulse shaper, peak-hold
circuit, and a threshold discriminator. Two ASICs are connected to each CZT module to acquire the entire 16 16 pixel array. The collimation system for this detector
insert is comprised of an MR-compatible heavy metal composite cylindrical sleeve
35 mm in diameter. This sleeve consists of a series of pinhole apertures, with each
pinhole corresponding to one of the CZT detector modules.
345
(a)
(b)
Figure 14.9 (a) CZT detector module used in the first MR-compatible SPECT detector.
(From Azman et al.)79 (b) SPECT/MR ring insert consisting of three rings of eight CZT detector modules. Each module uses a single pinhole collimator. (From Azman, S., IEEE Med.
Imag. Conf. Rec., 2007. With permission.)79
In the past, CZT production has been plagued with low production yields and
small crystal sizes. New production methods8082 offer the promise of delivering
higher-yield materials with higher grade, thus making the transition from scintillators to solid-state detectors possible. At least two manufacturers (GE and Spectrum
Dynamics) now offer CZT-based clinical gamma cameras for limited applications.
346
14.6Other Considerations
14.6.1Magnetic Compatibility
To reduce the influence of background radiation, PET and SPECT detectors are usually surrounded by a heavy metal material that acts as radiation shielding. On standalone devices, this material is typically lead because of its high stopping power and
relatively low cost. Because PET imaging uses coincidence detection, the required
shielding is not very significant and in fact is usually used more to shield the photo
multipliers from magnetic field effects. However, since SPECT detects a single
photon at a time, any background radioactivity can have a significant effect on the
imaging performance, so camera shielding is quite significant. In addition, SPECT
systems incorporate a heavy metal collimator attached to the scintillator, which limits the angle of acceptance of incoming gamma rays. Again, this collimator is typically constructed of lead or tungsten alloys.
To function in the high magnetic field of MR, it is necessary to consider the
appropriateness of the radiation shielding and collimator materials used in PET and
SPECT detectors.83 Ideal PET and SPECT shielding materials would have the following properties:

High stopping power for radiation

Magnetic compatibility (i.e., low magnetic susceptibility)
Nonconductivity
Thin and lightweight
The first and last properties are very much related, since, in general, for a material
to have high stopping power, it must have a high effective atomic number Z and high
density. Thus, there are trade-off considerations between weight and radiation stopping power that should be addressed for each application.
To measure magnetic compatibility, magnetic susceptibility is most often
used.84 This property measures the extent of an induced magnetic field produced by
the material when placed in a magnetic field and is defined as

M
H
where M is the volume magnetization (magnetic moment per unit volume), and H is
the magnetic field. Susceptibility is usually described in terms of parts per million
(ppm), with the sign denoting the direction of the magnetization vector and the absolute value denoting the strength. As an example, water has a low magnetic susceptibility of about 9 10 6, while pure iron has a susceptibility of around 2 105.
Depending on the application, it is common also to use relative permeability rather
than susceptibility. In free space, the two quantities are related by the expression

r = +1
347
In selecting appropriate MRI-compatible materials, it is important to select materials with both low absolute susceptibility values and similar susceptibilities to those
materials that are being imaged.85 In the former case, materials with high susceptibility will produce large magnetic field distortions and lead to severe image artifacts.
Materials with widely different susceptibilities also produce image artifacts, since
local perturbations in the magnetic field will be created from the different materials. These perturbations may result in spin dephasing between voxels or geometric
distortions by warping the imaging plane. In either case, so-called susceptibility
artifacts are produced.8688
It is also important to consider the generation of eddy currents in any material
introduced into the MR system. Eddy currents form when conductive materials are
placed in the changing magnetic fields produced by the gradient coils or B1 field.
These induced fields may alter the magnetization vector flip angle, thereby resulting
in image artifacts.
Along with artifacts produced in the MR images, it is also important to consider
the effect of the magnetic fields and the RFs on the performance of the radiation
detection equipment. If used in conjunction with PET imaging, the magnetic field
used in MR may actually have a beneficial effect on spatial resolution.89 This is
because the charged positrons produced as a result of the radioactive decay process
will experience a Lorentz force on them as a result of moving through the strong
magnetic field. This force will be directed perpendicular to the magnetic field so that
the net effect is to produce a helical path. The radius of this path is given by

R=
0.334
B
( 2m E ) + E
p
2
t
where B is the magnetic field strength, mp is the positron rest mass, and Et is the positron kinetic energy perpendicular to the magnetic field. While MRI systems have
multiple magnetic field gradients, the main static field is the only magnet that appreciably affects the positron path. Given that this magnetic field is oriented axially, it is
expected that spatial resolution improvements will only be seen along this direction.
Nevertheless, improvements in spatial resolution are seen to be appreciable when the
positron energy is significant, with little improvement with lower-energy positrons
such as those emitted from F-18.90 Table14.2 depicts some of the improvements seen
for different nuclides in the presence of high B 0 fields.
Aside from the theoretical improvements in spatial resolution, most of the other
consequences of the MR device will be detrimental to the operation of the radiation
detectors. As mentioned, the magnetic field will have a disastrous effect on PMT
performance, thereby rendering them useless. Even if alternative detectors may be
resilient to high magnetic fields, they may be susceptible to RF interference from the
RF pulses used in MRI. This is particularly true of highly sensitive components such
as APDs. The need to place additional electronics such as preamplifiers or ASICs in
the vicinity of the magnetic field may also introduce problems with electrical interference. To overcome some of these effects, copper shielding may be used to reduce
the amount of RF interference.
348
Table14.2
Spatial Resolution Measurements Expected for Some
of the Most Common PET Radionuclides in the Presence
of a 10-T Magnetic Field
Radionuclide
Maximum Energy
(MeV)
FWHM at 0 T
(mm)
FWHM at 10 T
(mm)
C
N
15O
18F
68Ga
82Rb
0.96
1.19
1.70
0.63
1.89
3.15
4.24 0.07
4.44 0.06
5.28 0.10
3.85 0.06
5.46 0.10
8.03 0.15
3.73 0.07
3.80 0.06
3.80 0.06
3.70 0.06
3.86 0.06
3.91 0.05
11
13
Source: From Raymann, R. R., Hammer, B. E., & Christensen, N. L., IEEE
Trans. Nucl. Sci., 43, 24062412, 1996. With permission.
For SPECT imaging, the choice of collimator material introduces some problems
of its own. The large lead collimator typically used for SPECT would probably not
be optimal for a SPECT-capable system, since the electrical conductivity results in
induced eddy currents when exposed to RF that may yield magnetic field inhomogeneities83 or susceptibility artifacts in the MR images.
To produce 3D images using SPECT, a typical parallel-hole collimator must rotate
around the object, collecting views at multiple angles. This motion would most likely
require the application of an electric motor, the exact placement of which, for obvious reasons, cannot be directly in the MR bore. Alternative approaches using a ring
of stationary cameras may alleviate some of the technical challenges involved in
mechanical rotation. Designs that incorporate full detector rings (e.g., Rowe et al.;
Goertzen et al.; Genna and Smith)9193 with multiple pinhole or coded aperture collimators would appear to be the most promising.
14.6.2System Design
Given the space constraints within an MRI bore, it important to consider the placement of the SPECT or PET system within the bore. In essence, three different designs
for integrated nuclear/MRI are possible:94 (a) separate systems, shared patient bed;
(b) removable insert for PET or SPECT; and (c) fully integrated system (as shown in
Figure14.10). The first system would necessarily involve sequential imaging rather
than simultaneous imaging, and while the resultant images would be reasonably
coregistered, it would still require at least twice the imaging time compared to simultaneous imaging. On the other hand, it has been shown that existing PMTs may be
used when coupled to PET detectors via fiberoptics, so the first system may currently be clinically feasible. The second and third systems, however, would involve
continued development of new radiation detection systems and, in the case of the
349
Gradient
PET-Ring
Gradie
nt
RF-Coil
RF-Screen
Magnet
Figure 14.10 Theoretical integrated PET/MR system showing PET detector ring integrated into the gradient coil subsystem. This system would appear externally as a conventional MR system and be capable of whole-body PET/MR imaging. (From http://www.
Philips.com.)
third system, a complete redesign of the MRI and PET/SPECT system to integrate
all components.
In addition, consideration must be made regarding where to place the respective
detectors (i.e., RF coils and scintillators) required for imaging. As mentioned, in
MRI the receiver RF coils are usually placed as close as possible to the patient to
maximize the SNR. To reduce costs and improve spatial resolution, PET or SPECT
detectors must also be placed close to the patient being imaged, thereby creating a
conflict regarding the exact placement of each detector system. It seems logical to
place the RF coils closest to the patient, but the addition of these coils may introduce
attenuation or scatter artifacts into the PET or SPECT data.
14.6.3MR Attenuation Correction

As described, the conversion of CT numbers to linear attenuation coefficients must
be performed to use the CT for attenuation correction.95 While not necessarily trivial, this conversion is still rather straightforward, since CT numbers do have a direct
relation to photon attenuation (i.e., both related to electron density). The conversion
of MR images to a linear attenuation coefficient is not as straightforward, however,
since MR images represent a measure of proton density and magnetic properties,
not necessarily correlated to electron density. As a result, alternative approaches to
derive linear attenuation coefficients must be devised.
350
The simplest approach is simply to segment MR images into air, soft tissue, and
bone and to apply known attenuation coefficients to these regions. This technique
has been successfully applied to neurological imaging96 but does require some user
intervention to aid in image segmentation. While working reasonably well for the
brain, problems exist in extending this approach to the rest of the body, which encompasses many more tissue types.97 Alternative approaches have mapped MR images
to standard attenuation coefficients98 or have mapped MR to CT images to derive
linear attenuation coefficients.99 Such techniques require additional image coregistration to align the patient-specific MR with a generalized atlas. Since PET/MR and
SPECT/MR are still relatively early in their development, ongoing research will no
doubt improve the accuracy of MR-based attenuation correction schemes.100
14.7Conclusions
The development of MR-compatible PET and SPECT imaging devices has rapidly increased with the development of new radiation detectors. The introduction
of Geiger-mode silicon photomultipliers and improved production yields of highquality CZT crystals has meant that there are now real alternatives to PMTs and
scintillators for PET and SPECT. While the development of PET/MR and SPECT/
MR may seem to offer many advantages over separate systems, with continued
development of these systems come other issues relating to health care costs101 and
patient safety102 that will need to be addressed before the mainstream clinical adoption of these technologies. Given this, it is not surprising that much of the early
development on these devices has been applied to preclinical imaging.103105 Many
soft tissue tumor models cannot be adequately visualized with X-ray CT, since soft
tissue contrast is not sufficient to delineate tumor from tissue. However, MRI, with
its superior soft tissue contrast, has proved itself valuable for many preclinical imaging studies. The ability to visualize the in vivo distribution of new radiotracers in
the context of high-resolution, high-contrast MRI has provided valuable insights into
drug metabolism and disease progression.
While many technical challenges still need to be overcome before PET/MR or
SPECT/MR devices are commonplace, significant progress has been made to the point
at which prototype systems have been developed, and initial imaging studies have been
performed (see Figure14.11a and 14.11b). Already, a number of potential clinical applications have been identified that will benefit from further SPECT/MR or PET/MR
development. Applications such as oncology,106 neurology,107,108 cardiology,109 abdominal imaging,110 and cell trafficking111,112 will keep pushing the need for these devices,
and we will no doubt continue to see great technological strides being made.
351
(a)
(b)
Figure 14.11 (a) First simultaneously acquired PET/MR images from the PSAPD PET
detector ring developed at the University of California at Davis. The PET image depicts
18F uptake within the skeletal system. (From Catana, C. et al., J. Nucl. Med., 47, 2006.)62
(b) Simultaneously acquired PET/MR image depicting 18FDG uptake in the brain of a human
subject. The MR image was acquired using a FLAIR sequence, and the entire MR plus PET
procedure was acquired in 15 min. (From Schlemmer, H. P. et al., Abdom. Imag., 2008. With
permission.)110
352
References

1. D. L. Bailey, B. F. Hutton, and P. J. Walker, Improved SPECT using simultaneous emission and transmission tomography, J. Nucl. Med., 28 (5), 844851, 1987.
2. D. L. Bailey, Transmission scanning in emission tomography, Eur. J. Nucl. Med., 25 (7),
774787, 1998.
3. R. A. deKemp and C. Nahmias, Attenuation correction in PET using single photon transmission measurement, Med. Phys., 21 (6), 771778, 1994.
4. A. Celler, A. Sitek, E. Stoub, P. Hawman, R. Harrop, and D. Lyster, Multiple line source
array for SPECT transmission scans: Simulation, phantom and patient studies, J. Nucl.
Med., 39 (12), 21832189, 1998.
5. A. C. Evans, S. Marrett, J. Torrescorzo, S. Ku, and L. Collins, MRI-PET Correlation
in Three Dimensions Using a Volume-of-interest (VOI) Atlas, J. Cereb. Blood Flow
Metab., 11 (2), A69A78, 1991.
6. R. L. Wahl, L. E. Quint, R. D. Cieslak, A. M. Aisen, R. A. Koeppe, and C. R. Meyer,
Anatometabolic tumor imaging: Fusion of FDG PET with CT or MRI to localize foci
of increased activity, J. Nucl. Med., 34 (7), 11901197, 1993.
7. T. Beyer, D. W. Townsend, T. Brun, P. E. Kinahan, M. Charron, R. Roddy, J. Jerin,
J. Young, L. Byars, and R. Nutt, A combined PET/CT scanner for clinical oncology,
J. Nucl. Med., 41, 13691379, 2000.
8. P. E. Kinahan, D. W. Townsend, T. Beyer, and D. Sashin, Attenuation correction for a
combined 3D PET/CT scanner, Med. Phys., 25, 20462053, 1998.
9. D. W. Townsend, T. Beyer, T. M. Blodget, PET/CT scanners: A hardware approach to
image fusion, Semin. Nucl. Med., 33(3), 193204, 2003.
10. T. F. Lang, B. H. Hasegawa, S. C. Liew, J. K. Brown, S. C. Blankespoor, S. M. Reilly, E.
L. Gingld, and C. E. Cann, Description of a prototype emission-transmission computed
tomography imaging system, J. Nucl. Med., 33, 18811887, 1992.
11. B. F. Hutton and M. Braun, Software for image registration: Algorithms, accuracy, efficacy, Semin. Nucl. Med., 33 (3), 180192, 2003.
12. J. P. W. Pluim, J. B. A. Maintz, and M. A. Viergever, Mutual information based registration of medical images: A survey, IEEE Trans. Med. Imaging, 22 (8), 9861004, 2003.
13. J. P. W. Pluim, J. B. A. Maintz, and M. A. Viergever, Image registration by maximization
of combined mutual information and gradient information, IEEE Trans. Med. Imaging,
19 (8), 809814, 2000.
14. R. Stokking, G. I. Zubal, and M. A. Viergever, Display of fused images: Methods, interpretation, and diagnostic improvements, Semin. Nucl. Med., 31(3), 219227, 2003.
15. P. E. Kinahan, B. H. Hasegawa, and T. Beyer, X-ray-based attenuation correction for
positron emission tomography/computed tomography scanners, Semin. Nucl. Med.,
33(3), 166179, 2003.
16. J. P. J. Carney, D. W. Townsend, V. Rappoport, and B. Bendriem, Method for transforming CT images for attenuation correction in PET/CT imaging, Med. Phys., 33 (4),
976983, 1996.
17. Burger C, Goerres G, Schoenes S, Buck A, Lonn AHR, von Schulthess GK, PET attenuation coefficients from CT images: Experimental evaluation of the transformation of CT
into PET 511-keV attenuation coefficients, Eur. J. Nucl. Med., 29, 922927, 2002.
18. K. J. LaCroix, B. M. W. Tsui, B. H. Hasegawa, and J. K. Brown, Investigation of the use
of X-ray CT images for attenuation compensation in SPECT, IEEE Trans. Nucl. Sci., 41
(6), 27932799, 1994.
19. G. H. Pantvaidya, J. P. Agarwal, M. S. Deshpande, V. Rangarajan, V. Singh, A. Kakade,
and A. K. DCruz, PET-CT in recurrent head neck cancers: A study to evaluate impact
on patient management, J. Surg. Oncol., 2009.
353
20. F. U. Chowdhury, K. M. Bradley, and F. V. Gleeson, The role of 18F-FDG PET/CT in the
evaluation of oesophageal carcinoma, Clin. Radiol., 63 (12), 12971309, 2008.
21. M. Horger, S. M. Eschmann, C. Pfannenberg, D. Storek, R. Vonthein, C. D. Claussen,
and R. Bares, Added value of SPECT/CT in patients suspected of having bone infection:
Preliminary results, Arch. Orthop. Trauma Surg., 127 (3), 211221, 2007.
22. J. Nathan, J. A. Crawford, D. B. Sodee, and G. Bakale, Fused SPECT/CT imaging of
peri-iliopsoas infection using indium-111-labeled leukocytes, Clin. Nucl. Med., 31 (12),
801802, 2006.
23. C. J. Ingui, N. P. Shah, and M. E. Oates, Infection scintigraphy: Added value of singlephoton emission computed tomography/computed tomography fusion compared with
traditional analysis, J. Comput. Assist. Tomogr., 31 (3), 375380, 2007.
24. R. Bar-Shalom, N. Yefremov, L. Guralnik, Z. Keidar, A. Engel, S. Nitecki, and O. Israel,
SPECT/CT using 67Ga and 111In-labeled leukocyte scintigraphy for diagnosis of infection, J. Nucl. Med., 47 (4), 587594, 2006.
25. L. Filippi and O. Schillaci, Usefulness of hybrid SPECT/CT in 99mTc-HMPAO-labeled
leukocyte scintigraphy for bone and joint infections, J. Nucl. Med., 47 (12), 19081913,
2006.
26. T. Schepis, O. Gaemperli, P. Koepfli, M. Namdar, I. Valenta, H. Scheffel, S. Leschka, L.
Husmann, F. R. Eberli, T. F. Luscher, H. Alkadhi, and P. A. Kaufmann, Added value of
coronary artery calcium score as an adjunct to gated SPECT for the evaluation of coronary artery disease in an intermediate-risk population, J. Nucl. Med., 48 (9), 14241430,
2007.
27. S. J. McQuaid and B. F. Hutton, Sources of attenuation-correction artefacts in cardiac
PET/CT and SPECT/CT, Eur. J. Nucl. Med. Mol. Imaging, 35 (6), 11171123, 2008.
28. L. Le Meunier, R. Maass-Moreno, J. A. Carrasquillo, W. Dieckmann, and S. L. Bacharach,
PET/CT imaging: Effect of respiratory motion on apparent myocardial uptake, J. Nucl.
Cardiol., 13 (6), 821830, 2006.
29. A. Saoudi, C. M. Pepin, and R. Lecomte, Study of light collection in multi-crystal detectors, J. Surg. Oncol., 100 (5), 401403, 2009.
30. S. Yamamoto, K. Kuroda, and M. Senda, Scintillator selection for MR-compatible
gamma detectors, IEEE Trans. Nucl. Sci., 50 (5), 16831685, 2003.
31. G. F. Knoll, Radiation detection and measurement, 3rd edition, New York: Wiley, 2000.
32. J. A. Bieszk, Performance changes of an anger camera in magnetic fields up to 10 G,
J. Nucl. Med., 27, 19021907, 1986.
33. C. Goetz, E. Breton, P. Choquet, V. Israel-Jost, and A. Constantinesco, SPECT low-field
MRI system for small-animal imaging, J. Nucl. Med., 49 (1), 8893, 2008.
34. K. M. Gilbert, W. B. Handler, T. J. Scholl, J. W. Odegaard, and B. A. Chronik, Design
of Field-Cycled Magnetic Resonance Systems for Small Animal Imaging, Phys. Med.
Biol., 51, 28252841, 2006.
35. W. B. Handler, K. M. Gilbert, H. Peng, and B. A. Chronik, simulation of scattering and
attenuation of 511 keV photons in a combined PET/field-cycled MRI system, Phys.
Med. Biol., 51, 24792491, 2006.
36. H. Peng, Investigation of new approaches to combined positron emission tomography and
magnetic resonance imaging systems. Ph.D. thesis. University of Western Ontario, 2006.
37. Y. Shao, S. R. Cherry, K. Farahani, R. Slates, R. W. Silverman, K. Meadors, A. Bowery,
and S. Siegel, Development of a PET detector system compatible with MRI/NMR systems, IEEE Trans. Nucl. Sci., 44 (3), 11671171, 1997.
38. G. M. Haak, B. E. Hammer, and N. L. Christensen, Coupling scintillation light into
optical fiber for use in a combined PET-MRI scanner, Nucl. Instrum. Methods A, 399,
455462, 1997.
354
39. Y. Shao, S. R. Cherry, K. Farahani, K. Meadors, S. Siegel, R. W. Silverman, and P. K.

Marsden, Simultaneous PET and MR imaging, Phys. Med. Biol., 42, 19651970, 1997.
40. R. Lecomte, Novel detector technology for clinical PET, Eur. J. Nucl. Med. Mol.
Imaging, 36 (suppl 1), S69S85, 2009.
41. M. Asghar and D. C. Imrie, Silicon photodiodes as photomultipler replacements in
industrial scintillation counters, J. Phys. E: Sci. Instrum., 22, 10121015, 1989.
42. C. S. Levin and E. J. Hoffman, Investigation of a new readout scheme for high resolution
scintillation crystal arrays using photodiodes. IEEE Trans. Nucl. Sci., 44 (3), 12081213,
1997.
43. D. E. Groom, Silicon photodiode detection of bismuth germanate scintillation light,
44. G. J. Gruber, W. W. Moses, S. E. Derenzo, N. W. Wang, E. Beuville, and M. H. Ho, A
discrete scintillation camera module using silicon photodiode readout of CsI(Tl) crystals
for breast cancer imaging, IEEE Trans. Nucl. Sci., 45, 10631068, 1998.
45. J. Kataoka, R. Sato, T. Ikagawa, J. Kotoku, Y. Kuramoto, Y. Tsubuku, T. Saito, Y. Yatsu,
N. Kawai, Y. Ishikawa, and N. Kawabata, An active gain-control system for avalanche
photo-diodes under moderate temperature variations, Nucl. Instrum. Methods A, 564,
300307, 2006.
46. Y. K. Su, C. Y. Change, and T. S. Wu, Temperature dependent characteristics of a PIN
avalanche photodiode (APD) in Ge, Si and GaAs, Opt. Quantum Elect., 11, 109117,
1979.
47. N. Z. Hakim, B. E. A. Saleh, and M. C. Teich, Generalized excess noise factor for
avalanche photodiodes of arbitrary structure, IEEE Trans. Elec. Dev., 37 (3), 599610,
1990.
48. A. W. Lighthouse, R. J. McIntyre, R. Lecomte, and D. Schmitt, A bismuth germanateavalanche photodiode module designed for use in high resolution positron emission
tomography, IEEE Trans. Nucl. Sci., 33 (1), 456459, 1986.
49. C. J. Marriott, J. E. Cadorette, R. Lecomte, V. Scasnar, J. Rousseau, and J. E. van Lier,
High resolution PET imaging and quantitation of pharmaceutical biodistributions in
a small animal using avalanche photodiode detectors, J. Nucl. Med., 35, 13901397,
1994.
50. R. Lecomte, J. Cadorette, S. Rodrigue, D. Lapointe, D. Rouleau, M. Bentourkia, R. Yao,
and P. Msaki, Initial results from the Sherbrooke avalanche photodiode positron tomograph, IEEE Trans. Nucl. Sci., 43 (3), 19521957, 1996.
51. R. Lecomte, J. Cadorette, A. Jouan, M. Heon, D. Rouleau, and G. Gauthier, High resolution positron emission tomography with a prototype camera based on solid state scintillation detectors, IEEE Trans. Nucl. Sci., 37 (2), 805811, 1990.
52. J. Cadorette, S. Rodrigue, and R. Lecomte, Tuning of avalanche photodiode PET camera, IEEE Trans. Nucl. Sci., 40 (4), 10621066, 1993.
53. R. Fontaine, F. Belanger, N. Viscogliosi, H. Semmaoui, M. A. Tetrault, J. B. Michaud,
C. Pepin, J. Cadorette, and R. Lecomte, The hardware and signal processing architecture
of LabPET, a small animal APD-based digital PET scanner, IEEE Trans. Nucl. Sci.,
56 (1), 39, 2009.
54. B. J. Pichler, M. S. Judenhofer, C. Catana, J. H. Walton, M. Kneilling, R. E. Nutt, S. B.
Siegel, C. D. Claussen, and S. R. Cherry, Performance test of an LSO-APD detector in
a 7-T MRI scanner for simultaneous PET/MRI, J. Nucl. Med., 47, 639647, 2006.
55. B. J. Pichler, B. K. Swann, J. Rochelle, R. E. Nutt, S. R. Cherry, and S. B. Siegel,
Lutetium oxyorthosilicate block detector readout by avalanche photodiode arrays for
high resolution animal PET, Phys. Med. Biol., 49, 43054319, 2004.
355
56. S. I. Ziegler, B. J. Pichler, G. Boening, M. Rafecas, W. Pimpl, E. Lorenz, N. Schmitz,

and M. Schwaiger, A prototype high-resolution animal positron tomograph with avalanche photodiode arrays and LSO crystals, Eur. J. Nucl. Med. Mol. Imaging, 28 (2),
136143, 2001.
57. K. S. Shah, R. Farrell, R. Grazioso, E. S. Harmon, and E. Karplus, Position-sensitive avalanche photodiodes for gamma-ray imaging, IEEE Trans. Nucl. Sci., 49 (4), 16871692,
2002.
58. P. A. Dokhale, R. W. Silverman, K. S. Shah, R. Grazioso, R. Farrell, J. Glodo, M. A.
McClish, G. Entine, V.-H. Tran, and S. R. Cherry, Performance measurements of a
depth-encoding PET detector module based on position-sensitive avalanche photodiode
read-out, Phys. Med. Biol., 49, 42934303, 2004.
59. C. S. Levin, A. M. K. Foudray, P. D. Olcott, and F. Habte, Investigation of position sensitive avalanche photodiodes for a new high-resolution PET detector design, IEEE Trans.
Nucl. Sci., 51 (3), 805810, 2004.
60. K. C. Burr, A. Ivan, J. LeBlanc, S. Zelakiewicz, D. L. McDaniel, C. L. Kim, A. Ganin,
K. S. Shah, R. Grazioso, R. Farrell, and J. Glodo, Evaluation of a position sensitive
avalanche photodiode for PET, IEEE Trans. Nucl. Sci., 50 (4), 792796, 2003.
61. P. Desprs, T. Funk, K. S. Shah, and B. H. Hasegawa, Monte Carlo simulations of
compact gamma cameras based on avalanche photodiodes, Phys. Med. Biol., 52 (11),
30573074, 2007.
62. C. Catana, Y. Wu, M. S. Judenhofer, J. Qi, B. J. Pichler, and S. R. Cherry, Simultaneous
acquisition of multislice PET and MR images: Initial results with a MR-compatible PET
scanner, J. Nucl. Med., 47, 19681976, 2006.
63. J. E. Mackewn, D. Struhl, W. A. Hallett, P. Halstead, R. A. Page, S. F. Keevil, S. C. R.
Williams, and S. R. Cherry, Design and development of an MR-compatible PET scanner
for imaging small animals, IEEE Trans. Nucl. Sci., 52 (5), 13761380, 2005.
64. M. S. Judenhofer, C. Catana, B. K. Swann, S. B. Siegel, W. Jung, R. E. Nutt, S. R.
Cherry, C. D. Claussen, and B. J. Pichler, PET/MR images acquired with a compact
MR-compatible PET detector in a 7-T magnet, Radiology, 244 (3), 807814, 2007.
65. C. Catana, D. Procissi, Y. Wu, M. S. Judenhofer, J. Qi, B. J. Pichler, R. E. Jacobs, and S.
R. Cherry, Simultaneous in vivo positron emission tomography and magnetic resonance
imaging, Proc. Natl. Acad. Sci. U. S. A., 105 (10), 37053710, 2008.
66. B. F. Aull, A. H. Loomis, D. J. Young, R. M. Heinrichs, B. J. Felton, P. J. Daniels, and D.
J. Landers, Geiger-mode avalanche photodiodes for three-dimensional imaging, Lincoln
Lab. J., 13 (2), 335350, 2002.
67. D. Pellion, V. Borrel, D. Esteve, F. Therez, F. Bony, A. R. Bazer-Bachi, and J. P. Gardou,
APD detectors in the Geiger photon counter mode, Nucl. Instrum. Methods A, 567, 4144,
2006.
68. D. Renker, Geiger-mode avalanche photodiodes, history, properties and problems, Nucl.
69. Y. Musienko, S. Reucroft, and J. Swain, The gain, photon detection efficiency and excess
noise factor of multi-pixel geiger-mode avalanche photodiodes, Nucl. Instrum. Methods
A, 567, 5761, 2006.
70. A. Nassalski, M. Moszynski, A. Syntfeld-Kazuch, T. Szczesniak, L. Swiderski, D.
Wolski, T. Batsch, and J. Baszak, Silicon photomultiplier as an alternative for APD in
PET/MRI applications, IEEE Med. Imaging Conf. Rec., 16201625, 2008.
71. G. Llosa, R. Battiston, N. Belcari, M. Boscardin, G. Collazuol, F. Corsi, G.-F. Dalla
Betta, A. Del Guerra, N. Dinu, G. Levi, S. Marcatili, S. Moehrs, C. Marzocca, C.
Piemonte, and A. Pozza, Novel silicon photomultipliers for PET applications, IEEE
Trans. Nucl. Sci., 55 (3), 877881, 2008.
356
72. D. J. Herbert, V. Saveliev, N. Belcari, N. DAscenzo, A. Del Guerra, and A. Golovin,

First results of scintillator readout with silicon photomultiplier, IEEE Trans. Nucl. Sci.,
53 (1), 389394, 2006.
73. P. Buzhan, B. Dolgoshein, L. Filatov, A. Ilyin, V. Kaplin, A. Karakash, S. Klemin, R.
Mirzoyan, A. N. Otte, E. Popova, V. Sosnovtsev, and M. Teshima, Large area silicon
photomultipliers: Performance and applications, Nucl. Instrum. Methods A, 567, 7882,
2006.
74. A. N. Otte, J. Barral, B. Dolgoshein, J. Hose, S. Klemin, E. Lorenz, R. Mirzoyan, E.
Popova, and M. Teshima, A Test of Silicon Photomultipliers as Readout for PET, Nucl.
75. G. Di Domenico, G. Zavattini, N. Cesca, N. Auricchio, R. Andritschke, F. Schopper, and
G. Kanbach, SiliPET: An ultra-high resolution design of a small animal PET scanner
based on stacks of double-sided silicon strip detector, Nucl. Instrum. Methods A, 571,
2225, 2007.
76. A. Drezet, O. Monnet, F. Mathy, G. Montemont, and L. Verger, CdZnTe detectors for
small field of view positron emission tomographic imaging, Nucl. Instrum. Methods A,
571, 465470, 2007.
77. P. Depres, E. W. Izaguirre, S. Liu, L. J. Cirignano, H. Kim, M. F. Wendland, K. S. Shah,
and B. H. Hasegawa, Evaluation of a MR-compatible CZT detector, IEEE Med. Imaging
Conf. Rec., 43244326, 2007.
78. D. J. Wagenaar, O. Nalcioglu, L. T. Muftuler, M. Szawlowski, M. Kapusta, N. Pavlov,
D. Meier, G. Maehlum, and B. Patt, Development of MRI-compatible nuclear medicine
imaging detectors, IEEE Med. Imaging Conf. Rec., 18251828, 2006.
79. S. Azman, J. Gjaerum, D. Meier, L. T. Muftuler, G. Maehlum, O. Nalcioglu, B. E. Patt,
B. Sundal, M. Szawlowski, B. M. W. Tsui, D. J. Wagenaar, and Y. Wang, A nuclear
radiation detector system with integrated readout for SPECT/MR small animal imaging,
IEEE Med. Imaging Conf. Rec., 23112317, 2007.
80. C. Szeles, S. E. Cameron, J. O. Ndap, and W. C. Chalmers, Advances in the crystal
growth of semi-insulating CdZnTe for radiation detector applications, IEEE Trans.
Nucl. Sci., 49 (5), 25352540, 2002.
81. H. Chen, S. A. Awadalla, J. Mackenzie, R. Redden, G. Bindley, A. E. Bolotnikov, G.
S. Camarda, G. Carini, and R. B. James, Characterization of traveling heater method
(THM) grown Cd0.9Zn0.1Te crystals, IEEE Trans. Nucl. Sci., 54 (4), 811816, 2007.
82. S. A. Awadalla, H. Chen, J. Mackenzie, P. Lu, K. Iniewski, P. Marthandam, R. Redden,
G. Bindley, Z. He, F. Zhang, M. Groza, A. Burger, D. R. Mayo, and C. L. Sullivan,
Thickness scalability of large volume cadmium zinc telluride high resolution radiation
detectors, IEEE Med. Imaging Conf. Rec., 5862, 2008.
83. D. Strul, D. Cash, S. F. Keevil, S. C. R. Williams, and P. K. Marsden, Gamma shielding
materials for MR-compatible PET, IEEE Trans. Nucl. Sci., 50(1), 6069, 2003.
84. D. Jiles, Introduction to Magnetism and Magnetic Materials, Boca Raton, FL: CRC
Press, 1998.
85. J. F. Schenck, The role of magnetic susceptibility in magnetic resonance imaging:
MRI magnetic compatibility of the first and second kinds, Med. Phys., 23(6), 815850,
1996.
86. K. M. Ludeke, P. Roschmann, and R. Tischler, Susceptibility artefacts in NMR imaging,
Magn. Reson. Imaging, 3, 329343, 1985.
87. C. J. G. Bakker, R. Bhagwandien, M. A. Moerland, and M. Fuderer, Susceptibility artifacts in 2DFT spin-echo and gradient echo imagingThe cylinder model revisited,
Magn. Reson. Imaging, 11, 539548, 1993.
88. C. R. Camacho, D. B. Plewes, and R. M. Henkelman, Nonsusceptibility artifacts due to
nonmetallic objects in MR imaging, J. Magn. Reson. Imaging, 5, 7588, 1995.
357
89. A. Wirrwar, H. Vosberg, H. Herzog, H. Halling, S. Weber, and H. Muller-Gartner, 4.5

Tesla magnetic field reduces range of high-energy positronsPotential implications for
positron emission tomography, IEEE Trans. Nucl. Sci., 44 (2), 184189, 1997.
90. R. R. Raylmann, B. E. Hammer, and N. L. Christensen, Combined MRI-PET scanner:
A Monte Carlo evaluation of the improvements in PET resolution due to the effects of a
static homeogeneous magnetic field, IEEE Trans. Nucl. Sci., 43, 24062412, 1996.
91. R. Rowe, J. N. Aarsvold, H. H. Barrett, J. C. Chen, W. P. Klien, B. A. Moore, I. W. Pang,
D. D. Patton, and T. A. White, A stationary hemispherical SPECT imager for threedimensional brain imaging, J. Nucl. Med., 34, 474480, 1993.
92. A. L. Goertzen, D. W. Jones, J. Seidel, K. Li, and M. V. Green, First results from the
high-resolution MouseSPECT annular scintillation camera, IEEE Trans. Med. Imaging,
24 (7), 863867, 2005.
93. S. Genna and A. P. Smith, The development of ASPECT, an annular single crystal brain
camera for high efficiency SPECT, IEEE Trans. Nucl. Sci., 35 (1), 654658, 1988.
94. G. Delso and S. Ziegler, PET/MRI system design, Eur. J. Nucl. Med. Mol. Imaging, 36,
S86S92, 2009.
95. H. Zaidi, Is MR-guided attenuation correction a viable option for dual-modality PET/MR
imaging? Radiology, 244(3), 639642, 2007.
96. H. Zaidi, M. Montandon, and D. O. Slosman, Magnetic resonance imaging-guided
attenuation and scatter corrections in three-dimensional brain positron emission tomography, Med. Phys., 30(5), 937948, 2003.
97. T. Beyer, M. Weigert, H. H. Quick, U. Pietrzyk, F. Vogt, C. Palm, G. Antoch, S. P. Muller,
and Bockisch, MR-based attenuation correction for torso-PET/MR imaging: Pitfalls in
mapping MR to CT data, Eur. J. Nucl. Med. Mol. Imaging, 35, 11421146, 2008.
98. M. Hofmann, B. J. Pichler, B. Scholkopf, T. Beyer, Towards quantitative PET/MRI: A
review of MR-based attenuation correction techniques, Eur. J. Nucl. Med. Mol. Imaging,
2008.
99. M. Hofman, F. Steinke, V. Scheel, G. Charpiat, J. Farquhar, P. Aschoff, M. Brady,
Scholkopf, and B. J. Pichler, MRI-based attenuation correction for PET/MRI: A novel
approach combining pattern recognition and atlas registration, J. Nucl. Med., 49,
18751883, 2008.
100. E. R. Kops and H. Herzog, Alternative methods for attenuation correction for PET
images in MR-PET scanners, IEEE Med. Imaging Conf. Rec., 43274330, 2007.
101. M. Goyen and J. F. Debatin, Healthcare costs for new technologies, Eur. J. Nucl. Med.
Mol. Imaging, 36 (suppl 1), S139S143, 2009.
102. G. Brix, E. A. Nekolla, D. Nosske, and J. Griebel, Risks and safety aspects related to
PET/MR examinations, Eur. J. Nucl. Med. Mol. Imaging, 36 (suppl 1), S131S138,
2009.
103. H. F. Wehrl, M. S. Judenhofer, S. Wiehr, and B. J. Pichler, Pre-clinical PET/MR:
Technological advances and new perspectives in biomedical research, Eur. J. Nucl. Med.
Mol. Imaging, 36 (suppl 1), S56S68, 2009.
104. S. R. Cherry, A. Y. Louie, and R. E. Jacobs, The integration of positron emission tomography with magnetic resonance imaging, Proc. IEEE, 96 (3), 416438, 2008.
105. S. Azman, J. Gjaerum, D. Meier, L. T. Muftuler, G. Maehlum, O. Nalcioglu, B. E. Patt,
B. Sundal, M. Szawlowski, B. M. W. Tsui, D. J. Wagenaar, and Y. Wang, A nuclear
radiation detector system with integrated readout for SPECT/MR small animal imaging,
IEEE Med. Imaging Conf. Rec., 23112317, 2007.
106. G. Antoch and A. Bockisch, Combined PET/MRI: A new dimension in whole-body
oncology imaging?, Eur. J. Nucl. Med. Mol. Imaging, 36 (suppl 1), S113S120, 2009.
107. W. D. Heiss, The potential of PET/MR for brain imaging, Eur. J. Nucl. Med. Mol.
Imaging, 36 (suppl 1), S105S112, 2009.
358
108. H. Schlemmer, B. Pichler, K. Wienhard, M. Schmand, C. Nahmias, D. Townsend, et al.

Simultaneous MR/PET for brain imaging: First Patient Scans, J. Nucl. Med., 48(6), 45P,
2007.
109. S. G. Nekolla, A. Matinez-Moeller, and A. Saraste, PET and MRI in cardiac imaging:
From validation studies to integrated applications, Eur. J. Nucl. Med. Mol. Imaging, 36
(suppl 1), S121S130, 2009.
110. H. P. Schlemmer, B. J. Pichler, R. Krieg, W. D. Heiss, An integrated MR/PET sytem:
Prospective applications, Abdom. Imaging 2008; http://www.SpringerLink.com.
111. H. Qiao, H. Zhang, Y. Zheng, D. E. Ponde, D. Shen, F. Gao, A. B. Bakken, A. Schmitz,
H. F. Kung, V. A. Ferrari, and R. Zhou, Embryonic stem cell grafting in normal and
infarcted myocardium: Serial assessment with MR imaging and PET dual detection,
Radiology, 250 (3), 821829, 2009.
112. H. Y. Lee, Z. Li, K. Chen, A. R. Hsu, C. Xu, J. Xie, S. Sun, and X. Chen, PET/MRI
dual-modality tumor imaging using arginine-glycine-aspartic (RGD)-conjugated radiolabeled iron oxide nanoparticles, J. Nucl. Med., 49 (8), 13711379, 2008.
X-Ray Imaging
15 Hard
Detectors Onboard
the Balloon-Borne
High-Energy Focusing
Telescope
C. M. Hubert Chen, Walter R. Cook,
Kristin Kruse Madsen, and Fiona A. Harrison
James C. Chonko, Jason E. Koglin,

and Charles J. Hailey
Columbia University
New York, New York
Finn E. Christensen and Carsten P. Jensen

Copenhagen, Denmark
William W. Craig and Klaus Ziock
Lawrence Livermore National Laboratory

Contents
15.1 Introduction...................................................................................................360
15.1.1 Detector Technology In-Orbit Today.................................................360
15.1.2 The Future: Direct Imaging by Focusing.......................................... 361
15.1.3 Focal Plane Detectors for Focusing Telescopes................................ 361
15.1.4 Scientific Ballooning......................................................................... 363
15.2 The High-Energy Focusing Telescope...........................................................364
15.2.1 Performance Objectives.....................................................................364
15.2.2 The Payload.......................................................................................364
359
360
15.2.3 The X-Ray Focal Plane...................................................................... 365

15.2.4 The Focal-Plane Modules.................................................................. 366
15.2.5 The X-Ray Detectors......................................................................... 366
15.2.6 Digital Circuitry................................................................................ 367
15.2.7 Background Shield Assembly............................................................ 367
15.3 Technology Development.............................................................................. 368
15.3.1 Contributions to the Spectral Resolution........................................... 368
15.3.2 The Small Pixel Effect....................................................................... 369
15.3.3 Depth Sensing.................................................................................... 370
15.3.4 Hybridization..................................................................................... 370
15.3.5 Anode Contact Size and Steering Electrodes.................................... 371
15.3.6 Bonding Techniques.......................................................................... 372
15.3.7 Material Defects and Characterization.............................................. 374
15.3.8 Processing Signals from the HEFT Detectors................................... 375
15.4 Deployment.................................................................................................... 376
15.4.1 General Characteristics of the Flight Data........................................ 377
15.4.2 In-Flight Background Measurement and Rejection.......................... 378
15.4.3 Observation of Celestial Sources....................................................... 379
15.4.3.1 Cyg X-1............................................................................... 379
15.4.3.2 The Crab Nebula...................................................................380
15.5 Future Outlook............................................................................................... 380
References............................................................................................................... 381
15.1Introduction
Telescope technology in the hard X-ray band of 10100 keV is undergoing rapid
transformation. This band is prime for the study of high-energy, nonthermal astrophysical processes, since thermal emission from stars and diffuse plasma dominates
most of the sky from the infrared up to soft X-rays at around 10 keV. Examples of
nonthermal phenomena of interest in hard X-ray astronomy include the so-called diffuse X-ray background, peaking in energy flux at 30 keV, inverse Compton scattering
of lower-energy photons by relativistic electrons in the haloes of clusters of galaxies
and the radio lobes of active galaxies, and synchrotron and both line emission and
absorption in stellar remnants such as young supernova remnants and pulsars. Our
understanding of these phenomena is limited by the imaging, spectral, and timing
capabilities of hard X-ray detectors currently in orbit. Further advances thus require
the development of new detection technologies specifically for hard X-ray astronomy,
for which most sources are extremely faint, and observation time is always limited.
In the past decade, several telescope projects have contributed to this goal using
the scientific balloon platform. In this chapter, we recount the detector development
effort in one representative project, the High Energy Focusing Telescope (HEFT).1
15.1.1Detector Technology In-Orbit Today

To date, instruments onboard hard X-ray observatories come in two main categories:
colimators and coded aperture imagers. They both operate in the same principle
Hard X-Ray Imaging Detectors Onboard the Balloon-Borne HEFT
361
as pinhole cameras, producing pixels of the sky by restricting the field of view of
the detectors, and imaging through time and spatial multiplexing, respectively.
Table 15.1 lists the systems currently in orbit and their properties. Although collimators and coded aperture imagers enable viable instruments with large collecting
areas and reasonable angular resolution, the detectors employed are inevitably large
when compared to the collecting area. Besides increasing the mass, and thus cost,
of a payload, a large detector volume also implies intrinsically high background in
the X-ray and gamma-ray bands, since spurious particles in the environment of an
astronomical instrument interact with matter within photon detectors to create noise
events. Thus, collimators and coded aperture imagers have only moderate signal-tonoise ratios (SNRs) and limited capabilities to detect faint sources.
15.1.2The Future: Direct Imaging by Focusing

In contrast, the new generation of hard X-ray telescopes in development are all based
on the principle of focusingthat is, the concentration of signal from a large collecting area into a relatively small collecting device by way of mirrors or lenses. The
principal merit of focusing is to decouple the signal-collecting area of an instrumentnow determined by the size of the mirrors or lensesfrom the dimensions
of a major noise source: the photon detector. This way, one increases the amount
of signal (light originating from the target pointing direction) collected by the photon detector without increasing the amount of local background noise. The result
is a substantial increase in SNR. In addition, with focal planes centimeters wide as
opposed to meters, focusing enables the use of photon detectors considerably more
compact than those employed in collimators and coded aperture imagers. Without
the need to cover a large collecting area with detectors, one can pursue alternative detector technologies that are not necessarily scalable in size but have lower
noise and better spectral performance, which leads to further improvement in SNR.
Focusing also makes direct imaging possible, which greatly reduces the complexity
of data reduction and background subtraction.
Readers can find a survey of general hard X-ray mirror designs in Joensen2 and a
description of the specifics of the mirrors on HEFT in Koglin et al.3 Here, we discuss
further only topics related to the focal plane detectors.
15.1.3Focal Plane Detectors for Focusing Telescopes

To take full advantage of the concentration of signals by focusing, focal plane detectors must possess several essential characteristics. They must be compact, they must
have pixel pitch small enough to oversample the arcminute-wide point-spread function (PSF) of focusing mirrors, and they must have high quantum efficiency to capture most of the concentrated signal. To meet these requirements, semiconductor
pixel detectors have been the technology of choice. With significant technological
advances in the semiconductor industry in the past two decades, one can now fabricate inch-size spectroscopy-grade detectors out of high electron density semiconductor materials and pattern them with submillimeter-scale pixels by lithography. At the
same time, very large-scale integration (VLSI) enables the design and fabrication
362
Table15.1
Hard X-Ray Astronomical Instruments Currently in Operation
Mission
Instrument
Detector
Technology
Continuum Sensitivity
(Nominal)
Collimator
Collimator
Coded aperture
Coded aperture
Coded aperture
Nal/Csl
Xe
Ge
CdTe + CsI
Xe
1 Crab
0.1 mCrab
8.8 10 cm4 ph/s/cm2/MeV
2.3 10 cm6 ph/s/cm2/keV
1.2 10 cm4 ph/s/cm2/keV
1 FWHM
1 FWHM
16 FWFC
8 FWFC
4.8 FWFC
Coded aperture
Coded aperture
Coded aperture
Xe
Xe
CdZnTe
30 mCrab
8 109 erg/s/cm2
1 mCrab
6 90 FWHM
90 90 FWZR
120
Field of View
Angular
Resolution
Energy Range
(Kev)
Energy
Resolution
150 FWHM
12 FWHM
3
15250
260
208,000
2010,000
335
15% at 60 keV
18% at 6 keV
0.2% at 1,330 keV
8% at 00 keV
13% at 10 keV
3 15
40
17
210
225
15150
25% at 6 keV
22% at 8 keV
6% throughout
1 FWHM
1 FWHM
For pointed observations:

RXTE
HEXTE
RXTE
PCA
INTEGRAL
SPI
INTEGRAL
IBIS
INTEGRAL
JEM-X
Wide-field monitors:
RXTE
ASM
HETE-2
WXM
Swift
BAT
Imaging
Technology
363
of focal plane electronics (e.g., for amplification and readout) within applicationspecific integration circuits (ASICs), significantly reducing the electronic noise level.
This allows us to measure photon energies to much higher precision than ever before.
The union of focusing technologies and the semiconductor revolution thus brings us
position-sensitive focal plane detectors with high quantum efficiency and high spectral resolution for the first time in the short history of hard X-ray astronomy.
15.1.4Scientific Ballooning
Because the atmosphere completely absorbs celestial X-rays well before they reach the
ground, hard X-ray telescopes can only operate above the bulk of the atmosphere. To
do so, we employ the scientific ballooning platform: An enormous, 40 million cubicfoot zero-pressure balloon carries a typical payload weighing 4,000 pounds (excluding
ballast) up to the top of the atmosphere, where, at an altitude of 40 km, 99.7% of the
atmosphere by mass is below. Twice a year, during a 2-week period called turnaround,
when wind in the upper atmosphere subsides as it reverses direction, a balloon flight
can last for 1030 h, enabling the observation of several targets per flight. Costing a
few hundred thousand dollars per launch, scientific ballooning is a relatively cheap
way to lift a payload above the atmosphere and is thus a common testbed for instruments geared toward future space missions. In the past decade, there have been three
projects in America to develop the first generation of focusing telescopes for hard
X-rays, all on the balloon platform. They are the High Energy Focusing Telescope
(HEFT), described next; the International Focusing Optics Collaboration for Crab
Sensitivity (InFOCS); and the High Energy Replicated Optics (HERO) experiment.
Table 15.2 lists the instrument capabilities in the three experiments. Baumgartner
et al.4 and Gaskin et al.5 describe the semiconductor development effort for InFOCS
and HERO, respectively.
Table15.2
Hard X-Ray Focusing Telescopes Currently Under Development
Mission
HEFT
InFOCS
HERO
Mirror technology
W/Si multilayer
coatings
CdZnTe detector
Pt/C multilayer
coatings
CdZnTe detector
Small grazing angle mirrors

High-pressure gas scintillator
2 mCrab for 20 ks
5 106 for 28.8 ks
2 105 for 10.8 ks
108 at 30 keV
30 at 69 keV
42 at 30 keV
80 for 2050 keV

20 at 70 keV
12.6
1.5
2069
0.7 keV at 14 keV
1.0 keV at 60 keV
9.6
2.2
2040
2.3 keV at 22 keV
4.8 keV at 60 keV
9 at 40 keV, 5 at 60 keV
0.25 (i.e., 15)
2070
5% for 2070 keV
Detector technology
Performance to date:
Continuum sensitivity
(ph/s/cm2/keV)
Collecting area (cm2)
Field of view
Angular resolution
Energy range (keV)
Energy resolution
(FWHM)
364
Table15.3
Performance Targets of HEFT
Performance
Property
HEFT (Target)
INTEGRAL/IBIS
Angular resolution
1 HPD
12 FWHM
Field of view
Energy resolution
Energy range
10
1 keV at 68 keV
20100 keV
8
8 keV at 100 keV
20 keV10 MeV
Continuum sensitivity
(ph/s/cm2/keV, 3)
Time resolution
Pointing stability
2 107 (14 modules,

40 keV)
1 s
20
2.3 106 (100 keV)

61 s
Science Motivation
Extended objects
(Supernova remnants,
clusters of galaxies),
resolving the XRB
Extended objects
CRSFs, 44Ti lines in SNRs
30-keV peak of the XRB,
68- and 78-keV 44Ti lines,
the Crab Nebula
Resolving the XRB
X-ray pulsars
15.2The High-Energy Focusing Telescope

15.2.1Performance Objectives
To achieve its science objectives, HEFT requires having a field of view at least 10
across and 1 or better angular resolution in half-power diameter (HPD), which translates to focal plane extent and pixel pitch, respectively. In addition, the focal plane
detectors must have 1.5-keV or finer energy resolution in the hard X-ray band (XRB) of
1069 keV and their quantum efficiency close to 100%. Table15.3 lists these requirements alongside the corresponding performance of the coded aperture imager IBIS on
INTEGRAL, the latest space-based instrument for pointed observations in the same
energy band.
15.2.2The Payload
Figure15.1 is an annotated photograph of the final HEFT instrument taken during its
first launch on Wednesday, May 18, 2005. It was hanging from a crane on the launch
vehicle on the right-hand side of the picture and connected to the inflating balloon
in the background on the left, through a parachute. The X-ray imaging part of the
HEFT instrument is a set of three coaligned telescopesmirrors and focal plane
detectors. These telescopes are placed on a roughly cylindrical telescope truss, with
the mirrors at one end and the focal plane detectors at the opposite end, separated
by a focal length of 6.0 m. The truss is connected to a gondola platform through an
azimuth-elevation mount that enables the telescopes to be pointed in any direction
in the sky above 20 elevation. Motors and flywheels control the motion of the truss
365
GPS antennae
Yaw ywheel
Pitch motor
and encoder
Oaxis star tracker

camera and bae
X-ray focal plane (within

insulated pressure vessel)
Onaxis star tracker

camera and bae
Magnetometer (1 of 2)
Parachute
X-ray mirrors
Zeropressure
balloon
Flight control computer,

batteries, power supplies
Yawroll gyroscope
Crush pads
Ballast
Figure15.1 An annotated photograph of the HEFT instrument.
with respect to the gondola; sensors detect the direction and motion of the telescope
truss as well as the location and altitude of the payload. A flight-control computer
(FCC) onboard provides attitude control by processing the sensor data and signaling the motors and flywheels in a negative-feedback loop at a frequency of 10 Hz.
The FCC also records and relays science data from the focal plane and attitude data
from the motion sensors down to the ground station in real time by radio transmission. Electrical power for the payload comes from lithium battery cells onboard the
payload, which sustains the entire instrument for at least 50 h. Details of the HEFT
mirrors can be found in Koglin et al.3 and the architecture of the pointing platform
is described in Gunderson et al.6
15.2.3The X-Ray Focal Plane

Hidden from view in Figure15.1 within the spherical, temperature-insulated pressure vessel is the X-ray focal plane flange. It hosts three X-ray detectors, each
housed within its own background-shield module, and positioned exactly behind the
three X-ray mirrors at the opposite end of the telescope truss. Figure15.2 shows a
cross-sectional view on the left of a detector module within its background shield
module. The focal plane flange also contains a microprocessor (the central MISC,
MISC is a minimal instruction set computer) that serves as a branching node between
the FCC and individual controllers in each detector module for instruction upload and
data download. In addition, the central MISC monitors and regulates the focal plane
366
PMT
Detector hybrid
Cu
Sn
Pb
Plastic scintillator
Figure15.2 Left: Cross-section of the background shield module and the focal plane module within. PMT, photomultiplier tube. Right: A HEFT focal-plane module without its shield.
environment by reading sensors of pressure, voltage, and temperature across the

flange and by operating two fans circulating air within the 1-atm pressure vessel.
15.2.4The Focal-Plane Modules

Tugged within each shield module is the focal plane module, consisting of the hard
X-ray detector, a MISC microprocessor that controls the detector, and support electronics. Figure15.2 shows a photograph on the right of a focal plane module without
its background shield; Figure15.3 is a schematic diagram of the same system.
15.2.5The X-Ray Detectors

At the heart of the focal plane module are two X-ray detectors placed side by side.
Each detector consists of a semiconductor detector made up of a 2-mm thick cadmium
zinc telluride (CdZnTe) detector crystal bonded to a readout circuit implemented as
an ASIC. The two components together form a hybrida semiconductor device
CdZnTe
CdZnTe
Bump bonding
ASIC
ASIC
12bit ADC
Wire bonding
ADC
MISC
Clk
Level shifter
3 128 kB SRAM
EIA422 serial line
Clk
MISC
RAM
Level shifter
Figure 15.3 Diagram and flow chart of the detector system. (Reproduced from Chen,
C. M. H. et al., Proceedings of SPIE, 5198, 2004. With permission.)7
367
made up of two different materials (CdZnTe and silicon in this case). Each hybrid
has an active area 12.9 by 23.6 mm on the focal plane, where we implement a raster
array of 24 44 pixels of pitch 0.498 mm. With a 6-m focal length, each pixel subtends 17.12 from the X-ray mirrors and oversamples the PSF of the mirrors roughly
by a factor of five [HPD/(pixel size)]. Although the size of the CdZnTe-ASIC hybrid
is smaller than the focal plane area that needs to be covered, the size of the CdZnTe
crystal was the maximum available from the vendor, given the specification of single-crystal, highly uniform material at the time of our first design. In addition, the
size of the ASIC is also the maximum permitted by the VLSI fabrication process we
use. We thus tile each focal plane with two hybrids. This results in dead area where
the two sensors butt against each other, which we seek to minimize by additional
measures.
The HEFT detector hybrid is sensitive to photons within the energy range
8200 keV (1,60040,000 electrons). The spectral resolution is 0.75 keV FWHM at
14 keV and 1 keV FWHM at 60 keV. The maximum event rate before saturation is
about 50 counts/s/module.
15.2.6Digital Circuitry
While the CdZnTe-ASIC detector hybrid detects incident X-rays and measures their
position and energy, it requires external digital logic for operation to handle tasks
such as clocking, state transitions, and signal readout. The MISC microprocessor in
each focal plane module performs these operations. It is a P24 microprocessora
24-bit MISC designed to interpret instructions written in the FORTH programming
language. Like the MISC on the focal plane flange, we implement the module MISC
on an Actel A54SX72A field-programmable gate array (FPGA). The MISC runs on
a 7.3728-MHz clock cycle, driven by a 14.7456-MHz oscillator chip. The number of
clock periods in each sampling period is programmable, and we set it to eight. This
setting gives the HEFT detectors a 1 s timing resolution, fine enough to sample
timing-sensitive celestial targets, including the many millisecond period accreting
X-ray pulsars. Three 128-kB SRAMs provide the MISC with 128 kB of 24-bit memory, where we store the kernel, the FORTH instructions, as well as events readout
from the ASIC. The output of the ASIC readout line is digitized by an 80 mW, 12-bit
ADC12062 analog-to-digital converter (ADC). At the end of every 1-s frame period,
the module MISC pipes the digitized data out to the central MISC on the flange on
an EIA-422 serial line through a level shifter at a maximum rate of 240 kBps. The
entire focal plane module consumes about 700 mW of power.
15.2.7Background Shield Assembly

On the balloon-borne platform in the upper atmosphere, much of the detector background originates locally from the interaction of cosmic rays with the atmosphere
and with passive materials on the payload itself. Because the upper atmosphere
glows in hard X-rays and gamma rays from these interactions, we place each focal
plane module inside a well-shaped background shield module to limit the exposure
of the detectors to the noisy ambient environment.
368
Each shield module is made up of an open-ended cylindrical well of passive

blocking materials surrounded by an active plastic scintillator, as shown in the crosssectional diagram in Figure15.2.
The passive shield reduces the intensity of background photons that can reach
the detector. It consists of three layers of materials with graded atomic numbers.
From the outside to the inside, they are a 0.040-inch layer of lead (Pb, Z = 82), a
1.274-inch layer of tin (Sn, Z = 50), and a 0.141-inch layer of copper (Cu, Z = 29).
When a background X-ray of energy below about 100 keV enters the passive shield,
it will be blocked by the lead layer. The lead layer releases the energy absorbed
in the form of characteristic lead X-rays, at 88 keV for K X-rays, 16 keV for L1
X-rays, and at lower energies. These characteristic lead X-rays are then stopped
by the tin layer, which in turn releases the absorbed energy via its characteristic
X-rays at 29 keV (K), 4.5 keV (L1), and lower energies. The same process repeats
within the copper layer, producing copper K X-rays at 9.0 keV. Because this energy is
below the threshold of the CdZnTe-ASIC detector hybrids, the copper K X-rays will
not affect the operation of the detector system and can thus be safely ignored. This
passive shield clearly reduces the background; however, the atmospheric spectrum
extends into the gamma-ray band, and Compton scatters in the shield and subsequently the detector are still the dominant source of background.
On the other hand, the outer scintillator detects and rejects charged particles traversing through the detector system. The passage of a charged particle through the
plastic produces scintillation light, which gets collected at the base of the shield
assembly, where a photomultiplier tube (PMT) amplifies the light and converts it into
an electrical signal. This process happens in just tens of nanoseconds, with most of
the delay occuring in the PMT and the associated electronics. The electrical signal
thus produced serves as a veto pulse, which is read by the MISC to filter out 1-s
clock periods coincident with the presence of charged particles through the shield.
The primary purpose of the plastic shield is to eliminate secondary X-rays and
gamma rays produced as a cosmic ray traverses the shield, which would otherwise
deposit energy in the detector in the HEFT band. Alkali halide scintillators, such as
NaI and CsI, provide more efficient background rejection, but for cost reasons we
adopted the combined active/passive shield approach.
15.3Technology Development
The HEFT project has undergone over a decade of technology development, during
which we faced many challenges building the low-noise detector-ASIC hybrids, and
we made various changes along the way. Here, we describe the major design trades
and alternatives we investigated, with references to point interested readers to further detail previously reported elsewhere.
15.3.1Contributions to the Spectral Resolution

To meet the spectral resolution required by the science objectives, we spent much of
our effort on finding a geometric configuration of the detector hybrid that yields the
lowest level of noise possible. In semiconductor detectors like the HEFT hybrid,
369
Table15.4
Properties of Various Common Semiconductor Detector Materials
Material
Atomic number(s) (Z)
Mass density ()
Bandgap (Egap)
Energy loss per e-h+ pair
Fano factor (F)
Si
Ge
CdTe
CdxZn1xTe
Hgl2
14
2.33
1.12
3.61
0.0840.16
(77 K)
32
5.32
0.74
2.98
0.0580.129
(77 K)
48, 52
6.06
1.47
4.43
0.11
48, 30, 52
5.86
1.572
4.64
0.14
(40 C)
53, 80
6.30
2.13
4.22
0.46
>1
1
>1
>1
3.3 103
2 104
<104
11.7
50
16
109
11
(35) 103
5 105
3 1010
10.9
104
4 105
1013
8.8
g/cm2
eV
eV
Mobility-lifetime products:
For electrons (e)
For holes (h)
Electric resistivity
Electric permittivity (e)
cm2/V
cm2/V
-cm
the major components of noise are thermal noise in the preamplifier, shot noise due
to leakage current, and incomplete charge collection.
Thermal noise is introduced in all stages of the amplifier and readout circuit and
amplified by the same amount as the signal. Thus, noise in early stages has the greatest
effect on the final waveform, and it is important to keep its contribution low at the input
of the preamplifier, which is the pixel anodes in the case of HEFT. Thermal noise at the
input acts like a random voltage source in series with the input signal. It is amplified,
independent of the signal strength, in proportion to the size of the input capacitance,
which is the total input-to-ground capacitance of all conductors seen by a pixel anode.
The reduction of thermal noise is thus achieved by reducing the input capacitance.
A continuous leakage current, due to thermal excitation of electrons from the valence
to the conduction band, flows between the grounded anodes and biased electrodes and
is strongly dependent on temperature. Leakage current contributes significantly to the
total electronic noise, especially for a detector operated at room temperature.
In compound semiconductors like CdZnTe, the mobility of holes is an order of
magnitude lower than that of electrons. Table15.4 lists the mobility-lifetime products of holes and electrons in common semiconductor detector materials. Within
the finite time of measurement, if the transport holes have traveled little from the site
of photon absorption, they appear as a deficit of induced charges at the collecting
electrodes and a surplus at the neighboring electrodes, distort the measured pulse
height value. In common detector geometries where collecting electrodes are positioned perpendicular to the optical axis, and thus the direction of photon incidence,
the main effect of hole trapping is a dependence of the measured pulse height on the
depth of photon interaction.
15.3.2The Small Pixel Effect

To reduce the effect of hole trapping, one can modify the pattern of charge induction
within a detector by proper design of the electrodes on the detector surfaces, thus
370
changing the boundary conditions that the electric field must satisfy. For single-sided
pixel detectors in particular, where an array of segmented pixel anodes covers one
surface of the semiconductor while a monolithic cathode covers the opposite surface,
the so-called small-pixel effect forces the bulk of charge induction to take place
when transport electrons drift near the anodes.8 This reduces the amplitude of hole
signals relative to electron signals, thus it also reduces the deficit of induced charges
due to hole trapping.
15.3.3Depth Sensing
Independent of the small-pixel effect, one can also read out and compare signals
induced at the collecting electrode as well as its neighbors. The relative amplitudes
of these signals give us information on the depth of photon interaction for each event.
By discarding events registering deep within the detector, one increases the SNR,
since true events tend to register near the surface where photons enter the detector,
while noise is produced across the entire volume. Comparison of signals induced at
multiple pixels also allows us to locate localized charge traps in the HEFT detectors,
since excess and abrupt charge loss at these traps shows a distinct signature in plots
of interaction depth as a function of event energy.9
15.3.4Hybridization
In a semiconductor detector system with the readout circuit implemented as an ASIC,
the largest component of input capacitance is the parasitic capacitance seen by long
leads and traces connecting the semiconductor crystal to the silicon ASIC. One can
significantly reduce the amplification of thermal noise at the electronics front end by
bonding the ASIC directly to the semiconductor, eliminating high-capacitance traces
and cables. In this configuration, called a hybrid, one positions the readout ASIC
directly underneath the semiconductor crystal, facing the pixel contacts (the anodes
in the case of HEFT), while X-rays enter the detector crystal through the opposite
face. Circuitry for each pixel is packed into an area that matches the corresponding
pixel on the semiconductor. The pixel electrode on the semiconductor is connected
to the input of the preamplifier, typically an opening on the ASIC tens of micrometers wide, using a malleable and conductive material through a general technique
called flip-chip bonding or hybridization. For a detector hybrid, the main parasitic
capacitance seen by each pixel electrode is that of the ASIC backplane underneath
and of the electrodes of opposite polarity on the opposite face of the semiconductor
volume. Typically, the ASIC backplane contributes much more, since its separation
from the collecting electrodes (i.e., the bond height) is on the order of micrometers,
hundreds to a thousand times smaller than the cathode-anode separation (i.e., the
thickness of the semiconductor), while the dielectric constants of typical semiconductor materials are only about 10 times that of air and vacuum, insufficient to compensate the difference. Thus, to further minimize the input capacitance of the HEFT
readout circuit, we develop techniques to maximize the bond height and minimize
the area of pixel electrodes.
371
15.3.5Anode Contact Size and Steering Electrodes

During the early development of HEFT, with the minimization of input capacitance
in mind, we designed and tested pixel detectors with anode contacts of various sizes,
ranging from 20% to 90% of the pixel pitch (500 m). Although detectors with small
anodes have low input capacitance, we found that when an X-ray event occurs at
a position below which there is no metal contact on the anode plane, the induced
charge is smaller than expected, indicating charge loss at the gaps between anodes
and incomplete charge collection.10 In a further attempt to keep the anode areas
small, we added steering electrodes to the anode plane. Figure15.4 shows a diagram of the anode plane with and without these steering electrodes. We placed these
steering electrodes in between all anode contacts on the previously bare CdZnTe surface. They form a rectangular grid delineating the boundaries of the pixels. We set
them at a slightly negative potential with respect to the anodes to direct the field lines
at the gaps toward the anodes. This measure successfully reduces charge loss at the
bare CdZnTe surfaces. However, practical limitations with the detector material pose
other problems for the steering electrode. The critical issue is excessive leakage current brought about by the potential difference between the steering electrode and the
anodes required to mitigate the charge loss at the bare CdZnTe gaps. Although the
surface resistivity of CdZnTe crystals we obtained in early years were low enough
for steering electrodes to be a practical idea, we found that material properties have
changed over the years, and that the crystals we obtained since about 2001 no longer
have a low-enough surface resistivity for us to operate the steering electrodes without
Gridded detectors:
1 mm guard ring on 3 sides
0.1 mm guard ring on mating edge
498 m pitch pixels
50 m gap from anode to grid
14 m wide steering-electrode grid
300 m pitch for pixels along mating edge
Final HEFT detectors (no grid):
1 mm guard ring on 3 sides
0.1 mm guard ring on mating edge
498 m pitch pixels
30 m gap from anode to anode
300 m pitch for pixels along mating edge
Figure15.4 Anode plane patterns. The diagram on the left is a mechanical drawing, to
scale, of the final anode plane, with a raster array of 44 24 pixels per detector. The top circle
shows an experimental pattern with steering electrodes (described in Section 15.3.5); the bottom circle shows the pattern on the final HEFT detectors. Note that the guard ring and the
first row of pixels from the mating edge in both designs are contracted for two detectors to be
placed side by side with minimal dead area in between. (Reproduced from Chen, C. M. H.
et al., Proceedings of SPIE, 5198, 2004. With permission.)7
372
saturating the readout electronics with excessive input current. To add to this practical barrier, the narrow width of our steering eletrodes (14 m wide) was causing
frequent breakage during the lithography that patterns our anode plane, significantly
reducing the yield and thus increasing both cost and lead time. As a result, we had no
choice but to abandon the steering electrode idea. While we did not adopt this design
in the final HEFT detectors, we have done a considerable amount of work on this
gridded design, including the characterization of their direct current (DC)voltage
relation11,12 and numerical modeling of charge sharing across pixel boundaries lined
with steering electrodes.13 These investigations produced results that significantly
influenced our final design decisions.
To minimize charge loss at the CdZnTe gaps without the steering electrodes, we
minimize the area of the gaps to 30 m, a more conservative feature size for the
lithography process. This means that we have to increase the size of the anode contacts, which has the adverse side effect of increasing the input capacitance from the
ASIC backplane, the primary component of the source of noise for the HEFT detectors. To counter this increase in capacitance, we look elsewhere for alternative ways
to alleviate the situation.
15.3.6Bonding Techniques
In the course of development of the HEFT detector hybrids, we experimented with
two bonding techniques: indium bump bonding and flip-chip bonding with conductive epoxy and gold studs.
Indium bump bonding was the first of the two methods we tried, entirely inhouse. In this process, we deposit indium bumps onto the two sides of the hybrid
the CdZnTe detector and the readout ASICat the intended positions of the physical
connections through masking and thermal evaporation. The indium bumps thus produced are 50 m in diameter and 1012 m in height. Figure 15.5 shows a photograph on the left of the indium bumps deposited on pixel contacts of a CdZnTe
detector. We then align the detector and the ASIC and press them together to form
the hybrid. The indium bumps on the two sides fuse together on pressure to form
conductive connections from the anode contacts to the input pads on the ASIC that
measure 810 m in height. Finally, we apply adhesive to the four corners for added
mechanical strength.
At the end of 3.5 years of bump-bonding experiments, mostly with 8-by-8 pixel
or smaller prototype hybrids, we produced three full-size hybrids via indium bump
bonding. However, one came apart due to overheating, while the other two had fewer
than 50% of their pixels electrically connected. Although indium bump bonding is a
proved practice that the consumer electronics industry employs routinely to produce,
for example, plasma television sets on a mass production scale, they do so with millions of dollars of investment into research and development. Our attempt to master
the same skill in-house to produce a handful of hybrids has proved to be impractical.
Therefore, we had to abandon our indium bump-bonding trials and look for alternative hybridization methods.
The hybridization approach eventually adopted for HEFT was derived from a
technique described in Takahashi et al.14 Instead of bump bonding the two sides of
373
Figure 15.5 Left: Indium bumps deposited on the pixel contacts of a CdZnTe detector
with the gridded anode geometry described in Section 15.3.5. The indium bumps are 50
m in diameter, while the pixel pitch is 498 m. Center: Stencil-printed conductive epoxy
bumps on the pixel contacts of a CdZnTe detector of the ungridded anode geometry prior to
flip-chip bonding. The bumps in both cases are intentionally offset from the center of the
square contacts to accommodate the row of narrow pixels adjacent to the mating edge (see
Figure15.4). The pixel pitches in the two photographs are identical. Right: Cross section of
a gold studconductive epoxy bond made by slicing a mechanical sample hybrid vertically
apart and imaging with a scanning electron microscope (SEM).
374
a hybrid directly, they first constructed gold studs measuring tens of micrometers
tall on the input pads on an ASIC. Then, they bump bonded the detector to the top of
these studs with indium, making an indium-stud connection between the two hybrid
components. The gold studs are essentially clipped ends of wire bonds; they are
sturdy enough to serve as pillars to uphold the detector, which sits on top of the ASIC
in the final configuration of the hybrid. Addition of the studs effectively increases
the separation between the anode plane and the ASIC backplane, thus reducing the
capacitance of the latter, as seen by each anode contact.
Our second hybridization method combines these gold studs with connections
made by conductive epoxy. By this time, about 2002, CdZnTe detector companies
had had experience using conductive epoxy to glue CdZnTe detectors directly onto
printed circuit boards. Working with industry partners, we developed a modified
hybridization process by which we first populated the ASIC side with gold studs,
just as in the process of Takahashi et al.,14 but instead of indium bump bonding, we
screened the detector side with viscous conductive epoxy, flipped the detector onto
the stud-populated ASIC, aligning the two components together so that the epoxy
made a physical connection at the tip of the studs.15 We again applied adhesive to the
four corners of the hybrid for additional mechanical strength.
This flip-chip bonding method with gold studs and conductive epoxy proved successful. For the first hybrid produced this way, 996 pixels of a total of 44 24 =
1,056 were connected. In addition, the difference in spectral performance between
detectors fabricated with the two types of hybrid connections proved to be negligible.7 We have since employed this method to hybridize all our detectors, including
all detectors fabricated for the first HEFT campaign. They numbered at 17 as of the
end of 2006.
Figure15.5 shows a photograph of the stencil-printed conductive epoxy bumps
on a CdZnTe detector prior to flip-chip bonding with a stud-populated ASIC with a
photograph of the old indium bumps for comparison. Also shown is an SEM photograph of the cross section of the resulting gold studconductive epoxy bond. From
this cross section, we measured the height of the overall connection to be 4550 m,
a fivefold improvement from indium bump bonding. In other words, the input capacitance contributed by the ASIC backplane is reduced by the same factor. When both
changes were in place, we increased the anode contact size by (468/384)2 = 149%,
and the anode-ASIC separation by a factor of five. The net change in input capacitance from the ASIC backplane was then a 70% decrease, still a substantial improvement over the previous prototype.
15.3.7Material Defects and Characterization

In addition to recombination between transport and residual charges, crystal defects
and tellurium inclusions produce localized charge traps as well as inhomogeneities
in the electric field that prevent normal charge transport locally. The main effect of
defects is a spatially inhomogeneous response of the detector across the focal plane.
To reduce material defects, one must prescreen detector materials for uniformity
and electrical properties prior to fabrication. After fabrication, extensive calibration
375
of the gain and efficiency of every pixel is still needed to compensate trapping by
defects not rejected during the initial material selection. Owing to time and cost limitations, we were unable to screen sensors prior to hybridization using X-ray scanning or infrared (IR) imaging, even though these are the most effective methods of
selecting uniform material. During the screening of CdZnTe detectors for HEFT, we
requested single-crystal material from the manufacturer, eV Products, and we also
placed specifications for resistivity and inclusion size at the wafer level. Because the
HEFT ASIC saturated at relatively low leakage current levels, we also screened the
detectors for overall leakage current and leakage current uniformity.
We screened every CdZnTe detector for leakage current, at the intended operating voltages but at room temperature, at an ensemble of pixels uniformly distributed
across the detector. This approach sufficed to reveal gross material defects and crystal boundaries but was insufficient in guaranteeing good spectroscopic performance
of the detectors, and it did not eliminate the need for extensive pixel-by-pixel calibration of the hybrids produced thereafter.16 We have also investigated properties of
the semiconductormetal interface at pixel contacts and their effects on spectral performance in detail at selected pixels,17,18 although it is not practical to perform such
detailed measurements routinely given the large number of pixels on each detector
requiring characterization.
15.3.8Processing Signals from the HEFT Detectors

The design of the custom ASIC circuitry, designed at Caltech for the HEFT program,
is the key to achieving low power, low noise, and good spectral resolution. To achieve
low power, we have chosen a design that is different from the conventional amplifier
chain. In our design, the signal-shaping and peak detection stages of the conventional chain are replaced by a bank of switch capacitors arranged to continuously
capture successive samples of the preamplifier output. The circuit was described in
detail in Cook et al.19
The implementation of the sample-and-store mechanism is illustrated in
Figure15.6. The preamplifier output is converted to a current and is integrated by
the switch capacitors, cyclically one by one, with a 1-s integration time. This process gives us a record of the current level during the previous 1516 s at any given
time. The preamplifier output is also fed into a simple shaping amplifier with 300-ns
shaping time and then to a simple peak detection circuit that generates event triggers.
When a trigger is detected, sampling of the preamplifier output continues for eight
more samples, after which the circuit freezes while the voltage levels at the switch
capacitors in the triggered pixel are read out of the ASIC chip. The off-chip ADC
then digitizes these 16 preamplifier output samples and sends them to the microprocessor. The pulse height of the event is recovered from these 16 samples and related
state information at a later time in software.
In addition to reading out the 16 samples from each triggered pixel, we also read out
samples from a collection of other pixels for additional information to help with pulse
height calculation, noise averaging, identification of charge sharing between adjacent
376

s(t)
Test pulse
s[t]
s(t)
t
Detector anode
Preamplifier
i(t)
Discriminator
Shaping amplifier
Subcircuit A
Clk
MUX
s[i]
16
Subcircuit B
Figure15.6 Schematic diagram of the HEFT ASIC readout chip. Subcircuit A has the
shaping and peak detection stages in a conventional amplifier chain; although this subcircuit exists in the HEFT ASIC, only a simple version is implemented for triggering. Instead,
accurate pulse height information is captured in the bank of 16 switch capacitors in subcircuit
B, from which the pulse height is determined. Each pixel in the ASIC contains a copy of
the circuit shown in this figure; all pixels share the same serial readout line for reading out
the bank of switch capacitors. (Reproduced from Chen, C. M. H. et al., Proceedings of SPIE,
5198, 2004. Wtih permission.)7
pixels, and background rejection by depth sensing. Under this scheme, each event
produces about 0.5-kB of information, and the read out process takes about 20 ms. As
implemented for HEFT, with one controller servicing both hybrids in the same module, each module can handle event rates of up to 50 counts/s before saturating.
The result of this sample-and-store mechanism is a large reduction in power dissipationfrom 250 to 50 W per pixelwhile allowing off-chip digital signal processing
to extract near-optimal energy resolution. For further details on the signal-processing
scheme, specifically how we transform the 16 waveform samples to a pulse height and
how we combine pulse heights at multiple pixels with housekeeping data to obtain the
event energy, refer to Chen,16 which also describes the spectral performance of the
detectors flown on HEFT, resulting from this data-processing scheme.
15.4Deployment
After over a decade of design, technology development, and fabrication, HEFT was
launched at 19:05 UTC on May 18, 2005, in Fort Sumner, New Mexico. Ascent
through the Pfotzer maximum of elevated secondary particle background at 20 km
and to the final altitude of 39.0 km took 3.5 h. To prevent a steady leakage current and frequent noise bursts from saturating the detectors at high temperatures,
we had to slowly ramp up the cathode bias over hours while the ambient temperature dropped with climbing altitude. After reaching flight altitude, we pointed the
telescopes at Cyg X-1, the brightest celestial point source of hard X-rays, to align
the X-ray telescopes and optical star trackers onboard. We also pointed at several
other celestial X-ray sources and detected the Crab Nebula securely with 15 min of
observation. HEFT stayed afloat for 24 h and 40 min, drifting slowly westward, and
landed in Holbrook, Arizona.
377
15.4.1General Characteristics of the Flight Data

Figure15.7 shows the time profiles of the payload altitude, pointing elevation, event
rate, and focal plane temperature in flight. The big drop in altitude at around 02:30
UTC was due to nightfall, and the big rise at about 12:00 UTC was due to daybreak.
The release of ballast at about 07:15 UTC also caused an abrupt rise in altitude. The
event rate profile includes only events with energy within 2070 keV, the bandpass
39
38
37
36
35
80
M 1 (Crab)
X Per
Cyg X1
3C 454.3
0.4
GRS 1915+105
0
0.6
Cyg X1
20
Her X1
M 1 (Crab)
40
Mkn 421
60
15:00
18:00
Fotzer
maximum
Event Rate (counts/sec)
Pointing Elevation (deg)
Payload Altitude (km)
40
0.2
Temperature (C)
0.0
30
20
10
0
10
20
21:00
00:00
20050519
03:00
06:00
09:00
12:00
Time (UTC)
Figure 15.7 Time profiles of the payload altitude, pointing elevation, event rate, and
temperature in flight. The observation times of the various targets and the time of passage
through the Pfotzer maximum are indicated.
378
bound by atmospheric absorption on the low-energy side, and by an abrupt drop in

reflectivity of the W/Si multilayer mirror coatings on the high-energy side. In addition, we have filtered out invalid events and most of the noise events due to bursts of
leakage current at individual pixels. From the event rate profile, we see clear elevation of counts during the time intervals when we observed Cyg X-1 and the Crab
Nebula. The relatively high event rate during the first 2 h of flight was due to the
payload crossing the Pfotzer maximum. From the temperature profile, we noted that
the mostly metal payload formed a large thermal mass, preventing the focal plane
from quickly cooling to the expected operating temperature range of [15, +5]C.
Once the temperature reached this range, it gradually fell at night and rose in the day
as a result of varying exposure to the sun.
Compared to data taken during ground operation, the flight data exhibited the
amplification of a number of systematics pertaining to the threshold of event trigger
at each pixel. In flight, we also found an increase in events triggering large numbers
of pixels. While this increase affected only a small percentage of all events, its presence was definite and consistent across all six flight detectors.
15.4.2In-Flight Background Measurement and Rejection

Figure15.8 shows our measured background spectrum at balloon altitude. This spectrum contains events recorded from most of the flight, excluding the initial ascent to
the final altitude and time periods when the telescopes were on a known target. The
CZT Background
Background (cts/sec/keV/cm2)
0.0015
Co57 122 keV

calibration
source
0.0010
0.0005
Observed HEFT balloon background
(plastic/graded=Z shield)
Simulated HEFT background
(plastic/graded=Z shield)
0.0000
50
100
keV
150
200
Figure 15.8 Background spectrum at balloon altitude as measured in flight (dark) and
simulated (light). The emission line at 122 keV is from a 57Co calibration source onboard
the payload. The emission features within 7090 keV are due to lead fluorescence from the
background shields.
379
measured background agreed well with our simulation of the focal plane and shield
modules.
The scintillator shield also functioned in flight per specification. Each shield
produced about 1,400 veto pulses/module/s during passage of the Pfotzer maximum and 600 veto pulses/module/s at flight altitude. The mean rates of coincidence
between (rejected) X-ray events and shield veto pulses were 5.98, 5.53, and 6.02
events/module/s in the three X-ray detector-and-shield modules, compared to typical source event rates of less than one event/module/s, showing the significance of
background rejection on the balloon platform. The lifetimes of the X-ray detectors
in all three modules averaged about 80% through the Pfotzer maximum and 85% at
flight altitude.
15.4.3Observation of Celestial Sources

Figure15.9 shows the sky images of Cyg X-1 and the Crab Nebula constructed from
all good events between 20 and 40 keV from two of the three X-ray focal plane modules. (Inferior quality of the X-ray detectors in the third module prevented us from
measuring photon energies as precisely as in the other two modules.)
15.4.3.1Cyg X-1
We observed Cyg X-1 for about 52 min with full aspect information. This observation exemplifies a scenario not uncommon to the balloon platform: During this
period, Cyg X-1 was at elevations above 60, where the carrying balloon blocked
the view of the on-axis optical star tracker but allowed hard X-rays to pass through.
Thus, attitude control relied on the off-axis star tracker tracking a single star in a
field about 30 offset in elevation relative to the X-ray target. In this single-star
mode, the star tracker was insensitive to rotations about the direction to the tracked
star, while this off-axis rotation manifested itself as translational movement in the
Crab pulsar by HEFT on
2005 519, 2040 keV, 15 min
35.35
35
22.15
35.3
30
22.1
35.25
25
35.2
20
15
35.15
10
35.1
35.05
299.7
299.6
RA (deg)
299.5
Dec (deg)
Dec (deg)
Cyg X1 by HEFT Module B on

2005 519, 2040 keV, 42 min
12
10
22.05
22
21.95
21.9
21.85
83.8
14
2
83.7
83.6
RA (deg)
83.5
Figure15.9 Left: Sky image of Cyg X-1 from a 42-min observation. Right: Sky image of
the Crab Nebula from a 15-min observation. Both images were constructed from 2040 keV
events registered on focal plane modules A and B during the deployment of HEFT on May 19,
2005.
380

50
Aspect selection < 1.5 arcmin

Aspect selection eciency: 71%
Fit energy: 19 keV < E < 71 keV
Fit spectral index: 2.37 + 0.18
Fit source events: 450
Background events: 29
Events
40
30
20
10
0
20
40
60
80
100
Energy (keV)
15
2/ODF = 1.1
(19 < E < 71)
Residuals
10
5
0
5
10
15
20
40
60
Energy (keV)
80
100
Figure15.10 Spectrum of Cyg X-1, measured by modules A and B of HEFT on May 19,
2005, for 52 min.
X-ray field of view. As a result, the X-ray pointing direction fluctuated about Cyg X-1
more than it would otherwise. Fortunately, other attitude sensors managed to keep
Cyg X-1 within the 10 field of view during most of the observation.
Figure15.10 shows our measured spectrum of Cyg X-1.
15.4.3.2The Crab Nebula
In contrast with Cyg X-1, we observed the Crab Nebula at elevations below 60,
which enabled the on-axis star tracker to track the same patch of sky as the X-ray
field of view. Pointing was thus more stable, allowing the X-ray detectors to image
the 10 wide vicinity about the Crab Nebula over a long exposure. This roughly constant X-ray field of view appears in Figure15.9 as the region of elevated background
about the Crab Nebula. Note that we intentionally offset the target slightly away from
the center to avoid the dead area between the two detector hybrids.
15.5Future Outlook
It has taken over a decade to develop and demonstrate the HEFT payload, from early
technology development to first launch. The performance of its focal plane detectors
to within specification in flight demonstrated the viability of semiconductor pixel
detectors as the focal plane instrument on a hard X-ray telescope. A future satellite mission called the Nuclear Spectroscopic Telescope Array (NuSTAR) is under
381
development as the successor to HEFT.1 Its hard X-ray telescopes and detectors build
on the heritage of HEFT technologies. NuSTAR is scheduled for launch in 2011 for a
mission lifetime of at least 2 years. It will open a new window for exciting discoveries in the hard X-ray sky.
References

1. F. A. Harrison, F. E. Christensen, W. Craig, C. Hailey, W. Baumgartner, C. M. H. Chen,

J. Chonko, W. R. Cook, J. Koglin, K.-K. Madsen, M. Pivavoroff, S. Boggs, and S. Smith,
Development of the HEFT and NuSTAR focusing telescopes, Exp. Astron., 20, 131137,
2005.
2. K. D. Joensen, Design, fabrication and characterization of multilayers for broadband, hard X-ray astrophysics instrumentation, PhD thesis, H. C. rsted Instituttet,
Copenhagen, Denmark, 1995.
3. J. E. Koglin, C. M. H. Chen, F. E. Christensen, W. W. Craig, T. R. Decker, K. S. Gunderson,
C. J. Hailey, F. A. Harrison, C. P. Jensen, K. Kruse-Madsen, M. Stern, D. L. Windt, H.
Yu, and E. Ziegler, Production and calibration of the first HEFT hard X-ray optics module, In O. Citterio and S. L. ODell, editors, Society of Photo-Optical Instrumentation
Engineers (SPIE) Conference Series, volume 5168, pages 100111, 2004.
4. W. H. Baumgartner, J. Tueller, H. Krimm, S. D. Barthelmy, F. Berendse, L. Ryan,
F. B. Birsa, T. Okajima, H. Kunieda, Y. Ogasaka, Y. Tawara, and K. Tamura, InFOCS
hard X-ray telescope: Pixellated CZT detector/shield performance and flight results.
In Joachim E. Truemper and Harvey D. Tananbaum, editors, X-Ray and Gamma-Ray
Telescopes and Instruments for Astronomy, Proc. SPIE, 4851:945956, March 2003.
5. J. A. Gaskin, G. A. Richardson, S. Mitchell, D. P. Sharma, B. D. Ramsey, and P. Seller.
Assessment of cadmium-zinc-telluride detectors for hard-X-ray astronomy. In 2003
IEEE Nucl. Sci. Sump. Conf. Rec., 5:34043407, October 2003.
6. K. S. Gunderson, C. M. H. Chen, F. E. Christensen, W. W. Craig, T. A. Decker, C. J.
Hailey, F. A. Harrison, R. McLean, R. E. Wurtz, and K. Ziock, Ground performance of
the High-Energy Focusing Telescope (HEFT) attitude control system. In K. A. Flanagan
and O. H. W. Siegmund, editors, Society of Photo-Optical Instrumentation Engineers
(SPIE) Conference Series, SPIE Conf. Series, 5165:158168, 2004.
7. C. M. H. Chen, W. R. Cook, F. A. Harrison, J. Y. Y. Lin, P. H. Mao, and S. M. Schindler,
Characterization of the HEFT CdZnTe pixel detectors. In L. A. Franks, A. Burger, R. B.
James, and P. L. Hink, editors, Hard X-Ray and Gamma-Ray Detector Physics V, Proc.
SPIE, 5198:918, 2004.
8. G. F. Knoll, Radiation Detection and Measurement. 3rd ed. Wiley, New York, 2000.
9. W. Baumgartner, F. Harrison, H. Chen, W. Cook, B. Craig, K. Kruse-Madsen, S. Boggs,
C. Wunderer, and A. Zoglauer, Flight performance and laboratory tests of CZT detectors
for hard X-ray focal planes, Bull. Am. Astron., Soc. 38:385, 2006.
10. A. E. Bolotnikov, W. R. Cook, F. A. Harrison, A.-S. Wong, S. M. Schindler, and A. C.
Eichelberger, Charge loss between contacts of CdZnTe pixel detectors, Nucl. Instr.
Methods A, 432(23):326331, 1999.
11. A. E. Bolotnikov, S. E. Boggs, C. M. H. Chen, W. R. Cook, F. A. Harrison, and S. M.
Schindler, Optimal contact geometry for CdZnTe pixel detectors. In R. B. James and
R. C. Schirato, editors, Hard X-Ray, Gamma-Ray, and Neutron Detector Physics II,
Proc. SPIE, 4141:243252, 2000.
12. A. E. Bolotnikov, C. M. H. Chen, W. R. Cook, F. A. Harrison, I. Kuvvetli, and S. M.
Schindler, Effects of bulk and surface conductivity on the performance of CdZnTe pixel
detectors, IEEE Trans. Nucl. Sci., 49(4):19411949, 2002.
382
13. C. M. H. Chen, S. E. Boggs, A. E. Bolotnikov, W. R. Cook, F. A. Harrison, and S. M.

Schindler, Numerical modeling of charge sharing in CdZnTe pixel detectors, IEEE Trans.
Nucl. Sci., 49(1):270276, 2002.
14. T. Takahashi, S. Watanabe, M. Kouda, G. Sato, Y. Okada, S. Kubo, Y. Kuroda, M. Onishi,
and R. Ohno, High-resolution CdTe detector and applications to imaging devices, IEEE
Trans. Nucl. Sci., 48(3):287291, 2001.
15. J. E. Clayton, C. M. H. Chen, W. R. Cook, and F. A. Harrison, Assembly technique
for a fine-pitch, low-noise interface: joining a CdZnTe pixel-array detector and custom
VLSI chip with Au stud bumps and conductive epoxy, 2003 IEEE Nucl. Sci. Symp. Med.
Imaging Conf. Rec., 5:35133517, 2003.
16. C. M. H. Chen, Development of hard X-ray imaging detectors for the High Energy
Focusing Telescope, PhD thesis, California Institute of Technology, Pasadena, CA,
2008.
17. A. E. Bolotnikov, S. E. Boggs, C. M. H. Chen, W. R. Cook, F. A. Harrison, and S. M.
Schindler, Properties of Pt Schottky type contacts on high-resistivity CdZnTe detectors,
Nucl. Instr. Methods, 482(12):395407, 2002.
18. A. E. Bolotnikov, C. M. H. Chen, W. R. Cook, F. A. Harrison, I. Kuvvetli, S. M. Schindler,
C. M. Stahle, and B. H. Parker, The effect of cathode bias (field effect) on the surface
leakage current of CdZnTe detectors, Nucl. Instr. Methods, 510(3):300308, 2003.
19. W. R. Cook, S. E. Boggs, A. E. Bolotnikov, J. A. Burnham, M. J. Fitzsimmons, F. A.
Harrison, B. Kecman, B. Matthews, and S. M. Schindler, High resolution CdZnTe pixel
detectors with VLSI readout, IEEE Trans. Nucl. Sci., 47:14541457, 2000.

Semiconductor

Uploaded by

Copyright:

Available Formats

Semiconductor

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Semiconductor

Uploaded by

Copyright:

Available Formats

Semiconductor Radiation

2010 by Taylor and Francis Group, LLC

Devices, Circuits, and Systems

Internet Networks: Wired, Wireless, and Optical Technologies

2010 by Taylor and Francis Group, LLC

Edited by Krzysztof Iniewski

Boca Raton London New York

CRC Press is an imprint of the

2010 by Taylor and Francis Group, LLC

2010 by Taylor and Francis Group, LLC

2010 by Taylor and Francis Group, LLC

Chapter 9. Caliste: Microcamera for Hard X-Ray Astronomy........................... 193

2010 by Taylor and Francis Group, LLC

2010 by Taylor and Francis Group, LLC

About the Editor

2010 by Taylor and Francis Group, LLC

Gian-Franco Dalla Betta

2010 by Taylor and Francis Group, LLC

Kristin Kruse Madsen

2010 by Taylor and Francis Group, LLC

2010 by Taylor and Francis Group, LLC

Semiconductor Radiation Detection Systems

Single photon emission

2010 by Taylor and Francis Group, LLC

Spatial and Spectral Resolution of Semiconductor Detectors in Medical Imaging 3

Detected output energy E

2010 by Taylor and Francis Group, LLC

Semiconductor Radiation Detection Systems

2010 by Taylor and Francis Group, LLC

Spatial and Spectral Resolution of Semiconductor Detectors in Medical Imaging 5

2010 by Taylor and Francis Group, LLC

Semiconductor Radiation Detection Systems

1.2.2Signal Transport Processes

1. Primary energy deposition. The primary energy deposition is modeled by

2010 by Taylor and Francis Group, LLC

Spatial and Spectral Resolution of Semiconductor Detectors in Medical Imaging 7

2010 by Taylor and Francis Group, LLC

Semiconductor Radiation Detection Systems

1.3.1Definition of the Modulation Transfer Function

2010 by Taylor and Francis Group, LLC

Spatial and Spectral Resolution of Semiconductor Detectors in Medical Imaging 9

Simulated Pixel Signal

1.3.2Simulation and Measurement of the MTF

1.3.3Properties of the MTF

2010 by Taylor and Francis Group, LLC

Semiconductor Radiation Detection Systems

MTF indirect converter

Spatial frequency in 1/mm

1.4.1Definition of the Detector Response Function (DRF)

2010 by Taylor and Francis Group, LLC

Spatial and Spectral Resolution of Semiconductor Detectors in Medical Imaging 11

2010 by Taylor and Francis Group, LLC

Semiconductor Radiation Detection Systems

1.4.2Comparisons of Detector Response Functions

2010 by Taylor and Francis Group, LLC

Spatial and Spectral Resolution of Semiconductor Detectors in Medical Imaging 13

1.4.3Integrating Indirect Conversion Detectors

2010 by Taylor and Francis Group, LLC

Semiconductor Radiation Detection Systems

1.2.2Signal Transport Processes

1.3.1Definition of the Modulation Transfer Function

1.3.2Simulation and Measurement of the MTF

1.3.3Properties of the MTF

1.4.1Definition of the Detector Response Function (DRF)

1.4.2Comparisons of Detector Response Functions

1.4.3Integrating Indirect Conversion Detectors

1.4.4Counting Direct Conversion Detectors

2.2.2Technology Computer-Aided Design Simulations

2.5.2Results for 3D Detectors from Other Manufacturers

2.63D Detector Developments in Trento