Op To Electronic Sensors

This page intentionally left blank
Optoelectronic Sensors
Optoelectronic
Sensors
Edited by
Didier Decoster
Joseph Harari
First published in France in 2002 by Hermes Science/Lavoisier entitled: Dtecteurs optolectroniques
LAVOISIER, 2002
First published in Great Britain and the United States in 2009 by ISTE Ltd and John Wiley & Sons, Inc.
Apart from any fair dealing for the purposes of research or private study, or criticism or review, as
permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced,
stored or transmitted, in any form or by any means, with the prior permission in writing of the publishers,
or in the case of reprographic reproduction in accordance with the terms and licenses issued by the CLA.
Enquiries concerning reproduction outside these terms should be sent to the publishers at the
undermentioned address:
ISTE Ltd John Wiley & Sons, Inc.
27-37 St Georges Road 111 River Street
London SW19 4EU Hoboken, NJ 07030
UK USA
www.iste.co.uk www.wiley.com
ISTE Ltd, 2009
The rights of Didier Decoster and Joseph Harari to be identified as the authors of this work have been
asserted by them in accordance with the Copyright, Designs and Patents Act 1988.
Library of Congress Cataloging-in-Publication Data
Dtecteurs optolectroniques. English
Optoelectronic sensors / edited by Didier Decoster, Joseph Harari.
p. cm.
Includes bibliographical references and index.
ISBN 978-1-84821-078-3
1. Optical detectors. 2. Image sensors. I. Decoster, Didier, 1948- II. Harari, Joseph, 1961- III. Title.
TK8360.O67D4813 2009
681'.25--dc22
2009011542
British Library Cataloguing-in-Publication Data
A CIP record for this book is available from the British Library
ISBN: 978-1-84821-078-3
Printed and bound in Great Britain by CPI Antony Rowe, Chippenham and Eastbourne.
Table of Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Chapter 1. Introduction to Semiconductor Photodetectors. . . . . . 1
Franck OMNES
1.1. Brief overview of semiconductor materials . . . . . . . . . . . . . 1
1.2. Photodetection with semiconductors: basic phenomena . . . . . 3
1.3. Semiconductor devices. . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4. p-n junctions and p-i-n structures . . . . . . . . . . . . . . . . . . . 5
1.5. Avalanche effect in p-i-n structures . . . . . . . . . . . . . . . . . . 7
1.6. Schottky junction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.7. Metal-semiconductor-metal (MSM) structures . . . . . . . . . . . 10
1.8. Operational parameters of photodetectors . . . . . . . . . . . . . . 11
1.8.1. Response coefficient, gain and quantum efficiency . . . . . 11
1.8.2. Temporal response and bandwidth . . . . . . . . . . . . . . . . 12
1.8.3. Noise equivalent power . . . . . . . . . . . . . . . . . . . . . . . 13
1.8.4. Detectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Chapter 2. PIN Photodiodes for the Visible and Near-Infrared. . . 15
Baudoin DE CREMOUX
2.1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2. Physical processes occurring in photodiodes . . . . . . . . . . . . 17
2.2.1. Electrostatics in PIN diodes: depleted region . . . . . . . . . 17
2.2.2. Mechanisms of electron-hole pair generation . . . . . . . . . 19
2.2.3. Transport mechanisms. . . . . . . . . . . . . . . . . . . . . . . . 23
2.3. Static characteristics of PIN photodiodes . . . . . . . . . . . . . . 25
2.3.1. I/V characteristics and definition of static parameters . . . . 25
2.3.2. External quantum efficiency . . . . . . . . . . . . . . . . . . . . 27
vi Optoelectronic Sensors
2.3.3. Dark current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.4. Breakdown voltage. . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3.5. Saturation current. . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.4. Dynamic characteristics of PIN photodiodes . . . . . . . . . . . . 34
2.4.1. Intrinsic limitations to the speed of response. . . . . . . . . . 34
2.4.2. Limitations due to the circuit . . . . . . . . . . . . . . . . . . . 37
2.4.3. Power-frequency compromise, Pf
2
law . . . . . . . . . . . . 41
2.5. Semiconductor materials used in PIN photodiodes
for the visible and near-infrared . . . . . . . . . . . . . . . . . . . . . . . 42
2.5.1. Absorption of semiconductors in the range 400-1,800 nm . 42
2.5.2. From 400 to 900 nm: silicon and the GaAlAs/GaAs family 43
2.5.3. From 900 to 1,800 nm: germanium, GaInAsP/InP . . . . 46
2.6. New photodiode structures . . . . . . . . . . . . . . . . . . . . . . . 49
2.6.1. Beyond the limits of conventional PIN . . . . . . . . . . . . . 49
2.6.2. Photodiodes with collinear geometry . . . . . . . . . . . . . . 50
2.6.3. Waveguide photodiodes. . . . . . . . . . . . . . . . . . . . . . . 52
2.6.4. Traveling-wave photodiodes. . . . . . . . . . . . . . . . . . . . 53
2.6.5. Beyond PIN structures . . . . . . . . . . . . . . . . . . . . . . . 54
2.7. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Chapter 3. Avalanche Photodiodes . . . . . . . . . . . . . . . . . . . . . . 57
Grard RIPOCHE and Joseph HARARI
3.1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.2. History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3. The avalanche effect . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.3.1. Ionization coefficients . . . . . . . . . . . . . . . . . . . . . . . . 61
3.3.2. Multiplication factors . . . . . . . . . . . . . . . . . . . . . . . . 62
3.3.3. Breakdown voltage. . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.4. Properties of avalanche photodiodes . . . . . . . . . . . . . . . . . 66
3.4.1. Current-voltage characteristics and photomultiplication . . 66
3.4.2. Noise in avalanche photodiodes. . . . . . . . . . . . . . . . . . 68
3.4.3. Signal-to-noise ratio in avalanche photodiodes . . . . . . . . 71
3.4.4. Speed, response time and frequency response
of avalanche photodiodes . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.5. Technological considerations. . . . . . . . . . . . . . . . . . . . . . 76
3.5.1. Guard ring junctions . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.5.2. Mesa structures. . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.5.3. Crystal defects and microplasmas . . . . . . . . . . . . . . . . 79
3.6. Silicon avalanche photodiodes . . . . . . . . . . . . . . . . . . . . . 80
3.6.1. Si N
+
P APDs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.6.2. Si N
+
PP
+
APDs . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.6.3. Si N
+
PP
+
APDs . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Table of Contents vii
3.6.4. SiPt-Si N Schottky APDs. . . . . . . . . . . . . . . . . . . . . . 87
3.7. Avalanche photodiodes based on gallium arsenide . . . . . . . . 88
3.8. Germanium avalanche photodiodes . . . . . . . . . . . . . . . . . . 90
3.8.1. Ge APDs with N
+
P, N
+
NP and P
+
N structures for
1.3 m communication . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.8.2. Ge APDs with P
+
NN
-
structures for 1.55 m
communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.9. Avalanche photodiodes based on indium phosphate (InP). . . . 95
3.9.1. InGaAs/InP APDs for optical communications
at 2.5 Gbit/s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.9.2. Fast InGaAs/InP APDs . . . . . . . . . . . . . . . . . . . . . . . 99
3.10. III-V low-noise avalanche photodiodes. . . . . . . . . . . . . . . 100
3.10.1. III-V super-lattice or MQW APDs . . . . . . . . . . . . . . . 101
3.10.2. Spin-orbit resonance APDs . . . . . . . . . . . . . . . . . . . . 102
3.11. Prospects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.11.1. Si/InGaAs APDs . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.11.2. Waveguide MQW APDs . . . . . . . . . . . . . . . . . . . . 104
3.11.3. Low-noise APDs with a very thin multiplication region. . 105
3.12. Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.13. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Chapter 4. Phototransistors . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Carmen GONZALEZ and Antoine MARTY
4.1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.2. Phototransistors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.2.1. Phototransistors according to their fabrication materials . . 112
4.2.2 Phototransistors classified by structure. . . . . . . . . . . . . . 114
4.3. The bipolar phototransistor: description and principles
of operation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.3.1. The phototransistor effect. . . . . . . . . . . . . . . . . . . . . . 119
4.3.2. The response coefficient of a phototransistor . . . . . . . . . 124
4.3.3. Static electrical and optical gains of the phototransistor. . . 125
4.3.4. Dynamic characteristics of phototransistors . . . . . . . . . . 126
4.3.5. Noise in phototransistors . . . . . . . . . . . . . . . . . . . . . . 138
4.4. Photodetector circuits based on phototransistors. . . . . . . . . . 140
4.4.1. Amplification circuits . . . . . . . . . . . . . . . . . . . . . . . . 140
4.4.2. Nonlinear circuits. . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.5. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.5.1. Galvanic isolation . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.5.2. Phototransistors for optical telecommunications . . . . . . . 145
4.6. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.7. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
viii Optoelectronic Sensors
Chapter 5. Metal-Semiconductor-Metal Photodiodes . . . . . . . . . 155
Joseph HARARI and Vincent MAGNIN
5.1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.2. Operation and structure . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.2.1. Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.2.2. Materials used . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
5.3. Static and dynamic characteristics. . . . . . . . . . . . . . . . . . . 165
5.3.1. Response coefficient . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.3.2. Dynamic behavior . . . . . . . . . . . . . . . . . . . . . . . . . . 172
5.3.3. Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
5.4. Integration possibilities and conclusion . . . . . . . . . . . . . . . 177
5.5. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
Chapter 6. Ultraviolet Photodetectors. . . . . . . . . . . . . . . . . . . . 181
Franck OMNES and Eva MONROY
6.1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
6.2. The UV-visible contrast . . . . . . . . . . . . . . . . . . . . . . . . . 189
6.3. Si and SiC photodetectors for UV photodetection . . . . . . . . . 190
6.3.1. UV photodiodes based on silicon. . . . . . . . . . . . . . . . . 191
6.3.2. SiC-based UV photodetectors . . . . . . . . . . . . . . . . . . . 194
6.4. UV detectors based on III-V nitrides . . . . . . . . . . . . . . . . . 195
6.4.1. Photoconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
6.4.2. Schottky barrier photodiodes based on AlGaN . . . . . . . . 202
6.4.3. MSM photodiodes . . . . . . . . . . . . . . . . . . . . . . . . . . 209
6.4.4. p-n and p-i-n photodiodes . . . . . . . . . . . . . . . . . . . . . 210
6.4.5. Phototransistors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
6.5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
6.6. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
Chapter 7. Noise in Photodiodes and Photoreceiver Systems . . . . 223
Robert ALABEDRA and Dominique RIGAUD
7.1. Mathematical tools for noise . . . . . . . . . . . . . . . . . . . . . . 224
7.1.1. Known signals with finite energy or power . . . . . . . . . . 224
7.1.2. Random signals and background noise . . . . . . . . . . . . . 226
7.2. Fundamental noise sources . . . . . . . . . . . . . . . . . . . . . . . 227
7.2.1. Thermal noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
7.2.2. Shot noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
7.2.3. Multiplication noise . . . . . . . . . . . . . . . . . . . . . . . . . 229
7.3. Excess noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
7.3.1. Generation-recombination noise . . . . . . . . . . . . . . . . . 232
7.3.2. 1/f noise. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Table of Contents ix
7.4. Analysis of noise electrical circuits . . . . . . . . . . . . . . . . . . 235
7.4.1. Representation of noise in bipoles . . . . . . . . . . . . . . . . 235
7.4.2. Representation of noise in quadripoles . . . . . . . . . . . . . 237
7.5. Noise in photodetectors . . . . . . . . . . . . . . . . . . . . . . . . . 239
7.5.1. Characteristic parameters . . . . . . . . . . . . . . . . . . . . . . 240
7.5.2. PIN photodiodes . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
7.5.3. Avalanche photodiodes . . . . . . . . . . . . . . . . . . . . . . . 244
7.6. Noise optimization of photodetectors. . . . . . . . . . . . . . . . . 245
7.6.1. Formulation of the problem . . . . . . . . . . . . . . . . . . . . 246
7.6.2. Concepts for photodetector-transistor matching. . . . . . . . 251
7.7. Calculation of the noise of a photoreceiver . . . . . . . . . . . . . 253
7.7.1. Basic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
7.7.2. Models of transistor noise . . . . . . . . . . . . . . . . . . . . . 255
7.7.3. Example calculation: a PIN-FET photoreceiver. . . . . . . . 259
7.8. Comments and conclusions . . . . . . . . . . . . . . . . . . . . . . . 266
7.9. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
List of Authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
Preface
Photodetection is found in a large number of professional and
mass-market systems. There are numerous applications including:
fiber-based and free-space optical telecommunications, galvanic
isolation, solar cells, proximity detectors, etc. All of these applications
are based on the same process: the transformation of optical power
into an electrical signal, with this signal needing to be as large as
possible for an optical flux as weak as possible. When information
needs to be transmitted quickly, potentially at very high speeds, the
photodetector must react very fast. These basic considerations imply
that a certain number of performance requirements should be met in
order to satisfy the demands of the intended application. It is in this
context that various photodetecting structures have been conceived:
photoconductors, p-n and p-i-n photodiodes, avalanche photodiodes,
phototransistors, Schottky photodiodes, MSM (metal semiconductor-
metal) photodetectors. The range of wavelengths relevant to the
application also plays an important role. It is often this, through the
intermediary of the semiconductor bandgap, which dictates the nature
of the material(s) used.
This book gathers together the most detailed and significant
contemporary thinking on photodetection for wavelengths from the
near-infrared to the ultraviolet
1
. Its content not only gives the reader
1. The very particular nature of photodetection in the mid- and far-infrared requires
specific treatment and thus is not covered here.
xii Optoelectronic Sensors
the grounding to design simple photodetectors with specified
performance characteristics, but also discusses the state of the art in
photodetection. Chapter 1 begins with an introduction to
photodetection and the most well-known photodetector structures.
Among these, some are more commonly used than others, or offer
more exciting possibilities.
Because of this, each of Chapters 2 to 5 concentrates in great depth
on a specific photodetector type. In sequence, the reader will be able
to learn about:
the PIN photodiode which is the photodetector on which the
majority of systems are based (Chapter 2);
the avalanche photodiode, which is a refinement of the PIN
photodiode and which, under a sufficiently high bias voltage, allows
us to achieve a gain (Chapter 3);
the phototransistor, which is another means of obtaining gain in
photodetection (Chapter 4);
the MSM photodiode which is well suited to monolithic
integrated circuits such as MMICs (Microwave Monolithic Integrated
Circuits) (Chapter 5).
Chapter 6, however, is different. It is dedicated to ultraviolet
photodetectors. Given the significant role that this wavelength range is
likely to take in the near future, notably for environmental
applications, this subject deserves a separate chapter. In this case, all
types of photodetector are concerned, and it is the material which is
the source of improvement, notably with the appearance of AlGaN
materials with a very large bandgap.
Finally, Chapter 7 is entirely dedicated to noise, as this concept is
absolutely fundamental for the photodetection of low-intensity signals.
Preface xiii
The chapters of this book have all been written by specialists and
we take this opportunity to thank them sincerely and warmly for their
contributions.
Didier DECOSTER and Joseph HARARI
Chapter 1
Introduction to
Semiconductor Photodetectors
1.1. Brief overview of semiconductor materials
A semiconductor material is a continuous crystalline medium
characterized by an energy band structure corresponding, in the case
of an infinite crystal, to a continuum of states (which, in practice,
means that the characteristic dimensions of the crystal are
significantly larger than the lattice parameter of the crystal structure;
this applies as long as the crystal dimensions are typically larger than
a few dozen nanometers). In general terms, the energy structure of a
semiconductor consists of a valence band corresponding to molecular
bonding states and a conduction band representing the molecular anti-
bonding states. The energy range lying between the top of the valence
band and the bottom of the conduction band is known as the forbidden
band, or more commonly the bandgap. An electron situated in the
valence band is in a ground state and remains localized to a particular
atom in the crystal structure, whereas an electron situated in the
conduction band exists in an excited state, in a regime where it
interacts very weakly with the crystalline structure. What
Chapter written by Franck OMNES.
2 Optoelectronic Sensors
differentiates semiconductors from insulators is essentially the size of
the bandgap: we refer to semiconductors where the bandgap of the
material is typically less than or equal to 6 eV, and to insulators when
the bandgap is more than 6 eV: above this, the solar spectrum arriving
on the Earths surface is unable to produce inter-band transitions of
electrons situated in the valence band of the material. Semiconductor
materials are mostly divided into two large classes: elemental
semiconductors (group IV of the periodic table): silicon, germanium,
diamond, etc. and compound semiconductors: IV-IV (SiC), III-V
(GaAs, InP, InSb, GaN) and II-VI (CdTe, ZnSe, ZnS, etc.). Impurities
can be introduced into the volume of the semiconductor material and
can modify its electrical conduction properties, sometimes
considerably. An impurity is known as a donor when it easily releases
a free electron into the conduction band. The characteristic energy
level of the impurity is therefore in the bandgap, slightly below the
conduction band. For example, in the case of compound
semiconductors in group IV of the periodic table such as silicon, the
main donor impurities are those which, being from group V of the
periodic table (arsenic, phosphorous, etc.), are substituted in place of a
silicon atom in the crystal structure: since silicon is tetravalent, these
atoms naturally form four covalent bonds with the silicon atoms
around them, and also easily give up their surplus electron to the
crystal structure. These electrons become free to move, subject to a
weak activation energy provided by thermal agitation. In this case we
refer to n-type doping. In the case of silicon, a group III element
incorporated into the crystal structure of silicon naturally forms three
covalent bonds around it, and then completes its own outer-shell
electronic structure by capturing an electron from its fourth nearest-
neighbor silicon atom, again subject to a weak thermal activation
energy. Such an impurity is known as an acceptor, and doping with
acceptors is known as p-type doping. A hole carrying a positive
elementary charge and corresponding to a vacant energy state in the
valence band is therefore left in the crystal structure of the silicon. In
the case of III-V composites, the donors are mostly atoms from group
IV (silicon) substituted in place of group III elements, or group VI
elements (S, Se, Te) substituted in place of group V elements, and
acceptors are group II (zinc, magnesium) substituted in place of group
Introduction to Semiconductor Photodetectors 3
III elements. In the case of II-VI composites, the most commonly-
encountered donors belong to group VII (chlorine, etc.) substituted in
place of group VI elements, and acceptors belong to either group I
(lithium, etc.) or to group V (nitrogen, arsenic, phosphorous, etc). In
this last case, the group V element is substituted in place of a group VI
element in the semiconductor crystal structure, whereas group I
acceptors are substituted in place of group II elements. The chemical
potential, or Fermi energy, of an intrinsic semiconductor (i.e. one free
from n and p impurities) is found in the middle of the bandgap of the
material. When a moderate n-type doping is added, the Fermi level
rises from the middle of the bandgap towards the conduction band, by
an increasing amount as the level of doping rises. When the level of n-
type doping becomes large, the Fermi level can cross the bottom of
the conduction band and be found inside this band (Mott transition).
The semiconductor then behaves like a metal and for this reason is
called a semi-metal. In this case it is referred to as degenerate. In the
case of p-type doping, the semiconductor is said to be degenerate
when the Fermi level is below the top of the valence band.
1.2. Photodetection with semiconductors: basic phenomena
Photodetection in semiconductors works on the general principle of
the creation of electron-hole pairs under the action of light. When a
semiconductor material is illuminated by photons of an energy greater
than or equal to its bandgap, the absorbed photons promote electrons
from the valence band into excited states in the conduction band,
where they behave like free electrons able to travel long distances
across the crystal structure under the influence of an intrinsic or
externally-applied electric field. In addition, the positively-charged
holes left in the valence band contribute to electrical conduction by
moving from one atomic site to another under the effects of the
electric field. In this way the separation of electron-hole pairs
generated by the absorption of light gives rise to a photocurrent,
which refers by definition to the fraction of the photogenerated free
charge-carriers collected at the edges of the material by the electrodes
of the photodetecting structure, and whose intensity at a given
wavelength is an increasing function of the incident light intensity. On
this level we can distinguish between two large categories of
photodetectors based on the nature of the electric field, which causes
the charge separation of photogenerated electron-hold pairs:
photoconductors, which consist of a simple layer of semiconductor
simply with two ohmic contacts, where the electric field leading to the
collection of the charge-carriers is provided by applying a bias voltage
between the contacts at either end, and photovoltaic photodetectors,
which use the internal electric field of a p-n or Schottky (metal-
semiconductor) junction to achieve the charge separation. This last
term covers p-n junction photodetectors (photovoltaic structures
consisting of a simple p-n junction, and p-i-n photodetectors which
include a thin layer of semiconductor material between the p and n
region which is not deliberately doped), as well as all Schottky
junction photodetectors (Schottky barrier photodiodes and metal-
semiconductor-metal (MSM) photodiodes).
We will now briefly introduce the main physical concepts at the
root of the operation of the different semiconductor photodetector
families. Here the emphasis is placed on a phenomenological
description of the working mechanisms of the devices in question; the
corresponding formalism has been deliberately kept to an absolute
minimum in the interests of clarity and concision.
1.3. Semiconductor devices
Photoconductors represent the simplest conceivable type of
photodetector: they consist of a finite-length semiconductor layer with
an ohmic contact at each end (Figure 1.1). A fixed voltage of
magnitude V
B
is applied between the two end contacts, in such a way
that a bias current I
B
flows through the semiconductor layer, simply
following Ohms law. The active optical surface is formed from the
region between the two collection electrodes. When it is illuminated,
the photogenerated changes produced under the effect of the applied
electric field lead to a photocurrent I
PH
which is added to the bias
current, effectively increasing the conductivity of the device.
Figure 1.1. Diagram of a photoconducting device
The main point of interest in a photoconducting device is its
increased gain, the response of photoconductors being typically
several orders of magnitude greater than that of photovoltaic detectors
for a given material. On the other hand, its other operational
parameters (bandwidth, UV/visible contrast, infrared sensitivity) are
generally below that of other types of photodetectors, which often
greatly limits the scope of its potential applications (this is particularly
the case for photoconductors based on III-V nitrides, as we will see
later on).
1.4. p-n junctions and p-i-n structures
In p-n diodes, the metallurgical linkage of a region of a p-type
doped semiconductor and a region of n-type doping forms a p-n
junction, where the joining of the Fermi levels in equilibrium mostly
occurs through a flow of charge between the n and p regions. In
equilibrium we therefore find a region with no free charge carriers
immediately around the junction, similar to a charged capacitor, where
there are, on the n side, positively ionized donors and, on the p side,
negatively ionized acceptors (this zone is known as the space charge
region (SCR), where ionized donors and acceptors provide fixed
charges). The presence of charged donors and acceptors produces an
electric field in that region which curves the energy bands and, in
equilibrium, forms an energy barrier between the two regions: the
bottom of the conduction band and the top of the valence band on the
n side are below the corresponding levels on the p side (Figure 1.2).
Figure 1.2. Curvature of the energy bands and mechanisms of
photocurrent generation in a p-n junction
The width of the SCR is a decreasing function of the level of
doping in the material, while the height of the energy barrier is an
increasing function of it. An electron-hole pair produced in this SCR
(situation 2 in Figure 1.2) is therefore separated by the effect of the
internal electric field of the junction, and so does not recombine.
These are the charge carriers which contribute to the photocurrent, to
which we can add, to some extent, those generated at a distance from
the junction less than or equal to the diffusion length (situations 1 and
3 in Figure 1.2). The band structure of the junction implies that the
photocurrent will consist of minority charge carriers. For this reason,
the photocurrent flows in the opposite direction to the bias on the
diode, where the forward direction is defined as the direction of flow
of the majority charge carriers (from the n to the p region in the case
of electrons, and vice versa for holes). Moreover, the application of an
opposing external electric field (V
p
V
n
< 0) allows us to increase the
height of the energy barrier in the vicinity of the junction, and also
increase the spatial extent of the SCR, which significantly improves
the efficiency of the separation of electron-hole pairs by increasing the
electric field within the junction.
We note that when the doping level is moderate, the width of the
SCR is important. This effect is beneficial in the case of p-n junction
photodetectors, where in order to increase the photoresponse it is
desirable to ensure that the mechanisms of electron-hole pair
generation through incident light take place predominately inside the
SCR. A simple means of increasing the spatial extent of the SCR is to
introduce between the n and p regions a thin layer of intrinsic
semiconductor material which is not intentionally doped: the structure
is therefore referred to as p-i-n. Such a structure is interesting because
it is possible to maintain high levels of doping in the n and p regions
without significantly reducing the extent of the SCR, whose width is
then largely determined by the thickness of the i layer. Additionally,
increasing the width of the SCR reduces the capacitance of the
structure, which makes p-i-n structures particularly well-suited for
high-speed operation.
1.5. Avalanche effect in p-i-n structures
When the reverse-bias voltage established at the terminals of a p-i-
n structure increases sufficiently that the electric field established in
the junction reaches values close to the breakdown field (in structures
of micron-scale thickness, this is generally the case when the bias
voltage at the terminals reaches a few dozen volts), the
photogenerated charge carriers in the SCR (which is effectively the
region that is not intentionally doped) are accelerated enough to
separate other secondary charge carriers from the atoms in the lattice
that they impact in the course of their motion: this is the avalanche
effect which results in a multiplication of the charge carriers in the
SCR. The gain is therefore greater than 1 for the generation of charge
carriers by light, and this gain can even typically reach 10 or 20 under
favorable conditions. This effect is exploited in what are called
avalanche photodiodes where the levels of n- and p-type doping are
generally adjusted to high values above 10
18
cm
-3
to maximize the
intrinsic electric field of the junction.
1.6. Schottky junction
A Schottky junction is formed by bringing a metal and a
semiconductor into contact. The basic phenomena which lead to the
formation of a Schottky junction with an n-type semiconductor are
summarized in Figure 1.3.
Figure 1.3. Formation of a Schottky junction (in an n-type semiconductor)
In thermal equilibrium, when the Fermi levels of the metal and the
semiconductor are equalized, a transfer of electronic charge occurs
from the semiconductor to the metal in the case where the work
function q.)
M
of the metal (q being the elementary charge) is greater
than the electron affinity X of the semiconductor, and a SCR appears
at the edge of the semiconductor of width x
d
next to the junction,
where the only charges present are the positively-ionized donors. A
curvature of the energy bands therefore occurs at the junction, which
leads to the appearance of an energy barrier between the metal and the
semiconductor, called a Schottky barrier, whose height is given to first
approximation by the expression:
q )
Bn
=q )
M
F
( )
[1.1]
In equilibrium, therefore, we find an intrinsic electric field
immediately next to the metal-semiconductor junction which is
comparable in form to that found in a p-n junction. Consequently, it is
the phenomenon of photogeneration of charge carriers inside and near
to the SCR which is responsible for the appearance of a photocurrent,
with the electron-hole pairs being separated by the effect of the
electric field in the Schottky junction. It is possible, as in the case of
the p-n junction, to modify the intensity of the internal electric field in
the junction by applying a bias voltage V between the semiconductor
and the metal of the Schottky contact (Figure 1.4).
Figure 1.4. Reverse-bias of a Schottky junction (n-type semiconductor material)
In the case of an n-type semiconductor, the application of a
negative voltage between the semiconductor and the metal electrode
of the Schottky contact has the effect of reverse-biasing the Schottky
junction, which leads to an increase in the height of the effective
barrier, along with an increase in the width of the SCR. This last effect
is of course favorable for photodetection. Indeed, it follows that the
majority charge carriers (electrons) cannot flow towards the Schottky
contact, and only the minority carriers (holes) generated by external
excitation (in particular photogeneration) can reach the Schottky
contact and hence produce an electric current: as in the case of the p-n
junction, we therefore find that the current flows in reverse through
the Schottky junction, that is, from the semiconductor towards the
Schottky contact. The illumination of Schottky photodiodes can occur
through the front or rear face (often this second option is chosen in the
case where the substrate material is transparent to the light to be
detected, as is the case for example with sapphire). In the case of
illumination through the front face, we resort to a semi-transparent
Schottky contact, characterized by a very small thickness of metal (of
the order of 100 ) selected to ensure sufficient optical transmission:
while a thin layer of gold of 100 thickness transmits up to 95% of
the incident light in the infrared, the percentage transmitted in the
ultraviolet is around 30% in the range 300-370 nm. The gain of p-i-n
photodiodes (other than the specific case of avalanche photodiodes)
and Schottky photodiodes is at most 1, which would be the case if all
the photogenerated charge carriers were collected by the electrodes at
the ends of the device.
1.7. Metal-semiconductor-metal (MSM) structures
An MSM structure consists of two Schottky electrodes, often
interlinked in the form of a comb structure, leaving a free
semiconductor surface between the two contacts which forms the
active region in which light will be absorbed. A bias voltage can be
applied between the two electrodes, in order to break the initial
electrical symmetry of the contacts: one of the Schottky junctions is
reverse-biased, producing a SCR of increased width, and the other
junction is forward-biased.
The absorption of light near the reverse-biased junction creates
electron-hole pairs which are separated under the effects of the electric
field present in the SCR, thus creating the photocurrent. The other
electrode, consisting of a forward-biased (and hence transmissive)
Schottky junction, simply acts as a collection electrode. The band
diagram of the device under increased bias voltage (V
B
) is represented
schematically in Figure 1.5, in which L is the distance between two
adjacent contact fingers, )
0
is the height of the Schottky barrier and
I
ph
is the photocurrent. MSM photodetectors normally use
semiconductor materials which are not intentionally doped, are
chemically very pure and electrically very resistive. The SCRs
associated with Schottky junctions made of these materials are hence
of significant width which, for a given bias voltage, allows the electric
field of the junction to extend more easily into semiconductor regions
some way from the contact. It follows that photogenerated electron-
hole pairs are more easily separated and collected by the electrodes at
either end.
Figure 1.5. Energy band diagram for an MSM structure
under electrical bias; effect of illumination
1.8. Operational parameters of photodetectors
The main parameters which define the behavior of an ultraviolet
photodetector are respectively the response coefficient, the gain, the
quantum efficiency, the bandwidth, the noise equivalent power (NEP)
and the detectivity.
1.8.1. Response coefficient, gain and quantum efficiency
The response coefficient of a photodetector, R
i
, links the
photocurrent I
ph
to the power of the incident light P
opt
through the
relationship:
I
ph
= R
i
P
opt
[1.2]
It is important to note in passing that the response coefficient is a
quantity independent of the active optical surface of the photodetector
structure: indeed, the photocurrent as well as the incident optical
power are both, in the ideal case, proportional to the active optical
surface. At a given wavelength O, the flux ) of photons arriving on
the semiconductor surface, which is defined as the number of photons
reaching the active surface per unit time, is given by:
)= P
opt
O / h c
( )
[1.3]
where h is the Planck constant and c is the speed of light.
The quantum efficiency K is defined as the probability of creating
an electron-hole pair from an absorbed photon. Considering that all
the incident light is absorbed in the semiconductor material, the rate G
of electron-hole pair generation per unit time is thus given by:
G =K )=K P
opt
O / h c
( )
[1.4]
If we now introduce the gain parameter g which corresponds to the
number of charge carriers detected relative to the number of
photogenerated electron-hole pairs, then the photocurrent is given by
the equation:
I
ph
=q G g =qK P
opt
O / h c
( )
g = qK O / hc
( )
g
( )
P
opt
[1.5]
where q is the elementary charge (1.602 x 10
-19
C), from which we
obtain the expression for the response coefficient of the detector:
R
i
=q g K O / hc
( )
[1.6]
1.8.2. Temporal response and bandwidth
The speed of response of a photodetector may be limited by
capacitative effects, by the trapping of charge carriers or by the
saturation speed of charge carriers in the semiconductor. These
phenomena all lead to a reduction in the response of the photodetector
in the high-frequency domain. The cutoff frequency f
C
of the
photodetector is defined as the frequency of optical signal for which
the response coefficient is half that for a continuous optical signal.
The temporal response of a photodetector is characterized by the fall
time W
f
(or the rise time W
r
), which is defined as the time needed for the
photocurrent to fall from 90% to 10% of its maximum (or to rise from
10% to 90% of it). In the case of a transient exponential response with
a time constant W, the following relationship links the bandwidth BW
and the temporal response of the photodetector:
BP =1/ 2S W
( )
=2.2 / 2SW
m
( )
=2.2 / 2S W
d
( )
[1.7]
1.8.3. Noise equivalent power
The NEP is defined as the incident optical power for which the
signal-to-noise ratio is 1, and hence the photocurrent I
ph
is equal to the
noise current I
b
. In other words, it is the smallest optical power which
can be measured. It follows that the NEP parameter is given by the
equation:
NEP = I
b
/ R
i
in W
( )
[1.8]
In the case of white noise, the noise current I
b
increases as the
square root of the bandwidth of the photodetector device. It follows
that it is preferable and customary to use the following expression for
the NEP, normalized with respect to the bandwidth BW:
NEP* = NEP BW
( )
1 2
in W Hz
1 2
( )
[1.9]
In semiconductors, there are five sources of noise:
shot noise, mainly due to the random nature of the collisions of
incident photons;
thermal noise, due to random collisions of charge carriers with
the atoms of the crystal lattice, in permanent vibration due to thermal
motion;
partition noise, caused by the separation of the electric current
into two parts flowing across separate electrical contacts;
generation-recombination noise, caused by the random
generation and recombination of charge carriers, either band to band
or via trapping levels situated in the bandgap;
1/f noise, associated with the presence of potential barriers at the
level of the electrical contacts. This last type of noise dominates at
low frequencies.
1.8.4. Detectivity
This figure of merit is defined by the equation:
D = NEP
( )
1
= R
i
/ I
b
in W
1
( )
[1.10]
In general terms, the photocurrent signal increases in proportion to
the active optical area A
opt
, and in addition the noise current increases
with the square root of the product of the active optical area with the
bandwidth BW. It follows that the preferred method of comparing
between different photodetectors is to use an expression for the
detectivity normalized with respect to these parameters, written:
D* = D A
opt
BP
( )
1 2
= R
i
I
b
( )
A
opt
BP
( )
1 2
in W
1
cm.Hz
1 2
( )
[1.11]
The normalized detectivity is the most important parameter for
characterizing a photodetector because it allows direct comparison of
the performance of photodetectors using technologies and methods of
operation which are at first glance very different. It is clear from the
preceding definitions that the determination of the NEP and the
detectivity requires measurement of three parameters: the response
coefficient, the bandwidth and the noise current of the photodetector
device. The measurement of the noise current must be made in
darkness. The device is biased using a very stable voltage source, and
the entire measurement system must itself have an intrinsic noise level
considerably lower than the intrinsic noise of the photodetector device.
Chapter 2
PIN Photodiodes for the
Visible and Near-Infrared
2.1. Introduction
Photodiodes are optoelectronic devices with two electrodes and
asymmetric electrical characteristics. They are normally reverse-
biased (biased in the non-conducting direction) and used to convert
optical fluxes into electrical currents. The term PIN refers to the three
doped semiconductor layers in their active part, of types P, intrinsic
and N respectively, a structure which aims to optimize the
characteristics compared to those of simple PN junction photodiodes.
Figure 2.1. Schematic layout of PIN photodiodes: collinear optical flux and charge
transport on the left, orthogonal on the right
Chapter written by Baudoin DE CREMOUX.
Figure 2.1 shows the basic processes taking place in a PIN
photodiode:
generation of an electron-hole pair through absorption of a
photon, most probably in the I region, either using radiative or non-
radiative processes;
transport under the effects of the electric field present in the I
region: of the electron towards the N region and of the hole towards
the P region, then to the external circuit through the metallic contacts
deposited on the surfaces, which have a window where required.
Thus, the absorption of a photon in the I region leads to the
transport of an electron in the external circuit. Figure 2.1 shows the
possible arrangements for the optical flux and the electrical current.
Collinear geometry is the most commonly used. We use it as the
reference type for the analysis of photodiode characteristics. After this,
we will consider orthogonal geometry.
The spectral domain of interest, mostly from 400 to 1,700 nm,
determines the materials used to fabricate photodiodes. This design is
the most mature of the semiconductors, and as a result almost all
photodiodes used in this domain have a PIN structure, or their central
layer is at least weakly doped compared to the P and N layers.
This chapter is organized in order to first provide the reader with
the basic tools to understand the concepts of PIN photodiodes for a
given application, then it describes the devices which are currently
available or which should be available soon. With this aim in mind,
section 2.2 recalls some of the physics of semiconductors,
emphasizing the processes taking place in photodiodes, the pertinent
characteristics of the materials and the notations used later in the text.
Sections 2.3 and 2.4 establish the simple quantitative relations
between the structure of devices with typical geometries and their
static and then dynamic characteristics. We will stress the
compromises required among these characteristics if we are to keep a
particular geometry. Section 2.5 then describes the properties of
photodiodes with collinear geometry, currently used in a numerous
professional and mass-market systems. Finally, section 2.6 describes
PIN Photodiodes for the Visible and Near-Infrared 17
advanced devices with novel structures which push the boundaries of
the previously-discussed compromises. These are for the most part
laboratory devices which could find applications in real-world systems.
2.2. Physical processes occurring in photodiodes
2.2.1. Electrostatics in PIN diodes: depleted region
Depleted region
Figure 2.2. From top to bottom, profiles of fixed charge density,
electric field and potential across a reverse-biased PIN diode
In the neighborhood of a PN junction there is a region devoid of
free charge carriers (the depleted region) where fixed ionized
impurities can be found, negative on the P side and positive of the N
side, forming a space charge region (SCR). By integrating the Poisson
equation twice in the z direction:
d
2
V
dz
2
=
U z
( )
H
[2.1]
Where U(z) is the charge density and H is the dielectric constant, we
can obtain the profiles of the electric field E and the potential V across
the structure. In the case of an abrupt PN junction, the thickness of the
depleted region varies as (V
BI
V) where V
BI
is the built-in or
diffusion potential present in the absence of external bias applied to
the diode, and V is the voltage across the terminals of the diode. As a
result, the capacitance of the junction decreases when the reverse-bias
voltage increases in absolute value. The case of a PIN diode structure
is considered in Figure 2.2, showing in sequence the profiles across
the structure of the density of fixed charges, of the electric field and of
the potential.
In the realistic approximation that the densities of ionized
impurities are much greater than the density of residual impurities in
the I region (typically by a factor of 100), we draw the following
conclusions:
the extension of the depleted region into the P and N regions is
negligible, and the depleted region is to all intents and purposes
coincident with the I region whose thickness, represented by d in the
rest of this chapter, is a very important parameter in the definition of a
PIN photodiode;
the capacitance of the junction is independent of the reverse-bias
voltage. It is given, as for a conventional capacitor, by:
C
J
=H
S
d
[2.2]
where S is the area of the junction;
the electric field is constant in the depleted region and, insofar as
V>>V
BI
, its value is
E =V d [2.3]
These approximations will be used throughout the remainder of the
chapter unless explicitly stated otherwise.
2.2.2. Mechanisms of electron-hole pair generation
2.2.2.1. Radiative generation and optical absorption
Electronic transitions between permitted levels in the valence band
(VB) and levels in the conduction band (CB) can occur under the
action of photons transferring their energy to the electrons. Thus, a
photon can be absorbed, creating an electron-hole pair. This process of
radiative generation, shown schematically in Figure 2.3, is the inverse
of the process of recombination which gives rise to stimulated
emission of photons in laser diodes.
It follows from this that the only photons absorbed efficiently are
those whose energy hQ is greater than the width of the bandgap E
G
of
the material under consideration:
hQ _ E
G
= E
C
E
V
[2.4]
Figure 2.3. Radiative generation
The material is transparent to photons with lower energies. The
parameter E
G
is thus a criterion in the choice of materials when
designing a photodetector. Optical absorption is characterized by the
coefficient of absorption D (cm
1
), according to the relationship:
dF
Fdz
= [2.5]
where F (cm
2
s
1
) is the photon flux propagating in the z direction.
The level of radiative generation G
R
(cm
3
s
1
), the number of electron-
hole pairs created per unit volume and time, is therefore:
G
R
=DF [2.6]
It follows from these definitions that the level of radiative
generation in a homogenous material, for a photon flux F
0
falling on a
surface at z=0, can be written:
G
R
=DF
0
exp Dz
( )
[2.7]
We see that 1/D (cm) gives an order of magnitude for the distance
required to absorb the radiation. Furthermore, the knowledge of
variation of D with photon energy (or with their wavelength) is
essential for the choice of materials for use in a photodetector. Figure
2.16 shows graphs for the main semiconductors used in the spectral
domain under consideration in this chapter.
The shape of these graphs and the magnitude of D depend on the
direct or indirect nature of the band structure of the material. This
nature is defined in Figure 2.4 which represents, for the two types of
material, the variations in energy of the electrons, with the wavevector
k describing their motion in the periodic crystal lattice.
Figure 2.4. Direct bandgap on the left and indirect bandgap on the right
For materials with a direct bandgap (such as gallium arsenide) the
main minimum in the conduction band is found at the same vector k
as the maximum in the valence band. Furthermore, radiative
transitions must simultaneously conserve the energy and momentum
of the particles involved (the photon and electron). The momentum of
the electron is much greater than that of the photon, and as a result
radiative transitions are effectively vertical and connect with a high
probability those states at the top of the VB and those at the bottom of
the CB.
The opposite is true for materials with an indirect bandgap (such as
silicon), and radiative transitions require the involvement of the third
particle, a phonon, to provide the difference in momentum between
the initial and final states. These transitions are much less probable
and D grows much less quickly with the energy of the photons above
E
G
.
2.2.2.2. Non-radiative generation
As well as the radiative processes there are also non-radiative
processes of generation-recombination. Thus, in the SRH (Shockley
Read Hall) process shown in Figure 2.5 in generation mode, the
required energy is borrowed from the thermal motion of the crystal
lattice. This is one of the contributions to the dark current of
photodetectors, a detrimental parasitic phenomenon.
Figure 2.5. Non-radiative SRH generation
The transition takes place in two parts, via a deep state of energy
E
T
, which is close to the middle of the forbidden band and is the result
of a crystal defect (impurity, vacancy, etc.). The rate G
NR
(cm
3
s
1
) of
SRH generation-recombination can be written [SZE 81, p. 35]:
G
NR
=
V
n
V
p
Q
th
N
T
n
i
2
np
( )
V
n
nn
i
exp
E
T
E
I
kT
(
(
)
)
l
l
l
l
l
V
p
p n
i
exp
E
T
E
I
kT
(
(
)
)
l
l
l
l
l
[2.8]
with the following notation:
V
n
and V
p
capture cross-sections for electrons and holes by the
deep state;
v
th
, thermal velocity Q
th
= 3k T m*
( )
( )
;
N
T
, density of deep states;
n
i
, intrinsic electron density;
n and p, electron and hole densities;
E
I
, middle of the forbidden band;
kT, Boltzmann factor.
In the depleted region of a photodiode we have np = n
i
2
. If we
further assume that V
n
=V
p
=V and E
T
= E
I
, and define an effective
lifetime W
eff
:
G
NR
= n
i
Q
th
N
T
2 = n
i
W
eff
[2.9]
We also know that:
n
i
2
= N
C
N
V
exp E
G
kT
( )
[2.10]
where N
C
and N
V
are the equivalent densities of state for the CB and
VB. We can draw the following conclusions from this:
G
NR
increases very rapidly as the ratio E
G
/kT decreases;
G
NR
is not an intrinsic characteristic of the material, but depends
for its purity on N
T
.
In an electrically neutral region of type p(n), where the electron
(hole) deficiency relative to thermodynamic equilibrium is Gn(Gp), if
we define Wn(Wp) as the lifetime of the minority electrons (holes), it
follows that:
G
NR
=GnVQ
th
N
T
=Gn W
n
[2.11]
SRH non-radiative generation is particularly active at the surface
of devices, where the density of defects is very high. It can be
reduced by a passivation treatment; for silicon the most effective is
the deposition of its natural oxide SiO
2
, silica. Another process of
non-radiative generation is shock ionization which occurs in a
semiconductor material under a strong electric field. It grows with
E
G
and the breakdown field is of the order of E
B
= 10
5
V/cm for the
materials considered. This effect is one cause of breakdown in
diodes, limiting the reverse-bias voltage that can be applied. It is
exploited as a gain mechanism in the avalanche photodiodes which
are described in Chapter 3, and whose structure is optimized with
this in mind.
2.2.3. Transport mechanisms
The charge carriers, electrons and holes generated by the
aforementioned processes, are transported in the device to the two
electrical contacts that transfer it to the external circuit specifically,
to the load resistance across the terminals, across which the electrical
signal is measured.
2.2.3.1. Transport under electric field and resultant current
In the depleted region of a photodiode (normally reverse-biased) an
elevated electric field is present, and the electrons and holes are driven
at a speed which depends on the intensity of the field, as shown in
Figure 2.6.
Figure 2.6. Characteristics Q
n
(E) and Q
p
(E) in logarithmic coordinates
For electric fields such that E < 10
3
V/cm, the velocities are
proportional to E (linear regime) and we have:
Q
n, p
E
( )
= P
n, p
E [2.12]
where P
n
and P
p
(cm
2
/Vs) are the respective mobilities of electrons
and holes. For electric fields such that E > 10
4
V/cm, the velocities are
fairly constant and equal to a saturation velocity of the order of
v
s
= 10
7
cm/s. Between these two regions, and for certain materials,
the electron velocity passes through a maximum (high mobility
regime).
2.2.3.2. Transport along the concentration gradient and diffusion
current
In the contiguous electrically neutral regions in the depleted region,
electrons and holes can be transported in the absence of an electric
field under the effect of their concentration gradients. Their fluxes F
n
and F
p
can be written:
F
n, p
= D
n, p
0n, p
0z
[2.13]
where D
n, p
= P
n, p
kT q cm
2
s
( )
are the diffusion coefficients.
The various recombination processes are also active in these
regions, such that the electrons and holes are only efficiently
transported over a distance known as the diffusion length and given
by:
I
n, p
= D
n, p
W
n, p
[2.14]
2.3. Static characteristics of PIN photodiodes
2.3.1. I/V characteristics and definition of static parameters
Figure 2.7 shows I/V curves, linking the current crossing a
photodiode to the voltage across its terminals for three illumination
conditions. From this we can extract the parameters commonly used to
characterize photodiode operation, whose relations to the structure of
the device are described in the following sections. The photocurrent I
P
produced by a photodiode is proportional to the illumination across a
large range of optical power P. The coefficient of proportionality is
the response coefficient R (A/W) (responsivity), not to be confused
with the sensitivity, which measures the minimum optical power
detectable by an optical detector, and which depends on the circuit
into which the photodiode is inserted and the specifications of the
signal to be received.
Figure 2.7. I/V characteristics of a photodiode, inserted in the circuit shown
and under several illumination conditions
The external quantum efficiency K
e
links the generated electron
flux to the incident photon flux. Thus, we have:
R =
I
p
P
=
q
hQ
K
e
[2.15]
where q is the charge of an electron and hQ is the energy of a photon.
The dark current I
0
, discussed in section 2.3.3, is caused by non-
radiative generation and/or the tunnel effect.
The saturation current I
M
is the photocurrent above which several
mechanisms limit the linearity of the response:
the intrinsic mechanism, due to the screening of the electric field
in the depleted region by the space charge that is produced by the
mobile charge carriers created by the illumination (see section 2.3.4);
the circuit-dependent mechanism, caused by the voltage drop
across the load resistance (see the circuit in Figure 2.7).
A gain, characterized by the multiplication coefficient M defined
by:
0 p
I
M
I I
=
[2.16]
can be obtained under a high reverse-bias voltage V
AV
in avalanche
photodiodes (see Chapter 3). The breakdown voltage V
B
is the highest
reverse-bias voltage the diode can support. For V=V
B
the dark current
grows very rapidly and the power dissipated in the diode can lead to
its destruction.
2.3.2. External quantum efficiency
The external quantum efficiency defined above can be thought of
as the product of an internal quantum efficiency K
i
and an optical
efficiency K
o
, the ratio of the photon flux entering the device to the
flux of photons incident on its surface. These two ratios, which are
both dependent on the wavelength, are considered in the following
text.
2.3.2.1. Optical efficiency
The optical refractive index n of the semiconductors under
consideration is in the region of 3, and as a result the incident light
rays are refracted and reflected at its surface. At normal incidence,
used as the reference situation, the Frensel transmission coefficient
determines the optical efficiency:
K
o
=
4n
n1
( )
2
[2.17]
which is of the order of 70% and depends little on the wavelength and
the angle of incidence. It is possible to achieve optical efficiencies
close to 100% at a given wavelength O, by depositing an antireflection
coating on the surface, for example of refractive index n and
thickness O 4 n
( )
. The optical efficiency therefore depends on O and
on the angle of incidence.
2.3.2.2. Internal quantum efficiency
The electron-hole pairs created by the absorption of photons
entering the device may, or may not, give rise to a current flow in the
external circuit, depending on where they are generated in the
structure. If generation is in the depleted region, the electrons and
holes are separated by the electric field, as shown in Figure 2.1,
leading to the passage of one electron for each absorbed photon: thus,
the internal quantum efficiency is 100%. In the case of generation in
the electrically neutral P and N regions, where the electric field is
weak, only the electrons (or holes) generated less than one diffusion
length away are transported into the depleted region, and the internal
quantum efficiency is in most cases weak.
It is therefore desirable to design a photodiode in such a way that
the radiation should be entirely absorbed in the depleted region.
Figure 2.8 shows a model which provides a good evaluation of the
internal quantum efficiency of a PIN photodiode, as a function of the
nature of the material on whose surface the radiation falls. The curves
1 and 2 represent the profiles of the radiative generation rate G
R
across
two structures. The structure for curve 1 is that of a homojunction PIN,
where all of the materials used have the same width of bandgap,
which is below the energy of the photons to be detected. The radiation
is assumed to be incident from the p side, and part of the radiation is
absorbed there, where the quantum efficiency is negligible. Another
part is absorbed in the i region, and has a quantum efficiency of 100%.
Finally, the contribution of the residual radiation entering the n region
is also negligible. The quantum efficiency can therefore be written:
K
i
=exp Ddp
( )
1exp Dd
( )
l
l
l
[2.18]
where the definitions of d
P
and d can be found in Figure 2.8.
Depleted region
Figure 2.8. Optical generation rate across the structure of two PIN
photodiodes: curve 1 shows a homojunction PIN photodiodes;
curve 2 shows a heterojunction PIN photodiode
Curve 2 corresponds to a PIN photodiode, with the capital P
indicating that the material in the P region has a bandgap width greater
than the energy of the photons to be detected and which is therefore
transparent to them. The same reasoning provides us with the quantum
efficiency:
K
i
=1exp Dd
( )
[2.19]
2.3.3. Dark current
The dark current is the sum of several contributions: non-radiative
generated currents requiring an input of energy (such as the SRH
process) and the tunneling current, which does not require such an
input. These two processes are represented in Figure 2.9 for a sharp
p
+
n homojunction.
2.3.3.1. Generation currents
According to equation [2.9], the dark current contribution of
generation in the depleted region is:
I
0
= SdG
NR
= Sdn
i
W
eff
[2.20]
Taking account of equation [2.11] of the fact that the shortage of
electrons (holes) in the p(n) material near the depleted region is
2
, ,
i AD
n p n N G = and the fact that the thickness of the p(n) material is
L
n,p
(see section 2.2.3.2) the generation contribution in the neutral p
and n regions can be written:
I
0
= Sqn
i
2
L
n
N
A
W
n
L
p
N
D
W
p
( )
[2.21]
This is the traditional expression for the reverse current of a diode
according to Shockley.
Figure 2.9. Profile of the VB and the CB of a reverse-biased p+n junction.
The generation process (shown here in the depleted region) requires an
input of energy; the tunnel process occurs at constant energy
Thus, we see from equations [2.20] and [2.21] that the generation
current is primarily a function of n
i
and varies very rapidly with E
G
/kT.
2.3.3.2. Tunneling current
Using the tunneling effect electrons can also be transported at a
constant energy across the triangular potential barrier present at the
junction, as long as it is narrow enough. A calculation of the current
gives the following expression [STI 82]:
I
T
= S
2m*q
3
EV
4S
3
h
2
E
G
1 2
exp
S 2m* E
G
3 2
4qhE
(
(
)
)
[2.22]
where E=V/d is the electric field in the depleted region, which is
constant for a PIN diode. From equation [2.22] we determine that, for
a given material, I
T
increases very rapidly with the voltage V across
the terminals of the diode, mostly as a result of the increase in E.
Additionally, I
T
increases when E
G
decreases, such that the
contribution of the tunneling current to the dark current is most
significant for photodiodes made of materials with a small bandgap.
2.3.4. Breakdown voltage
2.3.4.1. Zener breakdown via the tunneling effect
The very rapid variation of the tunneling dark current with the
reverse-bias voltage is the first process of breakdown. This mainly
applies to materials with a narrow depleted region and a small
bandgap; the two parameters which reduce the width and the height of
the potential barrier at the PI junction. The tunneling effect can be the
dominant one in the materials used for detection of the longest
wavelengths considered in this chapter (see section 2.5.3.2), if the
doping level of the i zone is too high [TAK 80].
2.3.4.2. Avalanche breakdown
We can define a critical electric field E
C
, of the order of
2 x 10
5
V/cm for the lightly-doped materials considered in this chapter,
where the multiplication coefficient becomes infinite and leads to
breakdown. In the case of PIN photodiodes, which are not designed to
take advantage of avalanche gain, breakdown can arise locally, most
often around the edge of the junction where the point effect increases
the electric field. The breakdown voltage is therefore approximated
below by:
V
C
= E
C
d [2.23]
2.3.5. Saturation current
2.3.5.1. Extrinsic limitation by the voltage drop across
the load resistance
In order to have a linear response, the photodiode must remain
reverse-biased (V<0) so that the carriers are separated by the electric
field in the depleted region. Thus, it can be seen in the simple bias
diagram of Figure 2.7 that for a bias voltage V
0
, the voltage drop in
the load resistance R
L
can cancel out the voltage V across the diode
terminals for a limiting current I
M
such that:
I
M
=V
0
R
L
[2.24]
2.3.5.2. Intrinsic limitation by the space charge of mobile
charge carriers
It is especially important to take into account the contribution of
mobile charge carriers to the electric field E in the depleted region for
high levels of incident optical power. Specifically, E must satisfy the
following conditions:
0 _ E _ E
B
[2.25]
in order to avoid breakdown of the junction.
We will determine the maximum value I
M
of the photocurrent that
simultaneously satisfies these conditions for a heterojunction PIN
structure, with a coefficient of absorption in the i region such that Dd>>1,
and all the illumination is absorbed close to the Pi junction. We further
assume that the mobile electrons in the i region are transported with the
saturation velocity v
s
, independent of the electric field.
Figure 2.10 shows the density profiles of the fixed and mobile
charges and the electric field profile across the structure. Given the
hypotheses set out in this section, it follows that the charge density
due to electrons is ,
s
J U Q = where J is the current density, the
electric field variation in the depleted region is G E = Ud H and the
maximum current I
M
, obtained when G E = E
B
, can be written:
I
M
=C
J
Q
s
E
B
[2.26]
where C
J
is the capacitance of the junction.
Depleted region
Depleted region
Figure 2.10. Space charge density and electric field, for I=0 (dotted line)
and for I=I
M
(solid line). For I=I
M
, the electric field varies
from 0 to the breakdown field E
B
Note that, for this value of current, the voltage across the terminals
of the junction is
V = E
B
d 2 =V
B
2 [2.27]
We can easily show that, regardless of the components in the bias
circuit, the part of the curve representing the variation of the electric
field in the depletion region (marked :) rotates around a fixed point
as the photocurrent varies. Its position depends on the values of V
0
and
RL; Figure 2.10 provides an example where R
L
= 0 and the voltage
across the diode terminals is constant and equal to the bias voltage V
0
= V
C
/2.
2.4. Dynamic characteristics of PIN photodiodes
2.4.1. Intrinsic limitations to the speed of response
2.4.1.1. Transit time in the depleted region
The transit time of photocarriers in the depleted region is the first
limit to the response time of the photodiode. Its analysis is carried out
here with the same approximations as in section 2.2.1, which is a good
representation of the properties of real PIN diodes. Figure 2.11 shows
the electric charge profile in the depleted region at a time t after a brief
impulse signal, of energy W
0
, which was applied to the photodiode at
time 0. The packet of generated electrons is transported at speed v
s
and takes time W
t
=dv
s
to cross the depleted region.
Depleted region
Figure 2.11. Density of mobile charges produced in the depleted region of a
PIN photodiode in response to an optical impulse G (z, t)
Figure 2.12. Impulse and frequency response of a
PIN photodiode limited by the transit time
The solution of Maxwells equations for the case R
L
=0 shows that
the current generated by the diode is the rectangular impulse shown in
Figure 2.12, of duration W
t
and of intensity:
I
0
= q hQ
( )
W
0
W
t
[2.28]
The frequency response can then be obtained by Fourier
transforming the impulse response. This is shown in Figure 2.12 and
given by:
I f
( )
= I
0
sin S f W
t
( )
S f W
t
[2.29]
The 3 dB cutoff frequency I f
( )
I
0
=1 2
( )
is:
f
t
=0.44 W
t
[2.30]
We see that this varies as 1/d. In the case where Dd <<1 and the
generation is uniform in the depleted region, the more accurate model
presented in section 2.4.2.2 shows that we obtain the following
numerical result:
f
t
=0.55 W
t
[2.31]
We see that regardless of the value of Dd, we can apply this
approximate numerical expression, which we will continue to use:
f
t
=1 2W
t
[2.32]
Also, equation [2.19] shows that if Dd <<1 we have:
K
i
=Dd [2.33]
Thus we see from equations [2.32] and [2.33] that the product K
i
f
t
only depends on the characteristics of the material in the i region:
K
i
f
t
=DQ
s
2 [2.34]
from which we conclude that, for the geometry considered and for a
particular wavelength and material, we must accept a compromise
between the internal quantum efficiency and the speed of response.
2.4.1.2. Diffusion time in the neutral regions
When the optical generation occurs in one of the electrically
neutral regions, n or p, the minority photocarriers must initially be
transported by diffusion to the edges of the depleted region, in which
they can be transported by the electric field. This process can be
characterized by a diffusion time W
d
which we can show has an order
of magnitude:
W
d
=d
2
D
n p
[2.35]
for electrons (holes) generated in the p(n) region. In practice the
diffusion time is always greater than the transit time in the depleted
region and has the effect of reducing the response speed of
photodiodes. Therefore, excluding some exceptions (see section 2.6.4),
it is desirable to design these devices such that the absorption of
photons takes place in the depleted region, both to optimize the speed
of response and the quantum efficiency.
2.4.1.3. Trapping at the Pi heterojunction
An additional parasitic mechanism can slow the speed of
heterojunction PIN photodiodes. The holes generated in the i
region, which has a small bandgap, can be trapped at the interface
with the P region by discontinuity in the valence band. They can
then be emitted by thermo-ionic effects, although with delay, into
the P region. This effect is particularly marked for holes, whose
effective mass is greater than that of electrons. The proposed
remedy [FOR 82] is the use of gradual heterojunction or layers
with an intermediate-sized bandgap in order to reduce the potential
barrier presented to the holes and to make it easier for them to
cross through tunneling or thermal effects. We will assume below
that the structures of the devices are designed in such a way that
these effects are rendered negligible.
2.4.2. Limitations due to the circuit
2.4.2.1. Response of the basic circuit, optimum thickness
of the i region
In order to produce a signal, a photodiode is inserted into a
circuit whose components also contribute to limit the bandwidth of
an optical detector. The most simple circuit is shown in Figure 2.13,
consisting of the capacitance of the junction C
J
in parallel with a
load resistance R
L
, with a negligible impedance of the source at
high frequencies. This circuit is driven by the photodiode generated
current, whose response is independent of the voltage at its
terminals.
Figure 2.13. 3 dB cutoff frequency of a PIN photodiode, as a function of the thickness
of the depleted region. The material is assumed to be GaInAs/InP, the width of the
active region 10m and the load resistance 50 :
The impulse and frequency response of such a resistor-capacitor
network are well known. The frequency response is given by the
expression:
I f
( )
=
I
0
1 2S fR
L
C
J
( )
2
[2.36]
and the 3 dB cutoff frequency is:
f
RC
=1 2SR
L
C
J
( )
[2.37]
We see that f
RC
varies with d. The cutoff frequencies f
W
and f
RC
vary inversely with d, thus there is an optimal thickness for the
depleted region which maximizes the global cutoff frequency f
3dB
.
Figure 2.13 shows the cutoff frequencies variation with d for a typical
PIN photodiode.
The optimal thickness of d is obtained when f
W
= f
RC
, giving:
W
t
=S R
L
C
J
[2.38]
The maximum cutoff frequency is then
f
3dB
=
Q
s
2 2d
[2.39]
2.4.2.2. Response of a real PIN photodiode
It is also interesting to consider the results of a more complete
analysis [BOW 87] of a photodiode made of GaInAs/InP, designed for
optical telecommunications at a wavelength of 1.3 m, taking into
account:
the finite value of the optical absorption coefficient: D = 1.3
10
4
cm
1
;
the different saturation velocities for electrons and holes Q
n
= 6
10
6
cm/s and Q
p
= 4.8 10
6
cm/s;
the inevitable parasitic capacitances of the circuit: C
P
= 50 fF;
a parasitic inductance L, caused by the connecting wires,
adjustable and kept as a free parameter.
The illumination is assumed to be incident at the N side of the
structure, and the diameter of the active region is assumed to equal
30 m. Figure 2.14 shows the response curves of the photodiode
inserted in its circuit, with five examples of the parameters d and L
listed in Table 2.1 along with the performances obtained, the cutoff
frequency and the internal quantum efficiency. The load resistance
was kept constant: R
L
= 50 :.
Table 2.1. Values of the parameters for the curves in Figure 2.17
Figure 2.14. Frequency response of a PIN photodiode made
of GaInAs/InP at 1.3m, inserted into the circuit shown.
The parameters of each curve are given in Table 2.1
Curves 1 and 2 show behaviors limited by the transit time and by
the circuit respectively. Curve 3 corresponds to a depleted region
whose width is close to the optimum, while curve 4 shows that its
response can be further improved by parasitic induction. For a higher
value of inductance (curve 5), the bandwidth is reduced, but a gain in
signal is obtained.
2.4.3. Power-frequency compromise, Pf
2
law
For certain optical transmission systems, it is desirable to use a
distortion-free increased electrical power in the sinusoidal regime
across the resistance R
L
loading the photodiode. In particular, this
allows us to simplify the electronics of the receiver, and is made
possible by the current availability of powerful modulated optical
sources or optical preamplifiers. Given the maximum frequency to be
transmitted, the problem is now to design a photodiode, defined
mainly by the thickness d of the depleted region and by the area S of
the active region, as well as select the bias voltage V
0
in a circuit, as in
Figure 2.7. We assume that the incident optical power is modulated at
a level of 100%.
Figure 2.15. Excursion from the operating point of the photodiode in a plot of its I/V
characteristics: allowing us to determine the maximum RF electrical power. Points A,
A and A are respectively the mean and extreme points
In order to maximize the RF power dissipated in the load, the
photocurrent must vary sinusoidally (see Figure 2.15) between 0 and
the maximum value I
M
given by equation [2.25], as well as the voltage
varying between the breakdown voltage V
B
given by equation [2.23]
and the voltage V
B
/2 (equation [2.27]). The maximum effective power
is therefore:
P
M
=
1
2R
L
V
C
4
(
(
)
)
2
=
E
B
2
d
2
32R
L
[2.40]
If we assume that the photodiode is optimized for its frequency
response, and we eliminate d from [2.39] and [2.40], it follows that:
P
M
f
3dB
2
R
C
=
E
C
2
Q
s
2
256
[2.41]
This expression shows that, for a fixed value of the load resistance,
the product P
M
f
3dB
2
only depends on the properties of the
semiconductor material used for the i region. We should note that the
numerical coefficient 1/256 is a maximal value which may in fact
depend on the detailed specifications of the nonlinearity of the
transmission system considered. Equation [2.41] is analogous to the
known result for electronic components such as transistors [JOH 64].
We can note that this limitation a photodiode can generate on the RF
power is of a purely electronic origin; in addition, thermal effects can
severely limit it, in particular for components designed for low
frequencies.
2.5. Semiconductor materials used in PIN photodiodes for the
visible and near-infrared
2.5.1. Absorption of semiconductors in the range 400-1,800 nm
The absorption functions of the main semiconductor materials or
semiconductor families used to build PIN photodiodes in the spectral
range considered in this chapter are shown in Figure 2.16. They can be
chosen as a function of the wavelengths to be detected. Other
materials have been studied in the past, but have not taken off in
practice for numerous reasons; these will be mentioned briefly later on.
, , , , ,
Figure 2.16. Absorption spectra for semiconductor materials used in the
visible to near-infrared range
2.5.2. From 400 to 900 nm: silicon and the GaAlAs/GaAs family
These two materials have achieved a good technological maturity
under the demands of the integrated circuit market, and as such they
are the natural choice for this spectral domain. However, the shape of
their absorption functions demonstrates their difference in bandgap
nature, which is indirect for silicon and direct for gallium arsenide.
2.5.2.1. Silicon
Silicon has an indirect bandgap and its absorption coefficient
slowly increases below the corresponding wavelength (O
G
=
1,100 nm), reaching the value of 10
3
cm
1
for O = 800 nm. The
interface with its natural oxide, SiO
2
, produces very few electronic
states in the bandgap, which is the main factor enabling the
development of planar and MOS technologies, and ultimately very
high density integrated circuits. On the other hand, this semiconductor
has not, as yet, contributed to the design of heterojunction
optoelectronics with competitive performance.
Figure 2.17 shows the spectral response of a silicon photodiode
designed for optical fiber telecommunications. The structure of the
device is chosen (with a depleted region thickness around 30 m) in
order to increase quantum efficiency to between 600 and 900 nm
where weakly-attenuating fibers and high performance electro-
luminescent emitters are available. It is possible to extend the spectral
response slightly towards the infrared by increasing the thickness of
the intrinsic region, in particular for detection of the O = 1,060 nm
radiation from solid-state neodymium-doped YAG lasers. The drop in
efficiency at short wavelengths, due to optical absorption in the
surface layer, can also be minimized in the near-ultraviolet by several
methods.
The response speed is often limited to less than 1 GHz by the
transit time, in particular for components whose response has been
extended into the infrared. The dark current is mostly caused by
current generation in the depleted region, thanks to the effectiveness
of the surface silicon passivation, which is of the order 10
7
A/cm
2
.
The maturity of silicon technology allows the construction of a
variety of photodiode types, potentially with multiple regions of
sensitivity, and offered at a low price. As a result, they are used for
short-range telecommunications in numerous professional systems
(automation, Lidar, etc.) or mass-market systems (optical disk readers,
photographic equipment, etc.). CCD and MOS imagers also form part
of the family of silicon-based photodetectors.
,
Figure 2.17. Spectral response of photodiodes, made of silicon
for optical telecommunications and of experimental
GaAlAs/GaAs without antireflection treatment
2.5.2.2. Gallium arsenide and similar materials
Gallium arsenide has a direct bandgap; its absorption coefficient is
approximately 10
4
cm
1
for O = 800 nm and increases very rapidly
below the corresponding wavelength O
G
(O
G
= 870 nm). Highly
developed for the construction of microwave integrated circuits and
diode lasers, it is available as a high-quality substrate for creating
defect-free heterojunctions with Ga
1x
Al
x
As, Ga
1x
In
x
As
1y
P
y
and
Al
x
Ga
y
In
1xy
As alloys.
In fact, there has been interest in heterojunctions from the
GaAs/GaAlAs system for photodiode and solar cell fabrication since
the 1970s [ALF 70], and components showing good performance have
been demonstrated [LAW 79]. A spectral response function is given in
Figure 2.17, showing the typical shape for a heterojunction photodiode:
rapid cutoff at long wavelengths, absorption in the direct-bandgap i
material, internal quantum efficiency close to 100%, and cutoff at
short wavelengths due to absorption from the large-bandgap window
layer. The density of the dark current is of the order of 10
8
Acm
2
at
V
B
/2 and a product K
i
f
t
of the order of 20 GHz can be achieved for
conventionally-structured devices. Alongside heterojunctions, the use
of Schottky diodes based on GaAs has also been demonstrated. In
addition, the deposition of ITO (indium tin oxide), a conductive and
transparent material, has allowed the development of fast photodiodes
based on GaAs [PAR 87]. MSM (metal-semiconductor-metal)
photodiodes, discussed in Chapter 5, can also be thought of as
Schottky photodiodes.
Despite these encouraging and already well-known characteristics,
GaAs-based photodiodes have not seen significant development since
2001, except in solar cells. This is probably due to the cheap cost of
silicon components whose speed of operation is adequate for real-
world systems. This situation may change if the need arises for optical
transmissions above 1 Gbit/s within the spectral range covered in this
section.
2.5.3. From 900 to 1,800 nm: germanium, GaInAsP/InP
This spectral range is mostly used for long distance transmissions
and broadband over silica fibers. These systems have driven the
development of photodetectors, initially in germanium which
predated the emergence of optical telecommunications and then in
various families of semiconductor compounds, with the GaInAs alloy
on InP substrate finally prevailing.
2.5.3.1. Germanium
Germanium has an indirect bandgap and its absorption coefficient
increases slowly below O
G
, then more rapidly for O < 1,500 nm where
its conduction band also has a direct minimum for k = 0 (see Figure
2.4). The need for photodetectors for telecommunications was the
drive for an important effort in the 1970s in the development of
germanium photodetectors [KAN 85]. However, several factors
finally brought an end to their application in this domain, along with
the increasing use of GaInAs/InP:
non-existence of materials allowing the construction of
heterojunctions;
limitation of the spectral response around 1,550 nm for fast
photodiodes, due to the shape of the absorption spectrum;
high dark current due to an insufficient control of the surface
preparation.
, , , , ,
Figure 2.18. Spectral response of state-of-the-art photodiodes in 2000: in germanium
for instrumentation and in GaInAs/InP for optical fiber telecommunications
Figure 2.18 shows the spectral response of a germanium
photodiode aimed at scientific instrumentation. The dark current
densities are of the order of 10
4
Acm
2
and the cutoff frequencies are
limited by the circuit to a few hundred MHz, depending on the active
surface.
2.5.3.2. Indium phosphate and related materials
Solid solution crystals from the Ga
x
In
1x
As
y
P
1y
family formed into
a crystal lattice on InP substrates (y # 2.2x) all have an indirect
bandgap, with Ga
0.43
In
0.57
As having the smallest bandgap of the
family (O
G
= 1,650 nm). Its absorption spectrum, as well as that of InP
(O
G
= 920 nm), is shown in Figure 2.16. The composition of the solid
solution can be selected to give a bandgap width anywhere between
those two values.
The technology of these materials was initially driven by the
development of electroluminescent emitters, electroluminescent
diodes and laser diodes, required for long distance transmissions. It
has been of benefit to photodetectors which have established
themselves as rivals to germanium photodetectors.
The main stages in the planar fabrication procedure for PIN
heterojunction photodiodes are shown in Figure 2.19. It is analogous
to that of the devices described above, but we note in particular that
the PIN junction emerges at the surface in the wider-bandgap InP
material, which contributes significantly to the reduction in the dark
current. The transparency of the substrate allows the possibility of the
illumination being in this direction (rear illumination).
Figure 2.19. Fabrication stages of PIN photodiodes in GaInAs/InP by planar
technology: 1) epitaxial deposition of the absorber and window layers,
2) deposition of the dielectric and etching of openings, 3) zinc diffusion
(p-type impurity), and 4) attachment of metallic contacts
Figure 2.18 shows the spectral response of a photodiode targeted
for optical telecommunications. This diode has had an antireflection
treatment applied which allows it to approach an external quantum
efficiency of 100% across a large spectral band. The dark current
density is of the order of 10
-5
Acm
-2
and, as we saw in section 2.4.2.2,
the cutoff frequency of a photodiode with an optimized structure can
surpass 20 GHz while retaining a high quantum efficiency. The
spectral response is limited to around 1,700 nm when the absorbing
layer is exactly matched to the InP substrate. This can be extended
beyond 2,000 nm with the use of unmatched materials, at the expense
of the dark current, for certain applications (imaging, spectroscopy,
etc.) [LIN 92].
2.5.3.3. Other materials [PEA 85]
The integration of the GaAlAsSb family into GaSb substrates was
also considered by several groups, towards the end of the 1970s, for
the fabrication of photodiodes for optical transmissions. GaSb is a
direct semiconductor whose bandgap is close to that of GaInAs
(O
G
= 1,720 nm). By combining it with GaAlAsSb, it was possible to
demonstrate photodiodes with competitive quantum efficiencies,
although with a higher dark current (by a factor of ~10) than
GaInAs/InP photodiodes. On the other hand, a weaker additional
avalanche noise was achieved (see Chapter 3). Nevertheless, the
industrial significance of GaInAsP/InP, also used for
electroluminescent emitters, seems to have prevented investment in
research which might have taken this technology to new levels.
The HgCdTe (MCT, Mercury Cadmium Telluride) integrated into
CdTe substrates is the dominant material for detection and imaging in
the mid-infrared. Attempts were made to re-use the technological
achievements in the infrared for the construction of detectors for
telecommunications. Although the performances obtained were
comparable to those of GaInAs/InP, as in the case of GaInAsSb/GaSb,
the potential gains were insufficient to justify its industrial
development.
2.6. New photodiode structures
2.6.1. Beyond the limits of conventional PIN
The development in the last decades of new epitaxial techniques
for the integration of very thin layers (only a few atoms thick) of
semiconductor compounds has enabled the design of optoelectronic
devices, including photodetectors with far more complex structures
than the simple PIN photodiodes with three useful layers. In this way
it has become possible to exceed the limit of the product K
i
f
t
, which
we examined in section 2.4.1.1 (for the PIN photodiode geometries
discussed), whose order of magnitude is 30 GHz for indirect bandgap
materials. Several avenues were explored with this aim, all based on
making the radiation pass through the narrow i absorption region
multiple times (as shown in Figure 2.20) in order to increase the
internal quantum efficiency while maintaining a short electronic
transit time.
2.6.2. Photodiodes with collinear geometry
A first approach to increase the absorption of radiation by a layer
of thickness Dd << 1 is to cause the light to pass through the absorbing
layer again, after reflection from a mirror at the back (see Figure 2.20),
which results in a doubling of the absorbing thickness. Such a mirror
can be fabricated either by metalizing the rear face of the structure or,
as shown in the figure, by a Bragg semiconductor mirror. These
mirrors made from a stack of layers with an alternating sequence of
bandgaps (and hence refractive indices) of thickness O/2n allow
reflection coefficients approaching 100% to be achieved. They were
first used in the fabrication of vertical cavity surface emitting lasers
(VCSELs) and more generally in microcavity devices.
One technique for increasing the number of crossings of the
absorbing region is to insert the i layer between two mirrors, possibly
Bragg mirrors, to form a Fabry Perot cavity (or interferometer), as
shown in Figure 2.20.
Figure 2.20. Photodiodes of collinear geometry with an
improvement in the K
i
f
t
tradeoff
The analysis of the behavior of such a structure [KIS 91] shows
that when the effective distance between the mirrors is an integer
number of half-wavelengths of light in the material, and the i layer is
placed at an antinode of the electromagnetic field, the absorption can
be considerably enhanced, as can the internal quantum efficiency.
This enhancement is obtained for a series of resonant wavelengths
satisfying the above condition, which gives rise to the name RCE
(Resonant Cavity Enhanced) photodiodes. However, this resonant
behavior has a marked effect on the characteristics of the device:
the spectral response consists of a series of peaks which become
even narrower as the fineness of the cavity is increased, which is in
contrast to the normally broad response of conventional photodiodes
(see Figures 2.17 and 2.18). RCE photodiodes can be thought of as the
monolithic integration of a photodiode and an optical filter whose
selectivity can be exploited beneficially, in particular if it can be made
tunable;
for a given wavelength, the resonance condition is satisfied only
for certain angles of incidence, such that the directivity is much higher
than that of conventional devices.
Analysis of the response speed [UNL 95] shows that a product K
i
f
t
approaching 100 GHz should be obtainable by RCE photodiodes. In
addition, other advantages of the use of very thin detector layers are
anticipated including:
the reduction of the contribution of pair generation to the dark
current (equation [2.20]);
the possibility of using partially incommensurate i materials
without loss of crystal quality if its thickness is below a critical value.
The experimental results obtained up to now for most of the
families of III-V materials have indeed confirmed the potential of
RCE photodiodes in terms of quantum efficiency and selectivity, but it
turns out to be difficult to show improvement in the K
i
f
t
product.
2.6.3. Waveguide photodiodes
Another approach consists of using an orthogonal geometry, giving
the device the shape of an optical waveguide, which is analogous to
that of a diode laser. This is the case in Figure 2.21, where we see an
absorbing i layer inserted between several other layers which have a
wider bandgap, are transparent and also have a lower refractive index.
Figure 2.21. Structure, width of the bandgap and doping of the layers, of a waveguide
photodiode with orthogonal geometry and with an improved K
i
f
t
tradeoff
In Figure 2.21, a geometric optics representation, the light rays
propagate in the structure through a series of total internal reflections,
and cross the i layer several times. In reality, since the dimensions of the
waveguide are of the order of the wavelength of light in the material, a
more rigorous analysis of the electromagnetic field shows that only
certain configurations (or modes) are supported. The radiation is
progressively absorbed with an effective coefficient of absorption
depending on the overlap between the guided mode and the absorbing
layer; this can be adjusted by the design of the waveguide, but is
difficult to reduce below 100 cm
1
, leading to an internal quantum
efficiency of 95% for a waveguide of length 300 m.
Numerous variations on this principle have been investigated, such
as the active part of the photodiode being localized on only part of the
waveguide. Figure 2.21 shows a structure with 5 active layers, allowing
independent adjustment of the transit time (by varying the thickness of
the i depleted region) and the waveguide thickness (which determines
the degree of coupling with the incident light and consequently the
external quantum efficiency [KAT 92]):
InP N layer (O
G
= 920 nm);
GaInAsP N layer (O
G
= 1,300 nm);
GaInAs i layer (O
G
= 1,650 nm);
GaInAsP P layer (O
G
= 1,300 nm);
InP P layer (O
G
= 920 nm).
In this way it has been possible to demonstrate a significant gain in
the merit factor K
i
f
t
# 30 for GaInAs. However, it should be noted that
the gain in internal quantum efficiency obtained, can in practice be
partially undone by the difficulty of coupling the radiation into the
mode(s) supported by the waveguide. Although it is in theory possible
to couple the mode from a single-mode fiber into a photodiode of
single mode with an efficiency of 100%, this can require coupling
optics of prohibitive cost.
2.6.4. Traveling-wave photodiodes
Waveguide photodiodes have the same limitation as conventional
structures, in terms of the power/frequency tradeoff discussed in
section 2.4.3. This limitation stems from the lumped-element circuit
analysis, which is a good approximation as long as the length of the
guide remains smaller than the wavelength of the electrical signal
(f = 100 GHz, O = 3 mm). Beyond this, it is necessary to use a
distributed element model and to think of a photodiode as a
waveguide coupled to a transmission line (see Figure 2.22),
terminated at both ends in its characteristic impedance in order to
maximize the bandwidth.
Such devices have been proposed and analyzed under the term
traveling-wave photodiodes [GIB 92]. It follows that, aside from the
transit time, the circuit bandwidth is limited by the mismatch between
the propagation speeds of the two guided waves, optical and
microwave a limit which is not fundamental in nature, but which is
difficult to overcome in practice. The analysis of [GIB 92] predicts
values for the product K
i
f
t
of greater than 100 GHz; however,
fabrication difficulties appear, until now, to have seriously limited the
practical performances relative to the potential performance: the
challenge still remains to conceive and then construct structures
allowing simultaneous improvement in K
i
f
t
and Pf
2
(the figure of
merit).
Figure 2.22. Traveling-wave photodiode
2.6.5. Beyond PIN structures
Whatever their geometry, the photodiodes considered up to this
point have had, by definition, three active regions, with the i region
fulfilling both the role of radiation absorption and also the role of the
depleted region. It is possible to imagine photodiodes where these
functions are separated so that they can be optimized separately,
possibly with the additional use of waveguide photodiode or resonant
cavity concepts.
2.6.5.1. P
+
p
+
N
N
+
structures
The absorbing region, here p+, can be doped without reducing the
internal quantum efficiency and the speed, on the condition that it is
thin enough to ensure both efficient and rapid transport of
photocarriers by diffusion (see sections 2.3.2.2 and 2.4.1.2). An initial
example is that of electroluminescent diodes whose optimization leads
to a P
+
p
+
N
N
+
double heterojunction structure, where the emitting
layer is designated p+. Under reverse-bias the depleted region grows
into the N- region and the photodetection, quantum efficiency and
speed characteristics are sufficient for some applications
K
e
=0.6, f
RC
=300 MHz
( )
[POU 80]. Very fast response speeds have
more recently been discovered to be at the expense of low quantum
efficiency K
e
=0.06, f
3dB
=300 GHz
( )
in a conventional geometry,
making more use of the high mobility regime of electron transport
[ITO 00].
2.6.5.2. P
+
IiIN
+
structures
The placement of the absorbing i region in a thicker I depleted
region, in order to equalize the transit time of electrons generated in
the i layer towards the N+ layer and of holes towards the P+ layer,
allows us in principle to gain a factor of 2 on the cutoff frequency f
t
[UNL 95].
2.7. Bibliography
[ALF 70] ALFEROV Z.I., ANDREEV V.M., ZIMOGOROVA N.S., TRETYAKOV D.N.,
Photoelectric properties of AlxGa1-xAs heterojunctions, Sov. Phys.-Semicond.,
vol. 3, no. 11, p. 1373-1376, May 1970.
[BOW 87] BOWERS J.E., BURRUS C.A., Ultrawide-band long-wavelength p-i-n
photodetectors, J. Lightwave Technology, vol. LT-5, no. 10, p. 1339-1350,
October 1987.
[FOR 82] FORREST S.R., KIM O.K., SMITH R.G., Optical response time of
In
0,53
Ga
0,47
7As/InP avalanche photodiodes, Appl. Phys. Lett., vol. 41, no. 1,
p. 95-98, July 1982.
[GIB 92] GIBONEY K.S., RODWELL M.J.W., BOWERS J.E., Travelling-wave
photodetectors, IEEE Photonics Technology Lett., vol. 4, no. 12, p. 1363-1365,
December 1992.
[ITO 00] ITO H., FURUTA T., KODAMA S., ISHIBASHI T., InP/InGaAs uni-travelling-
carrier photodiode with 310 GHz bandwidth, Electron. Lett., vol. 36, no. 21,
p. 1809-1810, 12 October 2000.
[JOH 64] JOHNSON E.O., Physical limitations on frequency and power parameters of
transistors, IEEE Int. Conf. Rec. pt.5, p. 27-34, 1965.
[KAN 85] KANEDA T., Semiconductors and semimetals, Lightwave Communication
Technology, vol. 22, part D, chap. 3, p. 247-328, Academic Press, 1985.
[KAT 92] KATO K., HATA S., KAWANO K, YOSHIDA J., KOZEN A., A high-efficiency
50 GHz InGaAs multimode waveguide photodetector, IEEE J. Quantum
Electronics, vol. 28, no. 12, p. 2728-2735, December 1992.
[KIS 91] KISHINO K., UNLU M. S., CHYI J-I., REED J., ARSENAULT L., MORKOC H.,
Resonant cavity enhanced (RCE) photodetectors, IEEE J. Quantum
Electronics, vol. 27, no. 8, p. 2025-2034, August 1991.
[LAW 79] LAW H.D., NAKANO K., TOMASETTA L.R., State-of-the-art performance of
GaAlAs/GaAs avalanche photodiodes, Appl. Phys. Lett., vol. 35, no. 2, p.
180-182, July 1979.
[LIN 92] LINGA K.R., OLSEN G.H., BAN V.S., JOSHI A.M., KOSONOCKI W.F., Dark
current analysis and characterization of InxGa1-x As/InAsyP1-y graded
photodiodes with x > 0.53 for response to longer wavelength (> 1.7 m), J.
Lightwave Technology, vol. 10, no. 8, p. 1050-1055, August 1992.
[MAT 87] MATHIEU H., Physique des semi-conducteurs et des composants
lectroniques, Masson, 1987.
[PAR 87] PARKER D. G., SAY P. G., HANSOM A.M., SIBBETT W., 110 GHz high-
efficiency photodiodes fabricated from indium tin oxide/GaAs, Electron. Lett.,
vol. 23, p. 527-528, 1987.
[PEA 85] PEARSALL T.P., POLLACK M.A., Semiconductors and semimetal,
Lightwave Communication Technology, vol. 22, Part D, Chapter 2, p. 173-247,
Academic Press, 1985.
[POU 80] POULAIN P., DE CREMOUX B., Diffusion limited transient response of
heterojunction photodiodes, Japanese J. Appl. Phys., vol. 19, no. 4, p. L189-
L192, April 1980.
[STI 82] STILLMAN G. E., COOK L.W., BULMAN G.E., TABATABAIE N., CHIN R.,
DAPKUS P.D., Long-wavelength (1.3 to 1.6 m) detectors for fiber-optical
communications, IEEE Trans. on Electron Devices, vol. ED-29, no. 9, p. 1355-
1371, September 1982.
[SZE 81] SZE S.M., Physics of Semiconductor Devices, 2
nd
edition, John Wiley &
Sons, 1981.
[TAK 80] TAKANASHI Y., KAWASHIMA M., HORIKOSHI Y., Required donor
concentration of epitaxial layers for efficient InGaAsP avalanche photodiodes,
Japanese J. Appl. Phys., vol. 19, no. 4, p. 693-701, April 1980.
[UNL 95] UNL S., ONAT B.M., LEBLECI Y., Transient simulation of heterojunction
photodiodes Part II: analysis of resonant cavity enhanced photodetectors, J.
Lightwave Technology, vol. 13, no. 3, p. 406-415, March 1995.
Chapter 3
Avalanche Photodiodes
3.1. Introduction
Photodetectors are one of two contrasting families of devices that
dominate the field of optoelectronics. Among these, avalanche
photodiodes (APDs) are very attractive devices because they benefit
from an internal gain due to the multiplication of the charge carriers
generated by absorption of incident light. This occurs when these
charge carriers cross a region of high electric field (>10
5
V/cm), thus
acquiring enough energy to ionize the atoms of the crystal lattice,
thereby creating new electron-hole pairs which are immediately
separated, and can themselves create other electron-hole pairs: this
leads step-by-step to an amplification of the primary photocurrent.
This mechanism, known by the term shock ionization or impact
ionization, is also at the root of the reversible breakdown of reverse-
biased p-n junctions.
In this chapter we will discuss the following subjects:
theoretical consideration of the avalanche effect (ionization
coefficients, multiplication factor, etc.);
technical conditions necessary for proper operation of APDs;
Chapter written by Grard RIPOCHE and Joseph HARARI.
APD structure;
photoelectric properties of APDs;
state of the art for APDs adapted to different ranges of
wavelength, based on Si, Ge, InGaAs, etc.;
new low-noise APD structures based on superlattices or
heterogenous structures (Si/InGaAs).
3.2. History
The discovery of solid-state, gas and semiconductor lasers between
1958 and 1962, and the creation at the same time of the first
electroluminescent diodes, led to the birth of optoelectronics and to
the subsequent boom that it has seen, particularly in the field of
optical communications. however to take advantage of the
considerable potential of these new light sources, mostly
semiconductor lasers and light emitting diodes (with direct modulation
at high modulation frequencies, and short response times), there is a
need for fast and sensitive detectors operating at the emitted
wavelengths. Among the possible candidates, reverse-biased p-n
junction photodiodes offer numerous advantages: low bias voltages,
small footprint, robustness, ease of use and reliability. These make
them very attractive to users compared to photomultipliers which,
although benefiting from a high sensitivity thanks to a noise-free
internal gain process, suffer from two serious drawbacks for many
applications: their size and high operating voltage.
However, the electrical signal generated by the absorption of
incident light by a photodiode is, in most cases, too weak to be
directly usable and must be amplified. In boosting the signal to a
sufficient level, the amplifier stage limits the performance of the
receiver in two ways: by reducing the bandwidth of the system and by
lowering the global sensitivity due to the additional noise that it
introduces.
Avalanche Photodiodes 59
In order to eliminate these problems and to address the need for
amplification of the electric signal generated by the light incident on
the photodiode, in 1964 Johnson [JOH 64] highlighted the advantages
of primary photocurrent multiplication through the avalanche effect;
which was reported by McKay and McAfee [KAY 53] in 1953 during
their study of reversible breakdown in Si p-n junctions. This internal
gain mechanism gives enough improvement to the signal-to-noise
ratio to make the use of avalanche photodiodes (APDs) very
attractive for many applications, in particular for optical
telecommunications, despite the additional noise due to the random
nature of the photomultiplication mechanism.
Studies carried out in the 1960s focused on the development of
silicon avalanche photodiodes (Si APDs) for 1.06 m telemetry and
for optical communications around 0.85 m, and also the development
of germanium avalanche photodiodes (Ge APDs) whose photosensitivity
extends to 1.65 m. The 1970s saw the optimization of Si APDs for
fiber-optic telecommunications at 0.85 m through the use of low-
noise designs, as well as the exploratory study of gallium arsenide
(GaAs) APDs and those based on GaAs compounds. Since the future
of fiber-optic telecommunications is very much based in the near-
infrared due to the weak attenuation of optical fibers around 1.3 m
(~2 dB/km) and at 1.55 m (~0.2 dB/km), several laboratories worked
on improving Ge APDs. As the performance of these was limited,
even at 1.3 m, the first investigations were carried out to evaluate the
possibilities offered by detectors based on III-V compounds, starting
with indium phosphate (InP). The development of working
InGaAs/InP APDs required important progress throughout the 1980s,
both in the epitaxial growth of heterostructures and in their design and
technology, which lead to the creation of devices tailored to the needs
of fiber-optic submarine links and in particular to those links that can
operate at bitrates of over 10 Gbit/s. From then on, investigations
focused on low-noise devices for high bitrates, such as superlattice
InGaAlAs/InAlAs APDs or heterogenous Si/InGaAs APDs.
3.3. The avalanche effect
The study of reversible p-n junction breakdown by McKay and
McAfee in silicon, and by Miller [MIL 55] in germanium, showed that
the growth of the dark current, observed when the bias voltage is
increased towards the breakdown voltage V
B
, which is a characteristic
of a given device, is caused by a mechanism commonly known as the
avalanche effect, similar to that observed in gas discharges.
The avalanche effect is shown schematically in Figure 3.1. A
primary charge carrier, electron or hole, which enters the space
charge region (SCR) where there is a fairly strong electric field E
(>10
5
V/cm), is strongly accelerated and can acquire a mean kinetic
energy greater than the characteristic ionization energy E
i
of the
semiconductor, before colliding with an atom of the crystal lattice,
creating an electron-hole pair through ionization and promoting an
electron from the valence band to the conduction band. The secondary
charge carriers (electron and hole) that are generated, as well as the
primary charge carrier, can then independently create new electron-
hole pairs, as long as the kinetic energy accumulated by each of the
charge carriers between successive collisions is greater than the
ionization energy E
i
. This process is cumulative and can under certain
conditions lead to the breakdown of the junction.
Figure 3.1. Diagram showing the principles of the shock ionization mechanism
3.3.1. Ionization coefficients
The ionization energy E
i
is, in the simplest case, that of a model
with two parabolic energy levels, greater than or equal to 3/2.E
g
,
where E
g
is the energy corresponding to the width of the bandgap. The
ionization energy E
i
depends on the type of charge carriers for a given
material. If it can be reduced by some means, this can lead to the
preferential ionization of one type of charge carrier, as we will see
later.
The ionization coefficients of electrons D
i
and of holes E
i
are
defined as the average number of ionizing collisions produced per unit
length (cm) by a charge carrier moving parallel to the constant electric
field E, and they quantify the ionizing power of the charge carriers.
Depending primarily on the electric field, they grow very rapidly with
the field. Their variations, determined experimentally, follow
empirical relationships of the form:
D
i
, E
i
= a
i
exp
b
i
E
(
(
)
)
[3.1]
where a
i
and b
i
are parameters that depend on the material and on the
type of charge carrier. Figures 3.2a and 3.2b show typical D
i
,
E
i
= f (1/E) curves, for silicon and germanium on the left and for InP
on the right, which are the base materials for APDs developed for
optical communications. Silicon is unusual in two ways: the electrons
are the most easily ionized carriers and the ratio of the ionization
coefficients k = E
i
/D
i
is very high for relatively weak electric fields
(E ~ 3 x 10
5
Vcm
1
). The nature of the carriers triggering the
multiplication and the ratio k play an important role in the design of
devices, and directly influence the photoelectric performance of APDs.
It is worth noting that the variation seen in the experimental curves
published in the literature is a good illustration of the difficulties
inherent in determining the coefficients D
i
and E
i
from measurements
of photomultiplication or noise.
Figure 3.2. Ionization coefficients in silicon, germanium and
InP as a function of the inverse of the electric field 1/E
3.3.2. Multiplication factors
The impact ionization mechanism leads to a multiplication in the
number of charge carriers injected into the SCR, with this taking place
in the avalanche region where there is a high electric field E. For each
type of charge carrier, the multiplication factor M
n
or M
p
is defined as
the ratio of the number of charge carriers extracted to the number of
charge carriers injected. The following approximations are normally
used in the calculation:
the ionization coefficients D
i
and E
i
only depend on the electric
field E, which itself is only a function of the x coordinate for a given
junction;
recombination of carriers is negligible in the depleted region;
mobile charge carriers do not disturb the electric field distribution;
the levels of thermal and optical carrier generation in the depleted
region are not included.
Calculation of the multiplication factors M
n
and M
p
is carried out
by solving the continuity equations for the electron current J
n
and the
hole current J
p
under a strong electric field. The system of equations
describing steady-state behavior is as follows:
dJ
n
x
( )
dx
=
dJ
p
x
( )
dx
=D
i
x
( )
J
n
x
( )
E
i
x
( )
J
p
x
( )
[3.2]
J = J
n
x
( )
J
p
x
( )
=C
ste
[3.3]
which can be rewritten as:
dJ
n
x
( )
dx
D
i
x
( )
E
i
x
( )
l
l
l
J
n
x
( )
= E
i
x
( )
J [3.4]
Solving this differential equation for the case of multiplication
triggered by a pure injection of electrons from the P+ region, by
integrating J
n
(x) over the avalanche region (between x = 0 and x = w),
leads to the expression:
1
1
M
n
= D
i
0
w
x
( )
exp D
i
x
( )
E
i
x
( ) ( )
0
x
dx '
l
l
l
l
l
l
.dx [3.5]
with M
n
J
n0
, where J
n0
is the current of injected electrons at x = 0.
In the case of an injection of holes from the N
+
region, the
multiplication factor M
p
is given by a similar expression:
1
1
M
p
= E
i
0
w
x
( )
exp D
i
x
( )
E
i
x
( ) ( )
x
w
dx '
l
l
l
l
l
l
.dx [3.6]
with M
n
J
pw
, J
pw
being the current of injected holes at x = w.
In reality, carriers of both types are normally injected into the
avalanche region, leading to a multiplication factor somewhere
between the two which takes into account the weighting of each
contribution. It also follows that the experimental determination of the
ionization coefficients D
i
and E
i
from measurements of multiplication
factors is very sensitive, since in addition to a precise knowledge of
the electric field distribution E(x), it requires the separation of the
effects of both types of charge carrier. In the case where the ionization
coefficients are equal (D
i
= E
i
), the two expressions take the form:
1
1
M
= D
i
0
w
x
( )
dx [3.7]
In the case of a PIN junction with a constant electric field in the
depleted region, for a given bias voltage V, the above equation
becomes:
1
1
M
=D
i
w [3.8]
where D
i
(E) is a function of the bias voltage V as w is constant, hence:
E =
V
w
[3.9]
3.3.3. Breakdown voltage
As the bias voltage increases, the electric field increases and
approaches the critical value for which the dark current grows
extremely fast: this is the volume breakdown of the junction, which
occurs for a bias voltage V
B
the breakdown voltage. These
conditions occur when the multiplications factors M
n
, M
n
of which
occurs for:
1= D
i
x
( )
exp D
i
x
( )
E x
( ) ( )
dx '
0
x
l
l
l
l
l
l
0
w
dx [3.10]
and:
1= E
i
x
( )
exp D
i
x
( )
E x
( ) ( )
dx '
x
w
l
l
l
l
l
l
0
w
dx [3.11]
respectively.
In addition, Sze and Gibbons [SZE 66], studying the behavior of
P
+
N and N
+
P junctions in the avalanche regime, established a simple
semi-empirical relationship between the breakdown voltage V
B
and
the concentration N in the weakly-doped region of an abrupt junction,
or the concentration gradient a in a gradual junction. These relations
can be applied to different materials. The experimental results are
well-matched by the following expressions:
V
B
=60
E
G
1.1
(
(
)
)
3
2
N
10
16
(
(
)
)
3
4
[3.12]
for an abrupt junction and:
V
A
=60
E
G
1.1
(
(
)
)
6
5
a
3.10
20
(
(
)
)
2
5
[3.13]
for a gradual junction, where E
g
is the width of the bandgap of the
material in question, expressed in eV.
Finally, we point out the empirical relationship established by
Miller [MIL 55] between the multiplication factor M and the
breakdown voltage V
B
:
1
1
M
=
V
V
B
(
(
)
)
n
[3.14]
where n is an exponent which depends on the material, the type of
junction and the nature of the charge carriers triggering the
multiplication. This expression satisfactorily describes the variation of
the multiplication factor M with the bias voltage V, and it is often
applied to classical P
+
N and N
+
P junctions.
3.4. Properties of avalanche photodiodes
The behavior of avalanche photodiodes differs from that of
standard PIN photodiodes only in terms of the properties that are
involved in the avalanche multiplication mechanism: the gain, the
noise linked to the photocurrent, and the frequency response, these
being at operating voltages in the pre-avalanche regime of the junction,
in contrast to the low voltages at which PIN photodiodes are normally
used.
3.4.1. Current-voltage characteristics and photomultiplication
For a PIN junction, the photocurrent I
ph
generated by the light
incident on the active region barely varies with the operating voltage.
In contrast, for APDs, above a certain bias voltage the photocurrent I
ph
grows steadily, reaching its maximum value when the bias voltage is
equal to the breakdown voltage V
B
of the junction. The gain M is
given by the equation:
M =
I
ph
V
( )
I
ph
V
0
( )
=
I V
( )
I
d
V
( )
I V
0
( )
I
d
V
0
( )
[3.15]
where I
ph
, I and I
d
are respectively the photocurrent, the total current
flowing across the junction, and the dark current; V and V
0
are the
operating voltage and the reference voltage for which M= 1. In the
following, we will refer to I
ph0
and I
d0
as the primary photocurrent and
the primary dark current (i.e. the non-multiplied currents). These are
denoted as I
ph
(V
0
) and I
d
(V
0
) in equation [3.15]. This equation allows
us to plot the M(V) characteristics of an APD measured under well-
defined experimental conditions. These may have a decreasing or
increasingly strong influence on the measured values. In particular,
although an APD may behave fairly linearly for relatively low values
of gain (M< 10), for which M depends little on the incident light
power, this contrasts with significant saturation effects that are seen at
high gains. The maximum gain obtained (M
max
) is very sensitive to
the conditions of the measurement. A reduction in M
max
can be
partially explained by a reduction in the avalanche electric field
caused by an opposing electric field created by the mobile charge
carriers, which are present in significant numbers in the depleted
region. Also, close to breakdown, the applied bias voltage is reduced
because of the voltage drop across the equivalent resistance R of the
diode.
Taking these considerations into account, equation [3.14], which
applies for weak values of the current I, can for high currents be
written:
1
1
M
=
V I
d0
I
ph0
( )
M R
V
B
l
l
l
l
l
l
n
[3.16]
for V= V
B
, M= M
max
and this can be simplified to:
M
max
~
V
B
n I
d0
I
ph0
( )
R
[3.17]
This expression shows the value of devices with a weak dark
current (I
d0
<< I
ph0
) in order to promote photomultiplication. In these
conditions, the maximum gain is inversely proportional to the square
root of the primary photocurrent I
ph0
, and hence the power of the
incident light P
0
:
M
max

1
I
ph0
or M
max

1
P
0
[3.18]
Taking into account equation [3.12], which expresses the breakdown
voltage V
B
as a function of the concentration N, we find:
M
max
N
3
8
[3.19]
Weak concentrations N in the depleted region, corresponding to
relatively high breakdown voltages V
B
, would be of benefit in
obtaining high gains M. However, for convenience of use, the diode
breakdown voltage is limited, where possible, to the range 50150 V.
Finally, for devices with a strong dark current (I
d0
>> I
ph0
), the
limitation on the gain is imposed by the dark current and in these
conditions, which can occur with Ge APDs, we have:
M
max
~
V
B
n I
d0
R
[3.20]
3.4.2. Noise in avalanche photodiodes
Noise in detectors is discussed in Chapter 7 and we will limit
ourselves here to revisiting the main results for noise associated with
the operation of APDs in the avalanche regime. In the absence of
multiplication, the APD behaves like a traditional photodiode whose
noise is essentially shot noise caused by the dark current I
d0
and the
primary photocurrent I
ph0
across the junction, and whose spectral
density is:
i
2
=2 q I
d0
I
ph0
( )
[3.21]
When the APD operates in the avalanche system, the currents
across the junction undergo a multiplication M, and hence the spectral
density of the noise is multiplied by M
2
. The random nature of the
impact ionization process results in a fluctuation in the gains
associated with the different carriers, contributing to the appearance of
extra noise linked to the multiplication. The spectral density of the
shot noise of the multiplied current can be written:
i
2
=2 q I
d0
I
ph0
( )
M
2
F M
( )
[3.22]
where F(M), the excess noise factor, represents the contribution of
multiplication to the noise. The value of F(M) depends on the
characteristics of the p-n junction and on the nature of the primary
carriers. F(M) expressions for the two types of primary injection,
electrons and holes, have been derived by McIntyre [INT 66] under
the assumption of an avalanche region with a constant electric field.
These are expressed as a function of M and the ratio of the ionization
coefficients k = Ei/Di, which is considered constant. Instances of a
pure injection of electrons or of holes are of particular interest. The
excess noise factor for a primary injection of electrons can be
expressed by the following:
F
n
M
( )
= M
n
1 1k
( )
M
n
1
M
n
(
(
)
)
l
l
l
l
l
l
[3.23]
and for a primary injection of holes:
F
p
M
( )
= M
p
1 1
1
k
(
(
)
)
M
p
1
M
p
(
(
)
)
l
l
l
l
l
l
[3.24]
Where M
n,p
>> 1, these equations can be expressed in the
following simplified forms:
F
n
M
( )
= kM
n
2 1k
( )
[3.25]
F
p
M
( )
=
1
k
M
p
2 1
1
k
(
(
)
)
[3.26]
and for very high values of multiplication:
F
n
M
( )
kM
n
and F
p
M
( )
1
k
M
p
[3.27]
In the specific case of ionization by only one type of carrier,
electrons or holes, the excess noise factor takes a minimum value of
F(M)=2, independent of M. Attempts can be made to approach this
value through modification of the ionization energy or by careful
design of the device. The typical variation of the noise factor as a
function of M for different values of k is shown in Figure 3.3.
Figure 3.3. Excess noise factor as a function of
different values of k for a pure injection of holes
In order to obtain low excess noise values, two conditions must be
met: the multiplication must be triggered by the more ionizing carrier,
and a material must be used whose ionization coefficients are very
different. Among the normal semiconductors (Si, Ge, GaAs, InP, etc.),
only silicon satisfies these two requirements. Furthermore,
consideration of the excess noise factor allows us to approach high
values of the ratio k = E
i
D
i
, with this parameter playing an important
role in the optimization of APDs. Finally, in the literature, the excess
noise factor F(M) is often written in the form:
F M
( )
= M
x
[3.28]
which comes directly from the spectral density of the noise, expressed
as:
i
2
=2 q I
ph0
M
2x
[3.29]
The exponent x, which depends on the value of the k ratio, allows
us to compare the different devices. This can often be found in device
specifications alongside the value of F(M) for a given M. For k = 1, the
excess noise factor is F(M) =M for the two types of injection, and the
exponent x therefore takes the value x = 1. For low values of k, the
multiplication is triggered by electrons, whereas for high values of k,
the injection of holes is favored.
3.4.3. Signal-to-noise ratio in avalanche photodiodes
In wideband applications, the limiting noise of traditional
photodiodes comes from the thermal noise associated with the load
resistance R
L
or, more commonly, from the noise of the first stage of
the amplification circuit required to make use of the electrical signal
produced by the photodiode. For some application types, the internal
gain of APDs allows us to improve the signal-to-noise ratio (SNR) of
the detector. This detector, consisting of the APD and amplifier,
includes two sources of noise outside the diode, independent of the
multiplication (the load resistance and the amplifying stage), given by
the equation:
i
ext
2
=
4k
B
T
e
R
L
'f [3.30]
where T
e
is the equivalent temperature of the noise, and also includes
sources of shot noise connected with the multiplication process:
i
M
2
= 2q I
d0
I
ph0
( )
M
2
F M
( )
'f [3.31]
where 'f is the bandwidth of the system.
The mean squared value of the photocurrent is given by:
i
ph
2
=
1
2
m
2
I
ph0
2
M
2
[3.32]
where m is the modulation index of the incident light signal. The SNR
for a system with bandwidth 'f therefore follows the relationship:
S
B
=
1
2
m
2
I
ph0
2
M
2
( )
2q I
d0
I
ph0
( )
M
2
F M
( )
4k
B
T
e
M
2
R
L
l
l
l
l
l
'f
[3.33]
which can be rearranged into the following form:
S
B
=
1
2
m
2
I
ph0
2
( )
2q I
d0
I
ph0
( )
F M
( )
4k
B
T
e
M
2
R
L
l
l
l
l
l
'f
[3.34]
demonstrating the advantage of APDs over PIN photodiodes, since for
low values of F(M) the thermal noise contribution is divided by M
2
.
Figure 3.4. Variation of the SNR as a function of the gain M
for two semiconductors, silicon and germanium
As the gain increases, the SNR grows and reaches a maximum for
an optimal gain value M
opt
, which depends closely on the point at
which the multiplied shot noise and the thermal noise are equal, such
that, in the case where I
d
0 << I
ph
0:
M
opt
~
2k
B
T
e
qI
ph0
R
L
(
(
)
)
1
2x
[3.35]
For higher multiplication values, the SNR drops, with the
contribution of shot noise becoming the dominant factor. The
variation of the SNR with the gain is shown in Figure 3.4. The optimal
gain value M
opt
depends on the incident radiation power P
0
through
the equation:
I
ph0
=
qK O
h c
P
0
[3.36]
As a result, M
opt
will increase as P
0
decreases. Thus, APDs allow
us to improve the sensitivity of photodetectors as long as the shot
noise remains lower than the thermal noise entering the amplifier
stage.
3.4.4. Speed, response time and frequency response of avalanche
photodiodes
The ability of avalanche photodiodes to detect and amplify optical
impulses with short rise and fall times, or light signals modulated at
high frequency, is a topic of research for numerous applications. The
response time of an APD operating in the avalanche regime is limited
by:
the transit time t
tr
of carriers in the SCR;
the diffusion time t
Dif
, the time taken by the electron-hole pairs
created outside the depleted region to reach the p-n junction;
the RC time constant of the APD/load system;
the avalanche build-up time t
av
, the time needed to reach a stable
multiplication value for a given bias voltage.
The transit time t
tr
, the diffusion time t
Dif
and the RC time constant
all depend on the width w of the bandgap. The transit time t
tr
can be
expressed by the equation:
t
tr
=
w
Q
sat
[3.37]
where v
sat
is the saturation velocity for the slowest charge carriers.
In order to increase the speed of APDs for high-frequency
applications or broadband applications, we will seek to reduce the
thickness w. This reduction will in turn produce an increase in the
capacitance C
j
of the junction and a reduction in the response
coefficient, because the light will be absorbed in a thinner width of
material. This results in a compromise between the quantum
efficiency and the speed which must be treated on a case by case basis,
in a similar way to the tradeoff in PIN photodiodes (see section 2.4).
The diffusion time t
Dif
of electron-hole pairs towards the junction
is dominated by the lifetime of the minority charge carriers in the
lightly-doped region of the junction. Its value (between a few s and a
few hundred s depending on the material) makes some devices un-
usable for many applications. This limitation appears for certain Si
and Ge APDs. In practice, as far as possible, the thickness w of the
depleted region is adjusted so that all the light is absorbed there. This
result is approximately achieved for:
w=
2
D
O
[3.38]
where D
O
is the absorption coefficient for light of a given wavelength
O in a semiconductor.
The RC time constant is only significant at high frequencies. In this
case R~50 :, the value of the load resistance. The capacitance of the
junction Cj is related to the area A of the junction and the width w of
the depleted region by:
C
j
=
H A
w
[3.39]
where H is the permittivity of the semiconductor. As a result, we need
to minimize the area of the active region and that of the guard ring,
and select a value of w compatible with an acceptable transit time t
tr
.
This optimization is equivalent to that for PIN photodiodes (see
section 2.4).
As the multiplication process is not instantaneous, the response
time of APDs can be greater than that of PIN photodiodes of
comparable structure, and is very sensitive to the ionization conditions.
Hence, in the case of ionization by only one type of carrier, which is
similar to that of silicon, the impulse response time is controlled by
the transit time across the depleted region of the non-ionizing
secondary carriers created by the primary carriers. Under these
conditions, the bandwidth of the APD, independent of the
multiplication M, is not very different from that of a PIN photodiode.
When the two types of carriers contribute to the multiplication, the
steady-state value of M will be reached after a time which will be
longer if M is higher and if the ionization coefficients of the carriers D
i
and E
i
are close to each other. This is because the high number of
carrier-lattice interactions means the carriers bounce back and forth,
prolonging the time that secondary carriers are present in the
avalanche region. The avalanche build-up time t
av
is therefore
expressed by the equation:
t
aQ
= MW [3.40]
where W is the effective transit time of a carrier in the avalanche region
between two ionizing collisions. This depends very strongly on the
value of k =E
i
/D
i
and can become very small if D
i
and E
i
are very
different. Emmons [EMM 67] analyzed the frequency response of PIN
photodiodes operating in the avalanche system and established the
following expression:
W = N
b
E
i
D
i
(
(
)
)
W
1
[3.41]
where N
b
is a number which varies slowly between 1/3 and 2 as the
ratio k =E
i
/D
i
varies from 10
3
to 1, and W
1
is the actual transit time in
the avalanche region. A low value of k minimizes the effective transit
time W and is also favorable for the response time and for the excess
noise factor. Emmons [EMM 67] showed that, for M>D
i
/E
i
, the gain-
bandwidth product is constant. The gain M varies with frequency
according to the following law:
M Z
( )
=
M
0
1 M
0
2
Z
2
W
2
l
l
l
1
2
[3.42]
where M(Z) and M
0
are the values of the gain as a function of the
frequency f = Z/(2S) and for a constant signal. The gain-bandwidth
product is given by:
M Z
( )
'f = N
b
k W
1
( )
1
[3.43]
and depends on the material through k and the characteristics of the
avalanche region through W
1
= l
a
/v
sat
, where l
a
is the length of the
avalanche region. It follows that the product will be much larger if the
ratio k of the coefficients of ionization is weak and if the avalanche
region is narrow. A high value of the gain-bandwidth product is
desirable as it allows a higher value of gain for a given frequency.
3.5. Technological considerations
APDs work best when they are biased with a voltage close to the
volume breakdown voltage in the active region. As a result there is a
very intense electric field (~4-5 x 10
5
V/cm) present in the avalanche
region, which must be as homogenous as possible throughout the
active region. Any inhomogenity present in the SCR that can cause a
local increase in the electric field will lead to a premature breakdown
which will degrade the APD performance. These inhomogenities can
be caused by the structure itself, as in a traditional planar junction
obtained by localized diffusion in a circular geometry, or linked to
imperfections in the crystal structure. Regardless of their origin, they
always need to be eliminated.
In the first case, the use of junctions with a guard ring or the use of
a mesa structure allows the suppression of peripheral breakdown, as
long as a geometry is used which is free from angular parts liable to
cause point effects. The suppression of crystal defects depends closely
on the progress of technology, particularly in the fields of
crystallogenesis and techniques of epitaxial growth and doping.
Figure 3.5. Cross-section of a planar photodiode
3.5.1. Guard ring junctions
For a junction constructed using planar technology (Figure 3.5),
the electrical field inside the depleted region is higher in the curved
region around the edge of the diffused area than in the active region,
as can be seen by solving Poissons equation [HAR 91a]. Breakdown
due to edge effects therefore occurs for a voltage sufficiently below
the breakdown voltage of the active region so that light falling on the
central region is not amplified or barely amplified. On the other hand,
if the illuminated region extends beyond the diffused area, the
increase in the injected global photocurrent due to the high electric
field present in a very narrow annular region around the edge can give
the illusion of a photomultiplication effect. Regardless, such a
photodiode cannot operate as an APD. In order to suppress peripheral
breakdown, two kinds of structures are commonly used, which have
annular guard rings modifying the electric field distribution around the
outer perimeter.
In the first structure (see Figure 3.6a), the guard ring is formed by
an annular diffused junction which is relatively weakly doped, and
quite deep in order to increase the radius of curvature and thus reduce
the peripheral field; the active region is obtained by a shallower
diffusion at a higher concentration, partially overlapping the guard
ring. The breakdown voltage of the gradual junction of the guard ring
is much larger than that of the active region, because for a given bias
voltage, the extent of the SCR is larger in the former than in the latter.
Figure 3.6. Cross-section of (a) an APD with a traditional
guard ring and (b) a PPN+ APD
In the second structure (see Figure 3.6b), the guard ring and active
region junctions are created by diffusion or ion implantation within a
weakly-doped substrate S. The n
+
S guard ring junction which
surrounds the n
+
p junction of the active region is much deeper than
the latter, but the weak concentration in the substrate precludes
breakdown. This is the PSPN
+
structure. When the n
+
S junction is
not very deep, or even superficial, the addition of a guard ring of the
first type is necessary to avoid peripheral breakdown. This is referred
to as a P
+
SPSN
+
structure.
3.5.2. Mesa structures
In mesa structures, the footprint of the active region of the APD is
defined by chemical etching or by dry etching followed by surface
reconstruction using a suitable chemical bath. This structure, shown in
Figure 3.7, offers the sometimes deceptive advantage of simplicity. It
allows us to reduce the capacitance of the device, while eliminating
the guard ring junction. However, the extension of the avalanche
region at high electric fields outside the edges of the etched region can
lead to migration of susceptible ion species, causing a growth in the
dark current which is often accompanied by instability of the diode. In
order to avoid these drawbacks, different passivation techniques have
been used with moderate success, which makes such a structure
difficult to commercialize on account of its lack of reliability.
Figure 3.7. Cross-section of a mesa APD
3.5.3. Crystal defects and microplasmas
Crystal defects such as dislocation, stacking faults, metallic
precipitation, etc., present in a PN junction introduce localized or
microplasma breakdowns: isolated regions of a few m
2
where high
current densities cross, and which appear at well-defined bias voltages
which may or may not be close to the volume breakdown voltage.
Unless these microplasmas appear at a voltage very close to the
breakdown voltage, their presence decreases APD performance by
limiting the photomultiplication factor to that of the breakdown
voltage of the first microplasma. At higher voltages, the noise linked
to electrical discharges triggered by the random injection of charge
carriers in the microplasma is so significant that it precludes any
interest in the use of such diodes. In the case of Si APDs, inspection
under a microscope allows the microplasmas to be visualized,
appearing in the form of bright red spots. In the absence of
microplasmas, a diffuse, pale light uniformly illuminates the whole
active surface when the breakdown voltage is reached. Observation of
the current induced in a scanning electron microscope also allows the
detection of inhomogenities in the electric field. In addition, the
presence of microplasmas leads to vertical jumps of several A in the
inverse characteristics I
d
(V), interspersed with resistive sections in the
relationship, thus enabling the electrical sorting of APDs through
systematic measurement of the I
d
(V) characteristics via probes.
Microplasmas can be eliminated using substrates with a low level of
dislocations and non-aggressive fabrication techniques tailored to the
specific materials used. The construction of high-performance,
reliable devices with a high fabrication yield depends closely on the
quality of the base material and on the standard of technology used in
each stage of the fabrication.
3.6. Silicon avalanche photodiodes
Silicon, an indirect-bandgap semiconductor, with a bandgap energy
E
g
= 1.1 eV at ambient temperature, is sensitive to light in the
wavelength range of 0.41.1 m. The absorption length, considerably
less than 1 m at O = 0.4 m, varies between 10 and 10 m in the
wavelength range 0.80.9 m, the first transmission window for
optical fibers (attenuation ~ 2 dB/km), and is more than 200 m at
O = 1.06 m, which is the emission wavelength of the Nd lasers that
are very popular in military applications. Also, the ionization
coefficients of electrons and holes are very different, with electrons
having the higher ionization energy (D
i
>> E
i
). These characteristics
play an important role in the design of devices targeted at different
applications, with the aim of achieving a weak excess noise factor and
an acceptable compromise between sensitivity and response time.
3.6.1. Si N
+
P APDs
Si APDs with an abrupt-junction N
+
P structure are fabricated on a
P substrate with a concentration around 10
16
cm
3
. As the width of the
depleted region is between 2 and 3 m, these APDs are well-suited to
the detection of light signals with wavelengths below 0.7 m (ruby
lasers emit at 0.63 m). Commercialized in 1966, they were initially
developed for military applications such as laser telemetry at 1.06 m,
taking advantage of an atmospheric transmission window allowing
targeting at 2025 km with high precision (of the order of a meter).
For the first APDs we will study, the guard ring and active region
junctions are made by a double diffusion of phosphor impurities,
masked using a layer of silica obtained by thermal oxidation; the
fabrication conditions for the different stages of the construction
procedure have been developed to ensure high performance in terms
of the dark current and photomultiplication, along with a high
fabrication yield. The homogenity and reproducibility of phosphor ion
implantation used as a predeposition technology have greatly helped
the achievement of microplasma-free APDs. Additionally, the choice
of substrates with fairly high dislocation levels (~3,000 cm
2
), but
which are stable under various thermal treatments at high
temperatures (T 1,0001,200C), has allowed the attainment of
fabrication yields considerably higher than 50%, whereas they were at
best around 20% with the zero dislocation substrates advocated in
the literature.
These APDs, with a breakdown voltage V
B
~ 60 80 V, offer M(V)
characteristics which vary extremely rapidly as breakdown is
approached, with the exponent n in the Miller equation taking a value
close to 4. At short wavelengths absorbed in the N
+
layer, the
multiplication is initiated by holes. For O > 0.7 m, the multiplication
is mostly initiated by electrons and takes higher values. At the
operating point, M ~ 50 to 100, with the gradient dM/dV very large.
To avoid issues with variations in M, the ratio V/V
B
must be kept
constant. Since the breakdown voltage V
B
follows variations in
temperature, some system of regulation is required. The involvement
of a Zener diode with a breakdown voltage close to that of the APD
has allowed the multiplication to be regulated within the temperature
range from 30C to +80C. Other regulation systems make use of the
noise linked with multiplication.
The impulse response of Si N
+
P APDs (see Figure 3.8) is
dominated by the diffusion time t
Diff
of the charge carriers generated
outside the depleted region. It consists of a brief rise to the signal
(<0.5 ns) followed by a relatively long tail, with the duration of the
impulse generally being insufficient to reach the steady-state value.
We then see a similar behavior for the fall of the signal. The diffusion
time limits the repetition frequency, and this can be improved by
limiting the thickness of the absorbing region with the use of epitaxial
P/P
+
structures or by localized thinning of the substrate.
Figure 3.8. Typical impulse response of a Si N+P photodiode
3.6.2. Si N
+
PP
+
APDs
To compensate for defects in the N
+
P structure, in 1966 Ruegg
[RUE 66] showed the potential of an N
+
PSP
+
structure, which
combines the advantages of an N
+
P structure (internal gain) with those
of a PIN structure (high quantum efficiency and speed). The result of
this is that the depleted region consists of two adjacent regions: one
relatively narrow, with an intense electric field, where the
multiplication of injected charge carriers takes place, and another
relatively extensive region, with a weak electric field, where light is
absorbed and the photocarriers collected. This separation of the roles
of multiplication and absorption gives the structure a high gain-
bandwidth product. Absorption takes place in the very weakly-doped
S region (<10
14
cm
3
), whose thickness is adjusted as a function of the
operating wavelength (0.85 m or 1.06 m) and the desired
performance. The doping and thickness of the P layer are absolutely
critical.
Figure 3.9. Operating principles of a Si P+PN+ APD: a) doping profile, b) typical
electric field, c) typical M(V) characteristics. V
rt
is the reach-through voltage
The operation of N
+
PSP
+
APD structures is shown in Figure 3.9,
which displays the doping profile, the electric field distribution for
three significant bias voltages and a typical multiplication curve. At
weak bias voltages, the depleted region is restricted to the P layer. As
the voltage increases, the edge of the depleted region moves and
enters the S layer for a bias voltage V
rt
known as the reach-through
voltage, for which the multiplication is around 10 to 20, just before the
breakdown of the N
+
P junction. Then, with the depleted region
extending across the whole S layer, the increase in the bias voltage
will only cause a weak increase 'E in the electric field. The
multiplication will vary relatively slowly, especially if the S layer is
wide, and the M(V) characteristic involves a more or less pronounced
plateau. Under normal operating conditions, the field in the
absorption region is sufficient to ensure carrier transport at their
saturation velocity. However, in order to benefit from a high gain, the
operating point must be beyond the plateau.
Consistent achievement of the desired M(V) multiplication
characteristics assumes that we can accurately control ( 10%) the
concentration profiles of the N
+
and P layers. Specifically, if the P
layer is too wide, the breakdown of the N
+
P junction occurs before the
depleted region extends in to the S layer, but if it is too narrow, the
extension into the S layer occurs too soon and the multiplication will
be weak (Figure 3.9c).
For applications at 0.85 m, the thickness of the S layer can be
adjusted between 15 and 50 m, as a function of the chosen level of
tradeoff between speed and response. In order to avoid the critical
stage of localized thinning, these APDs are best fabricated on an
epitaxial S/P
+
plate, with illumination of the N
+
P junction only
slightly degrading the excess noise factor.
3.6.3. Si N
+
PP
+
APDs
An N
+
SPSP+ structure with buried P layer is a refined version of
the N
+
PSP+ structure. Proposed and studied by Lecrosnier [LEC 75],
it offers a significant reduction in the excess noise factor F(M) thanks
to its narrow multiplication region with a constant electric field
limited by the buried P layer: the field is weaker when this is far
below the surface. The technique of implanting high-energy B
+
ions (1
to 2 MeV), chosen for constructing this P layer, is very well-suited
because by modifying the implantation energy and dose parameters it
is possible to change the breakdown voltage and excess noise behavior
of these APDs. For a given Si S/P
+
structure, the multiplication
electric field decreases as the implantation energy is increases, since
the P layer lies further from the surface, at distances of 2.75 and
3.1 m respectively for implantation energies of 1.5 MeV and
1.8 MeV.
The typical behavior of N
+
PP
+
APDs is only reached over a small
range ( 15%) on either side of the optimal implantation dose. If the
dose is too large, the behavior is like that of a N
+
P diode, or of a PIN
diode if it is too weak. Figure 3.10 shows a schematic doping profile,
the associated electric field distribution, and the M(V) characteristics
for different doses at a given implantation energy.
The fabrication procedure for APDs involves two diffusion phases
in order to construct stopper ring and guard ring junctions, followed
by two phases of ionic implantation and a phase of electrical
activation annealing: firstly the low-energy implantation of phosphor
ions to build the active N
+
S junction close to the surface (x
J
~ 0.2 m),
then the high energy implantation of Boron ions (1.5 or 1.8 MeV,
dosage 1.2 u 10
12
ions/cm
2
) for the buried P layer.
a)
b)
c)
Figure 3.10. Operation principles for a Si N+PP+ APD: a) doping profile,
b) typical electric field, c) typical M(V) characteristics for
different implantation doses of Boron (in cm
2
)
Control of the conditions for the two implantations ensures good
reproducibility of the procedure, which is particularly easy to do for
epitaxial S/P
+
wafers, and also a good fabrication yield. A model of
this structure, based on complete photoelectric characterization (I(V),
C(V), M(V), noise) for various APDs was developed by Maille [MAI
80].
APDs obtained in this way offer excellent photoelectric
performance, normally with 250 m diameter diodes:
weak dark current (I
0
< 1 nA) close to breakdown;
useful gain M
u
> 150, resulting in a response coefficient greater
than 75 A/W at S = 0.85 m and a rise time W
r
~ 0.5 ns for a S layer of
thickness 30 m;
an excess noise factor F(M) = M
x
typically with F(M = 100) =
3 3.5, corresponding to x = 0.25. This value, close to the lower limit
of x = 0.2 is a result of the low ratio k ~ 0.01 of the ionization
coefficients E
i
and D
i
, due to the fact that the multiplication electric
field is such that E
m
< 3 x 10
5
Vcm
1
;
a uniformity in the photoresponse of 5%.
Diodes with an active zone diameter of 800 m and even 1,500 m
have been achieved thanks to the excellent spatial homogenity offered
by the implantation technique. To compensate for the lack of suitable
high-energy implantation equipment, a number of solutions have been
researched. One of them consists of implanting doubly-ionized B
++
Boron ions at relatively high energies (~300 keV), or alternatively
implanting Boron ions at 800 keV in a preferred direction in a S/P+
epitaxial plate with <110> orientation, to take advantage of a
channeling effect, resulting in the P layer lying 2.2 m below the
surface, instead of 1 m as would be the case under normal conditions.
The excess noise factors at M = 100 are F = 45 with channeling,
instead of F = 67, with x = 0.300.35 and x = 0.40 respectively
[KAN 78]. However, although channeling implantation gives
promising results, these are not easily reproducible.
Another process exists where the buried P layer is fabricated in two
stages: first localized low-energy implantation of Boron ions on the
S/P
+
plate, followed by an epitaxial stage to obtain the upper S layer
for multiplication. The growth of this layer is delicate. APDs made in
this way have an excess noise factor F = 5 at M = 100, with x = 0.35
[YAM 76], which could be improved upon by reducing the doping
and increasing the thickness of the upper S layer. These results
confirm that the excess noise factor F(M) is weaker when the P layer
is buried more deeply, and hence the electric field is only marginally
increased.
3.6.4. SiPt-Si N Schottky APDs
SiPt-Si N Schottky APDs, obtained through the formation of a very
thin, transparent and non-absorbing layer (20 to 40 nm) of platinum
silicate on top of N-type silicon, with resistivity U ~ 0.51 :cm, are
well-suited to the detection of ultraviolet (UV) radiation around
O = 0.5 m, for which the absorption occurs very close to the surface
(absorption length of the order of 0.1 m). Their performance depends
closely on the quality of the SiPt-Si N interface, and hence on the
standard of the fabrication technology.
After diffusion of the guard ring, the silica protecting the active
central region is chemically removed. Then the residual native oxides
and the first few atomic layers of silicon (~10 nm) are removed by ion
bombardment during the chemical deposition phase. Then, a 10 to 20
nm platinum film is deposited by radio-frequency sputtering. The
platinum silicate is then formed by in situ annealing at 600C, with the
thickness of the silicate obtained being twice that of the platinum. The
metal present in the passivation layer is eliminated using aqua regia.
The upper contact is formed with a ring of aluminum deposited by
evaporation. The presence of an anti-reflection coating tuned to O =
0.5 m is needed to obtain a good response coefficient, as the
reflection coefficient of silicon is around 45% in the near ultraviolet.
Under these conditions, a response coefficient above 0.25 A/W at M= 1
can be obtained for O = 0.4 m.
APDs fabricated in this way possess excellent photoelectric
characteristics [ALA 74]. A barrier height I
BN
= 0.85 V and a good
ideality factor n = 1.02 measured from the I-V characteristics confirms
the good quality of the SiPt-Si N interface, despite a dark current
greater than that of traditional N
+
P or P
+
N junctions in their pre-
breakdown regime. At short wavelengths, multiplication is triggered by
electrons, and values of multiplication M ~ 300 have been measured at
O = 0.435 m with an excess noise factor M
x
such that x ~ 0.7. The
limit on the bandwidth is imposed by the time constant RC and is
several GHz. In contrast, at O = 0.8 m, the light is mostly absorbed
outside the depleted region and multiplication is thus triggered by
holes leading to performance degradation: a weaker M, higher excess
noise and a reduced bandwidth.
3.7. Avalanche photodiodes based on gallium arsenide
Si APDs are almost perfect photodetectors for the 0.800.85 m
transmission window in fiber-optics, simultaneously possessing a
weak dark current, a high gain and low noise, all thanks to the
maturity of the technology. Its only drawback stems from the response
coefficient/speed tradeoff imposed by the weak absorption coefficient,
which is in fact not a great problem because the applications are at
low or moderate bitrates (<500 Mbit/s).
Nevertheless, investigations have been carried out into gallium
arsenide (GaAs), a direct-bandgap semiconductor with a bandgap
E
G
= 1.40 eV and photosensitive up to 0.88 m, with respect to
possible use in integrated circuits and on ternary structures based on
GaAs, materials with a high absorption coefficient. GaAs APDs,
either of the Schottky type or diffused with an implanted guard ring
(H
+
protons or Mg
+
ions) offer performances below that of Si APDs
within the range 0.800.85 m, except in terms of speed. On the other
hand, Stilman et al. [STI 74] showed the presence of a very
pronounced Franz-Keldysh effect for APDs made of lightly-doped
material (n ~ 5 x 10
14
cm
3
), when biased in the multiplication regime.
a)
b)
Figure 3.11. Spectral response of electroabsorption GaAs Schottky APDs:
a) weak bias voltages, b) strong bias voltages
More recently, GaAs/Al
x
Ga
1-x
As APDs with low noise, using a
new concept (see section 3.10), have opened a new research avenue.
They offer a multi-quantum well (MQW) or staircase multiplication
region with a bandgap width varying continuously between GaAs and
Al
x
Ga
1-x
As. The need to cross discontinuities favors the ionization of
one type of charge carrier. Thus, for the x = 0.45 composition of Al,
the discontinuities are respectively 'E
C
= 0.48 eV and 'E
V
= 0.08 eV,
giving a value of k =E
I
/D
I
~ 0.15 and an excess noise factor F = 3 at
M = 10 [KAG 89]. The presence of an adjacent 1.5 m absorbing
layer of GaAs enables high quantum efficiency devices that are fast
and low-noise.
3.8. Germanium avalanche photodiodes
Germanium, an indirect-bandgap semiconductor with a small
bandgap E
g
= 0.67 eV at ambient temperature and photosensitive in
the spectral range 0.4 m to 1.65 m, became established in the
middle of the 1970s as the photodetector for the first generation of
receivers for fiber-optic links at 1.3 and 1.5 m, a region where silicon
does not work.
Starting in 1966, Melchior and Lynch [MEL 66] showed that Ge
APDs could be used as fast and sensitive detectors, and various
devices were brought to market. However, competition with Si APDs
limited their use to very specific areas of application. The problem
was that Ge APDs suffered a number of drawbacks linked to the
intrinsic properties of germanium, including:
a high dark current, dominated by the charge carrier diffusion
current, which is proportional to n
i
2
, n
i
being the intrinsic concentration
for germanium;
an insufficiently high ratio k of the ionization coefficients of
electrons and holes: k 2, with the holes being the more easily
ionized, in contrast with silicon;
the absence of stable native oxides, which makes the process of
surface passivation very difficult;
a difficulty growing high quality epitaxial structures.
On the other hand, due to the high absorption coefficient D
O
up to
1.50 m, depleted region widths w between 2 and 3 m are sufficient
to absorb the light signal, thus allowing a good response coefficient
and a high speed. In addition, the dark current does not include a
tunneling current contribution, despite the weak value of E
g
.
3.8.1. Ge APDs with N
+
P, N
+
NP and P
+
N structures for 1.3 m
communication
The three structures N
+
P, N
+
NP and P
+
N were developed in
succession for use in optical links at 1.3 m. They have a planar
construction with a guard ring to ensure good reliability. Figure 3.12
shows a cross-section view. The presence of a stopper ring decreases
the surface current.
N
+
P structures were the first to be used, since the diffusion of n-
type dopants (As, Sb) is easier to achieve. The conditions for the
surface preparation and the deposition of dielectric layers (SiO
2
,
Si
3
N
4
), used to limit the diffusion, for passivation, and also for the
antireflection coating, are very critical. For an active region of
diameter = 100 m, the total dark current measured at 0.9 V
B
is
around 150 nA and the dark current I
d0
around 50 nA at M = 1. The
excess noise factor for M = 10 is such that F(10) 10 11,
corresponding to an effective ratio between the ionization coefficients
of k
eff
1.3 m. This high value of F(M), due to the injection of
electrons instead of holes as primary charge carriers, increases still
further with the wavelength.
Figure 3.12. Cross-section of Ge APDs: N
+
P, N
+
NP and P
+
N structures
The N
+
NP structure was proposed to improve the noise properties
of Ge APDs, with multiplication taking place close to the NP junction
lying about 2.5 m from the surface. The N layer is obtained by
implantation of arsenic ions followed by annealing, leading to a
gradual junction. Since the light is mostly absorbed in the N layer, the
holes play the role of primary charge carriers triggering multiplication.
In addition, the electric field E at the junction is weaker (E 2 x 10
5
V/cm). This results in a sensitive increase in the excess noise factors:
F(10) = 7 instead of 10 as previously. On the other hand, the total dark
current at 0.9V
B
is around 1 m and the response time is limited by
the diffusion time in the N layer. The main drawback of the N
+
NP
layer is the lack of reproducibility caused by the fabrication conditions
for the N layer.
The P
+
N structure represents the final success of Ga APDs for 1.3
m applications. It possesses the respective advantages of the N
+
P and
N
+
NP structures in terms of speed and excess noise factor, with the
light being absorbed in the N-type depleted region and the
multiplication being triggered by holes. The active region and guard
ring junctions, as well as the stopper ring, are made by ionic
implantation followed by annealing at relatively low temperature. This
fully implanted technique offers good reproducibility of the
performance. For diodes with an active region diameter of 100 m
and 30 m, the excess noise factor F(10) is 8 and 7 respectively, and
the cutoff frequency is at least 1 GHz at M= 10.
3.8.2. Ge APDs with P
+
NN
-
structures for 1.55 m communication
Above L=1.52 m the N+P, N+NP and P+N photodiode structures
become blind if the temperature drops below -20C, because the
absorption coefficient of germanium falls dramatically. The
absorption length at ambient temperatures is around 25 m at O = 1.55
m, a window of weak absorption for optical fibers (0.2 dB/km).
Structures developed for 1.3 m communication are not suitable
because of their low depleted region width w 2 to 3 m. Only a
structure analogous to N
+
PSP
+
Si APDs, where absorption and
multiplication take place in separate regions, could attain the required
performance of the response coefficient and speed.
Figure 3.13. Cross-section of a P
+
NN
structure Ge APD
for 1.55 m communication
P
+
NN
-
structured Ge APDs on a weak-concentration N substrate
( 5 u 10
14
cm
3
) have been built to meet these requirements. The
active region of these planar-structure devices with guard rings
(Figure 3.13), of diameter = 80 m or = 30 m for coupling to
multimode or single-mode fibers, is best obtained by a double
implantation of boron or indium for the surface P
+
junction and of
arsenic for the N-type multiplication region, followed by activation
and diffusion annealing. The guard ring junction is made either by a
long period of zinc diffusion or by implantation of beryllium ions
followed by annealing. The best performances obtained for 80 m
diameter APDs at O = 1.55 m are as follows:
total dark current at 0.9 V
B
: 1.3 m;
dark current I
d0
at M = 1: 70 nA;
quantum efficiency K = 80% at M = 1;
excess noise factor F = 6.1 at M= 10;
cutoff frequency above 500 MHz at M = 10;
breakdown voltage V
B
7085 V.
The limitation on the response time is imposed by the transit time
in the depleted region. The impulse response of Ge P
+
NN
-
APDs at
1.55 m does not contain a lag due to the collection of charge
carriers created outside the depleted region. This is also the case for
Ge P
+
N photodiodes, whose cutoff frequency is only 10 MHz.
Additionally, Ge APDs often operate with an optimal gain between 5
and 10, with the response time depending little on the multiplication
factor.
Nevertheless, with the available performance gains for Ge APDs
limited by the fundamental properties of germanium, investigations
carried out by various laboratories have yielded APDs based on III-V
compounds with very promising performances as a second generation
of structures for optical communications at 1.3 and 1.55 m.
Figure 3.14. Cross-section of an InGaAs/InP planar APD
with front illumination, using a SAGM structure
3.9. Avalanche photodiodes based on indium phosphate (InP)
In
0.53
Ga
0.47
As/InP heterojunction APDs rapidly came to the
forefront in the 1980s as second generation detectors for optical
communications at 1.3 and 1.55 m. This was because the ternary
compound In
0.53
Ga
0.47
As, with a direct bandgap of E
G
= 0.75 eV,
offers strong absorption up to 1.65 m. Formed into a lattice using InP,
it is well-suited to epitaxial growth of high crystal quality layers.
However, the appearance of a high tunneling current for an electric
field E > 1.5 x 10
5
V/cm precludes all avalanche multiplication,
despite a more favorable ratio of ionization coefficients than that of
germanium (D
I
~ 2.E
I
).
On the other hand, P
+
N APDs based on InP, a material with a large
bandgap (E
G
= 1.35 eV) and transparent above 0.95 m, offer a low
dark current along with an ionization coefficient for holes that is
higher than that of electrons, such that E
I
~ 2D
I
.
Figure 3.15. Operation principles of an InGaAs/InP APD with a SAGM structure.
The electric field is given: 1) for a weak bias voltage, 2) for the punch-through
voltage, and 3) in the case of multiplication
The use of a structure analogous to the P
+
SPN
+
structure developed
for silicon, consisting of a separate absorption region (In
0.53
Ga
0.47
As)
and multiplication region (InP), led to the SAGM (Separated
Absorption Graded Multiplication) structure. Figure 3.14 shows a
cross-section through a planar InGaAs/InP APD with front-face
illumination which has a SAGM structure. The graduated
quaternary InGaAsP layers, with bandgap widths intermediate
between those of InGaAs and InP, fulfill a double function: to ensure
a gentle transition between InGaAs and InP, to facilitate the
transport of holes by reducing the valence band discontinuity at the
InGaAs/InP interface, and to avoid the decomposition of the ternary
layer during growth of the InP layers above it. Their presence is vital
for the growth of a vertical heterostructure by liquid phase epitaxy.
This structure, suited for illumination through front or back faces,
offers the advantage of a pure injection of holes, which are the most
ionizing carriers in InP, thus allowing a reduction in the excess noise
factor. It is suited to illumination through the front or rear faces.
Proper operation of these APDs, whose principle is summarized in
Figure 3.15, requires strict conditions to be met in terms of doping and
thickness of the epitaxial layers, especially of the multiplication layer,
since during avalanche the electric field E
QT
at the quaternary/ternary
absorption interface must be below 1.5 u 10
5
V/cm in order to retain a
low dark current.
Construction of the guard ring, required to reach the necessary
reliability levels for detectors targeted at submarine communications,
is even more delicate than for Si and Ge APDs. Specifically, the guard
ring junction, which is slightly deeper than the active region, must
remain within the InP, and the depleted region under the guard ring
must reach the ternary layer before that of the active regions junction.
This requires that the guard ring junction be gradual with a weak level
of doping. For a long time, the implantation of beryllium Be
+
ions,
followed by redistribution annealing [SHI 83], was the only method of
fabricating effective guard rings. More recently, better understanding
of epitaxial growth techniques has allowed good results to be obtained
by combining localized implantation and epitaxial regrowth [WEB 88].
A realistic predictive model for the behavior of planar InGaAs/InP
APDs was put forward by Harari [HAR 91b] after a detailed analysis
of their operating conditions.
3.9.1. InGaAs/InP APDs for optical communications at 2.5 Gbit/s
An APD heterostructure based on InGaAs/InP on a <100>
oriented InP-N
+
substrate consists of a stack of 6 layers: an InP-N
+
buffer layer (n = 5 u 10
17
cm
3
, 0.5 m), an InGaAs absorption layer
(n = 5 u 10
15
cm
3
, 2.5 m), 2 InGaAsP layers (not intentionally
doped, 0.2 m), the InP-N multiplication layer (n = 1.5 u 10
16
cm
3
),
and the InP-N
covering layer (n = 8 u 10
15
cm
3
, 2 m).
The guard ring junction is obtained by implantation of beryllium
ions (Be
+
, 150 keV, 5 u 10
13
cm
-2
) followed by annealing at 700C in
a sealed ampoule. The active junction is formed by diffusion of
cadmium at 620C in a sealed ampoule, for a duration which depends
on the required characteristics of the multiplication layer. A nitride
layer of Si
3
N
4
ensures surface passivation, and another acts as an
antireflection coating.
Typical performances of such 70 m diameter APDs are:
a response coefficient (at M = 1) of 0.85 A/W and 1 A/W at 1.3
m and 1.55 m respectively;
a dark current at 0.9 V
B
< 30 nA;
a maximum gain M
max
= 30;
an excess noise factor F = 5 at M= 10;
a bandwidth of 2.5 GHz.
Figure 3.16. M(V) and I
d
(V) characteristics of a fast InGaAs/InP APD
Experiments on accelerated aging under extreme conditions
(175C, 100 A) have demonstrated the robustness of these APDs,
whose median lifetime under these conditions is greater than 4,000
hours. In addition, thanks to their high degree of reliability, they are
one of the key devices for fiber-optic communications at 2.5Gbit/s. To
begin with, the variation in concentration and thickness of the layers
deposited through liquid phase epitaxy had a beneficial effect, with
the devices present on a treated wafer offering a wide range of
behaviors. This helped develop and understanding of their operating
conditions and to acquire a good understanding of the fabrication
requirements for such APDs. In this way, fabrication yields greater
than 50% were obtained.
More sophisticated growth techniques, such as molecular jet
epitaxy or vapor phase epitaxy, allowing the deposition of layers that
are very uniform in thickness and concentration have led to even
higher fabrication yields (>80%), while tightening the positioning
requirements for the junction layout with respect to the InP/InGaAs
interface.
3.9.2. Fast InGaAs/InP APDs
In order to meet the need for fast detectors for high-bitrate
communications at 5 or 10 Gbit/s, the normal bandwidth of APDs
(typically 2.5 GHz) must be increased. In order to do this, we must
reduce the accumulation of holes due to the large valence band
discontinuity at the InP/InGaAs interface, which is the main limitation
of the bandwidth of these APDs. It is also necessary to reduce the
transit time across the InGaAs absorption layer while keeping its
thickness between 1.0 and 1.5 m in order not to degrade the response
coefficient. Furthermore, the increase in doping of the InP
multiplication layer, along with a reduction in its thickness,
contributes to a reduction in the avalanche trigger time. Finally, an
active region of 30 m diameter, comparable with single-mode fiber
illumination, minimizes the capacitance of the junction. APDs with
one, two or three thin quaternary layers, or with a quaternary layer
whose composition varies continuously between InGaAs and InP,
have been fabricated with bandwidths between 7 and 9 GHz. With
only one thin 1.18 m quaternary layer (the wavelength corresponding
to the bandgap energy of the GaInAsP compound) ensuring
equipartition of the discontinuity of the valence band (0.19 eV instead
of 0.38 eV), a theoretical study [RIP 83] shows performances
equivalent to two- or three-layer structures which are more
complicated to fabricate. After fabrication, the measured
performances are as follows:
dark current < 60 nA at M = 10, with the best values being
around 10 nA;
dark current I
d0
lying between 0.8 and 4 nA;
maximum gain M
max
> 30 for the majority of diodes;
response coefficient at M = 1 of 0.9 A/W at 1.3 m and
0.95 A/W at 1.55 m;
bandwidth > 6 GHz at M = 8;
excess noise factor F = 6 at M = 10, with a fabrication yield
greater than 75%.
However, these InGaAs/InP APDs targeted at 10 Gbit/s
communications are reaching the limits of performance. System
designers often prefer to use InGaAs/InP APDs at 2.5 Gbit/s in a 4 u
2.5 Gbit/s configuration, when an infrastructure for 2.5 Gbit/s optical
communications is already in place.
3.10. III-V low-noise avalanche photodiodes
The performance of APDs in terms of noise and bandwidth
depends strongly on the ratio of the ionization coefficients of the
charge carriers. For a given material, they will be higher when the
ratio between the ionization coefficients of the least and more easily
ionized charge carrier is weaker. This is the case for silicon, where at
weak electric fields (E < 3 u 10
5
Vcm
1
), k ~ 0.01, which is close to
the behavior of ionization only by electrons. In contrast, in InP where
the holes are the more easily ionized, this ratio is around k ~ 0.5, a
value already corresponding to a significant noise factor.
In order to reduce the excess noise factor and increase the gain-
bandwidth product for III-V APDs based on InP or GaAs, it is
necessary to strongly increase the ionizability of one type of charge
carrier, using a modified structure or a material which has this
property under specific conditions. Chin et al. [CHI 80] suggested
reducing the ionization energy of holes or electrons by taking
advantage of the numerous discontinuities in the valence or
conduction band which can be found in multi-quantum-well (MQW)
structures. Alternatively, Hildebrandt et al. [HIL 81] showed that
ternary GaAlSb structures offer a significant increase in the
ionizability of holes close to a specific composition ratio. This is also
the case for the II-VI ternary compound HgCdTe.
3.10.1. III-V super-lattice or MQW APDs
The multiplication region of a MQW APD consists of a stack of
layers with alternately wide bandgaps (barriers) and narrow
bandgaps (wells) which produce discontinuities at the interface,
between conduction bands 'E
c
and between valence bands 'E
v
,
whose values depend on the nature of the materials used. The
barrier/well pairs favor the transport of electrons when 'E
c
> 'E
v
or
of holes when 'E
c
< 'E
v
. For example, in the case of the
InAlAs/InGaAs pair, an electron which undergoes a transition in a
barrier region crosses the interface and falls into the adjacent well,
gaining an energy 'E
c
, or alternatively benefits from an artificial
reduction in its ionization energy, thus favoring ionization in the wells.
Conversely, movement from a well to a barrier reduces the ionizing
power of the electron within the barrier. The crossing of numerous
barrier-well interfaces in succession translates into an increase in the
multiplication triggered by the electrons, equivalent to an increase in
D
I
. On the other hand, the ionizing power of the holes traveling in the
opposite direction is less favored, and overall this results in a
reduction of the effective ratio of ionization coefficients k =E
I
/D
I
,
leading to a reduction in the excess noise factor. When 'E
v
> 'E
c
,
such as in InP/InGaAs, it is the ionizing power of holes which is
favored, leading to a reduction in the ratio D
I
/E
I
for this structure.
Pairs of materials with very asymmetric bandgap discontinuities are
favored; thus, the combination of In
0.52
Al
0.48
As/In
0.53
Ga
0.47
As ('E
c
=
0.55 eV, 'E
v
= 0.2 eV) is more favorable than InP/In
0.53
Ga
0.47
As ('E
c
= 0.22 eV, 'E
v
= 0.38 eV). The number of barrier-well layers and the
thickness of those layers both have a significant influence on
performance.
Up to now, the best performances have been obtained [WAT 97]
for a planar APD with an implanted Ti guard ring with a MQW
multiplication layer of InAlGaAs/InAlAs (0.27 m, 13 repeats) and an
absorption region of InGaAsP (1.3 m, p = 8 u 10
15
cm
3
) separated
by a thin doped layer of InP-P
+
(0.04 m, p = 4 u 10
17
cm
3
) in order
to reduce the electric field between the two regions. These APDs are
characterized by a weak dark current of 0.36 A at M = 10, a
bandwidth of 15 GHz, a gain-bandwidth product of 110 GHz, a
response coefficient of 0.8 A/W at 1.55 m, an excess noise factor
F(M) ~ 3 at M = 10 and a high reliability. They are well fitted to the
specifications of high-sensitivity 10 Gbit/s receivers.
3.10.2. Spin-orbit resonance APDs
3.10.2.1. GaAlSb/GaSb APDs
The strong increase in the ionizing power of holes which is seen in
Ga
1x
Al
x
Sb/GaSb APDs with x ~ 0.05, sensitive at 1.3 and 1.55 m, is
linked, according to the interpretation of Hildebrand et al. [HIL 81], to
the ionization mechanism for holes which causes the valence band to
be split in two via spin-orbit coupling. The result is that, around this
composition, the structure of the energy bands specific to this material
has the unusual property of having a bandgap energy E
g
close to the
difference in energy ' separating the maximum values for the valence
band and the additional split band.
If 'E
g
o1, through the resonance effect, the ratio of the ionization
coefficients E/Do and the ionization energy of holes is equal to E
g
.
Subsequent construction of Ga
1x
Al
x
Sb/GaSb APDs with x ~ 0.05 has
made it possible to obtain devices with a weak excess noise factor
F(M) = 3.8 at M = 10, corresponding to a value of k =E
I
/D
I
~ 5, but
much lower than the value of 20 to 30 that has been anticipated, and a
gain-bandwidth product of 90 GHz [KAT 90]. These performances are
higher for these parameters than those obtained with comparable
InGaAs/InP APDs; however, the inability to reduce and stabilize their
dark current or to fabricate a planar APD has led to these
investigations being abandoned.
3.10.2.2. HgCdTe avalanche photodiodes
Research into ternary, mercury-rich alloys Hg
1x
Cd
x
Te, which had
already been explored with the aim of fabricating detectors for
military applications in the infrared windows at 3.5 m and 814 m,
was extended at the start of the 1980s for the development of
alternatives to Ge APD detectors for fiber-optic telecommunications
at 1.3 m and 1.55 m. The alloys with cadmium-rich compositions (x
~ 0.7 0.6) corresponding to these two transmission windows offer
the same electronic properties of spin-orbit coupling, which offers the
opportunity to develop APDs with performances similar to Ga
1x
Al
x
Sb.
A method has been developed for growing Hg
1x
Cd
x
Te wafers
with a given stochiometric composition, allowing the construction of
p-type substrates with doping levels around 2 u 10
16
cm
-3
, 40 mm in
diameter, with a high homogenity in terms of doping, and a low level
of dislocations. The devices are constructed in a planar structure using
traditional techniques of insulator deposition by cathode bombardment,
and photolithography. The n-type guard ring is obtained by mercury
diffusion. The active region junction is made by ionic implantation
(Al
++
, 65 keV, 10
14
ions cm
-2
) followed by redistribution annealing.
The result is a PIN junction. Compositions with x close to 0.6 are the
most suitable, because they cover both transmission windows and also
give the highest values of the ratio E
I
/D
I
[ORS 87], with the best being
the composition with x = 0.62, E
g
= 0.72 eV and E
I
/D
I
= 30.
Encouraging initial results obtained in terms of dark current, response
coefficient and excess noise factor have not been confirmed at the
predevelopment stage.
3.11. Prospects
A large proportion of the work in progress is aimed at the
development of devices for high-bitrate communications (20 to
40 Gbit/s) that are both fast and low-noise. Particularly significant is
the work on Si/InGaAs APDs, waveguide MQW APDs and low
noise APDs with a very thin multiplication region.
3.11.1. Si/InGaAs APDs
These APDs, combining the very low excess noise factor of Si
APDs with the strong absorption of InGaAs, should simultaneously
allow high quantum efficiency and a high bandwidth, a high gain-
bandwidth product and a low level of noise. Preliminary results
obtained for Si/InGaAs structures fabricated using the technique of
wafer fusion are very promising: bandwidths of 13 GHz for gains of
135, corresponding to a gain-bandwidth product of 315 GHz have
been measured. However, the dark current is raised very close to
breakdown and is unstable under a bias voltage [HAW 97].
Important work remains to be done to develop enough
understanding of wafer construction and the stages of APD fabrication
in order to overcome the limitations linked to the large discrepancy in
the lattice parameter between Si and InGaAs on the one hand and, on
the other hand, also linked to the difficulties of passivation.
3.11.2. Waveguide MQW APDs
In order to reduce the maximum transit time of charge carriers
without reducing the quantum efficiency, the light signal is injected
laterally, perpendicular to the electric field, in a vertical waveguide
structure whose InGaAs core is very thin (~0.5 m).
Since the MQW InAlAs/InGaAlAs or InAlAs/InGaAs multiplication
region is also very thin, the transit time of charge carriers is very short
and bandwidths of 25 GHz have been obtained. The gain-bandwidth
product is more than 5. For a diode of length 20 m, the response
coefficient at M = 1 is 0.9 A/W at 1.55 m for a fiber-diode coupling
factor of 80% [COH 97].
3.11.3. Low-noise APDs with a very thin multiplication region
The study of APDs with a very thin multiplication region (0.2 to
0.5 m) has demonstrated the possibility of making devices with a
low excess noise factor out of materials such as InP, GaAs, InAlAs
whose ionization coefficients are similar, even in the case of a primary
injection by the least ionizing charge carriers. Under such conditions,
McIntyres analysis is not relevant because the dead space, the
distance required for a charge carrier to acquire enough energy to
ionize, is no longer negligible compared to the thickness of the
multiplication region. Consequently, the ionization process is less
random, and as a result there is less noise.
Excess noise factors comparable to those measured for multi quantum
well APDs (F ~ 4 at M= 10) have been published for traditional [DRI 99]
and waveguide [KIN 00] In
0 53
Ga
0.47
As/In
0.52
Al
0.48
As APDs with a high
bandwidth (~20 to 30 GHz), whose multiplication region in the high-
electric-field monolithic InAlAs material is separated from the low-
field InGaAs absorption region by a very thin but heavily doped InP
layer. The primary injection of electrons makes the presence of
quaternary transition layers unnecessary. This makes such structures
very attractive for high-bitrate applications as they are simpler to
fabricate.
Additionally, a record gain-bandwidth value of 290 GHz was
obtained for InAlAs/InGaAs APDs with a vertical structure
incorporating a resonant cavity centered at O = 1.55 m, a high
response coefficient (above 0.8 A/W) and a weak dark current for a
thickness of 0.2 m for the InAlAs multiplication region [LEN 99].
However, these APDs require an excellent understanding of epitaxial
growth in order to achieve stacking of the mirror layers and the
active region, and are much more delicate to fabricate than the
traditional structure in [DRI 99].
3.12. Conclusion
Initially responding to the demands of military applications, the
development of avalanche photodiodes has occurred in parallel to that
of optical fiber telecommunications, with one not being possible
without the other. The result is that for each generation of fibers
corresponding to a spectral transmission window of silica, 0.85 m,
1.3 m and 1.55 m respectively, there is a specific APD associated,
be it P
+
SPN
+
Si APDs, Ge APDs, and then SAGM InGaAs/InP APDs,
tailored to the relevant emitter. The perfection of high-performance
APDs has been one of the important factors in the achievement of
competitive optical communications. An increase in the sensitivity of
the photoreceiver, consisting of the APD combined with an associated
amplifier, allows either a reduction in the power of the signal emitted
in the first place or a reduction in the number of repeaters along the
link.
In addition, the high level of reliability of SAGM InGaAs/InP
APDs has contributed to the expansion of submarine optical fiber
links over the last decade, which has seen the installation of many
intercontinental links at higher and higher bitrates (5 to 10 Gbit/s)
with shorter and shorter latencies. Traditional SAGM InGaAs/InP
APDs have met, and continue to meet, the needs of designers up to
5 Gbit/s and even 10 Gbit/s. However, the advent of optical links at
bitrates of 20 or 40 Gbit/s and beyond leaves no question that there is
no current APD structure which is really applicable, and further
research is required.
3.13. Bibliography
[ALA 74] ALABEDRA R., MAILLE C., LECOY G., RIPOCHE G., Photodiode Schottky
SiPt-Si N avalanche, ESSDERC, 1974.
[CHI 80] CHIN R., HOLONYAK N., STILLMAN G., TANG J., HESS K., Impact ionisation
in multilayered heterojunction structures, Electronics Letters, vol. 16, no. 12,
p. 467, 1980.
[COH 97] COHEN-JONATHAN C., GIRAUDET L., BONZO A., PRASEUTH J.P.,
Waveguide AlInAs/GaAlInAs avalanche photodiode with a gain-bandwidth
product over 160 GHz, Electronics Letters, vol. 33, no. 17, p. 1492, 1997.
[DRI 99] DRIES J., THOMSON K., FORREST S., In
0,53
Ga
0,47
As/In
0,52
Al
0,48
As separate
absorption, charge, and multiplication layer long wavelength avalanche
photodiode, Electronics Letters, vol. 35, no. 4, p. 334, 1999.
[EMM 67] EMMONS R., Avalanche photodiode frequency response, Journal of
Applied Physics, vol. 38, no. 9, p. 3705, 1967.
[HAR 91a] HARARI J., Etudes thoriques et exprimentales de photodiodes
avalanche planaires GaInAs-InP, PhD Thesis, Lille Flandre Artois University of
Science and Technology, 1991.
[HAR 91b] HARARI J., DECOSTER D., VILCOT J-P., KRAMER B., OGUEY C., SALSAC P.,
RIPOCHE G., Numerical simulation of avalanche photodiodes with guard ring,
IEEE Proceedings, vol. 138, no. 3, p. 211, 1991.
[HAW 97] HAWKINS A., WU W., ABRAHAMP., STREUBEL K., BOWERS J., High gain-
bandwidth product silicon heterointerface photodetector, Applied Physics
Letters, vol. 70, no. 3, p. 303, 1997.
[HIL 81] HILDEBRAND O., KUEBART W., BENZ K., PILKUHN M., Ga1-xAlxSb
avalanche photodiodes: resonant impact ionization with very high ratio of
ionization coefficients, IEEE Journal of Quantum Electronics, vol. QE-17, no. 2,
p. 284, 1981.
[INT 66] MCINTYRE R., Multiplication noise in uniform avalanche diodes, IEEE
Trans. Electron, Devices, vol. ED-13, no. 1, p. 987, 1966.
[JOH 64] JOHNSON K., High speed photodiode signal enhancement at avalanche
breakdown voltage, Digest of Technical Papers, International Solid State
Circuits Conference, vol. 7, p. 64, 1964.
[KAG 89] KAGAWA T., IWAMURA H., MIKAMI O., Dependence of the GaAs/AlGaAs
superlattice ionization rate of Al content, Applied Physics Letters, vol. 54, no. 1,
p. 33, 1989.
[KAN 78] KANEDA T., KAGAWA S., YAMAOKA T., NISHI H., INADA T., Low noise
avalanche photodiodes by channeling of 800 keV boron into <110> silicon,
Journal of Applied Physics, vol. 49, no. 12, p. 6199, 1978.
[KAT 90] KATSUWAKA H., MIKAWA T., MIURA S., YASUOKA N., TANAHASHI T.,
WADA O., An AlxGa1-xSb avalanche photodiode with a gain bandwidth product
of 90 GHz, IEEE Photonics Technology Letters, vol. 2, no. 1, p. 54, 1990.
[KAY 53] MCKAY K., MCAFEE K., Electron multiplication in silicon and
germanium, Physical Review, vol. 91, no. 5, p. 1079, 1953.
[KIN 00] KINSEY G., HANSING C., HOLMES A., STREETMAN B., CAMPBELL J., DENTAI
A., Waveguide In
0,53
Ga
0,47
As/In
0,48
Al
0,48
As avalanche photodiode, IEEE
Photonics Technology Letters, vol. 12, no. 4, p. 416, 2000.
[LEC 75] LECROSNIER D., PELOUS G., AMOUROUX C., BRILMAN M., RIPOCHE G.,
Optimization of avalanche silicon photodiodes: a new structure, Technical Digest
of International Electron Devices Meeting, Washington D.C., p. 595, 1975.
[LEN 99] LENOX C., NIE H., YUAN P., KINSEY G., HOMLES A., STREETMAN B.,
CAMPBELL J., Resonant cavity InGaAs- InAlAs avalanche photodiodes with
gain-bandwidth product of 290 GHz, IEEE Photonics Technology Letters, vol.
11, no. 9, p. 1162, 1999.
[MAI 80] MAILLE C., Modlisation lobscurit et sous clairement des
photodtecteurs avalanche N
+
SPSP au silicium, PhD Thesis, Languedoc
University of Science and Technology, Montpellier, July 1980.
[MEL 66] MELCHIOR H., LYNCH W., Signal and noise response of high speed
germanium avalanche photodiodes, IEEE Transactions on Electron Devices, vol.
ED-13, no. 12, p. 829, 1966.
[MIL 55] MILLER S., Avalanche breakdown in germanium, Physical Review, vol.
99, no. 4, p. 1234, 1955.
[ORS 87] ORSAL B., ALABEDRA R., VALENZA M., LECOY G., MESLAGE J.,
BOISROBERT C., Les photodiodes avalanche Hg
0,4
Cd
0,6
Te O = 1,55 Pm. Bruit
prs de la rsonance lie au couplage spin-orbite, Revue de Physique Applique,
vol. 22, p. 227, 1987.
[RIP 83] RIPOCHE G., PEYRE J.-L., LAMBERT M., MOTTET S., High speed ( 6 GHz)
InGaAs/InP avalanche photodiodes grown by gas source molecular beam epitaxy
with a thin quaternary grading layer for high bit rate (t 5 Gbit/s) systems,
Journal de Physique, III, vol. 3, p. 1761, September 1983.
[RUE 66] RUEGG H., A fast high gain silicon photodiode, Digest of Technical
Papers, International Solid State Circuits Conference, vol. 9, p. 56, 1966.
[SHI 83] SHIRAI T., MIKAWA T., KANEDA T., MIYAUCHI A., InGaAs avalanche
photodiodes for 1 m wavelength region, Electronics Letters, vol. 19, no. 14,
p. 534, 1983.
[STI 74] STILLMAN G.,WOLFE C., ROSSI D., DONNELLY J., Electroabsorption
avalanche photodiode, Applied Physics Letters, vol. 25, no. 11, p. 671, 1974.
[SZE 66] SZE S., GIBBONS G., Effect of junction curvature on breakdown voltage in
semiconductors, Solid State Electronics, vol. 9, p. 831, 1966.
[WAT 97] WATANABE I., NAKATA T., TSUJI M., MAKITA K., TAGUCHI K., High
reliability and low dark current 10 Gb/s planar superlattice avalanche
photodiodes, IEEE Photonics Technology Letters, vol. 9, no. 12, p. 1619, 1997.
[WEB 88] WEBB P., MCINTYRE R., SCHEIBLING J., HOLUNGA M., Planar InGaAs/InP
avalanche photodiode fabrication using vapor phase epitaxy and silicon
implantation techniques, Technical Digest of Optical Fiber Communications
Conference, OFC 88, 1988.
[YAM 76] YAMAOKA T., MATSUMOTO H., KANEDA T., Silicon avalanche photodiodes,
Fujitsu Scientific & Technical Journal, vol. 12, p. 87, September 1976.
Chapter 4
Phototransistors
4.1. Introduction
The effects of light on transistors have been studied since
transistors were first created, with the main motivation for this
research being to produce a combination of photodetection and signal
amplification in single device. The first work on phototransistors dates
from 1951 when Shockley et al. [SHO 51] proposed the use of a
bipolar n-p-n or p-n-p structure as a phototransistor operating with a
base current generated by optical means. The first demonstration of
this type of photodetector was achieved two years later when Shive
[SHI 53] described a Ge-based n-p-n phototransistor, operating in the
spectral region of 1.2 m wavelength, with an optical gain of the order
of 100. In the 1960s, it was mostly bipolar Si phototransistors,
operating in the near-infrared, which were developed. However, since
the optical absorption coefficient of silicon is weak (<10
3
cm
1
), the
foreseeable applications were not those requiring high speeds.
This interest in phototransistors has recently been revived with the
development of optical fiber transmission systems and the evolution
of the III-V materials technology. This is because, on the one hand,
Chapter written by Carmen GONZALEZ and Antoine MARTY.
the best attenuation and absorption properties of optical fibers are
found in the near-infrared, with the three minimal absorption windows
at 0.85, 1.3 and 1.55 m corresponding to the emission and absorption
ranges of semiconductor materials based on GaAs and InP. On the
other hand, III-V materials have absorption coefficients much higher
than that of silicon (>10
4
cm
1
). These two characteristics have
opened up the possibility of applying phototransistors based on III-V
materials to the field of optical telecommunications. Heterojunction
bipolar transistors and field effect transistors, made with these
materials, began to be studied as phototransistors in the 1970s [ALF
73, BAA 77].
In section 4.2, we present a summary of the different types of
phototransistors based on the materials used for their fabrication and
as a result their sensitivity to optical wavelengths. We will then
classify them according to their structure. In section 4.3 we describe
the mechanisms of operation of a bipolar phototransistor and its main
properties. Section 4.4 is dedicated to some examples of circuits based
around bipolar phototransistors. Finally, in section 4.5, we review the
main fields of application of this device.
4.2. Phototransistors
In terms of photodetectors, phototransistors can be classed according
to their fabrication material and, just like their transistor analogs, they
can also be sorted according to structure into two categories: unipolar
field effect devices and bipolar devices. This section reflects this dual
classification.
4.2.1. Phototransistors according to their fabrication materials
The first phototransistors were based on a bipolar homojunction
silicon transistor. In this indirect bandgap material, the energy of the
bandgap is 1.12 eV. As a result, silicon is photosensitive to
wavelengths in the near-infrared, 0.60.8 m. Later, starting in the
1970s, technological progress achieved with III-V compounds and
Phototransistors 113
their ternary and quaternary alloys enabled the development of
heterojunction devices: field effect transistors and bipolar transistors.
These materials have a direct band structure, and some of their ternary
compounds have bandgap energies which match the spectral windows
of lowest attenuation in optical fibers: 0.85, 1.3 and 1.55 m. This is
why two variants based on III-V materials have been developed: the
GaAs variant (for O = 0.85 m) and more recently the InP variant (for
O = 1.3 and 1.55 m). Heterojunction phototransistors based on
AlGaAs/GaAs are photosensitive to wavelengths between 0.8 and
0.9 m, with the GaAs bandgap being 1.43 eV. This value of bandgap
is compatible with the spectral window of optical fibers around 0.85
m, for which there is an attenuation of around 2 dB/km. The optical
absorption coefficient of GaAs for this wavelength is relatively high:
of the order of 10
4
cm
-1
. The first optical systems, operating initially
at 0.85 m, were later targeted at 1.3 and 1.55 m where the
attenuation in optical fibers is 0.5 and 0.2 dB/km respectively.
Thus, it is the ternary and quaternary alloys based on InP which are
best suited for photodetection at these wavelengths. In particular, the
ternary alloy In
x
Ga
1x
As with x = 0.53, which matches the lattice
spacing of InP, has a bandgap energy of 0.75 eV. This value of energy
is compatible with photodetection at 1.3 and 1.55 m. The optical
absorption coefficient of InGaAs is 1.16 u 10
4
cm
1
for O = 1.3 m
and 6.8 u 10
3
cm
-1
for O = 1.55 m. The bandgap energy (E
g
) and the
applicable spectral range for the main materials used in
phototransistor fabrication are shown in Table 4.1.
Table 4.1. Bandgap energy and applicable spectral range for various
materials used in the construction of phototransistors
4.2.2. Phototransistors classified by structure
Homojunction or heterojunction bipolar transistors (photo-HBTs)
and unipolar field effect transistors (photo-FETs) are three-terminal
devices. When illuminated, the optical input acts as an additional
terminal across which the device can be controlled optically. These
devices can be integrated into MMICs (Monolithic Microwave
Integrated Circuits) to achieve optically-controlled amplification or
switching, or even more sophisticated microwave functions such as
optical locking of oscillators and the mixing of optical and electrical
signals [SEE 90].
4.2.2.1. Unipolar field effect transistors
Unipolar field effect transistors (FETs), based on III-V materials,
only create a single type of charge carrier. Out of the family of field
effect transistors, most research has been focused on heterojunction
phototransistors which are Schottky gate transistors made with GaAs
(MESFETs, Metal-Semiconductor Field Effect Transistors) and high
electron mobility transistors (HEMT). This popularity is mostly due to
the possibilities of using the phototransistor effect in MMICs
(Monolithic Microwave Integrated Circuits).
Figure 4.1 shows the cross-section of an illuminated MESFET
device (photo-MESFET). Two distinct phenomena occur in the photo-
MESFET during illumination by light: a photoconductor effect which
is the result of the increase in conductivity due to the creation of
photocarriers, and a photovoltaic effect which occurs close to the
gate/channel and channel/substrate junctions. Several investigations
[GAU 85, MAD 92] have shown not only that the photovoltaic effect
dominates the photoresponse of the device, but also that this effect is
the cause of the mediocre dynamic performance of photo-MOSFETs.
Physically, we can explain this degradation by the poor coupling
between the incident light and the gate junction (poor overlap between
the optical absorption region and the region with an electric field). A
large proportion of the photocarriers end up trapped in low field
regions (e.g. the barrier), thus distorting the static characteristics of the
device.
Compared to a photo-MESFET without illumination, an important
reduction in performance has been observed for the transition
frequency f
T
and the maximum frequency of oscillation f
MAX
when
illuminated [SIM 86]. The best dynamic performance obtained under
illumination is for a bandwidth less than 100 MHz [BAR 97].
Dynamic optical performances are very poor compared to those
obtained electrically.
Figure 4.1. Structure of a GaAs MESFET under illumination. The channel is an
n-type semiconducting region sandwiched between a semi-insulating substrate and
the space charge region (depletion region) of the reverse-biased Schottky junction
MESFET-based phototransistors have shown themselves to be a
poor approach for developing photodetectors in the microwave and
millimeter range. Other devices in the FET family, notably HEMT,
suffer from the same limitation [ROM 96]
4.2.2.2. Bipolar phototransistors
An alternative to the photo-MOSFET is the silicon-based
homojunction bipolar phototransistor or the heterojunction bipolar
phototransistor (based on GaAs or InP). In the bipolar phototransistor,
there is an overlap between the region of light absorption and the
high-electric-field depletion region. This overlap is as good as in PIN
photodiodes. As will be shown in section 4.3, the frequency response
to a modulated optical signal is directly related to its purely electrical
properties. To make the best use of the performance of a transistor,
different illumination approaches have been considered, such as
vertical illumination through the front or rear face, lateral illumination,
or even the use of a waveguide.
4.2.2.2.1. Traditional surface illumination
The first phototransistors were surface-illuminated, which is the
traditional situation, particularly for silicon phototransistors. This
vertical illumination allows easy coupling between the optical fiber
and the phototransistor, making the integration of the device easier.
Several solutions have been proposed: in the case of an opening in the
metallic contact of the emitter, the incident light flux crosses the
emitter without being absorbed, and the electron-hole pairs are created
in the active (or intrinsic) base-collector region of the photo-HBT, that
is, the region which is just below the emitter. If a base contact is
removed, as shown in Figure 4.2a, the optical flux directly illuminates
the base and the collector in the extrinsic region of the
phototransistor.
4.2.2.2.2. Rear-face illumination
The aim of rear-face illumination of a transistor is the creation of
photocarriers in the active region of the transistor (see Figure 4.2b)
without modifying the emitter contact. The traditional HBT structure
is used with a substrate which does not absorb the incident light. The
absorption of light takes place in the collector and the base.
Furthermore, the response coefficient is improved because the
metallic contact of the emitter acts as a mirror.
Figure 4.2. Cross-section of a HBT phototransistor with
(a) front-face illumination and (b) rear-face illumination
4.2.2.2.3. Lateral illumination
Lateral illumination (parallel to the layers) offers another approach
for the injection of light into a transistor. It is inspired by the
technological processes developed for photodiodes. This type of
illumination requires a face very vertical to the base-collector island,
in order to obtain the best possible injection efficiency. In laterally-
illuminated phototransistors, the photons and charge carries no longer
propagate in the same direction, as is the case of vertically illuminated
photo-HBTs. The aim of lateral illumination is to simultaneously
improve the conversion of optical power into electrical power
(quantum efficiency) and the speed performance (high operating
frequencies) of the device.
4.2.2.2.4. Lateral illumination with an integrated waveguide
However, lateral illumination does pose significant difficulties in
terms of achieving a good optical coupling between the fiber and the
device. This is why one trick involves the fabrication of a waveguide
integrated into the structure of the device, as shown in Figure 4.3.
With this approach, the matching of the mode leaving the optical fiber
to the propagation mode inside the device is made easier [FRE 96]. In
the best cases, an improvement to the injection efficiency is observed
which rises from 50% to 90%, with an improved frequency response.
4.3. The bipolar phototransistor: description and principles of
operation
The bipolar phototransistor is a transistor designed such that the
signal current which feeds the base terminal is mostly provided by
photoelectric effects. This signal is then amplified by the transistor
effect of the device. First we revisit the principle of operation of a
bipolar n-p-n phototransistor, which consists of two distinct p-n
junctions, as shown in Figure 4.4. Then, we shall discuss the parameters
characterizing the phototransistor: the response coefficient, the static
and dynamic gains, the response time, the conversion gain and the noise.
A cross-section of an n-p-n type bipolar phototransistor, using mesa
technology, is shown in Figure 4.4, along with its bias circuit.
Figure 4.3. Cross-section of an integrated waveguide
phototransistor under illumination
Figure 4.4. Schematic diagram of an n-p-n phototransistor, along with its bias circuit
in common-emitter configuration. The shaded regions are free from mobile charges.
The base, collector and sub-collector parts can be thought of as a PIN photodiode
4.3.1. The phototransistor effect
In Figure 4.4, the n-p-n phototransistor is biased in the common-
emitter configuration, with voltages V
BE
> 0 and V
CE
> 0. It differs
from traditional bipolar structures, having a relatively large side area
of the base-collector junction; this is called the optical window and is
what becomes illuminated. This part of the device is effectively a
photodiode, connected between the collector and base contacts of the
active transistor. When illuminated, electron-hole pairs are created by
the photoelectric effect, in the base and in the space charge region
(SCR) associated with the base-collector junction (we assume for the
sake of simplicity that no absorption takes place in the sub-collector
and that the collector is entirely free of mobile charges). A
photocurrent I
, known as the primary photocurrent, is established

between the base and collector regions; it is transported by minority
electrons from the base which diffuse towards the collector, and by the
carriers generated in the SCR, which are separated and moved by the
electric field. The flow of this photocurrent I
generates a voltage
across the base-emitter and base-collector junctions such that the
transistor finds itself in its normal operating system (V
BE
> 0 and
V
CE
> 0).
The holes attracted by the base will therefore find themselves
blocked by the emitter-base junction. This excess of holes will cause a
reduction in the emitter-base potential barrier, which results in an
injection of electrons from the emitter into the base, from where the
majority will diffuse until they are at the level of the collector. Thus,
this is the traditional behavior of a bipolar transistor. The
amplification of the photocurrent is a purely electrical phenomenon
due to the transistor effect.
4.3.1.1. The main currents in the phototransistor
In the normal operating system, the phototransistor can be
characterized in terms of the following currents (see Figure 4.5):
Firstly, the emitter injects electrons into the base region. These,
minority carriers in this region, diffuse perpendicular to the junction
layout and, if the base is thin enough that recombination can be
ignored, they reach the depletion region of the base-collector junction,
where the high electric field present in this region clears them out
towards the collector region. The flow of these charge carriers gives
the contribution I
ne
, which is shown in Figure 4.5.
Conversely, a current I
pe
of holes, majority carriers in the base, is
injected from the base towards the emitter.
Generation-recombination phenomena mostly occur at the level
of the emitter-base junction, I
reb
, and in the base, I
rb
. I
reb
comes from
the recombination of electrons in the SCR of the emitter-base
junction. I
rb
is caused by the recombination of electrons with holes,
majority carriers, in the base.
Illumination produces the primary photocurrent I
. Mostly created
in the SCR of the base-collector junction, I
originates from a current

of electrons which migrate directly toward the collector contact, and
from a current of holes which accumulate at the level of the base.
Thus, balancing the different contributions to the current crossing
the two junctions, listed above, allows us to calculate the total currents
at the emitter, collector and base; in this way, we can use the continuity
of the electron and hole currents across the transition regions to write
the following equations. For the base:
I
E
= I
ne
I
pe
I
reb
[4.1]
the emitter:
I
B
= I
pe
I
rb
I
reb
I
)
[4.2]
the collector:
I
C
= I
ne
I
rb
I
)
[4.3]
Figure 4.5. Distribution of the different currents of electrical and optical origin
in the phototransistor. The SCRs around the emitter-base
and base-collector junctions are shown by dotted lines
Equations [4.1], [4.2] and [4.3] will be used in the following
sections to define the optical gain and the electrical gain of the
phototransistor.
4.3.1.2. Injection efficiency from the emitter in a homojunction
and a heterojunction
The standard parameters for a bipolar transistor, the injection
efficiency from the emitter J, and the transport factor in the base D
T
,
can be extended to the phototransistor with the help of equations [4.1],
[4.2] and [4.3] where the photocurrent term I
was introduced. The

injection efficiency J relates the ratio of the electron current I
ne
injected into the base to the total emitter current I
E
, by the following
equation:
J =
I
ne
I
ne
I
pe
I
reb
[4.4]
Recombination currents reduce the injection efficiency and they
must be minimized. If in equation [4.4] we ignore the effect of
recombination in the emitter-base SCR, we obtain:
J =
1
1
I
pe
I
ne
[4.5]
This ratio must be as close as possible to 1 (I
pe
<< I
ne
) in order to
achieve the maximum injection efficiency. Starting from the transport
equations established for a bipolar phototransistor, it has been shown
that the injection efficiency closely follows the following expression
[CAM 85, MOR 72]:
J =
A.J
0
1 AJ
0
[4.6]
where A is a constant, a function of the thickness of the base and the
diffusion length of electrons in the base, and J
0
is the Kroemer factor
[KRO 57a and b] established for an emitter-base heterojunction. The
Kroemer factor is a function of the physical parameters of the
materials making up the heterojunction, and can be expressed in the
following way:
J
0
=
D
nb
L
pe
n
e
D
pe
L
nb
p
b

m
nb
*
m
pb
*
m
ne
*
m
pe
*
(
(
)
)
3 2
exp
'E
g
kT
(
(
)
)
[4.7]
with:
'E
g
the difference in bandgap between the emitter and the base;
D
nb
the diffusion coefficient of electrons in the base;
D
pe
the diffusion coefficient for holes in the emitter;
L
pe
the diffusion length for holes in the emitter;
L
nb
the diffusion length for electrons in the base;
n
e
the electron density in the emitter;
p
b
the density of holes in the base;
m
*
nb
and m
*
pb
the effective masses of the electrons and holes in
the base;
m
*
ne
and m
*
pe
the effective masses of the electrons and holes in
the emitter.
For a homojunction bipolar transistor, there is no variation in the
bandgap between the emitter and the base, 'E
g
is zero, and the
exponential factor in equation [4.7] is equal to 1. It is clear from [4.6]
and [4.7] that to obtain a high injection efficiency, J 1, it is crucial
that the emitter should be much more heavily doped than the base.
Conversely, for a heterojunction, J depends mostly on 'E
g
, and the
term exp('E
g
/kT) becomes dominant compared to the n
e
/p
b
ratio; thus,
to obtain a J close to 1, it is no longer necessary to under-dope the
base relative to the emitter and/or overdope the emitter relative to the
base. In heterojunctions based on the materials GaAs/AlGaAs and
InGaAs/InP, the size of the emitter bandgap (AlGaAs and InP) is more
important than that of the base (GaAs and InGaAs), thus offering a
reduction in the injection of majority carriers from the base into the
emitter. For the heterojunction phototransistor, the base can be doped
at high levels without compromising the efficiency of the junction,
leading to a reduction in the resistance of the base. The doping of the
emitter can remain within relatively low limits, thus reducing the
capacitance of the emitter. These two effects combine to give an
improvement in the current gain and an increase in the high frequency
performance of the heterojunction bipolar phototransistor.
4.3.1.3. Transport factor in the base
The transport factor in the base B is defined by the ratio between
the electron current gathered by the collector and the electron current
injected from the emitted into the base:
B =
I
nc
I
ne
=
I
ne
I
rb
I
ne
[4.8]
Due to the recombination current, B is always less than 1.
Beginning with equation [4.8], B can be expressed as a function of the
transit time in the base t
B
and the lifetime of electrons in the base W
n
using the following equation [CAM 85, CAS 89, POU 94]:
B =1
t
B
W
n
[4.9]
According to [4.9], B is closer to 1 when the transit time t
B
is small
compared to the electron lifetime W
n
. The transit time t
B
is smaller
when the base thickness is small. As a result, it is necessary for the
base thickness to be smaller than the diffusion length of the electrons.
For a phototransistor, as for a transistor, the base should therefore be
as thin as possible.
4.3.2. The response coefficient of a phototransistor
The response coefficient of the photodiode base-collector part of
the phototransistor is defined as the ration of the primary photocurrent
I
to the received optical power P

opt
:
S
0
=
I
)
P
opt
A W
( )
[4.10]
The quantum efficiency, which is the ratio of the number of
electrons collected to the number of incident photons, is also used to
characterize the optical-electrical conversion of the base-collector
photodiode. This efficiency, often called the external quantum
efficiency, is expressed in the following way:
K =
I
)
q
P
opt
hQ
=
hQ
q
.S
0
[4.11]
and the response coefficient of the phototransistor is characterized by
the ratio of the component, due to optical excitation (I
C
)
opt
of the
current leaving the device I
C
to that same incident optical power:
S =
I
C
( )
opt
P
opt
=
I
C
( )
opt
I
)
I
)
P
opt
A W
( )
[4.12]
As we will see in the following section, the ratio (I
C
)
opt
/I
defines
the optical gain of the phototransistor.
Taking into account equations [4.11] and [4.12] we can express
(I
C
)
opt
in the following way:
I
C
( )
opt
=
q
hQ
K G
opt
P
opt
=G
opt
S
0
P
opt
[4.13]
This last equation clearly shows that the phototransistor has an
effective response coefficient G
opt
times greater than that associated
with the base-collector photodiode.
4.3.3. Static electrical and optical gains of the phototransistor
4.3.3.1. Static electrical gains E
0
and D
0
In the static system, the current gain of the phototransistor is
obtained without illumination and, as a consequence, it is defined in
the same way as for a bipolar transistor. In the common-emitter
configuration, the static electrical gain E
0
is given by:
E
0
=
I
C
I
B
=
I
ne
I
rb
I
pe
I
rb
[4.14]
This gain can also be expressed as a function of the injection
efficiency J and of the transport factor in the base B taking into
account equations [4.5], [4.8] and [4.14]:
E
0
=
BJ
1BJ
[4.15]
Because of this dependence on J, the current gain in the common-
emitter configuration, E
0
, can reach very high values for the
heterojunction bipolar transistor. We also define the current gain in
common-base configuration D
0
, which is the ratio of the collector
current to the emitter current. Its relationship with the gain E
0
is:
D
0
=
I
C
I
E
=
E
0
1E
0
[4.16]
4.3.3.2. Static optical gain G
opt
The static optical gain of a phototransistor is obtained under
continuous illumination. It is defined as the ratio between the
component linked to the collector current (I
C
)
opt
and the primary
photocurrent I
)
, obtained at the level of the base-collector photodiode:
G
opt
=
I
C
( )
opt
I
)
=
hQ
qK

I
C
( )
opt
P
opt
[4.17]
G
opt
is the equivalent of the electrical gain E
0
under illumination. It
links the gain obtained through amplification, due to the transistor
effect, to the primary photocurrent I
. Several authors have shown

that, for a bipolar transistor, G
opt
is proportional to the product of the
external quantum efficiency and the electrical current gain [CAM 85,
CHA 85]:
G
opt
=K 1E
0
( )
[4.18]
Equation [4.18] shows that to achieve a given value of G
opt
, it is
necessary to establish a compromise between the quantum efficiency
K and the electrical gain E
0
, depending on the intended application.
The result is that the larger the value of K, the smaller the value of E
0
,
and vice versa. In addition, G
opt
is always smaller than E
0
.
4.3.4. Dynamic characteristics of phototransistors
In the dynamic system, just as for a normal transistor, the
phototransistor effect can be put to good use in the quasi-linear
small-signal mode, in the non-linear mode to achieve multiplication
and mixing behavior and in the large signal switching system. In
these three cases, the illumination varies with time and we are
interested either in (amplitude) modulated light or in an impulse.
These three dynamic modes of operation of a phototransistor will be
analyzed below.
4.3.4.1. Small-signal operation
In the context of small-signal applications, the excitation by light
may correspond directly to the signal to be amplified, or alternatively
it may include a continuous component (modulated signals) which can
be put to the particular use of ensuring a pre-bias in addition to that
obtained by electrical access to the base. Phototransistors whose base
is electrically accessible have the advantage of allowing, by selecting
the bias to the base, the choice of an operating point which ensures
optimal linearity. The global response time of a phototransistor is no
different to the electrical response time of the transistor element
triggered by the current I
induced in the base-collector photodiode.

Two time constants are therefore associated with the dynamic
operating system of a phototransistor: the first is due to the intrinsic
transport of photocurrent carriers, while the second is due to the
electrical response of the phototransistor. The first time constant may
be more or less important depending on the geometry of the device;
however, to begin with, we can presume that it is the electrical
response time which is most significant. The electrical response time
of a phototransistor is linked to its transition frequency f
T0
, which is
the frequency at which the optical gain is equal to 1. In an equivalent
manner to a transistor, this sets the limit for the use of a
phototransistor as a photocurrent amplifier. This transition frequency
can be calculated in a similar way to the calculation of the equivalent
small-signal method for the hybrid configuration.
4.3.4.1.1. Transition frequency calculated using the hybrid model
The behavior of a phototransistor in the small-signal system is
shown in the equivalent circuit diagram of Figure 4.6b. This is based
on an electrical model of a bipolar transistor known as the Giacoletto
circuit, which is established for the common-emitter arrangement
[LET 78]. The incident illumination is modeled by a photocurrent
source i
placed between the base B and the collector C. The output

signal is taken across the terminals of a load resistance R
L
connected
to the collector.
Figure 4.6. (a) Electrical model and (b) its equivalent electrical circuit in the small-
signal system of a phototransistor, for a common-emitter configuration
The elements of the above equivalent circuit can be described as
follows:
R
BE
is the dynamic input resistance of the phototransistor;
R
BE
= E
0
U
T
/I
C
;
C
TBE
and C
TBC
are the transition capacitances of the emitter-base
and base-collector junctions;
C
SBE
is the diffusion capacitance, which represents the effects of
accumulation of minority charge carriers in the base, effects linked to
the transit time W
B
in this region, C
SBE
= W
B
.I
C
/U
T
;
g
m
defines the transconductance, g
m
= I
C
/U
T
.
Under these conditions the input photocurrent i
, called the
primary, can be expressed in the following form:
i
)
=v
BE

1
R
BE
jZ C
TBE
C
SBE
( )
l
l
l
l
l
v
BE
R
L
i
C
l
l
l
jZ C
TBC
[4.19]
in which the second term (corresponding to the current component i
indicated in Figure 4.6) represents the internal reaction mechanism
induced by the transition capacitance of the base-collector junction
C
TBC
and by the load resistance R
L
, commonly referred to as the
Miller effect. If we now ignore the current contributions i and i
compared with the amplified current g
m
v
BE
, the output collector
current is found to be:
i
C
= g
m
v
BE
[4.20]
Given this, the ratio i
C
/i
, which defines the dynamic current gain

of the phototransistor g
opt
, known as the optical gain in the small-
signal system, can be written in the form:
i
C
i
)
jZ
( )
= g
opt
jZ
( )
=
E
0
1 j Z E
0

C
TBE
C
TBC
g
m
W
B
R
C
C
TBC
(
(
)
)
[4.21]
At high frequencies, the imaginary part of the denominator of
equation [4.21] becomes dominant and we can write:
g
opt
jZ
( )
=
1
jZ
C
TBE
C
TBC
g
m
W
B
R
L
C
TBC
(
(
)
)
[4.22]
If we assume that the product R
C
.C
TBC
is weak compared to the
two first terms, |g
opt
| is obtained when:
1
Z
=
1
Z
TO
=
C
TBE
C
TBC
g
m
W
B
[4.23]
Equation [4.23] defines the angular frequency of the transition Z
T0
,
with the transition frequency f
T0
being Z
T0
/2S. Physically, the
transition frequency, f
T0
, corresponds to the total transit time, W
T
, of
photocarriers from the emitter to the collector. It is interesting to
analyze this time constant W
T
, associated with f
T0
. It is defined as:
W
T
=
1
Z
TO
=
1
2S f
TO
[4.24]
If we substitute equation [4.23] into [4.24] we find:
W
T
=
C
TBE
g
m

C
TBC
g
m
W
B
[4.25]
The first two terms of equation [4.24] correspond to the charging
times of the base-emitter and base-collector junctions respectively.
Equation [4.25] shows that W
T
depends on the collector current I
C
(via
g
m
) up to a certain threshold and, if we take into account equation
[4.13], W
T
is inversely proportional to the incident optical power P
opt
.
For high values of P
opt
, W
T
approaches the limiting value W
B
. On the
other hand, at low values of P
opt
, the terms in C
TBE
and C
TBC
become
dominant and the transition frequency f
T0
is proportional to P
opt
.
This behavior of f
T0
as a function of the incident optical power has
been experimentally observed, as shown in Figure 4.7. This
dependence makes the use of relatively high optical powers (>10 W)
necessary. Alternatively, we can use an electrical current via the
metallic base contact to pre-bias the structure, thus reducing the
threshold optical power needed to reach the maximum value of f
T0
.
With the help of equation [4.22] we can also define the cutoff
angular frequency Z
E
as being the angular frequency at which |g
opt
|
falls by 3 dB. This is equal to:
1
Z
B
= E
0
C
TBE
C
TBC
g
m
W
B
R
L
C
TBC
(
(
)
)
[4.26]
Figure 4.7. Variation in transition frequency f
TO
with incident optical power P
opt
.
At weak P
opt
, f
TO
varies rapidly up to a certain limiting value of
P
opt
, after which it remains constant
Figure 4.8. Frequency response of a phototransistor for different load resistances
It can be clearly seen from equation [4.26] that the cutoff angular
frequency Z
E
depends on the operating point (as does Z
TO
) and on the
value of the load resistance R
C
, as shown in Figure 4.8.
4.3.4.2. Nonlinear operation
A bipolar phototransistor behaves similarly to a bipolar transistor,
and as a result possesses nonlinear characteristics which allow it to act
as an optical-electrical mixer. It can mix a modulated optical signal,
carrying information, with an electrical signal from a local oscillator.
This mixing behavior has been the subject of recent research [BET 98,
GON 98, SUE 96] and has been developed for telecommunication
applications. Two mixing configurations are possible:
transposition of a low frequency input signal into a higher
frequency signal. This mode of operation will be referred to below as
up-conversion;
transposition of a high frequency input signal into a lower
frequency signal. This mode will be referred to as down-conversion;
4.3.4.2.1. Mixing principles
Any device able to transpose an input signal from a frequency f
E
into another higher or lower frequency is known as a mixer. This
change in frequency originates in the nonlinear properties of the mixer
and can be explained in the following manner: consider a circuit
element whose I/V current-voltage characteristics are nonlinear (a
nonlinear resistance, a Schottky diode, etc.). If it is subjected to a
voltage V
OL
= v
OL
sin(Z
OL
t), its I/V nonlinearity can be described by
expanding the current flowing through the element in terms of a
discrete series that is a function of the voltage V:
I = I V
( )
= I
0
a
1
V a
2
V
2
a
3
V
3
... [4.27]
where I
0
is a continuous bias current and a
1
, a
2
and a
3
are real
constant coefficients. [4.27] shows that we see an infinite number of
powers of V appearing at the output of the circuit, these will enrich the
spectrum of the input signal V. If we add a second signal to the input,
V
FI
= v
FI
sin(Z
FT
t), the spectrum of the output signal becomes even
more complicated due to the presence of the products of V
OL
and V
IF
and their respective harmonics.
If we now replace V with V
OL
+ V
FI
in [4.27], the current (V)
becomes:
I =I
0
a
1
V
OL
V
FI
( )
a
2
V
OL
V
FI
( )
2
...
=I
0
a
1
v
OL
sinZ
OL
t v
FI
sinZ
FI
t
( )
a
2
v
OL
sinZ
OL
t v
FI
sinZ
FI
t
( )
2
...
=I
0
a
1
v
OL
sinZ
OL
t v
FI
sinZ
FI
t
( )
a
2
1
2
v
OL
2
1cos2Z
OL
t
( )
v
OL
v
FI
cos Z
OL
Z
FI
( )
t cos Z
OL
Z
FI
( )
t
l
l
l
'
|
1
1
+
1
1
1
2
v
FI
2
1cos2Z
FI
t
( )
'
1
1
+
1
1
...
[4.28]
The interesting frequencies are (Z
OL
+ Z
FI
) and (Z
OL
Z
FI
) and
they can be extracted by a filter tuned to the target frequency. In a
bipolar mixing transistor, the main nonlinearity contributing to the
mixing behavior is transconductance. For a bipolar phototransistor, the
nonlinearity in the current gain is the most important parameter.
4.3.4.2.2. Performance criteria
As a first approximation, we can say that the performance criteria
established for a mixing transistor can be applied to a mixing
phototransistor. We will discuss two of these criteria: conversion gain
and insulation.
Figure 4.9. Optical and electric inputs of the phototransistor
(in place of the mixing phototransistor)
Conversion gain
In a mixing phototransistor, the optical input is physically
separated from the electrical input (see Figure 4.9). The optical signal
enters through the optical window and the electrical signal is
applied to the metallic contact of the base. In the common-emitter
configuration, the mixed signal is obtained at the output of the
collector. The traditional definition of the conversion gain G
con
,
applied to a phototransistor, can be expressed as the ratio between the
output electrical power at the mixed frequency and the effective
available input optical power at the modulation frequency of the light.
Taking the case of up-conversion, the light is modulated at an
intermediate frequency F
IF
, the frequency of the local oscillator F
OL
is
much higher than F
IF
, and the mixed frequencies F
RF
are:
F
RF
= F
OL
F
FI
The conversion gain can be written as:
G
con
=
P
elec
F
RF
( )
P
opt
F
FI
( )
[4.29]
In equation [4.29], the denominator refers to an optical power. In
order to determine the electrical power generated by the incident
optical power, we must take into account the primary photocurrent I
generated by the illumination at the level of the base-collector junction,

and of the input impedance associated with this junction, Z
BC
. The
electrical power can thus be expressed by the equation:
P
elec
=
1
2
Real Z
BC
( )
I
)
2
[4.30]
and the electrical conversion gain can be expressed as:
G
con
=
P
RF
P
elec
( )
FI
=
2P
RF
Reel Z
BC
( )
I
)
2
[4.31]
Another definition of the conversion gain is possible. It is defined
as being the ratio between the power of the mixed signal F
RF
and the
primary power of the incident signal F
IF
, in other words, the power
of F
IF
obtained when the base-emitter junction of the phototransistor
is short-circuited (base-emitter voltage V
BE
= 0 V).
G
con
=
P
RF
P
FI
( )
V
BE 0
[4.32]
When describing experimental results this second definition of
G
con
is the most widely used.
Insulation
All the frequencies generated by the mixing are present at each of
the terminals of the mixer, and this has a significantly detrimental
affect on its conversion gain performance. As Figure 4.9 shows, the
phototransistor has three points of access, two electrical, the base
contact (OL) and the collector contact (RF), and one optical, the
optical window (IF). The insulation is measured between two of the
mixers points access, IF-OL and IF-RF, by the ratio between the
power at the frequency F
OL
(or F
RF
) present at the access point OL (or
RF) and the power at the frequency F
OL
(or F
RF
) present at the IF
access point. Let us consider the OL and IF access points at which the
F
OL
and F
IF
signals are injected. If we hypothesize that no electrical
signal originating from the access point of the base can be converted
into an optical signal leaving through the optical window, the IF-OL
insulation is infinite. The same reasoning can be applied to the IF-RF
access point. As a result, we obtain a maximum insulation between
these two pairs of terminals without needing to use external circuitry.
However, for insulations in the opposite direction, OL-IF and RF-IF,
there can be a loss of F
IF
signal power in each of these two electrical
access points.
4.3.4.3. Operation in the strong-signal system
As was mentioned previously, the other application of a
phototransistor concerns the detection of radiation in the strong-signal
system. In this case, the presence or not of the illumination induces a
switching type of behavior: the device initially blocks a signal, or is in
the weak conduction system if it is pre-biased. It then enters the
normal, or even saturated, system when light is applied; it will then
return to the initial mode of operation when the input signal ceases.
These system changes give rise to transitory phases, as shown in
Figure 4.10, which defines a response time for each: the lag time t
W
is
inherent to the time it takes the transistor to start conducting (in the
case of operation without pre-bias) and is linked to the charging of the
transition capacitances C
TBE
and C
TBC
. By considering the mean
values of these parameters, this time can be calculated using the
expression:
t
r
=
0.7 C
TBE
C
TBC
( )
I
)
[4.33]
Figure 4.10. Representation of the different switching
response times of phototransistors
The rise time t
r
in the normal active system (up to saturation, if
applicable) is more difficult to evaluate, bearing in mind the variation
in numerous electrical parameters that apply in this transitory system;
nevertheless, the dynamic behavior of the phototransistor can be
approached in a simple manner, using the small-signal model. If we
assume that the current gain and its cutoff frequency are independent
of the level of injection; the result is that the first-order model, which
relates I
C
to I
in the context of this assumption, suggests that the

collector current response to a step change to the primary photocurrent
in the base follows an exponential law with time constant 1/Z
E
. Thus,
the rise time which separates the start of the growth in the collector
current (conventionally, I
C
equal to 10% of its final value) from the
end of this first transient phase (I
C
equal to 90% of its steady-state
value), being either E
0
.I
or I
Csat
, can be calculated as:
t
m
=
2.9
Z
E
[4.34]
when saturation is not reached and:
t
m
=
0.8
Z
T

I
C
sat
I
)
[4.35]
in the alternative case.
For opposite switching, the desaturation time corresponds to the
removal of the accumulated surplus charge in the form of the minority
carriers (electrons) in the vicinity of the base, and is therefore defined
by the lifetime of these charge carriers.
Over the course of the decaying phase of the collector current, the
transistor is once more in the normal active system, up to the point it
stops conducting. The fall time t
f
can again be expressed by equations
[4.34] and [4.35].
,
Figure 4.11. Response time of a phototransistor as a function of the
collector current and for different values of the load resistance
Here once more, in the switching system, the rise time t
r
and the
fall time t
f
increase with the load resistance R
L
. We also observe that,
while it leads to an increase in response time, the presence of
saturation leads to a reduction in these rise and fall times. Finally, the
addition of a pre-bias also contributes to an improvement in t
r
and t
f
,
as can be seen in Figure 4.11. Overall, depending on the type of
phototransistor and the circuit used, along with its load resistance and
operating point, these response times vary from a few tenths to tens of
microseconds.
4.3.5. Noise in phototransistors
The phototransistor is the first element in a photoreceiver system.
This means that there is a certain threshold power below which the
photoreceiver cannot detect a signal: a power which is determined by
the power of the noise in the phototransistor. The minimum detectable
power in the photoreceiver system is limited by the noise in the
phototransistor and its load circuit. The noise in the phototransistor is
hence an important criterion for judging the performance of the whole
photodetection apparatus. The most commonly-used approach for
characterizing the noise of a photodetector is the noise equivalent
current generator, applied at the input [CAM 82, SMI 80, WAN 86].
This is the approach that we will develop over the rest of this section.
Four main noise sources can be associated with a HPT:
the shot noise due to the base current I
B
entering the phototransistor,
this current is the sum of the photocurrent I
and the electrical bias

current I
Belec
;
the shot noise due to the collector current I
C
, at the output of the
HPT;
the thermal noise due to the load resistance R
C
;
the thermal noise generated in the base region and, in contrast to
the other quasi-neutral regions, the base, although heavily doped,
presents an effective base resistance R
B
which can reach values as
high as hundreds of Ohms.
The spectral densities of the noise power, i
2
'f in A
2
/Hz,
associated with each source of noise are, for the base, the following:
i
base
2
'f
= 2 2q I
)
I
Belec
( ) ( )
4kT
R
BE
[4.36]
The factor of 2 in equation [4.36] is due to the correlation which
exists between the base-emitter and base-collector junctions [MON
71]. Physically, this arises from the fact that each fluctuation in the
base current 'I
B
simultaneously produces another equivalent
fluctuation in the emitter current to compensate for 'I
B
. These two
currents, in the opposite direction, cancel through recombination, but
their shot noises are independent and add up. For the collector:
i
C
2
'f
output
i
th
2
'f
output
=2qI
C

4kT
R
L
[4.37]
These two spectral densities associated with the collector at the
output of the transistor are linked to the optical input of the
phototransistor in the following manner [THU 99]:
i
C
2
'f
input
i
th
2
'f
input
=
1
g
opt
2
2qI
C

4kT
R
L
(
(
)
)
[4.38]
where g
opt
is the dynamic optical gain of the phototransistor, defined
in equation [4.21].
Some degree of correlation can exist between the sources of shot
noise in the base and collector. Nevertheless, for frequencies above
the noise cutoff frequency 1/f, the correlation between these two
sources can be assumed to be negligible [ESC 95]. As a result, the
total spectral density of noise power of the phototransistor, in terms of
its optical input, can be expressed as the sum of all the spectral
densities, equations [4.36] and [4.38]:
i
Tot
2
'f
=4 q I
)
I
Belec
( )
4kT
R
B

1
g
opt
2
2 q I
C

4kT
R
L
(
(
)
)
A
2
Hz
(
(
)
)
[4.39]
According to equation [4.39], the total spectral noise density of the
phototransistor is dominated at low frequencies by the sources of
noise linked to the photocurrent I
, to the bias current of the phase

I
Belec
and to the base resistance R
BE
; at high frequencies, the noise
source due to the collector current becomes dominant.
4.4. Photodetector circuits based on phototransistors
4.4.1. Amplification circuits
Phototransistors can be used either in the linear detection system
(continuously modulated signals or pulsed signals) or in the switching
system, in circuits with (see Figure 4.12a) or without (see Figure
4.12b) pre-bias, with of course the possibility of adding an additional
level of amplification (see Figure 4.12c). It has been shown that in
every application detection of weak or strong signals the speed is
limited by the value of the load resistance of the phototransistor. In
order to improve sensitivity, it is common nowadays to use either an
active load with low input impedance (a common-base transistor) (see
Figure 4.12d) or an operational amplifier providing the current-
voltage conversion (see Figure 4.12e). Finally, we should note that the
phototransistor can be integrated with a second transistor to form a
photo-Darlington pair (see Figure 4.12f). This structure clearly gives a
greater response coefficient at the cost of deterioration in the operation
speed.
Figure 4.12. Main circuit diagrams for phototransistors: a) structure with two
electrodes E and C, b) pre-bias using access to the base, c) supplementary
amplification with a second stage, d) improvement in speed using a low impedance
load (common-base transistor) or e) by current-voltage conversion, f) integration on
silicon of a second stage: the photo-Darlington pair. Note: structures c) and d) are
used in fast circuits
Structures c) and d) have been used to make monolithic amplification
circuits integrating the phototransistor and several HBTs, for wideband
[CHA 92, KAM 00, WAN 86a] and narrowband [GON 00a, KAM 95]
applications.
4.4.2. Nonlinear circuits
Phototransistors can also be used in optoelectric mixing and auto-
oscillation configurations. The simplest circuit is shown in Figure
4.13a and consists of a phototransistor and two matching cells. The
first cell is placed between the source OL and the input to the base of
the phototransistor. The second cell is placed between the output of
the collector and the load resistance R
L
. The optical signal is
modulated at the intermediate frequency IF, and the mixed signals at
frequencies OL IF are recovered at the output of the collector. The
second circuit, described by Sawada et al. [SAW 99] and depicted in
Figure 4.13b, consists of a phototransistor and a circuit associated
with its base which allows it to auto-oscillate at the OL frequency of
9.567 GHz. This frequency is stabilized using a dielectric resonator
DR. The optical signal is modulated at the frequency of 0.2 GHz, and
the mixed signals recovered with up-conversion are at frequencies of
9.367 and 9.767 GHz. The third circuit, described by Lasri et al. [LAS
00], is shown in Figure 4.13c. It consists of two phototransistors with
their emitters joined together. One phototransistor is used as an auto-
oscillator at 30 GHz locked optically to an optical signal O
1
modulated
at the same frequency OL. This signal OL is coupled with the second
phototransistor, which acts as an optoelectrical mixer between the
30 GHz signal and a second optical signal, O
2
, modulated at 300 MHz.
The two optical signals O
1
and O
2
could be at wavelengths
corresponding to a WDM (wavelength division multiplexing) scheme.
These circuits demonstrate the abilities of phototransistors to combine
the functions of photodetection, amplification, mixing and auto-
oscillation.
4.5. Applications
4.5.1. Galvanic isolation
Silicon phototransistors have been used for a long time in optical
isolators. These devices may or may not have a base contact. Figure
4.14 shows an optical isolator based on a phototransistor. For this
application, all that is required is to connect a structure upstream
which takes care of the conversion of an input electrical signal into a
light beam. This is in order to achieve a transfer of information which
avoids any form of electrical connection. In this way, galvanic
isolation between the controller and the receiver is ensured, and the
achievable voltage difference, which in the static system is determined
by the distances which separate the two components or the
connections to the different electrodes, can reach several kilovolts.
Figure 4.13. Various circuit diagrams for phototransistors acting as (a) a simple
optoelectrical mixer, (b) auto-oscillation and mixing with only one phototransistor
and (c) auto-oscillation and mixing with two phototransistors
Figure 4.14. Schematic representation of an optical isolator
based on a phototransistor
The device providing the electron-photon conversion at the input is
normally a light-emitting diode (LED) based on GaAs, whose
emission line of around 850 nm is well suited to optical detection by
the phototransistor or the photo-Darlington pair. The current
conversion gain can reach 500%. An example of its variation as a
function of the input current is shown in Figure 4.15.
,
Figure 4.15. Current transfer level (current from the collector of the
phototransistor/current in the input LED) as a function of the
input current to the optocoupler
Another application of silicon phototransistors involves power
electronics. These types of devices must often meet strict requirements
of galvanic isolation, particularly for control applications and energy
conversion from the industrial electricity grid. Control through optical
means allows this specific problem to be solved in an efficient and
elegant manner: with the thyristor and the triac which are the two
basic bipolar components in this application domain constructed
around a photodiode, whose response replaces the current from the
control electrode, the gate. A schematic cross-section for a
photothyristor and the equivalent circuit for a phototriac are shown in
Figure 4.16.
Photothyristors present very interesting possibilities for direct current
power transmission at high voltages (1,000-1,500 A and 5-8 kV)
[ARN 92].
The phototriac is constructed by combining two photothyristors in
parallel, in opposing directions. The near-obligation, for monolithic
integration of this device onto a silicon substrate, of using only one
surface of the substrate minimizes the current capacities and reduces
the voltage stability. Because of this, these devices are applied in the
domain of static relays which involve currents and voltages not
exceeding 1 A and 1 kV.
4.5.2. Phototransistors for optical telecommunications
This application, which requires both a good response coefficient
and speed, mostly involves heterojunction bipolar transistors based on
InP and InGaAs.
Figure 4.16. (a) Schematic cross-section of a photothyristor
and (b) the equivalent electric circuit for a phototriac
In the basic process of an optical transmission system, shown in
Figure 4.17, the phototransistor is the device which converts the
optical signal into an electrical signal. It is situated at the input of the
receiver, acting as an optical detector. In the transmitter, the light is
modulated (directly or indirectly via an optical modulator) with the
signal carrying the information. This signal can be of an analog or
digital. After transport through the optical fiber, the optical signal is
demodulated by the photodetector in order to recover the electrical
signal. This signal is subject to noise and distortion, and so the
receiving circuit may need to amplify and reconstruct the signal in
order to extract the original information.
Figure 4.17. Schematic diagram of an optical transmission system
with an analog transported signal
An InP/InGaAs phototransistor is an alternative to traditional
photodetection, consisting of a PIN photodiode associated with a
preamplifier. As was previously stated, this device is of interest
because of its combination of photodetection and amplification
behavior and, since its structure is similar to that of a heterojunction
bipolar transistor, it can attain a frequency performance comparable to
that of a HBT based on InP [BLA 00]. In addition, when acting as an
optical-electrical mixer, it can convert the demodulated signal to other
frequency ranges depending on the required output.
4.5.2.1. InP/InGaAs phototransistors as pre-amplifying
photodetectors
The transport of analog and digital data takes place at frequencies
and bitrates that are constantly increasing. In digital transmission, for
example, currently 10 and even 50 Gbit/s can be achieved. As a result,
phototransistors must become faster and faster while maintaining a
good level of sensitivity. The structure of a phototransistor based on
InP/InGaAs is shown in Figure 4.18. This device is sensitive to optical
wavelengths, O = 1.3 and 1.55 m. The base and collector layers are
made of InGaAs and the emitter layer is InP. The optical window sits
directly above the base. This has been optimized to achieve an optical
gain of 30 dB and a transition frequency of 88 GHz [GON 00]. These
high performances are the result of many refinements, particularly to
reduce the dimensions of the optical window, the base-emitter and base-
collector junctions (in order to reduce their capacitances), as well as the
use of a base layer with a graded composition, in order to increase the
gain and speed.
,
Figure 4.18. Vertical structure of a phototransistor based on InP/InGaAs
on a semi-insulating substrate of InP doped with Fe
Figure 4.19 shows the response coefficient of the phototransistor in
its two modes of operation, transistor (Tr Mode) and photodiode (PIN
mode), as a function of the modulation frequency of light. The
response coefficient (R) is expressed in dB, R (dB) = 20 log S, with S
expressed in A/W. The PIN mode corresponds with the
phototransistor operating as a photodiode (V
BE
= 0V) and the Tr mode
operating with the transistor effect (V
BE
> 0V). As discussed in section
4.3.2, the optical gain G
opt
, expressed in dB, is equal to the difference
between the response coefficient in the Tr mode and that of the PIN
mode. The response coefficient in the PIN mode, S
0
, is 0.25 A/W.
This response coefficient S
0
has been increased to 0.44 A/W in
another phototransistor, while maintaining an equivalent optical gain
(32 dB). However, as predicted by equation [4.18], the transition
frequency drops to 56 GHz [THU 99a].
Figure 4.19. Frequency response of an InP/InGaAs photodiode.
The response coefficient (photoresponse) is expressed in dB
Monolithic photoreceiver circuits for wideband amplification
involving photo-HBTs and HBTs have been constructed. The
amplifier circuit proposed and built by [KAM 00] exhibits a
bandwidth of 40 GHz. The circuit diagram is shown in Figure 4.20.
The circuit consists of a phototransistor and two HBT transistors. The
two types of device have a double heterojunction and use the same
epitaxial layers.
Figure 4.20. Monolithic circuit diagram for wideband amplification consisting of a
phototransistor and two double heterojunction transistors, using InP technology
[KAM 00]. The structure of the epitaxial layers is the same for the two devices
4.5.2.2. InP/InGaAs phototransistors as an optoelectrical mixer
The optoelectrical mixing transistor can be used in mixed
radio/fiber network access, or in satellite communications, to achieve
frequency conversion (up- or down-conversion). It can also be used in
high bitrate communication systems as an integral part of timing-
signal extraction circuits. Suematsu et al. [SUE 96] were among the
first to use a photo-HBT based on GaAs, to achieve conversion from
the modulation frequency of the light at 2 GHz to a frequency band at
50 GHz. Later, other investigations were reported [GON 98, GON 00],
using InP/InGaAs technology, to achieve conversion from an
intermediate frequency (2 GHz) to frequency bands at 30 and 40
GHz. An example of an experimental setup used to achieve this high
frequency conversion is shown in Figure 4.21. The light emitted by a
laser source at 1.55 m is amplitude modulated at an intermediate
frequency IF (via an external EOM modulator) in the frequency range
of 200 MHz to 2 GHz. The modulated optical signal is injected into
the optical window of the phototransistor via a single-mode optical
fiber. The base contact is connected to a local oscillator LO at a
frequency of 30 GHz. After demodulation of the optical signal, the
phototransistor mixes the OL and IF signals and the result is
visualized with the help of a spectrum analyzer.
Figure 4.22 shows the output electrical powers of the IF signal (in
PIN mode) and the mixed signal OL + IF, as a function of the IF
frequency. The conversion gain, using the definition given in [4.32], is
the ratio of the output power of the LO + IF signal to that of the IF
signal detected in PIN mode. In the frequency range of IF, from 0.2 to
2 GHz, the conversion gain for the LO + IF signal is relatively
constant, of the order of 7 dB. This positive conversion gain is high
because of the direct amplification of the mixed signal by the
phototransistor.
Figure 4.21. Experimental setup for up-conversion from the intermediate
frequency IF to a band at 30 GHz, using the HPT as an O/E mixer
Figure 4.22. Conversion gain for the IF signal transposed to a frequency
of 30 + IF GHz, using the InP/InGaAs phototransistor as an O/E mixer
4.6. Conclusion
In this chapter we have covered the main characteristics of the
bipolar phototransistor. We have also presented its applications in the
field of galvanic isolation for silicon phototransistors, and in the field
of optical telecommunications for InP/InGaAs heterojunction
phototransistors. The study of monolithic circuits incorporating
InP/InGaAs phototransistors, and providing more complex functions
such as optical-electrical mixing and auto-oscillation, is the particular
aim of research intended to exploit the synergy between photonics and
microwaves, with the goal of meeting the ever-growing requirements
in sensitivity, speed and functionality of photodetectors.
4.7. Bibliography
[ALF 73] ALFAROV Z.I., AKHMEDOV F., KOROLKOV V., NIKETIN V., Phototransistor
utilizing a GaAs-AlAs heterojunction, Sov. Phys. Semicond., vol 7, no. 6, p. 780,
1973.
[ARN 92] ARNOULD J., MERLE P., Dispositifs de llectronique de puissance, Hermes,
Paris, 1992.
[BAA 77] BAAK C., ELZE G., WALF G., GaAs MESFET: a high-speed optical
detector, Electron. Lett., vol. 13, p. 139, 1977.
[BAR 97] DE BARROS L.E., PAOLELLA A., FRANKEL M., ROMER M., HERCZFELD P.,
MADJAR A., Photoreponse of microwave transistors to high-frequency modulated
lightwave carrier signal, IEEE Trans. on Microwave Theory Tech., vol. 45, no. 8,
p. 1368-1374, 1997.
[BET 98] BETSER Y., RITTER D., LIU C.P., SEEDS A. J., MADJAR A., A single stage
three-terminal heterojunction bipolar transistor optoelectronic mixer, J. Light.
Tech., vol. 16, no. 4, p. 605-609, 1998.
[BLA 00] BLAYAC S., RIET M., BENCHIMOL J.L., BERDAGUER P., KAUFFMAN N.,
GODIN J., SCAVENNEC A., Lateral design of InP/InGaAs DHBTs for 40 Gbit/s,
Proc. of the Indium Phosphide and Related Materials, IPRM2000, p. 481-484,
2000.
[CAM 82] CAMPBELL J.C., OGAWA K., Heterojunction phototransistors for long-
wavelength optical receivers, J. Appl. Phys., vol. 53, no. 2, p. 1203-1208, 1982.
[CAM 85] CAMPBELL J.C., Phototransistors for lightwave communications,
Semiconductors and Semimetals, vol. 22D, p. 389-447, 1985.
[CAS 89] CASTAGNE R., DUCHEMIN J.P., GLOANEC M., RUMELHARD C., Circuits
intgrs en arsniure de gallium, Masson, 1989.
[CHA 85] CHAND N., HOUSTON P.A., ROBSON P.N., Gain of a heterojunction bipolar
phototransistor, IEEE Transactions on Electron Devices, vol. ED-32, no. 3, p.
622-627, 1985.
[CHA 92] CHANDRASEKHAR S., LUNARDI L.M., GNAUCK A.H., RITTER D., HAMM
R.A., PANISH G.L., QUA G.L., A 10 Gbit/s OEIC photoreceiver using
InP/InGaAs heterojunction bipolar transistors, Electronics Letters, vol. 28, no. 5,
p. 466-468, 1992.
[ESC 95] ESCOTTE L., ROUX J. P., PLANA R., GRAFFEUIL J., GRUHLE A., Noise
modelling of microwave heterojunction bipolar transistors, IEEE Transactions
on Electron Devices, vol. 42, no. 5, p. 883-888, 1995.
[FRE 96] FREEMAN P., ZHANG X., VURGAFTMAN I., SINGH J., BHATTACHARYA P.,
Optical control of 14 GHz MMIC oscillators based on InAlAs: InGaAs HBTs
with monolithically integrated optical wave guides, IEEE Transactions
Electrons Devices, vol. 43, no. 3, p. 373-379, 1996.
[GAU 85] GAUTIER J.L., PASQUET D., POUVIL P., Optical effects on the static and
dynamic characteristics of a GaAs MESFET, IEEE Trans. on Microwave Theory
Tech., vol. 33, no. 9, p. 819-822, 1985.
[GON 98] GONZALEZ C., THURET J., RIET M., BENCHIMOL J.L., Optoelectronic up-
converter to millimetre-wave band using an heterojunction bipolar
phototransistor, Proc. of the European Conference on Optical Communication,
ECOC98, vol. 1, p. 443-444, 1998.
[GON 00] GONZALEZ C., MULLER M., BENCHIMOL J.L., RIET M., JAFFR P., LEGAUD
P., HBT phototransistor for remote upconversion in hybrid fibre radio
distribution systems, Proc. of the European Conference on Optical
Communication, ECOC 2000, vol. 2, p. 103-104, 2000.
[GON 00a] GONZALEZ C., MULLER M., BENCHIMOL J.L., RIET M., JAFFR P., LEGAUD
P., A 28 GHz HPT/HBT monolithically integrated photoreceiver for hybrid fibre
radio distribution systems, Proc. of the 8
th
IEEE International Symposium on
Electron Devices for Microwave and Optoelectronic Applications, EDMO 2000,
p. 55-60, 2000.
[KAM 95] KAMITSUNA H., Ultra-widebande monolithic photoreceivers using HBT
compatible HPTs with novel base circuits and simultaneously integrated with an
HBT amplifier, J. Lightwave Tech., vol 13, no. 12, p. 2301-2307, 1995.
[KAM 00] KAMITSUNA H., MATSUOKA Y., YAMAHATA S. AND SHIGEKAWA N., A 82
GHz-Optical-gain-cutoff-frequency InP/InGaAs double-hetero-structure
phototransistor (DHPT) and its application to a 40 GHz band OEMMIC
photoreceiver, Proc. of the European Microwave week 2000, EuMC 2000,
EuMC36, 2000.
[KRO 57a] KROEMER H., Quasi-electric and quasi-magnetic fields in non-uniform
semiconductors, RCA Rev. vol. 18, p. 332, 1957.
[KRO 57b] KROEMER H., Theory of a wide-gap-emitter for transistors, Proc. IRE,
vol. 45, p. 1535, 1957.
[LAS 00] LASRI J., BILENCA A., EISENSTEIN G., RITTER D., ORENSTEIN M., SIDEROV
V., GOLDGEIER S., COHEN S., A two heterojunction bipolar photo-transistor
configuration for millimeter wave generation and modulation, Proc. of the
International Topical Meeting on Microwave Photonics, MWP00, p. 62-65,
2000.
[LET 78] LETURCQ P., REY G., Physique des composants actifs semiconducteurs,
Dunod University, 1978.
[MAD 92] MADJAR A., PAOLELLA A., HERCZFELD P.R. Analytical model for
optically generated currents in GaAs MESFET, IEEE Trans. on Microwave
Theory Tech., vol. 40, p. 1681-1691, 1992.
[MON 71] de LA MONEDA F.H., CHENETTE E.R., VAN DER ZIEL A., Noise in
phototransistors, IEEE Transactions on Electron Devices, vol. ED-18, no. 6, p.
340-346, 1971.
[MOR 72] MORIIZUMI T., TAKAHASHI K., Theoretical analysis of heterojunction
phototransistors, IEEE Transactions on Electron Devices, vol. 19, no. 2, p. 152-
159, 1972.
[POU 94] POUVIL P., Composants semiconducteurs micro-ondes, Masson, 1994.
[ROM 96] ROMERO M., MARTINEZ M., HERCZFELD, An analytical model for the
photodetection mechanisms in high-electron mobility transistors, IEEE Trans. on
Microwave Theory Tech., vol. 44, no. 12, p. 2279-2287, 1996.
[SAW 99] SAWADA H., IMAI N., An experimental study on a self-oscillating
optoelectronic up-converter that uses a heterojunction bipolar transistor, IEEE
Trans. on Microwave Theory Tech., vol. 47, no. 8, p.1515-1521, 1999.
[SEE 90] SEEDS A. J., DE SALLES A.A.A., Optical control of microwave devices,
IEEE Trans. on Microwave Theory Tech., vol. 38, no. 5, p. 577-585, 1990.
[SHI 53] SHIVE J.N., The properties of germanium phototransistors, J. Opt. Soc.
Am., vol. 43, p. 239, 1953.
[SHO 51] SHOOCKLEY W., SPARKS M., TEAL G.K., p-n junctions transistors, Phys.
Rev., vol. 83, p. 151, 1951.
[SIM 86] SIMONS R., BHASIN K., Analysis of optically controlled
microwave/millimeter- wave device structures, IEEE Trans. on Microwave
Theory Tech., vol. MTT-34, no. 12, p. 1349-1355, 1986.
[SMI 80] SMITH R.D., PERSONICK S.D., Semiconductor Devices for Optical
Communications, Springer, Berlin, Heidelberg, p. 89, 1980.
[SUE 96] SUEMATSU E., IMAI N., A fiber optic: millimeter-wave radio transmission
link using HBT as direct photodetector and an optoelectronic upconverter, IEEE
Trans. on Microwave Theory Tech.,vol. 44, no. 1, p. 133-142, 1996.
[THU 99] THURET J., Phototransistor bipolaire htrojonction InP/InGaAs pour
conversion optique/bande millimtrique dans les rseaux de distribution hybride
radio sur fibre, Thesis, University of Paris VI, 1999.
[THU 99a] THURET J., GONZALEZ C., BENCHIMOL J.L., RIET M., BERDAGUER P.,
High-speed InP/InGaAs heterojunction phototransistor for millimetre-wave fibre
radio communications, Proc. of the Indium Phosphide and Related Materials,
IPRM99, p. 389-392.
[WAN 86] WANG H., Photorcepteur monolithique intgrant un phototransistor et des
transistors bipolaires heterojonction GaAlAs/GaAs pour transmission par fibre
optique, Thesis, University of Paris XI, 1986.
[WAN 86a] WANG H., ANKRI D., Monolithic integrated photoreceiver implemented
with GaAs/GaAlAs heterojunction bipolar phototransistor and transistors,
Electronics Letters, vol. 22, no. 7, p. 391-393, 1986.
Chapter 5
Metal-Semiconductor-Metal
Photodiodes
5.1. Introduction
Among the different semiconductor photodetectors, the metal-
semiconductor-metal (MSM) photodiode has a special place. It is a
planar photodiode based on simple technology, it is easy to integrate,
but it has not, up to now, gained the interest of the large
telecommunication firms. Currently, to the best of our knowledge,
only one device for millimeter applications has been commercialized.
However, we will see that this photodetector is not without benefits,
and indeed has very specific properties linked to its interdigitated
structure. In this discussion, we will first consider the operation and
structure of this photodiode before discussing its different
characteristics: its response coefficient, dynamic behavior and noise.
We will conclude with integration possibilities for the device.
Chapter written by Joseph HARARI and Vincent MAGNIN.
5.2. Operation and structure
5.2.1. Fundamentals
An MSM photodiode consists of two Schottky contacts on a
semiconductor. Because of this, with each contact having a current-
voltage characteristic I(V) similar to that of a traditional junction, the
MSM photodiode behaves like two diodes in series, facing in opposite
directions. Briefly, under bias, with one of the diodes necessarily
being reverse-biased, the dark current crossing the structure is weak
and, to the extent that the two Schottky contacts are identical, the
symmetry of the device gives rise to a symmetric global I(V)
characteristic, with each half (positive or negative bias) corresponding
to the behavior of a reverse-biased metal-semiconductor junction.
Figure 5.1. MSM photodetector with its typical interdigitated structure.
Inset: current lines in the structure
If the bias voltage is sufficiently high, the region between the
electrodes is completely depleted and this produces an electric field.
The basic principle of this photodetector is that light is absorbed in a
semiconductor layer forming part of the depletion region, such that the
electron-hole pairs generated are separated under the effects of the
electric field, with each carrier then being directly collected by an
electrode. This effect, familiar in photodiodes, takes place in a planar
structure whose geometry is shown in Figure 5.1. In general, the
electrodes have an interdigitated shape, which allows the carriers to be
collected over a large semiconductor area while maintaining a short
Metal-Semiconductor-Metal Photodiodes 157
inter-electrode distance. If, for example, illumination is from above,
this allows carriers to be collected over the whole surface of the light
beam emitted by a fiber in the near field. The inset in Figure 5.1
shows the current lines between several fingers of the electrode, and
shows that, due to the planar structure, the deeper a carrier is
photogenerated in the material, the longer it will take to reach the
electrode. This leads to a very specific distribution of transit times for
the carriers. As with all photodiodes, an MSM photodetector under
illumination can be characterized by its response coefficient, its cutoff
frequency and its photodetection noise. However, before we consider
these different aspects, we will discuss the behavior of the device in
the dark.
Figure 5.2. Behavior of an MSM photodetector in the dark: a) band structure
in thermodynamic equilibrium, b) band structure under bias,
c) corresponding electric field under bias
Figure 5.2 shows the evolution of the band structure of an MSM
structure under bias. The two electrodes are assumed to be of different
metals, so that the band structure of each metal-semiconductor
junction differs. On each side, I
n
is the height of the barrier seen by
the electrons in the metal, I
p
is the height for holes and V
b
is the built-
in bias of each junction.
In thermodynamic equilibrium (see Figure 5.2a), the two depletion
regions depend on the doping of the semiconductor and the work
function which separates the Fermi level in the metal from the vacuum
level in the semiconductor. Since the semiconductor is normally
weakly doped, the carriers mostly cross the potential barriers through
thermoelectric effects [SZE 96], in other words only those carriers
with sufficient energy cross the barrier. The sum of all the currents
across the barriers is zero. Under bias, as shown in Figure 5.2b, the
situation is more complicated, with one of the junctions (number 1 in
this case) being reverse-biased (voltage V
1
), and the other being
forward-biased (voltage V
2
). It is the current densities of electrons J
n1
and J
n2
, and of holes J
p1
and J
p2
which determine the behavior [SZE
71]. If we ignore recombination in the region between the electrodes,
we have:
1 2 1 2
and
n n p p
J J J J = = [5.1]
These relationships show that the forward bias voltage of junction
2 is such that the current flowing across it is equal to the current
allowed across junction 1 in reverse. Under these conditions, it is the
height of the barrier I
n1
which determines the value of J
n1
(and hence
J
n2
), and it is I
p2
which determines the value of J
p2
(and J
p1
). The
value of the total dark current thus depends on the heights of the
barriers, which are linked to the nature of the metal used for each
electrode. In parallel to this process, the depletion region of junction 1
expands, while that of junction 2 contracts slightly. Under these
conditions we can define two particular bias voltages which depend on
the structure of the MSM under consideration: the first is the total
depletion voltage V
TD
, which is reached when the depletion region of
the reverse-biased junction reaches that of the forward-biased junction,
that is, when:
W
1
W
2
= s [5.2]
where s is the inter-electrode distance.
The second is the flat band voltage V
FB
, reached when W
1
= s. The
electric field present in the inter-electrode region is thus only zero at
electrode 2 (the anode), and it is at a maximum at the other side (at the
cathode). Clearly, an MSM photodetector is only useful when the bias
voltage is greater than V
FB
. If we assume a homogenous N-type
semiconductor, with doping level N
d
and permittivity H
s
, we have:
V
FB
=
q N
d
s
2 H
s
V
b1
V
b2
[5.3]
When the applied bias is increased, the electric field increases at
the cathode, which leads gradually to a breakdown process linked
either to shock ionization in the semiconductor material or to
tunneling through the potential barriers of the metal-semiconductor
barriers from band to band. All this leads to the typical current-voltage
characteristics displayed in Figure 5.3. For V < V
FB
, the dark current
crossing the photodiode is weak and grows slowly with the bias, until
it reaches a saturation value which is maintained until breakdown,
which can be more or less gradual with voltage. If the two electrodes
are made of the same metal, the I(V) characteristics in the dark are, in
principle, symmetric with respect to the V = 0 V axis. Generally, since
a photodetector must operate with a weak dark current, the best bias
conditions for a photodiode are:
a bias voltage considerably below the breakdown voltage, in
order to limit the dark current (V < V
C
);
an inter-electrode region that is completely depleted, with a
sufficiently high internal electric field so that the velocities of the
photogenerated charge carriers are in the saturation system (in any
case, V > V
FB
).
Of course, given the planar structure of the device, some elements
must be added to this general description.
Figure 5.3. Typical current-voltage characteristics of an
MSM photodetector in the dark
The biasing of the semiconductor, using electrodes placed above it,
leads to a particular electric field distribution, as shown in Figure 5.4.
Initially, the field decreases with depth in the device, and then there
are local maxima in the electric field just under the edge of each
electrode (or each finger for an interdigitated structure). Consequently,
the breakdown of the device always begins at the edge of the electrode,
and a very high electric field may exist close to the electrodes, but the
field is weaker in the middle of the inter-electrode space and deep in
the photodetector. Finally, as the field is weaker deeper in the
structure, the carriers that are photogenerated deep in the structure not
only have a great distance to travel to reach the electrode (as we have
already seen in Figure 5.1), but also have a small drift speed, which
increases the transit delay and decreases the dynamic performance of
the photodetector. All this confirms that the optimal bias voltage of
such a device must be selected with care in order to obtain good
dynamic performances without degradation.
Figure 5.4. Distribution of the modulus of the electric field in an MSM photodetector.
The applied bias is 4 V, and the values of the field are in kV/cm.
The semiconductor is weakly doped (Nd = 5 u 10
14
cm
3
)
5.2.2. Materials used
5.2.2.1. Electrode composition
Since the composition of the electrodes has a direct influence on
the band structure of the photodetector due to the magnitudes of I
n1
,
I
p1
, I
n2
and I
p2
, as shown in Figure 5.2, it will have an effect on the
dark current resulting from the thermoelectric effect. If we bias far
from the breakdown voltage, the phenomena limiting the dark current
are the thermoelectric passage of electrons across junction 1, and the
thermoelectric passage of holes across junction 2. The total dark
current is thus commonly written as:
J = A
n
*
T
2
exp
q I
n1
'I
n1
( )
kT
l
l
l
l
l
A
p
*
T
2
exp
q I
p2
'I
p2
( )
kT
l
l
l
l
l
l
l
l
l
l
l
l
1exp
qV
kT
(
(
)
)
(
(
)
)
[5.4]
where A
n
*
and A
p
*
are the Richardson constants associated with
electrons and holes, T is the absolute temperature, V is the applied
voltage, I
n1
and I
p2
are the heights of the barriers seen by electrons
and holes respectively, corrected with the terms introduced by the
effects of the electric field. This results in [SZE 96]:
'I
n1, p2
=
qE
1 2
4SH
s
[5.5]
where H
s
is the permittivity of the semiconductor and E
1,2
is the value
of the electric field at the metal-semiconductor junction. Greater
barrier heights thus minimize the dark current. If we assume that the
two electrodes are made of the same material, we have (see Figure
5.2):
I
n1
= E
g
I
p2
[5.6]
Simultaneously achieving maximum-height barriers for electrons
and holes requires us to choose the contact metal for which I
n
is as
close as possible to E
g
/2. Measurements of the dark current [ITO 86],
made for a gallium arsenide (GaAs) MSM photodetector for different
electrode metals, are presented in Table 5.1.
Dark current
Table 5.1. Dark current measured for GaAs MSM photodetectors (E
g
= 1.42 eV)
for different electrode metals and photodetectors with the same surface
Because of this, for this type of photodetector it is desirable to
decide on the cathode and anode at the fabrication stage, choosing a
metal with a high electron barrier I
n
for the first electrode and a high
I
p
barrier for the second. Under these conditions, the values of the
dark current are of the order of nano-amps, which are compatible with
all possible uses of the device.
Clearly, all this assumes that no other effect, associated with
fabrication defects, surpasses the thermoelectric effect which is
normally the main cause of the dark current. This can occur if the
band structure is perturbed close to the metal-semiconductor junction.
In this case, a narrowing of the potential barrier favors the passage of
carriers through the tunnel effect, to the detriment of the
thermoelectric effect. The tunnel effect can be associated with the
presence of a native oxide layer at the interface, which leads to an
accumulation of electrons or holes; alternatively, the effect can be
assisted by trapping states present in the semiconductor material,
which create energy levels inside the bandgap. As a result, attention
must be paid to the epitaxy, the cleanliness of the semiconductor
surface before the metal deposition, and above all to the passivation of
the structure. In the case of a photodetector illuminated from above,
the dielectric layer used for this final operation can operate as an anti-
reflection coating.
5.2.2.2. Epitaxial structure
The epitaxial structure of an MSM is designed starting with the
absorbing layer, which itself depends on the wavelength to be
absorbed according to the basic law:
h c
O
E
g
[5.7]
where h is the Planck constant and c is the speed of light. The
different materials already reported in this application are given in
Table 5.2 along with the substrate they are grown on (an asterisk
indicates growth with a lattice mismatch) and the associated
wavelengths.
Table 5.2. Absorbing materials already used to fabricate MSM photodetectors,
along with their growth substrate and their associated wavelengths
As Figure 5.5 shows, around the absorption layer there are
normally two layers with a higher bandgap. The adaptation layer acts
to isolate the absorbing layer from the substrate during growth, and
the barrier layer acts to increase the potential barrier at the level of the
metal-semiconductor junction [BUR 91, WOH 97]. A barrier layer is
required, for example, when the absorbing material is N-type GaInAs,
on which the Schottky contacts are poor quality. Furthermore, to
improve the passage of carriers from the absorbing layer to the
electrodes, we also introduce one or more transition layers which
reduce the discrepancy in the band structure as seen by free carriers,
between the barrier layer and the absorbing layer [WAD 89, ZHA 96].
These very thin layers allow a gentle transition in the band structure
and prevent the accumulation of carriers in the upper part of the
absorbing layer under the electrodes. This accumulation leads to a
drop in the external quantum efficiency of the device, as a non-
negligible number of free carriers recombine while they are blocked at
the interface. It also simultaneously leads to an increase in the transit
time for charge carriers, which has an adverse effect on the frequency
response of the photodetector.
Figure 5.5. Typical epitaxial structure of an MSM photodetector.
The semiconductor is weakly doped (Nd = 5 u 10
14
cm
3
)
The thickness of the absorbing layer depends mostly on its
absorption coefficient because, as we will see, this directly influences
the response coefficient of the photodetector when it is illuminated
from above or below. The thickness of the adaptation layer is
normally greater than one micron, while that of the barrier layer is as
thin as possible (<0.2 m in general). In some cases, for example, that
of a gallium arsenide MSM or a silicon MSM, the substrate and the
adaptation layer are also absorbing, but this requires the collection of
carriers photogenerated deep in the structure which, as we have
already seen, decreases the dynamic performance of the photodiode.
To avoid this problem, the adaptation layer of the GaAs MSM can be
made of non-absorbing AlGaAs matched to the lattice of the GaAs
substrate. In the case of silicon, SIMOX wafers are used, which have a
layer of silicon underneath.
5.3. Static and dynamic characteristics
5.3.1. Response coefficient
The response coefficient, which provides the photocurrent
generated per Watt of incident light (in units of A/W as a result),
initially depends on the mode of illumination. The photodetector can
be illuminated from above, across the interdigitated metallic structure,
from below or side-on, through an optical waveguide. The case of
lateral illumination must be treated on a case-by-case basis because
the response coefficient depends both on the optical coupling between
the waveguide and the detector, and between the fiber and the
waveguide. This requires a detailed study of the propagation of the
light in the multi-layer structure of the device. Illumination from
below is similar to illumination from above, apart from the absence of
electrodes to partially occlude the light entering the device; in addition,
the absorption associated with the substrate must be taken into account
if it is not thin and not perfectly transparent. The absorption
coefficient of the substrate is generally weak at the wavelengths under
consideration, but the thickness of the material to be crossed can be
sufficiently large to have an effect on the response coefficient.
Illumination from above is common. In this case, the response
coefficient is:
R=
q
hQ
K [5.8]
where q is the electronic charge, hQ the energy of a photon of the
incident light and K the quantum efficiency, which is itself given by:
K = 1R
( )
s
s d
( )
1exp D W
a
( )
l
l
l
[5.9]
where R is the reflection coefficient at the surface of the device, s is
the distance between two fingers, d the width of a finger, D the
absorption coefficient and W
a
the thickness of the absorbing layer.
The ratio s/(s + d) introduces the shadowing effect of the electrodes
which is typical of this photodetector. In brief, the response coefficient
is directly linked to the characteristics of the absorbing layer and the
geometry of the interdigitated electrodes.
As far as the characteristics of the absorbing layer are concerned,
in the case of a semi-infinite layer, the absorption coefficient must be
sufficiently high so that all the carriers are photogenerated in the
depletion region. This is the case for GaInAs at wavelengths of 1.3
and 1.55 m, and for GaAs around 0.8 m. In contrast, for silicon, the
absorption coefficient at 0.8 m is of the order of 10 m
1
. This
reduces the quantum efficiency of the photodetector and causes a
reduction in the bandwidth, because at least some of the carriers
photogenerated far from the depletion region reach it through
diffusion, which is much slower here than the transport under the
effects of the electric field. This clearly depends on the inter-electrode
distance, as we will see. Finally, we can see that optimizing the
transport of the carriers across the planar structure implies a
relationship between the thickness W
a
of the absorbing region and the
inter-electrode distance s.
If we now consider the geometry of the electrodes, we find that
attempts to reduce the shadowing effect leads to two approaches. The
first consists of using opaque electrodes while reducing the finger
width relative to the inter-electrode distance. However, this increases
the resistance of the finger, which has a detrimental effect on the
bandwidth. Measurements made on electrodes deposited on silica
show that the effective resistance of a metallic finger is greater than
that which would be calculated based on the volume of the material.
This effect is more important the narrower the finger, importance
being obtained at around 0.1 m. This may be due to metallic
limitations that the structure places on electron transport becoming
important. In any case, for fingers made of titanium (150 ) and gold
(350 ), and of width 0.06 m, the measured resistance is around 80
:/m instead of 11 :/m for d = 0.5 m. This gives an idea of the
sizes to be considered when designing the electrodes. As a result, s
and d normally take similar values, within the range d s 4d. The
second approach consists of research into electrodes made of
transparent materials. Several possibilities exist, notably thin layers of
gold [MAT 96] and doped oxides: indium tin oxide (ITO) or cadmium
tin oxide (CTO). In each case, there is a tradeoff between the
transparency at the wavelength under consideration and the resistance
of the material. For example, doped ITO is remarkably transparent at
short wavelengths (O < 0.85 m), but it requires a compromise
between resistance and transparency at longer wavelengths as doping
reduces its transparency [GAO 94, SEO 93]. The same is true for CTO.
In any case, an optical transmission of 100% cannot be achieved at
long wavelengths. Furthermore, attention must be paid to the
conditions of the deposition, which for the oxides is normally through
sputtering, which does not favor particularly good characteristics of
the electrode-semiconductor interface. In short, while the response
coefficient is an absolute priority, the deposition of transparent
electrodes clearly reduces the shadowing effect, but it is not without
consequences in terms of the dynamic behavior of the photodetector.
Firstly, it leads to an increase in the finger resistance, the
consequences of which can be seen in the equivalent circuit of the
MSM in the dynamic system. Also, the electric field diagram (Figure
5.4) shows that this is weak in the medium under each finger. The
drift velocity of the photocarriers will therefore be weak at these
points. As a result, the electrons photogenerated just under the
transparent cathode, and the holes photogenerated under the anode,
will take a long time to reach the opposite electrode. Not only is their
drift velocity low as they travel through areas of weak field, but they
also have a long way to travel (see Figure 5.1). Thus, transparent
electrodes increase the mean transit time and decrease the bandwidth.
A similar effect on the mean transit time is seen when the device is
illuminated from below. In this case, the majority of the carriers are
photogenerated in the lower part of the absorbing layer, as much
underneath the fingers as between them. Since this is a region of weak
electric field, this also leads to a longer mean transit time.
Still on the subject of the electrode geometry, a property that
cannot be ignored in the case of illumination from above is the
transparency of the network of fingers when its period p (p = d + s)
becomes smaller than the wavelength of light. Then we observe
typical optical behavior which is more complicated than a simple
shadowing effect. The few studies carried out on this subject [KUT
94] show that the transmission of light through the metallic network
depends on the ratios of O/p and O/d, on the finger thickness, on the
refractive index n
s
of the semiconductor, and on whether the optical
polarization is parallel or perpendicular to the direction of the fingers.
It is higher in the perpendicular case than in the parallel case, where
only a few percent of the optical energy reaches the semiconductor.
Furthermore, it is a minimum for O/p = 1 and O/p = n
s
. The finger
thickness becomes relevant when it is close to its width. All these
complex properties, which must be modeled on a case-by-case basis,
are sharp enough to make it possible to built a wavelength
discriminator using two parallel MSMs with slightly different periods,
illuminated from the same beam and measuring the ratio of the
photocurrents [CHE 97].
Figure 5.6. Example of the epitaxial structure of an MSM photodetector
on top of an optical waveguide
In order to understand all the modes of illumination, we will now
consider the case of lateral illumination. The electrode structure of the
MSM is particularly favorable to the injection of light through an
optical waveguide [SOO 88, VIN 89]. Indeed, the direct lateral
illumination, for which absorption occurs in the first few microns after
the cut face, is difficult to achieve, as is the case for a PIN photodiode.
The example in Figure 5.6, of an MSM structure on top of a
waveguide, is based on an N-type InP/GaInAsP/InP optical waveguide,
the standard for this type of technology. The quaternary GaInAsP
layer (with bandgap E
g
= 1.05 eV), formed into a lattice on InP, is
what forms the guiding layer, with InP acting as the cladding.
Coupling with the photodetector occurs through evanescent waves.
Modeling of the optical propagation through such a structure (see
Figure 5.7) shows how the injected beam couples with the absorbing
layer through the intermediary of the thin quaternary adaptation layer
whose characteristics (thickness and composition) determine the
absorbed wavelength. Of course, the photocarriers are still generated
in the lower part of the absorbing layer, but the thickness of this layer
can be selected to be quite thin without significantly reducing the
quantum efficiency. For a fixed adaptation layer, the effect of the
thickness of GaInAs on the internal quantum efficiency (see Figure
5.8) shows that, for TE and TM optical polarizations at a wavelength
of 1.55 m, a certain number of optimum cases exist, even for small
thicknesses.
At a conceptual level, to achieve the shortest possible absorption
length, it is necessary to carefully choose both the thickness of the
adaptation layer and that of the absorbing layer. The response
coefficient is:
R= 1R
( )
K
q
hQ
K
internal
[5.10]
where K
internal
is the internal quantum efficiency determined by the
epitaxial structure, R is the reflection coefficient at the entrance of the
waveguide and K is the coupling coefficient between the beam leaving
the optical fiber and the semiconductor waveguide. The internal
quantum efficiency can only be calculated through detailed modeling
of the optical transport in the material structure of the
photodetector/waveguide. The dynamic behavior of such a structure is
comparable to that seen with illumination from below; however, the
thin width of the absorbing layer allows a short transit time to be
retained. Thus, this type of structure allows a high response coefficient
to be combined with a small absorber thickness.
Figure 5.7. Propagation and absorption of light in a MSM
on a waveguide with evanescent coupling
Figure 5.8. Evolution of the quantum efficiency of the detector/waveguide
as a function of the thickness of the absorbing material
To conclude, we will discuss all the internal electronic effects in
the device which affect the response coefficient. Typically, this is
reduced under the effect of recombination and increases with internal
gain. Recombination occurs either at the surface of the semiconductor
between to fingers or at the interface between two epitaxial layers,
wherever there is a potential barrier associated with the transition
between different materials, which blocks the transport of carriers and
leads to their accumulation. In the first case, careful passivation of the
surface of the photodetector eliminates the problem. The second case
is relevant when a barrier layer is introduced to improve the quality of
the contact with the electrode. A transition layer between the absorber
and the barrier is thus required, made of an epitaxial layer with
variable bandgap or even a super-lattice [WAD 89, ZHA 96].
There are many effects which lead to gain [KLI 94]. They mostly
consist of two types: those connected with the trapping of carriers,
which operate at low frequencies (below a few tens of megahertz), and
shock ionization close to the edges of the fingers, whose cutoff
frequency is higher, as can be seen in the case of avalanche
photodiodes. In the case of trapping, the gain effect is due to a
photocarrier becoming trapped in the semiconductor for a time W
p
greater than the inter-electrode transit time W
T
which triggers the
injection of additional carriers in order to maintain electrical neutrality.
As long as this carrier remains trapped, a carrier with the opposite
charge will cross the device to provide electrical compensation for it.
The gain is then given by:
G
trapping
=
W
p
W
T
[5.11]
This trapping can occur at the surface of the semiconductor, as in
the case of GaAs photoconductors, or at the interface between two
epitaxial layers. When it occurs at the surface, the holes are mostly
trapped close to the cathode and electrons trapped close to the anode,
this creates an asymmetric charge distribution. Hetero-interface
trapping is associated with epitaxial growth defects, and normally a
barrier layer is used to eliminate surface effects. Generally, attempts
are made to limit both recombination and trapping in order to obtain a
photodetector with constant dynamic behavior up to the high cutoff
frequency, while ensuring a good reliability.
5.3.2. Dynamic behavior
As with all photodetectors, the dynamic behavior of MSMs is
determined by their capacitance and their input and output resistances,
in other words their equivalent electric circuit under modulated
electric power, and the transit time of the photocarriers. The
equivalent small-signal circuit (see Figure 5.9) shows:
the intrinsic capacitance C
PD
and resistance R
PD
of the device;
the finger resistance R
F
, the capacitance and the resistance of the
contact, C
CONTACT
and R
CONTACT
;
the connection wires with inductance L
F
and capacitance C
W
.
The finger resistance is given by:
R
F
=2 R
0

L
N
[5.12]
where R
0
is the resistance per unit length of each finger, L the length
and N the number of fingers. In the equivalent circuit, R
L
is the load
resistance and we see how the finger resistance can, in some cases,
increase the time constant RC of the device. One of the important
terms is the intrinsic capacitance of the MSM. A standard analysis of
the planar structure [LIM 68] shows that:
C
PD
=
A
d s
( )
H
0
1H
R
( )
K k
( )
K k '
( )
[5.13]
where A is the surface area of the active region, H
0
and H
R
are the
vacuum and relative permittivities, and finally:
K k
( )
=
dM
1k
2
sin
2
M
( )
0
S 2
with k = tan
2
S.d
4 d s
( )
(
(
)
)
and k ' = 1k
2
The calculation shows two details connected to the interdigitated
planar structure of the MSM. Firstly, the capacitance of the device
falls when the ratio s/d increases; the decrease in the shadow effect is
thus inextricably linked to the capacitance of the photodetector.
Secondly, for a photosensitive surface and an identical mean transit
time, the capacitance of the MSM is 3 to 4 times weaker than that of a
PIN photodiode. This aspect favors the MSM for high modulation
frequencies (millimeter wavelengths), where the size of the devices is
very small.
Typical values for the equivalent circuit can be given as a guide.
The capacitance of the MSM lies between 10 and 1000 fF, its
resistance between 1 and 100 :, and the capacitance of the contact is
a significant parasitic element, generally less than 100 fF (and in all
cases less than C
PD
). Finally, the connection wires have an auto-
inductance of the order of 10 pH and a capacitance of a few fF linking
them to the body of the device. If we ignore all the parasitic elements,
that is to say the connection wire, the capacitance of the contact, and
the finger resistance, we find a standard time constant: R
L
C
PD
. The
cutoff frequency associated with the capacitance effects is thus given
by:
f
CAPA
=
1
2S R
L
C
PD
[5.14]
Now turning to the transit time, a thin, two-dimensional physical
model is required to account for the effects of the geometry of the
electrode, the thickness of the absorbing layer, of the hetero-interfaces
to be crossed, and of the electric field distribution produced by the
bias voltage [ASH 95].
Nevertheless, a number of rules allow us to evaluate the cutoff
linked to the transit time of the carriers. Firstly, the presence of high
electric field areas just at the edges of a finger mean that it is difficult
to appropriately bias a device with a large inter-electrode distance (s >
3 m). What happens in this case is that a local electric field is
reached which is above the breakdown voltage, while the field in the
middle of the inter-electrode space is still too weak to give the
photocarriers their saturation velocity. This effect, visible in Figure
5.4, is more serious for holes, as they move more slowly than
electrons. Even for small inter-electrode distances, the variation of the
field in the device must be taken into account. The bias voltage is thus
an important parameter that must be adjusted to limit the mean transit
time while avoiding breakdown.
Figure 5.9. Equivalent circuit for an MSM photodetector in the case of a small signal
Secondly, the distribution of the current lines in the device favors a
certain ratio between the thickness of the absorbing layer and the
inter-electrode distance. The effect of transit time is smallest when
this value is close to 2 (W
a
/s = 2). If this is achieved, the mean transit
time in the structure is:
W =G
s
2 v
sat
[5.15]
where v
sat
is the saturation velocity of the carriers (we will take the
same velocity for holes as for electrons), s is the inter-electrode
distance and G accounts for the effects of the planar structure (1 G
2). The saturation velocity used is taken as equal to that of the holes
for the most pessimistic results. The cutoff frequency associated with
the transit time is thus given by:
f
T
=
1
2S W
[5.16]
The cutoff frequency of the MSM photodiode under these
conditions is:
f =
1
f
CAPA
2

1
f
T
2
(
(
)
)
1 2
[5.17]
An increase in the inter-electrode distance will decrease the
capacitance and increase the transit time. For every structure, then,
there is a compromise to be made.
5.3.3. Noise
The different sources of noise in MSMs, as in other photodetectors,
are thermal noise, 1/f
D
noise and shot noise. Thermal noise is linked to
the variations in energy of the carriers in resistive regions. It is
independent of frequency and its spectral density is linked to the
absolute temperature T and the resistance R by:
i
TH
2
=
4kT
R
in A
2
Hz
( )
[5.18]
The 1/f
D
noise is associated with all the random generation and
recombination effects in the semiconductor material. These effects are
themselves linked to the different imperfections, in terms of
irregularities in the structure of the crystal lattice or in terms of foreign
atoms. This generates a specific noise whose spectral power density
varies as 1/f
D
, where D is normally close to 1. This spectral density is
often higher in horizontal transport devices, such as MSMs or field
effect transistors. Finally, the shot noise, familiar from PIN and
avalanche photodiodes, is linked to the photogeneration process and
the charge transport. Its spectral power density is:
i
SHOT
2
= 2 q I
DARK
I
PH
( )
in A
2
Hz
( )
[5.19]
In principle, this is white noise, i.e. its spectral density is
independent of frequency; however, in practice it follows the
frequency response curve of the device, making it possible to obtain
the frequency response of a photodetector using a continuous signal,
by measuring the variation of its shot noise as a function of frequency.
Having covered these items, we will now describe the variation of
the spectral noise density of the MSM photodetector in two situations:
in normal operating conditions and close to breakdown in the
conditions where a gain in the photocurrent, linked to the shock
ionization effect, is possible.
In general, if the finger and contact resistances are sufficiently low,
the thermal noise is much weaker than the shot noise, so that we can
ignore its effects. Under normal operating conditions [SOO 91, WAD
88], the 1/f
D
noise dominates for frequencies of the order of 1 to
10 MHz, after which it is the shot noise which dominates and which
must be considered when performing calculations for photoreceivers.
This shot noise increases with the dark current, which is why a low
dark current is desirable, normally below about 100 nano-amps.
However, as breakdown is approached [SCH 90], two changes are
observed. Firstly, the 1/f
D
noise increases, so much so that it
dominates for frequencies up to the order of 100 MHz; secondly, the
shot noise also increases, in association with the multiplication
process connected with shock ionization. The spectral density of the
noise becomes:
i
SHOT
2
=2 q I
DARK
I
PH0
( )
G
2
F in A
2
Hz
( )
where G is the gain, I
PH0
the primary photocurrent and F the noise
factor, which can be expressed as a function of the gain G, as is done
in the case of avalanche photodiodes: F = G
x
. If, as is often the case,
the part linked to the dark current is negligible compared to the part
linked to the photocurrent, we have:
i
SHOT
2
=2 q I
PH0
G
2 x
in A
2
Hz
( )
However, values of x measured up to now for MSMs are somewhat
higher: between 1.5 and 3, which is not desirable. Due to these noise
characteristics, MSM photodetectors are not used close to breakdown.
5.4. Integration possibilities and conclusion
The functionalities offered by MSM photodiodes are considerable.
It is with these photodetectors that the highest current cutoff
frequencies, above terahertz levels, have been achieved [CHO 92],
while maintaining an acceptable response coefficient. Of course, the
inter-electrode distance required for such a performance is small
(25 nm) and requires the use of an electronic mask, but this shows the
interest in a device whose main advantages are:
a planar configuration of the electrodes allowing, in particular,
the integration of the photodiode into a coplanar microwave
transmission line;
weakly doped active layers which remain consistent with
technologies based on semi-insulating substrates;
a capacitance significantly below that of all other photodetectors
with equivalent transit time characteristics;
electrode technologies allowing the dark current to be
minimized;
a photodetection noise equal to that of a PIN photodiode.
These properties allow the MSM to be integrated without difficulty
into all integrated circuit technologies based on III-V materials,
whether on GaAs [BUR 91, HAR 88, ITO 84] or on InP [HOR 96],
with MESFET or HEMT transistors. In general, the weakly-doped
epitaxial structure of the detector is grown before the various layers
required to make the transistors and the metallic deposition of the
interdigitated electrode is the same as that of the transistor gates.
In conclusion, the optical properties of thin metallic lattices, which
are still poorly-understood, offer interesting and original characteristics,
even if the power limitations are more restrictive than those for PIN
photodiodes.
5.5. Bibliography
[ASH 95] ASHOUR I.S., HARARI J., VILCOT J.P., DECOSTER D., High optical power
nonlinear dynamic response of AlInAs/GaInAs MSM photodiode, IEEE Trans.
on Electron Devices, vol. 42, no. 5, p. 828-834, May 1995.
[BUR 91] BURROUGHES J.H., H-Mesfet compatible GaAs/AlGaAs MSM
photodetector, IEEE Photonics Technology Letters, vol. 3, no. 7, p. 660-662,
July 1991.
[CHE 97] CHEN E., CHOU S.Y., A wavelength detector using monolithically integrated
subwavelength metal-semiconductor-metal photodetectors, Proceedings SPIE, vol.
3006, p. 61-67, 1997.
[CHO 92] CHOU S.Y., LIU Y., FISCHER P.B., Tera-hertz GaAs metal-semiconductor-
metal photodetectors with 25 nm finger spacing and width, Applied Physics
Letters, vol. 61, no. 4, p. 477-479, July 1992.
[GAO 94] GAO W., KHAN A., BURGER P.R., HUNSPERGER R.G., ZYDZIK G., OBRIAN
H.M., SIVCO D., CHO A.Y., InGaAs metal-semiconductor-metal photodiodes
with transparent cadmium tin oxide Schottky contacts, Applied Physics Letters,
vol. 65, no.15, p. 1930-1932, October 1994.
[HAR 88] HARDER C.S., VAN ZEGHBROECK B.J., MEIER H., PATRICK W., VETTIGER P.,
5.2 GHz bandwidth monolithic GaAs optoelectronic receiver, IEEE Electron
Device Letters, vol. 9, no. 4, p. 171-173, April 1988.
[HOR 96] HORSTMANN M., SCHIMPF K., MARSO M., FOX A., KORDOS P., 16 GHz
bandwidth MSM photodetector and 45/85GHz ft/fmax HEMT prepared on an
identical InGaAs/InP layer structure, Electronics Letters, vol. 32, no. 8, p. 763-
764, April 1996.
[ITO 84] ITO M., WADA. O., NAKAI K., SAKURAI T., Monolithic integration of a
metal-semiconductor-metal photodiode and a GaAs preamplifier, IEEE Electron
Device Letters, vol. 5, no. 12, p. 531-532, December 1984.
[ITO 86] ITO M., WADA O., Low dark current GaAs metal-semiconductor-metal
photodiodes using Wsix contacts, IEEE Journal of Quantum Electronics, vol. 22,
no. 7, p. 1073-1077, July 1986.
[KLI 94] KLINGENSTEIN M., KUHL J., ROSENZWEIG J., MOGLESTUE C., HLSMANN A.,
SCHNEIDER J., KOHLER K., Photocurrent gain mechanisms in metal-
semiconductor-metal photodetectors, Solid State Electronics, vol. 37, no. 2, p.
333-340, 1994.
[KUT 94] KUTA J.J., VAN DRIEL H.M., LANDHEER D., ADAMS J.A., Polarization and
wavelength dependence of metal-semiconductor-metal photodetector response,
Applied Physics Letters, vol. 64, no. 2, p. 140-142, January 1994.
[LIM 68] LIMY.C., MOORE R.A., Properties of alternately charged coplanar parallel
strips by conformal mapping, IEEE Transactions on Electron Devices, vol. 15,
p. 173-180, 1968.
[MAT 96] MATIN M.A., SONG K.C., ROBINSON B.J., SIMMONS J.G., THOMPSON D.A,
GOUIN F., Very low dark current InGaP/GaAs MSM Photodetector using semi
transparent and opaque contacts, Electronics Letters, vol. 32, no. 8, p. 766-767,
April 1996.
[SCH 90] SCHUMACHER H., SOOLE J.B.D., LEBLANC H.P., BHAT R., KOZA M.A.,
Noise behaviour of InAlAs/GaInAs MSM photodetectors, Electronics Letters,
vol. 26, no. 9, p. 612-613, April 1990.
[SEO 93] SEO J.W., CANEAU C., BHAT R., ADESIDA I., Application of indium-tin-
oxide with improved transmittance at 1.3m for MSM photodetectors, IEEE
Photonics Technology Letters, vol. 5, no. 11, p. 1313-1315, November 1993.
[SOO 88] SOOLE J.B.D., SCHUMACHER H., ESAGUI R., KOZA M.A., BHAT R.,
Waveguide integrated MSM photodetector on InP, Electronics Letters, vol. 24,
no. 24, p. 1478-1480, April 1988.
[SOO 91] SOOLE J.B.D., SCHUMACHER H., InGaAs metal-semiconductor-metal
photodetectors for long wavelength optical communications, IEEE Journal of
Quantum Electronics, vol. 27, no. 3, p. 737-752, March 1997.
[SZE 71] SZE S.M., COLEMAN D.J., LOYA J.R.A., Current transport in metal-
semiconductor-metal structure, Solid State Electronics, Pergamon Press, New
York, vol. 14, p. 1209-1218, 1971.
[SZE 96] SZE S.M., Physics of Semiconductor Devices, John Wiley & Sons Ltd., New
York, 1996.
[VIN 89] VINCHANT J.F., VILCOT J.P., LORIAUX J.L., DECOSTER D., Monolithic
integration of a thin and short metal-semiconductor-metal photodetector with a
GaAlAs optical inverted rib waveguide on a GaAs semi insulating substrate,
Applied Physics Letters, vol. 55, no. 19, p. 1966-1968, May 1989.
[WAD 88] WADA O., HAMAGUCHI H., LE BELLER L., BOISROBERT C.Y., Noise
characteristics of GaAs metal-semiconductor-metal photodiodes, Electronics
Letters, vol. 24, no. 25, p. 1574-1575, December 1988.
[WAD 89] WADA O., NOBUHARA H., HAMAGUCHI H., MIKAWA T., TACKEUCHI A.,
FUJII T., Very high speed GaInAs metal-semiconductor-metal photodiode
incorporating an AlInAs/GaInAs graded superlattice, Applied Physics Letters,
vol. 54, no. 1, p. 16-17, January 1989.
[WOH 97] WOHLMUTH W., ARAFA M., MAHAJAN A., FAY P., ABESIDA I.,
Engineering the Schottky barrier heights in InGaAs metal-semiconductor-metal
photodetectors, Proceedings SPIE, vol. 3006, p. 52-60, 1997.
[ZHA 96] ZHANG Y.G., LI A.Z., CHEN J.X., Improved performance of InAlAs-
InGaAs-InP MSM photodetectors with graded superlattice structure grown by gas
source MBE, IEEE Photonics Technology Letters, vol. 8, no. 6, p. 830-832, June
1996.
Chapter 6
Ultraviolet Photodetectors
6.1. Introduction
The first mention of the existence of ultraviolet radiation was in
1801 when J.W. Ritter observed that certain types of chemical
reactions were accelerated by a radiation which was not visible to the
eye and whose wavelength was shorter than that of violet light. It was
quickly shown that this particular radiation was subject to the laws of
interference just like visible light (T. Young, 1804). Thus this
observation, along with various others, made it possible to establish at
the start of the 19
th
century that visible and ultraviolet radiation
exhibited the same electromagnetic properties, and differed only in
their wavelength.
We know today that ultraviolet radiation (UV) covers a wide
spectral range which extends from the limit of the visible, around
400 nm (3.1 eV) up to the X-ray boundary at 10 nm (124 eV) (see
Figure 6.1), covering a total spectral range which in terms of energy is
about 80 times as large as that of the visible spectrum. It is
traditionally divided into four categories:
Chapter written by Franck OMNES and Eva MONROY.
UVA, covering wavelengths between 400 and 320 nm (from 3.1
to 3.87 eV);
UVB, whose wavelengths range from 320 to 280 nm (from 3.87
to 4.43 eV);
UVC, for wavelengths between 280 and 200 nm (from 4.43 to
6.20 eV);
the far UV, for wavelengths between 200 and 10 nm (form 6.2 to
124 eV).
Due to its own radiation absorption properties, the terrestrial
atmosphere does not allow the free propagation of light with
wavelengths below 200 nm: as a result, the spectral end of the UV
between 200 and 10 nm cannot really propagate except in a vacuum,
and because of this it is often known as vacuum UV (we will use the
term VUV for the rest of the chapter when referring to this particular
spectral range).
Figure 6.1. Spectral ranges of the main types of electromagnetic radiation
The fields of application for UV detectors, of which there is a great
variety, mostly consist of:
biological and chemical sensors: detection of ozone, pollutants
and biological agents;
fire detection: fire alarms, detection of missile exhausts, flame
control;
optical inter-satellite spatial communications: secure transmission
of data in space;
Ultraviolet Photodetectors 183
the calibration of emitters and UV imaging: instrumentation,
measurement of solar UV, astronomical research.
In general, photodetectors suited to these areas of application must
simultaneously possess a high response coefficient, a good linearity of
the photocurrent as a function of the incident optical power, a large
bandgap energy (in the case of semiconductors), a low level of noise
and a high degree of spectral selectivity. The level of importance and
the priority ascribed to these different requirements clearly depends on
the type of application intended. A high response speed can be
particularly useful in applications requiring rapid real-time processing
of the signal, such as UV imaging.
The sun represents the most important natural source of UV
radiation. It is therefore not surprising that most of the applications of
UV detectors are connected with the measurement of solar radiation.
The solar radiation spectrum is shown in Figure 6.2.
,
,
,
,
,
Figure 6.2. Solar radiation spectrum. Inset: UVA and UVB solar irradiance,
outside the Earths atmosphere and at ground level
For the vast majority of applications, it is helpful if the UV
photodetectors are not sensitive to visible and infrared radiation. They
basically operate as high-pass filters, and they are classed as visible blind
detectors for cutoff wavelengths between 400 and 280 nm, or solar blind,
assuming that the UVC radiation is blocked by atmospheric absorption
for cutoff wavelengths typically less than or equal to 280 nm. The visible
blind detectors are nowadays quite widely used for flame detection
1
, and
have potentially very promising growth possibilities for applications in
space to optical inter-satellite telecommunications, for UV wavelengths
below 280 nm which clearly cannot be detected at ground level. Outside
the Earths atmosphere, UV light represents 9% of the total luminous
power emitted by the Sun. The stratospheric ozone layer completely
absorbs the UVC part of the solar spectrum, and only allows UVA and
UVB to be transmitted to the ground after attenuation (see the inset in
Figure 6.2). It is generally considered that solar radiation of wavelengths
below 280 nm does not reach the surface of the Earth. UVA and B
radiation, on the other hand, are mostly responsible for the biological
effects of solar radiation, since they reach the ground (see Figure 6.3).
Damage to
human DNA
Figure 6.3. Spectral dependence of the harmful effects of
UVA and UVB radiation on human health
1. The applications of flame detection mostly cover the following areas: fire detector
heads, forest fire detection systems, missile track detection systems, etc. Gas flames,
for example, have an emission peak in the UV around a wavelength of 320 nm.
UVB radiation, being particularly energetic, is the direct cause of
sunburn and skin cancers, and it has also been shown more recently
that UVA radiation, although considered less dangerous, is able to
penetrate the skin and has enough ionizing power to cause deep
damage to living cells
2
. It has been found that UVA radiation can in
particular alter the genetic structure of cells and, thus, also produce a
carcinogenic effect comparable to that of UVB, particularly in the
case of prolonged or repeated exposure. This represents a clear threat
to human heath from UV radiation, but also allows it to play an
extremely useful role in a number of different fields. Among others,
UV radiations are involved in the synthesis of Vitamin D in the human
body, which means that a moderate amount of exposure to sunlight is
important for everybody, and the ionizing properties of these
radiations make them useful for the activation of numerous chemical
reactions, both organic and inorganic. Also, their bactericidal action
makes them useful for many types of biological cleaning and
sterilization procedures
3
. Finally, the ozone layer present in the upper
atmosphere is naturally regenerated through the conversion of oxygen
in the air into ozone, a reaction which is photo-assisted by UV
radiation.
On the other hand, the rapid depletion of the stratospheric ozone
layer, caused in particular by human activity
4
, and the resultant
increase in exposure of the Earths surface to UV radiation has raised
considerable concern among scientists and professionals interested in
health and the environment. This has recently driven a significant
2. This issue is particularly interesting since it has public health implications because
of the popularity of tanning salons with the general public.
3. The illumination of objects with high intensity UVB and C radiation can be used
for this, particularly in techniques for the preservation of perishable foodstuffs, the
sterilization of surgical instruments, etc. Housewives have also known since time
immemorial that the drying of washing in the sun in fair weather makes it healthier,
etc.
4. The emission of aerosols and gases containing fluoro-, chloro- and bromo-carbons
is largely responsible for this. Large-scale natural phenomena such as volcanic
eruptions also contribute in a manner which can be very significant. One of the most
significant holes in the ozone layer is currently situated above the continent of
Antarctica, but the effect is very widespread and nowadays has a global effect.
increase in efforts to develop many specialized optoelectronic devices
which in particular allow the detection and measurement of solar UV
illumination. Thus, visible blind detectors have recently found a
growing number of applications in UVA and B monitoring systems
for the environmental study of ground-level solar illumination and UV
dosimetry in fields which are mostly connected with biology, medical
science or research into cosmetic products (the development of sun
creams or other products which modify the photosensitivity of human
skin to UV radiation, etc.). Personal solar dosimeters able to detect
direct or diffuse UV light are also currently available to the general
public for UVB dosimetry. The spectral response curve of these
sensors reproduces as closely as possible (as we will see later) the
erythema (sunburn) action spectrum on human skin.
There are a wide range of different UV photodetectors suitable for
all the different applications listed above. Traditionally,
photomultiplier tubes have always been used for UV detection and
their use is still widespread, particularly in laboratories. However,
they generally involve detection systems which are cumbersome,
expensive and fragile, although they offer a sensitivity which is
difficult to match with the other types of photodetectors applicable to
these ranges of wavelength. The increasing need for miniaturization
and reliability of these systems for the development of portable or
remote applications has naturally led to the development of UV
photodetectors based on semiconductors, whose active components
can be photoconductors, Schottky junction photodiodes, p-i-n
photodiodes, MSM photodiodes, phototransistors or avalanche
photodiodes.
The use of semiconductors for UV photodetection historically
began with the use of narrow bandgap materials such as silicon and
certain III-V materials (GaP, GaAsP, etc.) whose bandgaps are mostly
situated in the near-infrared (Si) or the red (GaAsP). In exceptional
cases they may be in the yellow, as in the specific case of GaP, which
represents their upper limit. The main drawback of silicon, and small
bandgap semiconductors in general, is that it is impossible to use them
directly for UV photodetection. Firstly, the direct exposure of such
sensors to daylight clearly results in them being flooded with visible
wavelengths and sometimes even infrared, so that it is simply
impossible to extract information pertaining to the UV spectral
component from this white light. Also, the quantum efficiency of a
semiconductor photodetector (defined as the probability of electron-
hole creation by the absorption of an incident photon) is greatest for
energies above the bandgap of the material, but limited to a spectral
range relatively close to that value. For energies much higher than the
bandgap of materials, such as those that we are interested in for UV
light, part of the energy of the incident light is effectively lost through
heating of the semiconductor (excitation of vibration modes, or
phonons in the crystal lattice), so that the quantum efficiency of the
photodetector decreases in the short wavelength part of the optical
spectrum. The use of semiconductor materials with a small bandgap
energy for UV photodetection therefore rapidly drove the
development of filtering devices, which normally use layers of
selectively-absorbing material, and sometimes a phosphor-based high-
pass filter which absorbs the UV light and re-emits towards the
semiconductor photodetector a light whose spectral maximum is not
far from the bandgap of the semiconductor material being used [GOL
99]. It still needs to be joined on to detection systems that are
relatively complex and also expensive, whose behavior is also
increasingly liable to drift in the long term, due to aging of the filters,
resulting in a certain degree of performance loss.
A number of years ago, a new generation of UV photodetectors
appeared, made from semiconductor materials with large bandgaps
such as silicon carbide (SiC), diamond or gallium nitride (GaN) and
AlGaN alloys. Silicon carbide, whose bandgap at ambient temperature
(T = 300 K) is 2.86 eV (6H-SiC), has allowed sensitive detection of
the spectral range in question. In particular, it has recently given rise
to industrial applications such as the construction of flame detectors
that are visible-blind, operating in the UVC range. The only issue is
that even here the use of high-pass optical filters is standard in order
to tune the photodetection system to the desired spectral window,
meaning that this only offers a partial solution to the problem.
Thanks to the development of epitaxial crystal growth techniques,
the recent use of gallium nitride (GaN), and still more recently
aluminum gallium nitride (Al
x
Ga
1-x
N, with x being between 0 and 1),
has helped reduce the problems listed earlier, enabling the fabrication
of high performance and robust visible blind photodetectors. The
direct bandgap of Al
x
Ga
1x
N materials, whose size is an increasing
function of the concentration of aluminum, is found at ambient
temperature to extend from 3.42 eV (O = 362 nm) (x = 0) to 62 eV (O
= 200 nm) (x = 1). Thus, the spectral range that is covered includes in
particular the cutoff values for UVB (3.87 eV or O = 320 nm) and
UVC (4.43 eV or O = 280 nm). It follows that it becomes possible to
fabricate UV photodetectors using these materials which are free from
any need for intermediate spectral filtering, opening up the possibility
of simplifying photodetection systems through miniaturization,
accompanied by an improved reliability of the devices. In addition, the
high chemical stability of GaN and Al
x
Ga
1x
N compounds,
comparable to that of SiC, means that these materials are also suitable
for use in the fabrication of photodetectors which are naturally
resistant to extreme operating conditions (high temperatures and high
values of inverse bias voltage in particular). Due to their direct
bandgap at all levels of aluminum, the optoelectronic properties of
Al
x
Ga
1x
N alloys are also clearly superior to those of SiC, whose
indirect bandgap does not so strongly favor the effects associated with
photo-generation of carriers. For all these reasons, group III nitrides
therefore offer a technical solution for the future, as we will see
throughout the rest of this chapter, a solution which is at the same time
flexible, robust and very well-suited for all visible-blind or solar-blind
photodetection applications, because they allow relatively low-cost
fabrication of effective and reliable UV photodetectors suitable for all
the application windows listed above.
In this chapter, we will first introduce the UV-visible contrast and
UV detectors based on silicon and SiC. We will then describe in detail
several types of UV detectors based on nitrides: photoconductors,
Schottky photodiodes, MSM photodiodes, and p-i-n photodiodes. The
most sophisticated technological advances involved with the recent
construction of avalanche photodiodes and phototransistors based on
nitrides are also briefly described
5
.
6.2. The UV-visible contrast
The absorption of light is the fundamental effect governing the
dependence of the spectral response of a photodetector as a function
of the energy of the incident photons. In direct-bandgap
semiconductors, we therefore observe a dramatic transition in the
value of the spectral response close to the bandgap of the
semiconductor material, dropping all at once to a very low value when
we drop below the bandgap. In the case of UV photodetectors,
assigning an arbitrary limit to the wavelength of visible light of
400 nm, the UV-visible contrast is defined by the ratio
R
imax
(O = O
G
)/R
i
(O = 400 nm), where R
imax
is the maximum value of
the response coefficient at the wavelength O
G
of the bandgap and
R
i
(O = 400 nm) represents the value of the response coefficient at the
limit of visible light.
The UV-visible contrast is mostly limited by the two following
factors:
the presence in the bandgap of the semiconductor of deep energy
states which are electrically and optically active. These originate in
particular from extended or localized defects in the crystal structure of
the semiconductor, but can also be attributed to the presence of deep
energy levels due to impurities, or deep levels caused by
imperfections in the surface of the semiconductor;
in the case of Schottky junction photodetectors with semi-
transparent contacts, the effects of photo-emission of carriers by the
metal.
5. The work carried out by the authors on UV photodetectors, described in this
chapter, was carried out with the support of the European Community, under the
Environment and Climate grant number #ENV4-CT97-0539.
6.3. Si and SiC photodetectors for UV photodetection
Contemporary semiconductor-based commercial technical
solutions proposed for UV photodetection mostly rely on Si and SiC-
based technologies. Other applications, a little less widespread and
well-known, use other semiconductor compounds such as GaAs, GaP
and GaAsP. These are mainly used for Schottky barrier photodiodes.
This particular family of devices is in particular very well-suited for
the detection of VUV radiation, for which they offer a remarkable
stability of operation. It is also worth mentioning the recent
emergence of UV photodetectors based on diamond, which allows the
construction of photoconductors with a short cutoff wavelength (225
nm) with a UV/visible contrast of 10
6
[WHI 96]. Although diamond-
based devices are potentially very interesting and well-suited for UV
photodetection applications and the detection of high-energy particles,
due in particular to diamonds very high resistance to the effects of
irradiation, the development of this last type of photodetector has
nevertheless been rather restricted by the current technical challenges
of achieving epitaxial films with adequate homogenity and
monocrystallinity: the best material obtained to date is a crystal
mosaic with grain boundaries which have a fairly strong mismatch
between one another. Clearly, this results in a high density of
electrically and optically active defects which reduces the
performance of the photoconductors, both in terms of response
coefficient and UV/visible contrast and in terms of response time.
Nevertheless, it is important to emphasize that UV photodetectors
based on thin epitaxial films of diamond have recently become
commercially available, which is the sign of a definite and growing
interest in these devices for UV photodetection and its applications
[MAI 00].
Next we will give, by way of illustration, a brief description of the
main families of Si and SiC-based photodetectors.
6.3.1. UV photodiodes based on silicon
Silicon p-n junction photodiodes are currently by far the most
economical and widespread commercial solution for UV
photodetection. Silicon gives rise to low-cost detectors which are
suitable for the vast majority of UV detection applications across a
very wide spectral range extending from near-UV to VUV, and even
extends to the high-energy system of soft X-rays. However, the
general shape of their spectral response is not uniform, and in
particular it reaches into the infrared, silicons bandgap being 1.1 eV
at ambient temperature (300 K). As a result, their use in selective UV
photodetection applications requires, generally speaking, the use of
absorbing filters between the light source and the photodetector, in
order to select only the spectral window of direct interest. We note in
passing that this often significantly increases the cost of the final
detection system, the filters normally requiring complex technologies
which make them very expensive.
UV photodiodes based on silicon can be subdivided into two large
families: p-n junction photodiodes and metal-oxide-semiconductor
photodiodes with a charge inversion layer [GOL 99, RAZ 96]. The
first family of Si UV photodetectors comprises photodiodes whose p-n
junction is situated at a shallow depth below the surface (typically of
the order of 0.2 m), and which are covered with a thin layer of SiO
2
.
This insulating layer plays the double role of passivation of the
surface of the semiconductor and of antireflection coating. The depth
of the junction is an extremely important parameter, because
absorption of light is more superficial the more energy the photon has.
In the case of p-n junction photodiodes, it is desirable, in order to
optimize its behavior and obtain the maximum photocurrent, that as
many UV-photogenerated carriers as possible reach the junction
region without recombining. For the same reason, control of surface
recombination effects is of fundamental importance for silicon UV
photodetection. Detailed control and optimization of the Si/SiO
2
interface properties makes it possible to reduce the density of surface
traps, while naturally producing an electric field at the surface of the
silicon which limits, or even eliminates, recombination effects at the
interface.
The p-n junction UV photodiode with record performance was
initially developed by Korde et al. [CAN 89, KOR 87]. In order to
form the junction, these authors used phosphor (P) diffusion onto an
Si (p) substrate free from dislocations. A SiO
2
layer of thickness 60
nm was built on the surface of the device. Due to its thickness, this
layer absorbs all radiation of wavelengths below 120 nm, which
unfortunately makes this type of photodiode unusable in applications
involving the detection of high energy UV (VUV in particular).
Nevertheless, this type of photodiode is characterized in particular by
an internal quantum efficiency of 100% in the spectral range of 350-
600 nm. The internal quantum efficiency is greater than 1 in the short-
wavelength UV region (<350 nm), due to secondary impact ionization
effects: a high kinetic energy is imparted to the electron-hole pair
photogenerated by high-energy photons, and this is sufficient to create
other secondary electron-hole pairs through impacts with atoms in the
crystal lattice. On the other hand, the efficiency becomes less than 1
beyond a wavelength of 600 nm, because this long-wavelength light is
absorbed in the deeper layers of the material which are far from the p-
n junction, creating electron-hole pairs which are unable to reach the
junction since their diffusion lengths are too short. As a result, they
recombine in the volume of the semiconductor and do not contribute
to the photocurrent.
It is interesting to dwell for a moment on UV photodiodes with a p-
n junction designed for the detection of VUV up to an energy of
124 eV. Although the base structure of the p-n junction remains the
same as that which we have already described, conversely the oxide
layer present on the surface is much thinner than in other types of UV
Si photodiodes. Its characteristic thickness is in this case no larger
than 4-5 nm, making it a factor of 10 or 20 times smaller than that
normally found in more conventional applications. Transparent to
short wavelengths of light, its useful volume is on the other hand
reduced, in such a way that the possibilities for UV light and humidity
to create high densities of recombinant traps in the oxide (which is
normally one of the main sources of aging in Si UV photodiodes) is
very significantly reduced compared to the normal situation. It follows
that the stability of operation of these devices is very good when they
are exposed to very high energy photons. The process of photo-
generation produces carriers with very high kinetic energies, which
multiply through secondary impact ionization. In the system of high
photon energies, the internal quantum efficiency K of Si VUV
photodiodes typically follows a linear variation as a function of the
energy E of the photon, which can be written: K = E/3.63 [RAZ 96].
To give a better idea of the orders of magnitude, the internal quantum
efficiency of a Si VUV photodiode is typically of the order of 30, for a
photon energy of 124 eV. These photodiodes are commonly equipped
with a bandpass filter consisting of thin metallic layers, picking out
the 10-50 nm band (Al/C, Al/C/Sc, Ti, Sn, Ag, etc.) [RAZ 96].
Photodiodes with a charge inversion layer [GOL 99, RAZ 96] are
similar in their architecture to metal-oxide-semiconductor structures
used to build field effect transistors: they make use of effects
associated with the presence of the two-dimensional charge induced
through inversion at the Si/SiO
2
interface. Through the effects of the
energy band curvature in the semiconductor, a thin deposited layer of
silica on the surface of the Si (p) material leads to the creation of a
charge bilayer with the opposite sign to the doping of the volume,
which consists of electrons in the case we will describe. This very
localized accumulation of charge results in a high electric field being
established in the immediate vicinity of the semiconductor surface
which, in the case of UV photodetection, offers the particular
advantage that this is the exact region where the main absorption of
the high-energy incident light takes place. As a result, this
configuration is ideal for maximizing the quantum efficiency of the
photodiodes, something which is also helped by the effectively
complete absence of a dead region of recombination at the surface,
which tended to be found in the first generations of p-n junction UV
Si photodiodes. Si UV photodiodes with an inversion layer offer a
high quantum efficiency in the spectral band from 250-500 nm and a
cutoff wavelength of 120 nm, which is fundamentally linked to the
absorption of the UV light by the oxide layer above this wavelength. It
should, however, be noted that the behavior of this type of photodiode
tends to change over time under UV illumination, since UV light has a
tendency to alter the surface oxide layer. Additionally, the range of
linearity of these photodetectors remains reduced compared to the
possibilities offered by p-n junction UV Si photodiodes, an effect
which seems to be mostly explained by the effects of the high
electrical resistance presented by the charge inversion bilayer at the
SiO
2
/Si interface.
6.3.2. SiC-based UV photodetectors
The best UV photodetectors based on 6H-SiC are p-n junction
photodiodes [RAZ 96]. The optimized structure consists of a layer of
SiC (n+) with a thickness between 0.2 and 0.3 m, sitting on a SiC (p)
substrate. The doping levels are 510 u 10
18
cm
3
for SiC (n+) and
58 u 10
17
cm
-3
for SiC (p). The structure is topped with a SiO
2
passivation layer on the surface. An ohmic contact based on nickel is
present on the n+ side. The response coefficient of p-n SiC
photodiodes made in this way lies in the range from 0.15 to 0.175
A/W at a wavelength of 270 nm, which corresponds to an internal
quantum efficiency of 70 to 85%. At O = 200 nm, the response
coefficient of these same structures is typically 0.05 A/W. The
spectral response of these photodiodes contains a peak lying between
O = 268 and O = 299 nm. The dark current of these devices is very
weak. A dark current value of 10
11
A for a reverse-bias of 1 V has
been measured for SiC photodiodes, at a temperature of 473 K [EDM
94].
Schottky junction 6H-SiC photodiodes are also currently being
developed, operating in the 200400 nm spectral range [RAZ 96]. The
best performances have traditionally been obtained for Schottky
devices made from SiC (p). The height of the Schottky barrier is about
twice as large as that normally achieved for n-type SiC. In addition,
the diffusion length for electrons is significantly larger than that for
holes. Very low leakage currents can be obtained for Schottky diodes
made using SiC (p). Leakage currents below 1 pA are currently
measured for a reverse-bias of 10 V. However, Anikin et al. [ANI
92] report the fairly recent construction of a high quality Schottky
photodiode in SiC (n). In this the Schottky junction is made through
deposition of gold, which gives a Schottky barrier height in SiC which
is typically between 1.4 and 1.63 eV. The behavior of these
photodetectors is characterized by a very low leakage current, of the
order of 100 pA for a reverse-bias voltage of 100 to 170 V, and a
high value of the response coefficient, which reaches 150 mA/W at a
wavelength of 215 nm [RAZ 96].
As we have already emphasized, one of the interests in SiC-based
devices is their remarkable resilience under extreme operating
conditions. This is why only a very small deterioration in their
operational characteristics (spectral response and I(V) characteristics)
is seen when these photodetectors are used at high temperatures,
which can reach or even surpass 300C without seeing a significant
degradation of the devices.
6.4. UV detectors based on III-V nitrides
Among the various families of materials suitable for UV
photodetection, III-V nitrides represent one of the most interesting and
flexible technical solutions at present. The first work in characterizing
the physical properties of polycrystalline gallium nitride (GaN) dates
back more than 30 years. The development of epitaxial techniques for
this material has for its part been held back for several decades,
mainly due to the poor suitability of the techniques used and the
absence of a substrate with a suitably matching lattice parameter,
which for a long time made the development of applications
impossible. Despite a strong lattice mismatch (it is 16% that of GaN)
and despite a strong difference in thermal expansion coefficient
compared to GaN, sapphire substrates are currently very widely
employed for the growth of thin layers of III-V nitrides and the
resultant applications, for which they are the substrate of choice.
Organometallic vapor phase epitaxy (OMVPE) and, more recently
introduced for nitride growth, molecular jet epitaxy with a plasma or
gas source (MJE) currently allow the growth of thin layers of GaN and
related compounds which, despite a high density of dislocations (of
the order of 10
9
to 10
10
cm
2
in GaN), exhibit structural, optical and
electrical properties on sapphire which are good enough to make them
suitable for many applications, mostly involving light-emitting diodes
at short emission wavelengths (from green to blue, and even into the
near-UV), blue lasers, field effect transistors and UV photodetectors.
6.4.1. Photoconductors
6.4.1.1. Spectral response
Photoconductors based on AlGaN typically consist of an epitaxial
layer of AlGaN on sapphire, doped with silicon, with a thickness of
1 m, onto which two ohmic contacts are attached (see Figure 6.4).
The electrical bias circuit includes a DC voltage source connected
to the photoconductor, and the device is in series with a low-value
load resistance. The current induced in the device can be easily
deduced from measurement of the voltage drop across the load
resistance when the photoconductor is illuminated. Photoconductors
based on GaN and AlGaN have a gain under continuous illumination
which can reach 10
5
10
6
and which strongly depends on the incident
optical power. Its variation is a function which decreases in proportion
to P
J
over a range of five decades, with 0.5 < J < 0.95, where P is the
incident optical power [MON 99a]. This behavior is independent of
the wavelength of the light. The value of J is a decreasing function of
the electrical resistance of the layer, and decreases with temperature.
Figure 6.4. Diagram of a photoconductor based on Al
x
Ga
1x
N(Si)
The spectral response of photoconductors based on Al
x
Ga
1-x
N
exhibits a cutoff at 298 nm for x = 0.319 nm for x = 0.22 and 368 nm
for x = 0.35. The UV/visible contrast is significantly smaller than the
value obtained from a simple analysis based on the absorption
coefficient of these materials. The mechanism which explains the
variation of the spectral response of the photoconductor must
therefore involve more than just the optical absorption of the
semiconductor. In addition, a sub-linear dependence of the
photocurrent is observed as a function of the power of the incident
light. In the model developed by Garrido et al. [GAR 98], the
photoconductors response coefficient R
i
is the combination of two
terms, one due to the photogenerated free carriers, 'n, and the other
due to the modification by the light of the effective conducting cross-
section, 'S. The response coefficient can be written:
R
i
=
'I
P
opt
=
qV
B
e
LP
opt
S'nn'S
( )
[6.1]
where q is the elementary charge of the electron, P
e
is the electron
mobility, L is the distance between the contacts, S is the conductive
cross-section, V
B
is the bias voltage and n is the concentration of free
carriers.
The transverse conduction cross-section in the model of Garrido et
al. is not the same as the geometric cross-section of the device, due to
the presence of space charge regions (SCRs) [HAN 98] around lattice
discontinuities (dislocations that reach the surface, grain boundaries
and interface). The absorption of light produces a reduction in these
depletion regions, leading to a modification of the conduction cross-
section. Given that 'n can be expressed in the form [RAZ 96]:
'
n
=KgW
P
opt
hc O
[6.2]
where K is the quantum efficiency, g the photoconduction gain, W the
lifetime of the free carriers and hc/O the photon energy; by
substituting [6.2] into [6.1], it follows that:
R
i
=
'I
P
opt
=
qV
B
P
e
L
KgWS
hc O

n
P
opt
'S
(
(
)
)
[6.3]
The spectral variation of the first term in this equation is given by
O.K(O). Since the quantum efficiency is a direct function of the
absorption, this first term leads to a significant UV/visible contrast.
However, experiments show that the dominant mechanism in these
photodetectors is represented by the second term of equation [6.3], in
other words, it depends strongly on the modification to the effective
transverse conduction cross-section. This mechanism can also explain
the high value of the spectral response for photon energies above the
bandgap energy: the energy levels responsible for the absorption of
light in the visible can originate either from point defects distributed
homogenously within the semiconductor, either dopants or vacancies,
or in extended defects associated with lattice discontinuities. If these
defects are electrically charged, they create a depletion region around
them, which reduces the effective conduction cross-section of the
device. Complete absorption of light through these defects can be
treated as negligible, so that their presence has little effect on the
absorption coefficient, and hence on the first term of equation [6.3].
However, their effect on the photoconduction spectral response is
large, since the electrical charge concentrated at these lattice
discontinuities is modified by the light, which alters the effective
transverse conduction cross-section.
6.4.1.2. Response time
AlGaN-based photoconductors display significant persistent
photoconductivity (PPC), effects. This means they have a very slow
transitory temporal response (several thousandths of a second) and a
non-exponential tail, which is consistent with the frequency
dependence of their spectral response [SHE 99].
The slow drop-off in the photocurrent can be explained by a model
based on the modulation of the transverse conduction cross-section
[MON 99a]. The curvature of the bands around defects and
dislocations leads to a spatial separation of electrons and holes, which
can be the cause of the PPC. When the light is extinguished, the
electrons are inclined to recombine, but they must first cross the
potential barrier separating them from their recombination sites. The
height of this barrier depends on the charge present at the defect, so
that it will change over time as recombination occurs. Through this
mechanism, the SCR returns slowly to its original extent in the dark,
and as a result the resistance of the device also slowly returns, in a
non-exponential manner, towards its dark value. The dominant defect
type may be that associated with dislocations at the AlGaN-air
interface or the AlGaN-substrate interface [SHE 99]. However, it is
difficult to discriminate between these two regions, and their relative
significance will depend on the crystalline quality of the material.
Quantitatively, the temporal response of photoconductors can be
calculated in the following manner, if 'Q
SS
is the increase in charge at
the defects, connected to the variation in the width of the SCR, 'x
SCR
.
'Q
SS
can be expressed in the form:
'QSS =qN
D
'x
SCR
= 2N
D
H
S
H
0
<
0
<
il
( )
[6.4]
where N
D
is the level of n-type doping, H
S
and H
0
are the relative
dielectric constant of the semiconductor and the vacuum permittivity
respectively, and <
0
and <
il
are the heights of the barrier around the
defect, in the dark and under illumination respectively. It follows that
the response coefficient of the photoconductor is given by:
R
i
=
A
opt
V
B
P
e
P
opt
L
2
'QSS [6.5]
where A
opt
is the optical surface area of the device.
In the dark, 'Q
SS
originates from the current crossing the potential
barriers around the defects. Including the thermionic emission current
J
th
, as well as the tunneling current J
tunnel
, it follows that:
dQ
SS
dt
= J
th
J
tunnel
[6.6]
The thermionic and tunneling currents can be expressed respectively
in the following forms:
J
th
= A*T
2
exp
q<
0
kT
(
(
)
)
exp
qV
ph
kT
(
(
)
)
l
l
l
l
l
[6.7]
J
tunnel
= A*T
2
S E
00
<
0
V
ph
( )
kT
q
cosh
qE
00
kT
(
(
)
)
exp
q <
0
V
ph
( )
E
0
(
(
)
)
1exp
qV
ph
kT
(
(
)
)
l
l
l
l
l
[6.8]
with E
00
= (h/4S).(N
D
/m*
e
) and E
0
= coth(qE
00
/kT). V
ph
is the voltage
induced by the light. The drop in voltage across the barrier is V
ph
(t),
which can be expressed as:
V
ph
t
()
= <
0

Q
2
SS t
()
2qH
S
H
0
N
D
[6.9]
To illustrate this, Figure 6.5 shows the temporal response of
AlGaN photoconductors, both measured, and calculated by numerical
solutions of equations [6.4] to [6.9].
, , , , ,
Figure 6.5. Decrease in the normalized photocurrent in Al
0.23
Ga
0.77
N
photoconductors, along with numerical simulation using equations [6.4][6.10]
Figure 6.6. Normalized spectral response of a photoconductor, measured at
different modulation frequencies of the incident optical signal.
The spectral response of a Schottky photodiode made from
the same batch is also shown for comparison
6.4.1.3. Effects of frequency modulation of the incident optical signal
The effect of frequency on the spectral response of
photoconductors is shown in Figure 6.6. When the frequency
increases, the SCRs associated with the defects do not have time to
respond. The result is that the cutoff slope becomes sharper and the
spectrum of the response coefficient tends to approach that of a
photovoltaic detector. As a result, the UV/visible contrast is much
better at higher frequencies. The mechanism at the root of the
response of III-V nitride-based photoconductors therefore makes them
of little interest for applications which requires speed or a high
UV/visible contrast.
6.4.2. Schottky barrier photodiodes based on AlGaN
6.4.2.1. Electrical properties
Work published on Schottky barrier photodiodes involves planar or
vertical structures (Figure 6.7), which in theory both give extremely
similar operating characteristics. Although the vertical structure is
particularly interesting for the construction of fast devices with a high
response coefficient, the current limitations in III-V nitride
technologies (connected in particular with damage to the material
during the engraving of the mesa) result in a performance degradation,
particularly in terms of the bandwidth and the noise level.
Figure 6.7. Schematic diagram of AlGaN Schottky photodiodes in
(a) vertical and (b) planar configurations
Figure 6.8 shows the typical current-voltage (I-V) characteristics of
planar Schottky photodiodes. In GaN-based devices, the ideality factor
is of the order of 1.2, with a series resistance of 20-50 : and a leakage
resistance greater than 1 G:. The leakage current increases as a
function of the fraction of aluminum, and the ideality factor also
increases and can reach values close to 4. Consequently, the measured
I-V characteristic of AlGaN-based diodes does not make it possible to
deduce precise and reliable information about the height of the
Schottky barrier, due to the high values of the ideality factor. The
variation of the capacitance as a function of the bias voltage leads to
barrier heights of 0.84 eV and 1.02 eV respectively for semi-
transparent Ni and Au contacts on GaN. The height of the barrier as a
function of the molar fraction of aluminum for Schottky diodes with
nickel and gold contacts is shown in Figure 6.9.
Figure 6.8. Typical current-voltage characteristics for Schottky diodes
made of Al
x
Ga
1x
N on sapphire
6.4.2.2. Spectral response
In contrast to photoconductors, the photocurrent of Schottky
barrier devices depends linearly on the incident light power (inset in
Figure 6.10). This linearity is independent of the size of the diode, the
nature of the metal used to fabricate the Schottky diode and the
excitation wavelength, for values situated both above and below the
bandgap energy of the semiconductor. The decrease in the response
coefficient observed when the fraction x of aluminum increases results
from two effects. Firstly, at a constant quantum efficiency, the
response coefficient decreases at the shortest wavelengths. Also, the
penetration depth of the light is larger than the SCR, and so the
response coefficient depends on the diffusion length of the minority
charge carriers, which decreases with the concentration of aluminum.
Figure 6.9. Variation of the height of the Schottky barrier
as a function of the molar fraction of Al
Figure 6.10 shows that the UV/visible contrast is more than three
orders of magnitude. The cutoff wavelength develops from 362 nm to
293 nm. The steepness of the cutoff slope of Schottky photodetectors
shows that the only limitation in the UV/visible contrast is that caused
by absorption by deep defects, in contrast to the results obtained for
photoconductors. The spectral response is flat for wavelengths smaller
than those of the bandgap, which is an advantage that Schottky
photodiodes have over other types of photodetectors, particularly p-n
junction photodetectors. Since the SCR lies just below the surface of
the semiconductor, there is no decrease in the quantum efficiency at
short wavelengths as is normally seen in p-n and p-i-n photodiodes,
where the photogenerated carriers must diffuse across the layer above
before reaching the SCR at the level of the junction. In junction
devices based on GaN, this problem is accentuated by the high density
of dopants required to achieve p-type conductivity, which results in a
very thin SCR and a material which has a very low mobility. From
this point of view, the choice of Schottky photodiodes based on
Al
x
Ga
1-x
N has more potential than p-n or p-i-n devices for
constructing wideband UV detectors.
Figure 6.10. Normalized spectral response at ambient temperature of
Schottky photodiodes for different values of the molar fraction of Al.
Inset:: variation of the photocurrent as a
function of the incident light power
The maximum value of the response coefficient obtained in AlGaN
Schottky photodiodes is limited by the optical transmission of the
semi-transparent Schottky contact, and by the diffusion length L
h
of
the minority carriers in this material. For radiation with an energy
close to the bandgap of the semiconductor, a certain proportion of the
photons may cross the SCR without absorption and penetrate deep
into the volume of the material. In order to be collected, the electron-
hole pairs must therefore diffuse towards the SCR at the surface, and a
low value of L
h
results in a decrease in the response coefficient. L
h
is
a function of the mobility and the lifetime of the minority charge
carriers, and is an increasing function of the quality of the material. It
follows that the use of thick AlGaN layers (thickness greater than
1 m) is preferred because they allow a reduction in the density of
dislocations, and hence an increase in the diffusion length and the
response coefficient; it also results in a reduction of the dark current.
The response time of these photodetectors is mainly limited by the
RC time constant of the device, where C is the sum of the internal
capacitance of the diode and the load capacitance, and R is the sum of
the load resistance and the series resistance of the device. Response
times around 50 nanoseconds are typical for diodes of diameters
between 200 and 300 m. The capacitance of the depletion region is
given by the expression:
C
SCR
= A
qN
D
H
S
H
0
2 <
0
V
( )
[6.10]
where A is the area of the Schottky contact, N
D
is the density of
dopants, H
s
is the dielectric constant of the semiconductor, H
0
is the
vacuum permittivity, <
0
is the height of the Schottky barrier seen
from the semiconductor side, and V is the bias voltage. Since the
series resistance does not depend on the bias voltage, the response
time of the device decreases in proportion to (<
0
+V)
1/2
.
For planar devices, R
S
and C
SCR
are proportional to (PN
D
)
1
and
N
D
1/2
respectively, where is the electronic mobility. The maximum
bandwidth of these devices is therefore proportional to PN
D
1/2
. Since
the mobility increases as the doping level falls, there exists an
optimum doping level, of the order of 10
18
cm
3
, for which the
response time is minimized. This effect is accentuated for AlGaN
alloys with high concentrations of aluminum, because of the difficulty
in achieving good quality ohmic contacts.
6.4.2.4. Noise and detectivity
The 1/f noise dominates at low frequencies. Its spectral noise
power density, S
n
, satisfies the relationship:
S
n
=s
0
I
dark
D
f
J
[6.11]
where I
dark
is the dark current, f is the frequency, and s
0
, J and D are
dimensionless parameters whose values are close to 1 and 2
respectively. As well as the 1/f noise, the contribution of the shot noise
must be included. The noise equivalent power (NEP) can therefore be
written in the form:
NEP =
i
shot
2
i
1 f
2
R
i
[6.12]
where i
shot
is the shot noise current, i
1/f
is the 1/f noise current and R
i
is
the response coefficient of the device. The value of <i
shot
2
> is given
by:
i
shot
2
= 2qI
dark
'f [6.13]
where 'f is the bandwidth of the photodiode, and I
dark
represents the
dark current. NEPs of 8 pW/Hz
1/2
and 41 pW/Hz
1/2
have been
obtained for Schottky GaN/Au and Al
0.22
Ga
0.78
N/Au photodiodes
respectively, at a reverse-bias voltage of 2 V. The corresponding
detectivity values are 6.1 u 10
9
W
1
H
1/2
cm and 1.2 u 10
9
W
1
Hz
1/2
cm respectively.
6.4.2.5. Schottky barrier photodiodes made from GaN grown using
epitaxial lateral overgrowth (ELOG)
Despite these promising characteristics, heteroepitaxial growth of
GaN results in a high density of dislocations (~ 10
8
10
9
cm
3
), which
limits the UV/visible contrast in GaN-based photodetectors. The
recent development of the technique of epitaxial lateral overgrowth
(ELOG) has reduced this density of dislocations by at least two orders
of magnitude [NAM 97, USU 97].
Schottky barrier photodiodes made with epitaxial GaN layers
grown using the ELOG technique offer an order of magnitude increase
in the UV/visible contrast compared to GaN-based devices grown on
sapphire using standard methods. The dark current, whose value lies
below 1 nA/cm
2
for a reverse-bias voltage of 1 V, is significantly
smaller than that of Schottky photodiodes based on GaN/sapphire. As
the response time of the device is limited by the time constant RC, the
weaker residual doping in these epitaxial layers makes it possible to
obtain bandwidths greater than 30 MHz, 12 MHz and 8 MHz for
devices with diameters equal to 200 m, 400 m and 600 m
respectively. A detectivity as high as 5 u 10
11
W
1
Hz
1/2
cm has been
measured for GaN ELOG photodiodes with a diameter of 400 m,
with a bias voltage of 3.4 V.
6.4.2.6. Application of AlGaN Schottky barrier photodetectors for the
simulation of the biological effects of UV light
UV radiation produces a wide range of biological effects [DEG 94,
MCK 87]:
pigmentation: maximum effect at O = 360400 nm;
erythema (sunburn): maximum sensitivity at O < 297 nm;
synthesis of vitamins D2 and D3: O = 249 315 nm. Maximum
yield at 290 nm;
damage to plants: O < 317 nm;
bactericidal action: O = 210310 nm. Maximum effects at
254 nm;
carcinogenic effects: UVB and UVC. Maximum at 310 nm;
DNA damage: O < 320 nm. Effect grows rapidly as the
wavelength decreases.
Simple, accurate, reliable and low-cost devices are hence required
to evaluate the biological effects of UV radiation. UV detectors with a
wide spectral band have been developed to study the effects of UVA
and UVB on the skin [MCK 87]. Commercial UV dosimeters use
photodiodes with a small bandgap (Si, GaAs, GaP) combined with a
series of filters inserted into the path of the incident light. Muoz et al.
[MUN 00] showed that the spectral response of AlGaN Schottky
photodiodes can match the UVA and B erythema action spectrum,
with the appropriate selection of an appropriate molar fraction and
carefully selected growth conditions (Figure 6.11). In this way, the
response coefficients of these devices can give direct information on
the biological effects of UV light. For this application, AlGaN
photodetectors therefore offer a flexible solution with a low cost and
minimum footprint, since they require neither filters nor temperature
stabilization.
Figure 6.11. Normalized spectral response of an AlGaN Schottky photodiode
compared with the sunburn action spectrum (international standard
set by the International Commission on Illumination (CIE))
6.4.3. MSM photodiodes
MSM photodiodes consist of two interdigitated Schottky contacts
deposited onto a planar surface (see Chapter 5). The main advantage
of these structures, based on AlGaN, is their ease of fabrication. The
typical structure of such an MSM diode involves two interdigitated
Ni-Au Schottky contacts deposited on an epitaxial layer of AlGaN,
not intentionally doped, of micron-scale thickness. Their dark current
is very weak and follows the thermionic emission model with a value
of the Richardson constant equal to 26 Acm
-2
K
-2
and a barrier height
of 1.04 eV for GaN.
Their spectral response is relatively flat above the bandgap, with a
steep cutoff slope which shifts towards shorter wavelengths as the
proportion of aluminum increases. The photocurrent varies in a quasi-
linear manner with the incident optical power. The UV-visible
contrast varies from around three orders of magnitude for the weakest
bias voltages (typically less than 5V) to four or five orders for the
highest bias voltages. This accentuation of the contrast with the bias
voltage can be explained by the appearance of a gain. At present, there
is no satisfactory explanation available for the origin of this gain.
Despite the existence of this gain, the dynamic behavior of these
photodetectors is for the most part limited by the RC time constant and
the transit time of the photocarriers [MON 99c].
Their NEP lies between a few pW/Hz
1/2
and a few tens of
pW/Hz
1/2
depending on the geometry of the device and the bias
voltage [MON 99b].
6.4.4. p-n and p-i-n photodiodes
The performance limitations of the first GaN p-n junction
photodiodes were mainly caused by a high resistivity of the p-type
layer, connected with difficulties associated with magnesium p doping,
and the high resistivity of the associated ohmic contact. Progress made
in understanding the growth conditions of the materials, and in the
technology of the ohmic contacts, then made it possible to make p-n
junction GaN photodiodes with a fast temporal response (response
time of the order of 100 ns) and a very low noise level. The
performance of these devices was further improved by the
introduction of a region that was not intentionally doped (p-i-n
photodiodes) and through the use of AlGaN/GaN heterostructures.
Arrays of heterostructure Al
0.20
Ga
0.80
N/GaN p-i-n photodiodes with
dimensions 32 u 32 and 128 u 128 pixels [BRO 00] have been made
with good performance in terms of their response coefficient and
detectivity. p-i-n photodiodes in Al
x
Ga
1x
N, sometimes with high
levels of aluminum (up to 76% [SAN 00]), have also been described.
The optimization of these last devices is however limited by the
intrinsic difficulties in achieving a high level of p doping in
Al
x
Ga
1x
N materials, which can only be made with difficulty to
exceed 10
17
cm
3
, and by the associated ohmic contact whose
resistivity remains high. The behavior of the device is thus doubly
penalized: firstly, the response time of a Al
x
Ga
1-x
N p-i-n
photodetector cannot be made small enough, because the series
resistance component due to the resistance of the electrical contacts
remains large, and the device also contributes a significant level of
noise associated with an ohmic contact of insufficient quality, which
leads to detectivity values which are notably smaller than those that
can be obtained with p-n and p-i-n photodiodes based on GaN.
6.4.4.1. Static characteristics
In p-n and p-i-n photodiodes based on Al
x
Ga
1x
N, the linearity of
the photocurrent as a function of the incident light power extends over
a scale of more than five orders of magnitude [WAL 99]. The
response coefficient of p-n and p-i-n homojunction GaN photodiodes
typically lies in the range 0.100.15 A/W, which corresponds to
external quantum efficiencies of between 30 and 44%. These results
are improved by the use of a Al
x
Ga
1x
N layer in the illuminated region
(which can be n- or p-doped), in order to avoid the loss of carriers
through diffusion. A response coefficient of 0.2 A/W at O = 365 nm
has been measured for a sapphire-Al
0.28
Ga
0.72
N(n)-GaN(i)-GaN(p)
heterostructure illuminated by the rear face through the sapphire
[YAN 98]. A maximum in the response coefficient of 57 mA/W was
reported at O = 287 nm, which corresponds to a quantum efficiency of
25% [PAR 99]. An increase in this coefficient is observed in all cases
as a function of the bias voltage [WAL 99]. This result confirms that
the sensitivity is limited by the diffusion length of the charge carriers.
As the bias voltage increases, the width of the SCR increases, such
that the carriers created through photo-ionization processes in the
regions further from the junction are collected.
A UV/visible contrast of more than four orders of magnitude is
observed for all these devices. The small diffusion length of the
carriers can affect the spectral response above the bandgap of p-i-n
photodiodes based on Al
x
Ga
1x
N [XU 97]. The absorption coefficient
increases as the wavelength decreases, meaning that the absorption of
the light occurs closer to the surface. The result is that electron-hole
pairs are generated above the junction in the upper layer and must
diffuse to the junction in order to be collected. A short diffusion
length therefore leads to a decrease in the response coefficient for
shorter wavelengths, which is all the more marked when the surface
layers are thicker (of the order of 0.2 m). p-i-n detectors based on
group III nitrides are characterized by a sharp cutoff at long
wavelengths, whose energy value is very closely linked to the
bandgap of the material forming the active region. For certain
applications, however, an additional cutoff wavelength is also required
at short wavelengths, defining a spectral window for UV detection. In
their theoretical study, Pulfrey and Nener [PUL 98] investigated the
possibility of using p-i-n Al
x
Ga
1x
N/GaN heterostructures as bandpass
UV detectors, with a high energy cutoff wavelength at the bandgap of
Al
x
Ga
1x
N. In this case, the Al
x
Ga
1x
N layer must be thick enough not
only to accentuate the absolute response peak, but also to act as an
integral high-pass filter. Devices equipped with a 1 m thick
Al
0.10
Ga
0.90
N layer [KRI 98] have demonstrated that it is possible to
achieve rejection on a level of more than two orders of magnitude in
the short wavelength region.
GaN-based p-n and p-i-n photodiodes have a temporal response
which is in general limited by the RC time constant, and show an
exponential fall in the photocurrent. Response times of the order of 25
nm have been measured with zero bias voltage on devices whose
optical surface area is 200 u 200 m
2
[OSI 97]. This response time
drops to 10 ns for a bias voltage of 6 V, which can be explained by a
reduction in the capacitance of the junction associated with the applied
voltage. Faster response times can be obtained with AlGaN/GaN
heterojunctions [XU 97]. However, the response time of these devices
is often non-exponential, with characteristic fall times that are more
complicated than a simple RC effect [CHE 95, VAN 97, WAL 98].
This behavior can be attributed to the presence of a defect in the
semiconductor, with an activation energy of 99 meV and a high
concentration (of the order of 10
18
cm
3
) which can probably be
explained by substituted magnesium atoms.
GaN p-i-n photodiodes with response times below 1 ns have been
constructed [CAR 99]. This result is obtained by increasing the
thickness of the i layer to 1 m, which makes it possible to build a
device which has a low capacitance. In this case, the factor limiting
the temporal response is the transit time of the carriers. The only
drawback of this particular type of device is its low response
coefficient, which is around 0.03 A/W. Such devices are of course
ideal for the detection of rapid signals.
6.4.4.3. Noise
Under illumination, the spectral density of noise under reverse-bias
is well described by the equation S
n
=2.q.I
ph
(shot noise). A very low
normalized noise equivalent power, of the order of 7 fW/Hz
1/2
, has
been obtained in p-n junction photodiodes for a reverse-bias voltage of
3 V [OSI 97]. Since these devices have an optical area of 200 u 200
m, their corresponding normalized detectivity is high, of the order of
3u10
12
W
1
Hz
1/2
cm.
6.4.4.4. p-i-n photodiodes on laterally-grown GaN
Al
0.33
Ga
0.77
N p-i-n photodiodes on GaN have been built using
lateral growth, by means of a SiO
2
mask consisting of 35 m
widebands with 5 m openings [PAR 99]. Diode mesas, with
dimensions of 10 u 100 m
2
, have been made in the regions of the
material corresponding to the wings of the lateral growth (outside
the coalescence region, and outside the openings in the SiO
2
mask).
Additional devices with mesa dimensions of 30 u 100 m
2
have also
been built by way of comparison. These particular devices also cover
the coalescence region situated at the junction of the wings of lateral
growth. Finally, devices with a similar form with square 300 u
300 m
2
mesas have been made from GaN on standard sapphire. Low
dark current densities, of the order of 10 nA/cm
2
, have been measured
for the smallest devices lying on the wings of lateral growth. This
leakage current density is an order of magnitude smaller than that
measured for devices which include the coalescence regions, and more
than 6 orders of magnitude smaller than that of devices made from
GaN on standard sapphire. This low value of the leakage current can
be linked to the strong reduction in the density of dislocations, since
these defects lead to a strengthening of the transport mechanism,
assisted by the tunneling effect [YU 98b]. In addition, the spectral
response has a more abrupt cutoff in the case of devices built on the
wings of the growth, in a similar way both when the coalescence
region is covered and when it is not. All these responses are consistent
with similar observations made for Schottky photodiodes made with
laterally-grown GaN [MON 99d].
6.4.4.5. Avalanche photodiodes based on GaN
The mechanism of avalanche photodiodes is analyzed in detail in
Chapter 3. The unusual requirement for GaN-based materials is that
very high fields are required to achieve multiplication, as calculations
of the ionization coefficients for electrons and holes in GaN show
[KOL 95, OGU 97]. Although a homogenous avalanche multiplication
of carriers has recently been shown for PIN photodiodes based on
GaN [MCI 00], with a response time between 100 and 500 ns, the
high density of defects in these materials makes it difficult to achieve
a homogenous multiplication across the whole active area of the
device. Also, to overcome the limitations in noise due to ionization
coefficients for electrons and holes that are too close together, Ruden
[RUD 99] proposed a hybrid structure consisting of an AlGaN/Si
heterostructure, with multiplication taking place in the silicon.
6.4.5. Phototransistors
The operation of phototransistors is described in Chapter 4. As far
as GaN-based devices are concerned, the literature refers to bipolar
and field effect transistors [KHA 95, YAN 98].
6.4.5.1. Bipolar phototransistors
In a bipolar transistor, the reverse-biased base-collector
junction acts like a p-n junction photodiode, and its photocurrent
is amplified by the transistor effect. In its most up to date
configuration, the base contact is not connected (floating-base
mode of operation). Yang et al. [YAN 98] report the fabrication of a
GaN(n)/GaN(p)/GaN(i)/Al
0.20
Ga
0.80
N(n) heterojunction phototransistor.
Electrical contacts were present on the collector and the emitter, and
the base was left floating. The UV light enters the device through the
sapphire substrate, crosses the Al
0.20
Ga
0.80
N(n) layer, and is absorbed
in the GaN(i) layer. The photogenerated electron-hole pairs are
separated by the electric field in the i region, and the electrons and
holes travel to the collector and the base respectively. The
accumulation of holes in the floating base increases the injection of
electrons from the emitter, which results in a current gain. A gain of
more than 10
5
has been demonstrated.
GaN-based phototransistors exhibit a sub-linear behavior of the
photocurrent as a function of the incident light power as well as
showing persistent photoconductivity (PPC), analogous to that also
recently observed in photoconductors. In their normal operating
regime, holes ought to recombine in the base with electrons injected
from the emitter. However, the holes are trapped at defects which
reduces their recombination rate and leads to persistent charge effects.
Yang et al. [YAN 98] showed that this recombination is increased by
the application of a bias to the detector in order to force the holes to
enter the emitter. The voltage impulse acts as a sort of electrical reset
to zero.
In order to avoid persistent effects, the phototransistor is subjected
to a reset to zero voltage impulse before each measurement of the
response coefficient (a voltage between 7 and 10 V is used for this),
and the photocurrent is then measured at a bias voltage of 3 to 4 V.
The gain is strongly dependent on the frequency, so that the gain-
bandwidth product remains constant. An excellent UV/visible contrast
of 8 orders of magnitude was obtained [YAN 98].
6.4.5.2. Field effect phototransistors
AlGaN(n)/GaN(n)/GaN(i) heterostructure field effect transistors
have also been used in photodetection mode [KHA 95], with
illumination through the substrate (sapphire). The photogenerated
holes travel towards the channel, where they are rapidly drawn
towards the drain by the high electric field. These devices display high
response coefficients which can reach 3,000 A/W, with a steep cutoff
and a temporal response of the order of 200 s.
6.5. Conclusion
The development of visible blind ultraviolet photodetectors is at
present largely motivated by the large number of possible applications
in fields as diverse as UV astronomy, studies of the ozone layer, motor
control, missile plume detection, flame detection, secure
communications in space, etc. With these goals in mind,
photoconductors, Schottky photodiodes, MSM photodiodes, p-n and
p-i-n photodetectors and phototransistors based on Al
x
Ga
1x
N
materials have been developed in recent years, complementing older
widely-available technologies for Si and SiC-based UV photodiodes.
It is important to emphasize that this range of nitride-based sensors is
currently bringing a new degree of improved flexibility to UV detector
fabrication in general, due mainly to the large bandgaps available with
Al
x
Ga
1x
N materials which make it possible to build detection
systems which are entirely free from intermediate spectral filters, with
performances that are at least equivalent in terms of reliability and
robustness to other currently-available technologies that are
themselves already very high performance.
Al
x
Ga
1x
N photoconductors offer a high internal gain at ambient
temperatures (~100 for P
opt
= 1 W/m
2
). However, this gain is
associated with a sub-linear behavior as a function of the incident
optical power, a low UV/visible contrast and significant and
undesirable persistent photoconductivity effects. These drawbacks
make them less suitable for most applications. The use of a modulated
optical signal makes it possible to considerably improve the linearity
and UV/visible contrast. However, these photoconductors lose much
of their advantage in this configuration, above all because the
response coefficients of the devices is significantly reduced in such a
configuration, and the measurement system becomes much more
complex and as a result much more expensive.
Schottky photodiodes have the advantage of a uniform and flat
spectral response when they are excited at energies above the bandgap,
independent of the incident optical power and the temperature. They
also offer a steep cutoff slope, with a UV/visible contrast which is of
the order of 10
3
. Their temporal response is limited by the RC product,
with minimum response times typically of the order of nanoseconds.
This clearly demonstrates that these devices are suited to
environmental applications and the construction of UV photodetector
arrays.
MSM photodiodes with very low dark currents have been built.
These devices have a photocurrent which varies linearly as a function
of the incident light power and makes it possible to achieve a
UV/visible contrast of 10
4
. Given their significant bandwidth and their
low noise level, these devices can be a judicious choice to meet the
needs of detection in the domain of visible blind optical
communications. Furthermore, the possibility could be considered of
integrating Al
x
Ga
1x
N MSM photodiodes with field effect transistors
based on group III nitrides in the context of the monolithic fabrication
of optical receivers.
p-n and p-i-n junction photodiodes are linear as a function of the
optical power, and can achieve a UV/visible contrast of 10
4
. However,
their response time is normally limited by effects associated with the
presence of trapping levels linked to magnesium, to which a degree of
degradation in the spectral response can also be attributed. The
minimum cutoff wavelength is currently limited by the difficulty of
achieving a high p doping in Al
x
Ga
1x
N alloys at high concentrations
of aluminum. In spite of very promising published results,
improvements are still currently needed in the p doping of these
materials, in order to increase the performance of these devices, as
well as their reliability.
Phototransistors offer the combination of a very high gain with a
record UV/visible contrast of 10
8
. As a result, these devices are
promising for applications requiring a high spectral resolution, in spite
of the narrowness of their bandwidth which makes them poorly-suited
to high-frequency operation.
To summarize, current results confirm that Al
x
Ga
1x
N alloys are
the best choice in terms of semiconductors suited to UV
photodetection. However, the performance of these devices is still
currently limited by the high density of defects present in the
heteroepitaxial layers. In consequence, very significant improvements
in photodetector devices are anticipated from work currently in
progress in the fields of epitaxial growth and the technology of group
III nitrides, such as the technique of lateral growth.
6.6. Bibliography
[ANI 92] ANIKIN M.M., ANDREEV A.N., PYATKO S.N., SAVKINA N.S., STRELCHUK
A.M., SYRKIN A.L., CHELNOKOV V.E., UV photodetectors in 6HSiC, Sensors
and Actuators: A. Physical 33 (1-2), pp. 91-93, 1992.
[ARU 00] ARULKUMARAN S., EGAWA T., ZHAO G.Y., ISHIKAWA H., JIMBO T., UMENO
M., Electrical characteristics of Schottky contacts on GaN and Al. 0:11. Ga. 0:89.
N, Japanese Journal of Applied Physics. 39, L351, 2000.
[BRO 00] BROWN J.D., BONEY J., MATTHEWS J., SRINIVASAN P., SCHETZINA J.F.,
NOHAVA T., YANG W., KRISHNANKUTTY S., UV-specific digital camera based on
a 128x128 focal plane array of GaNAlGaN PIN photodiodes, MRS Internet
Journal of Nitride Semiconductor Research, 5, article 6, 2000.
[CAN 89] CANFIELD L.R., KERNER J., KORDE R., Stability and quantum efficiency
performance of silicon photodiode detectors in the far ultraviolet, Applied Optics,
28, 3940, 1989.
[CAR 99] CARRANO J.C., LI T., EITING C.J., DUPUIS R.D., CAMPBELL J.C., Very fast
metal-semiconductor-metal ultraviolet photodetectors on GaN with submicron
finger width, J. of Electronic Materials, 28, 325, 1999.
[CHE 95] CHEN Q., KHAN M.A., SUN C.J., YANG J.W., Visible-blind ultraviolet
photodetectors based on GaNp-n junctions, Electronics Lett. 31, 1995.
[DEG 94] DE GRUJI F.R., VAN DER LEUN J.C., Estimate of wavelength dependency
of ultraviolet carcinogenesis and its relevance to the risk assessment of a
stratospheric ozone depletion, Health Physics, 67, 314, 1994.
[EDM 94] EDMOND J.A., KONG H.S., CARTER JR C.H., Blue LEDs, photodiodes and
high-temperature rectifiers in 6HSiC, Physica B 185, 453, 1993.
[GAR 98] GARRIDO J.A., MONROY E., IZPURA I., MUOZ E., Photoconductive gain
modelling of GaN photodetectors, Semiconductor Science and Technology, 13,
563, 1998.
[GOL 99] GOLDBERG Y.A., Semiconductor near-ultraviolet photoelectronics,
Semiconductor Science and Technology, 14, R41, 1999.
[HAN 98] HANSEN P.J., STRAUSSER Y.E, ERICKSON A.N., TARSA E.J., KOZODOY P.,
BRAZEL E.G., IBBETSON J.P., MISHRA U., NARAYAMURTI V., DEN BAARS S.P.,
SPECK J.S., Scanning capacitance microscopy imaging of threading dislocations
in GaN films grown on (0001) sapphire by metalorganic chemical vapor
deposition, Applied Physics Lett., 72, 2247, 1998.
[KHA 95] KHAN M.A., SHUR M.S., CHEN Q., KUZNIA J.N., SUN C.J., Gate
photodetector based on GaN/AlGaN heterostructure field effect transistor,
Electronics Lett., 31, 398, 1995.
[KHA 97] KHAN M.R.H., NAKAYAMA H., DETCHPROHM T., HIRAMATSU K., SAWAKI N.,
A study on barrier height of Au-Al
x
Ga
1-x
N Schottky diodes in the range 0, Solid
State Electronics, 41, 287, 1997.
[KOL 95] KOLNIK J., H.OGUZMAN I., BRENNAN K.F., WANG R., RUDEN P.P., WANG Y.,
Electronic transport studies of bulk Zincblende and Wurzite phases of GaN
based on an ensemble Monte Carlo calculation including a full zone band
structure, J. of Applied Physics, 78, 1033, 1995.
[KOR 87] KORDE R., GEIST J., Quantum efficiency stability of silicon photodiodes,
Applied Optics, 26, 5284, 1987.
[KRI 98] KRISHNANKUTTY S., YANG W., NOHAVA T., RUDEN P.P., Fabrication and
characterization of GaN/AlGaN ultraviolet-band heterojunction photodiodes,
MRS Internet Journal of Nitride Semiconductor Research, 3, article 7, 1998.
[KUK 98] KUKSENKOV D.W., TEMKIN H., OSINSKY A., GASKA R., KHAN M.A.,
Origin of conductivity and low-frequency noise in reverse-biased GaN p-n
junction, Applied Physics Lett., 72, (11), 1365-1367, 1998.
[MAI 00] MAINWOOD A., Recent developments of diamond detectors for particles
and UV radiation, Semiconductor Science and Technology, 15, R55, 2000.
[MCI 00] MCINTOSH K.A., MOLNAR R.J., MAHONEY L.J., MOLVAR K.M., EFREMOWN.
JR., VERGHESE S., Ultraviolet photon counting with GaN avalanche photodiodes,
Applied Physics Lett., 76, 3938, 2000.
[MCK 87] MCKINLEY A.F., DIFFEY B.L., A reference action spectrum for ultraviolet
induced erythema in human skin, CIE Journal 6, 17, 1987.
[MON 99a] MONROY E., CALLE F., MUNOZ E., OMNES F., BEAUMONT B., GIBART P.,
Visible-blindness in photoconductive and photovoltaic AlGaN ultraviolet
detectors, J. of Electronic Materials, 28, 238, 1999.
[MON 99b] MONROY E., CALLE F., MUNOZ E., OMNES F., GIBART P., High
UV/visible contrast photodiodes based on epitaxial lateral overgrown GaN layers,
Electronics Lett., 35, 240, 1999.
[MON 99c] MONROY E., CALLE F., MUNOZ E., OMNES F., AlGaN metal-
semiconductor-metal photodiodes, Applied Physics Lett., 74, 3401,1999.
[MON 99d] MONROY E., CALLE F., MUNOZ E., BEAUMONT B., OMNES F., GIBART P.,
High UV/visible contrast photodiodes based on epitaxial lateral overgrown GaN
layers, Electronics Lett., 35, 1488, 1999.
[MON 00] MONROY E., CALLE F., PAU J.L., SANCHEZ F.J., MUOZ E., OMNES F.,
BEAUMONT B., GIBART P., Analysis and modeling of Al
x
Ga
1-x
N-based Schottky
barrier photodiodes, J. of Applied Physics, 88, 2081, 2000.
[MUN 00] MUOZ E., MONROY E., CALLE F., OMNES F., GIBART P., AlGaN
photodiodes for monitoring the solar UV radiation, Journal of Geophysical
Research, 105, 4865, 2000.
[NAM 97] NAM O.H., BREMSER M.D., ZHELEVA T., DAVIS R.F., Lateral epitaxy of
low defect density GaN layers via organometallic vapor phase epitaxy, Applied
Physics Lett., 71, 2638, 1997.
[OGU 97] OGUZMAN I.H., BELLOTTI E., BRENNAN K.F., KOLNIK J., WANG R., RUDEN
P.P., Theory of hole initiated impact ionization in bulk zincblende and Wurtzite
GaN, J. of Applied Physics, 81, 7827, 1997.
[OSI 97] OSINSKY A., GANGOPADHYAY S., GASKA R., WILLIAMS B., KHAN M.A.,
KUKSENKOV D., TEMKIN H., Low noise p--n ultraviolet photodetectors, Appl.
Phys. Lett., 71, 2334, 1997.
[PAR 99] PARISH G., KELLER S., KOZODOY P., IBBETSON J.P., MARCHAND H., FINI P.T.,
FLEISCHER S.B., DEN BAARS S.P., MISHRA U.K., TARSA E.J., High-performance
(Al,Ga)N-based solar-blind ultraviolet p-i-n detectors on laterally epitaxially
overgrown GaN, Applied Physics Lett. 75, 247, 1999.
[POL 98] POLYAKOV A.Y., SMIRNOV N.B., USIKOV A.S., GOVORKOV A.V., PUSHNIY B.,
Studies of the origin of the yellow luminescence band, the nature of nonradiative
recombination and the origin of persistent photoconductivity in n-GaN films,
Solid State Electronics, 42, 1959, 1998.
[PUL 98] PULFREY D.L., NENER B.D., Suggestions for the development of GaN-
based photodiodes, Solid State Electronics, 42, 1731, 1998.
[RAZ 96] RAZEGHI M., ROGALSKI A., Semiconductor ultraviolet detectors, J. of
Applied Physics, 79, 7433, 1996.
[RHO 88] RHODERICK E.H., WILLIAMS R.H., Metal-Semiconductor Contacts,
Clarendon Press, Oxford, 1988.
[RUD 99] RUDEN P.P., Materials theory based modeling for III-nitride devices,
SPIE Proceedings, 3629, 174, 1999.
[SAN 00] SANDVIK P., WALKER D., KUNG P., K.MI, SHAHEDIPOUR F., KUMAR V.,
ZHANG X., J.DIAZ, C.JELEN, M.RAZEGHI, Solar-blind AlxGa1-xN p-i-n
photodetectors grown on LEO and non-LEO GaN, SPIE Proceedings, 3948, 265,
2000.
[SHE 99] SHEN B., YANG K., ZHAN L., CHEN Z., ZHOU Y., CHEN P., ZHAN R., HUANG
Z., ZHOU H., ZHENG Y., Study of photocurrent properties of GaN ultraviolet
photoconductor grown on 6H-SiC substrate, Japanese Journal of Applied
Physics, 38, 767, 1999.
[USU 97] USUI A., SUNAKAWA H., SAKAI A., YAMAGUCHI A., Thick GaN epitaxial
growth with low dislocation density by hydride vapor phase epitaxy, Japanese
Journal of Applied Physics, 36, L899, 1997.
[VAN 97] VAN HOVE J.M., HICKMAN R., KLAASSEN J.J., CHOW P.P., RUDEN P.P.,
Ultraviolet-sensitive, visible blind GaN photodiodes fabricated by molecular
beam epitaxy, Applied Physics Lett., 70, 2282, 1997.
[WAL 98] WALKER D., SAXLER A., KUNG P., ZHANG X., HAMILTON M., DIAZ J.,
RAZEGHI M., Visible blind GaN p-i-n photodiodes, Applied Physics Lett., 72,
3303, 1998.
[WAL 99] WALKER D., KUNG P., SANDVIK P., WU J., HAMILTON M., LEE I.H., DIAZ J.,
RAZEGHI M., AlxGa1-xN p-i-n photodiodes on sapphire substrates, Proc. SPIE,
3629, 193, 1999.
[WHI 96] WHITFIELD M.D., MCKEAG R.D., CHAN S.S.M., JACKMAN R.B., A thin film
diamond photodiode for ultra-violet light detection, Applied Physics Lett., 68,
290,1996.
[XU 97] XU G.Y., SALVADOR A., KIMW., FAN Z., LU C., TANG H., MORKOC H., SMITH
G., ESTES M., GOLDENBERG B., YANG W., KRISHNANKUTTY S., High speed, low
noise ultraviolet photodetectors based on GaN p-i-n and AlGaN (p)-GaN(i)-
GaN(n) structures, Applied Physics Lett., 71, 2154, 1997.
[YAN 98] YANG W., NOHAVA T., KRISHNANKUTTY S., TORREANO R., MCPHERSON S.,
MARSH H., Back-illuminated GaN/AlGaN heterojunction photodiodes with high
quantum efficiency and low noise, Applied Physics Lett., 73, 978,1998.
[YU 98a] YU L.S., QIAO D.J., XING Q.J., LAU S.S., BOUTROS K.S., REDWING J.M.,
Analysis and modeling of Al
x
Ga
1-x
N-based Schottky barrier photodiodes,
Applied Physics Lett., 73, 238, 1998.
[YU 98b] YU L.S., LIU Q.Z., XING Q.J., QIAO D.J., LAU S.S., REDWING J.D., The role
of the tunneling component in the current-voltage characteristics of metal-GaN
Schottky diodes, J. of Applied Physics, 84, 2099, 1998.
Chapter 7
Noise in Photodiodes and
Photoreceiver Systems
In systems transmitting information by means of photons, the
recovery of a useful signal must necessarily involve an optical-
electrical (photon-electron) transformation. This transformation,
which makes use of an optoelectronic system, is, as with any system,
fundamentally limited by the background noise. The aim of this
chapter is to discuss the origin of this noise and also to study various
means of minimizing it in order to obtain the best signal-to-noise ratio,
a value which influences the quality of the optical detection. In section
7.1 we will recall the basics of signal analysis, leading to the
frequency analysis of the power of random signals. After recalling the
different sorts of noise associated with the most relevant physical
processes (sections7.2 and 7.3), in section 7.4 we will present the
main results concerned with the analysis of noisy electrical circuits.
Section 7.5 is dedicated to the study of noise in photodetectors, which
we choose to limit to PIN and avalanche photodiodes. Section 7.6
discusses the problems of background noise associated with a
photodiode-amplifier pair forming the photoreceiver, and lays out the
methods of obtaining the best signal-to-noise ratio. Finally, in order to
Chapter written by Robert ALABEDRA and Dominique RIGAUD.
illustrate these techniques, we present the complete background noise
calculation for a PIN-FET photoreceiver.
7.1. Mathematical tools for noise
Before discussing the treatment of random signals, we will recall a
summary of the treatment of known signals in an attempt to justify a
frequency approach as compared to a temporal approach for the
treatment of the power or the energy of random signals.
7.1.1. Known signals with finite energy or power
All systems are governed by a differential equation in the time
domain, and we will characterize them by their impulse response h(t).
The output signal s(t) for an input signal i(t) can be calculated in the
time domain either by solving this differential equation or by
calculating the convolution of h(t) with i(t):
s t
()
=i t
()
* h t
()
= e W
( )
h t W
( )
dW =
h W
( )
e t W
( )
dW
[7.1]
The signal s(t) has the same dimensions as i(t) which implies that
h(t) has the dimensions of inverse time. Thus, in the case of an RC
circuit, the impulse response is:
h t
()
=
1
RC
exp
t
RC
(
(
)
)
in s
1
[7.2]
In order to avoid calculating the convolution, we use Fourier
analysis which transforms the convolution in the time domain to a
simple product in the frequency domain, and we can recover the
original s(t) through the inverse transform. On the other hand, if we
Noise in Photodiodes and Photoreceiver Systems 225
want to analyze the same system from the point of view of energy or
power, where phase information no longer has any meaning, we will
then use the autocorrelation function R
s
(W) of the signal. The
relationship between R
i
(W), the autocorrelation of the input signal, and
R
s
(W), that of the output signal, is:
R
s
W
( )
= h* W
( )
* h W
( )
* Re W
( )
[7.3]
where h*(W) is the complex conjugate of h(W).
This expression will only occasionally be used, given its double
convolution product. We will repeat this same analysis in terms of
frequency, using the concept of spectral density and the Wiener-
Khinchin theorem which shows that the spectral density G
x
(f ) of a
signal x(t) is the Fourier transform of the autocorrelation R
x
(W) of x(t)
such that:
G
x
f
( )
=T F R
x
W
( )
l
l
l
= R
x
W
( )
exp j2S f W
( )
dW
[7.4]
and its inverse
R
x
W
( )
=T F
1
G
x
f
( )
l
l
l
= G
x
f
( )
exp j2S f W
( )
df
[7.5]
If we take W = 0, which is the maximum of the autocorrelation
function, R
x
(0) represents the energy or the power of the signal. We
find:
R
x
0
( )
= G
x
f
( )
df
[7.6]
G
x
(f ) represents the energy or power distribution of the signal in
the frequency domain. We take the Fourier transform of equations
[7.1] to [7.3]:
F T R
s
W
( )
= h* W
( )
* h W
( )
* Re W
( )
l
l
l
[7.7]
which can be written as:
G
s
f
( )
= H f
( )
H * f
( )
G
i
f
( )
[7.8]
or:
G
s
f
( )
= H f
( )
2
G
i
f
( )
[7.9]
and which is much more easy to use than equation [7.3].
What we need to do now is to work out how to transfer what we
have already set out for deterministic signals to the random signals
which represent the sources of noise.
7.1.2. Random signals and background noise
In the case of random signals, we have no choice but to treat them
in terms of power or energy, and of course without any consideration
of their phase. As we did for the known signals in section 7.1.1, we
will use the autocorrelation function, the spectral power density and
the same Wiener-Khinchin theorem. This will lead us to define a class
of random signals which, fortunately for us, will cover the majority of
practical situations. This class is characterized by ergodicity and time-
stability.
DEFINITION OF ERGODICITY The average resulting from many
simultaneous experiments is the same as that obtained by the temporal
mean across a single experiment.
DEFINITION OF TIME STABILITY In its strict sense, this means that
the mean value is independent of time and the autocorrelation function
only depends on W = t
1
t
2
.
In this case, we obtain for the random signal x(t):
R
x
W
( )
= E x t
()
x t W
( )
l
l
l
[7.10]
where E is the expected value associated with the probability density
of the process under study (Poisson, Gaussian systems, etc.). Under
these conditions the autocorrelation function is practically analogous
to that for known signals which have the same properties. Now, using
this formalism, we can treat the random signals appearing in the
expression in terms of energy or power, and calculate the signal-to-
noise ratio which characterizes the receiving quality.
7.2. Fundamental noise sources
Here we are interested in noise generated by solid devices in
electronic circuits. As we have seen in section 7.1, we will remain in
the frequency domain in order to avoid convolving the input and
output autocorrelation functions. We will use the spectral densities as
our measurement of scale, and in order to recover a power or an
energy we will integrate these spectral densities over the useful
bandwidth. These spectral densities are expressed in A
2
/Hz or V
2
/Hz.
7.2.1. Thermal noise
At the terminals of a metallic resistance of value R at temperature T
in degrees Kelvin, the Brownian motion of electrons produces a
randomly-varying voltage v(t) at its open-circuit terminals, with a
mean value of zero. Bearing in mind the large number of electrons (of
the order of 10
22
cm
3
), the voltage v(t) obeys a centered Gaussian law.
For frequencies below the inverse of the mean time between two
collisions (W
r
10
12
s), the spectral density associated with this
random signal can be written in the basis of positive frequencies as:
S
v
f
( )
= 4 kTR V
2
Hz
( )
[7.11]
where k is the Boltzmann constant (1.38 u 10
23
J.K
1
). We note that
this spectral density is independent of the frequency (white noise).
7.2.2. Shot noise
Shot noise stems from the discrete nature of electrons and their
random emission and collection over time. Thus, it takes the form of a
random variable governed by the Poisson distribution law. Frequency
domain analysis of this process leads to a spectral noise current
density (short circuit current) of:
S
i
f
( )
= 2q I A
2
Hz
( )
[7.12]
where q is the elementary charge (q = 1.6 u 10
19
C). Here also we
find a white noise as long as we ignore the transit time of the carriers.
Shot noise is always encountered when the current is caused by
carriers emitted from a reservoir or across a potential energy barrier
greater than kT. This is the case for reverse-biased junctions. For PIN
photodiodes, for example, we have:
S
i total
f
( )
= 2 q I
dark
I
ph
l
l
l
[7.13]
where I
dark
is the intrinsic dark current of the photodiode and I
ph
the
photocurrent due to absorbed photons and representing the useful
information. For modern devices, I
dark
<< I
ph
and the noise is mainly
due to the photocurrent. In contrast, for detection of low light
intensities, the dark current will make a significant contribution to the
signal-to-noise ratio.
7.2.3. Multiplication noise
In a similar way to vacuum photomultipliers, avalanche
photodiodes with internal gain are based on the process of
multiplication through the impact of carriers in the space charge
region of the photodiode. In the case of vacuum photomultipliers,
there is only one type of carrier: electrons locally multiplied from one
dynode to the next. If we ignore the multiplication inertia and the
transit time of the electrons, the noise generated by this
photomultiplier is amplified shot noise. The variance in the total
multiplication factor is:
M
2
= M
2
[7.14]
and for large enough M we have:
S
i
f
( )
= 2qI
pho
M
2
[7.15]
where M is the mean value of the multiplication factor and I
pho
is the
primary photocurrent. Setting:
M =
I
total
I
pho
[7.16]
we have:
S
i
f
( )
=2qI
pho
M
2
= 2qI
pho
I
total
2
I
pho
2
[7.17]
in other words, we have a noise proportional to the total current
squared. In the case of semiconductor avalanche photodiodes, the
multiplication process involves both types of carrier across the whole
multiplication region, i.e. in the high electric field region in the space
charge region. Numerous theoretical approaches have been proposed
in the literature. To summarize these approaches, we recall that the
two types of carrier are characterized by their ionization coefficient (in
cm
1
) D for electrons and E for holes, and that these depend on the
physical characteristics of the semiconductor materials. The important
parameter is the ratio k =
D
E
. The ideal material would be one in
which only one of the two types of carrier could be ionized: either
electrons, in which case E = 0 and k = f, or holes, in which case D = 0
and k = 0. The noise characteristics would then approach that of a
vacuum photomultiplier.
For avalanche photodiodes, the expression for the spectral noise
density can be written in the general form:
S
i
f
( )
=2qI
pho
M
x
[7.18]
where x lies between 2 and 4.
Taking the noise of a vacuum photomultiplier as our reference, we
define the excess noise factor F as:
F M
( )
=
M
2
M
2
=
M
x
M
2
= M
x 2
[7.19]
in other words as the ratio of the variance of the multiplication process
in the semiconductor to the variance of the same process in the
vacuum tube. In the case of an initial injection of electrons, and for
values of k close to 1, we can approximate F M
( )
by the expression:
F M
( )
= k M 2 M
1 (
(
)
)
1k
( )
[7.20]
In particular, for k = 1 (D = E), as in germanium for example, we
have F M
( )
= M and:
S
i
f
( )
=2qI
pho
M
3
[7.21]
For silicon avalanche photodiodes, D is 50 times larger than E. For
multiplication initiated by electrons, we arrive at the expression:
S
i
f
( )
=2qI
pho
M
2 2
[7.22]
which leads to an excess noise factor of:
F M
( )
= M
0 2
[7.23]
In the limiting case where only electrons are responsible for
multiplication (E = 0), the spectral density must approach that of a
vacuum photomultiplier and:
F M
( )
= M
0
=1 [7.24]
However, multiplication initiated by the less ionizing carriers is of
no practical interest. In this case, we find excess noise factors F(M) of
the order of M
2
, which gives:
S
i
f
( )
=2qI
pho
M
2
F M
( )
=2qI
pho
M
4
[7.25]
These results are all depicted in Figure 7.1.
In order to achieve behavior for solid avalanche devices
approaching that of vacuum photomultipliers, several devices have
been conceived based on engineering of the band structure with the
help of heterojunctions. The aim was to obtain impact ionization by
electrons only. Despite the sophistication of the technologies used, the
results did not live up to the performances initially promised, because
the interfaces resulted in an additional 1/f noise.
Figure 7.1. Characteristic plots of the variations in the spectral noise density for
multiplication in avalanche photodiodes as a function of the multiplied current
7.3. Excess noise
In addition to the sources of white noise, there are two other
sources of noise which depend on frequency. These are generation-
recombination (GR) noise and scintillation noise, which is referred to
as the 1/f noise. These two sources of noise are referred to together as
excess noise, as compared to the fundamental sources which are
thermal noise and shot noise.
7.3.1. Generation-recombination noise
This noise originates in the fluctuations of the number of carriers,
linked to the presence of different trapping states in the bandgap of
semiconductors. The spectral density associated with these
fluctuations can be written as:
S
N
f
( )
= 'N
2
4W
1Z
2
W
2
[7.26]
where 'N
2
is the second order moment of the random variable N
which represents the number of carriers under consideration. W is the
time constant of the generation-recombination process taking place.
This spectrum, of Lorentzian form, displays a plateau at low
frequencies followed by a 1/f
2
dropoff at high frequencies. These
fluctuations lead to current fluctuations proportional to the square of
the mean current:
S
i
f
( )
=
'N
2
N
2
I
2
4W
1Z
2
W
2
A
2
Hz
( )
[7.27]
Generally, the observed time constants lie between 10
3
and 10
6
s.
However, it should be noted that in some specific cases, these values
can be smaller.
7.3.2. 1/f noise
The origin of this noise is poorly understood. Its spectral density
takes the form:
S
i
f
( )
=
A
f
J
I
2
A
2
Hz
( )
[7.28]
where J usually takes a value close to 1, which leads to the term 1/f
noise, and A is an experimentally-determined constant which varies
from one device to another. We note that the spectral density varies as
the square of the bias current. It is not possible to calculate the
autocorrelation function of the process by the use of the inverse
Fourier transform of S
i
(f) because the integral diverges. There are two
interpretations of the origin of this noise. The first was proposed by
McWhorter [MCW 57] who attributed the origin of this noise to the
surfaces and interfaces of the material where a continuum of trapping
states for carriers can be found (surface effects), a situation which
produces an overlapping series of generation-recombination events.
The second interpretation is due to Hooge [HOO 76], who attributes
this noise source to a volume effect in materials, due to fluctuations in
the mobility of the carriers. Whichever of these is correct, it is almost
certain that the amplitude of this 1/f noise is affected by the quality of
the device and its fabrication technology. The frequency f
c
is defined
as that for which the spectral density of the 1/f noise is equal to that of
the fundamental noise (shot noise or thermal noise).
For the rest of this chapter, the main sources of noise that we will
take into account when evaluating the noise characteristics of a
photoreceiver are thermal noise, shot noise and its potential
amplification through multiplication processes, and the 1/f noise when
the optical signal requires a treatment that includes low frequencies. In
practical terms, Table 7.1 summarizes the expressions for the spectral
density for the different sources of noise involved with devices and
systems.
Table 7.1. Expressions and characteristics of the main noise sources
7.4. Analysis of noise electrical circuits
As is the case for deterministic signals, the behavior of a system is
described in terms of its equivalent circuit, with which a certain
number of noise sources must be associated. These sources are
represented using the Thevenin or Norton formalisms. In contrast to
generators of deterministic signals, noise generators will be
characterized by the spectral density associated with the quantities that
they represent. In this way we can avoid all the problems of temporal
analysis while retaining the concept of impendence or admittance as
traditionally defined in circuit analysis. The concepts presented below
will be used later in the evaluation of the noise and the signal-to-noise
ratio of photoreceiver systems.
7.4.1. Representation of noise in bipoles
A real bipole (for example a photodetector) is thus represented by a
random voltage generator (Thevenin), which has a spectral density
S
v
(f ) in series with its impedance Z(f ), which is assumed not to be
noisy; or by a random current generator (Norton) with spectral density
S
i
(f ) in parallel with its admittance Y f
( )
=
1
Z f
( )
.
These representations are illustrated in terms of electrical circuits
in Figure 7.2.
Figure 7.2. Representation of a noisy bipole using a frequency
spectrum approach for random processes
Henceforth, all noise generators will be marked with hatching in
order to distinguish them from normal signal generators. The
equivalence of these two representations can be expressed in the
following equation:
S
V
f
( )
= Z f
( )
2
S
i
f
( )
[7.29]
which connects the two spectral densities that we have used.
Other parameters can be used to characterize the noise of a bipole.
If we take the example of shot noise, we can define the equivalent
noise current I
eq
as:
I
eq
=
S
i
f
( )
2q
[7.30]
In the case of thermal noise, we can define:
either the equivalent resistance R
eq
of the noise:
R
eq
=
S
v
f
( )
4kT
[7.31]
or the equivalent conductance G
eq
of the noise:
G
eq
=
S
i
f
( )
4kT
[7.32]
or the equivalent temperature T
eq
of the noise:
T
eq
=
S
v
f
( )
4kRe Z f
( )
=
S
i
f
( )
4kRe Y f
( )
[7.33]
The representations of the noise given in Figure 7.2 make it
possible to directly treat series or parallel combinations of noisy
bipoles: two bipoles in series are equivalent to a single bipole whose
spectral density of voltage noise can be written:
Sv
t
f
( )
= Sv
1
f
( )
Sv
2
f
( )
2 Re Sv
1
v
2
f
( )
[7.34]
where Sv
1
v
2
(f ) represents the cross-spectrum which encapsulates the
statistical dependence of the two random processes associated with the
two noise sources. This dependence can be represented in terms of a
correlation coefficient defined by:
C f
( )
=
Sv
1
v
2
f
( )
Sv
1
f
( )
Sv
2
f
( )
[7.35]
with 0 |C| 1. Thus, Sv
T
(f ) can then be written:
Sv
T
f
( )
= Sv
1
f
( )
Sv
2
f
( )
2 Re C f
( )
Sv
1
f
( )
Sv
2
f
( )

[7.36]
In the normal case where the two sources of noise are uncorrelated
(C = 0), the resultant spectral density is then simply the sum of the
spectral densities associated with each bipole. In a similar manner, in
the case of a parallel combination of two bipoles, we have, using the
parallel representation:
Si
T
f
( )
= Si
1
f
( )
Si
2
f
( )
2 Re C f
( )
Si
1
f
( )
Si
2
f
( )

[7.37]
7.4.2. Representation of noise in quadripoles
For quadripoles (the case of amplifiers placed downstream of
photodetectors), the noise is generally represented by a minimum of
two partially-correlated noise sources. These sources, of voltage
and/or current, can be placed at the input and/or output of the
quadripole. This leads to six equivalent representations, each one
associated with one of six matrices describing the properties of the
quadripole. In practice two of these representations are the most
commonly used (see Figure 7.3).
Figure 7.3. The two most commonly-used configurations for
representation of a noisy quadripole
In the first case, the noise is represented by a voltage generator and
a current generator both present at the input of the quadripole (which
is itself assumed noise-free), with spectral densities S
en
(f ) and S
in
(f ).
This representation is well suited to describe noise in amplifiers. In the
second case, two current generators are used, one at the input and one
at the output of the quadripole, with spectral densities Si
1
(f ) and Si
2
(f ).
This representation is useful for describing the intrinsic noise of
bipolar and field effect transistors.
An important quantity for characterizing the noise of a system is its
noise factor, which is defined as the quotient of the input and output
signal-to-noise ratios:
F f
( )
=
S B
( )
e
S B
( )
s
[7.38]
This equation shows that the noise factor can also be written as the
ratio of the total output noise power to what the output noise power
would be if the quadripole was noise-free. In this way it can be shown
that the noise factor is always greater than 1, and that the presence of
the quadripole amplifier degrades the signal-to-noise ratio. Bearing in
mind the first representation given in Figure 7.3a, and assuming a
resistive load R
g
at the input to the quadripole which only generates
thermal noise, we have, for a given frequency f,
F f
( )
=1
S
en
f
( )
R
g
2
S
in
f
( )
2R
g
Re S
enin
f
( )
4kTR
g
[7.39]
As R
g
varies, the noise factor passes through a minimum for the
value:
R
gopt
=
S
en
f
( )
S
in
f
( )
Finally, for a series of quadripoles, it can be shown that under
certain conditions the noise factor of the resultant quadripole can be
written as a function of the individual noise factors F
i
and the power
gains G
i
:
F 1= F
1
1
F
2
1
G
1

F
3
1
G
1
G
2
... [7.40]
This equation (the Friis formula) shows that the first quadripole
controls the noise of the whole system as long as it is not very noisy
and its gain is significant. Thus, when designing a photoreceiver, it
will be necessary to match the photodetector to the first amplifying
component in order to minimize its noise factor and give it a sufficient
gain that the noise of the successive stages can be ignored. These
issues will be covered when we study the PIN-FET system.
7.5. Noise in photodetectors
In order to describe the noise of a photodetector in the presence of
a weak signal, we will need to evaluate a signal-to-noise ratio for the
information carried by the photons. In order to do this, we need to
define a number of optoelectronic quantities and characteristic
parameters of photodetectors, which will allow us to express the noise
associated with the photocurrent generated by the conversion of
photons into electron-hole pairs.
7.5.1. Characteristic parameters
In all optoelectronic transducers, there is an electric field present
which separates the two types of carriers generated by absorption of
the energy E
ph
= hQ of the incident photons, of wavelength O. For this
generation to take place, we require that:
E
ph
= hv =
hc
O
_ E
g
[7.41]
where h is the Planck constant, v =
c
O
is the optical frequency and E
g
the
size of the bandgap of the semiconductor material. For hQ = E
g
we
define the cutoff wavelength O
C
given by:
O
c
=
1.24
E
g
in Pm with E
g
in eV
( )
[7.42]
The response coefficient R of the detector allows us to relate the
generated photocurrent I
ph
to the incident optical power P
opt
:
R =
I
ph
P
opt
A W
( )
The calculation of R involves the external quantum efficiency K
ext
and the wavelength of the optical signal. We have:
R =K
ext
qO
hc
=0.8K
ext
O A W
( )
[7.43]
if we express O in m and c in m/s. In calculations of noise in
photodetectors, only knowledge of the photoresponse R is required.
The photodetector exhibits a basic noise S
i
(f) with which the dark
current is associated:
I
obs
=
S
i
f
( )
2q
We will now consider the incident optical power which would
produce a photocurrent I
ph
such that the associated shot noise 2qI
ph
would be equal to the basic noise referred to above (a signal-to-noise
ratio of 1). This optical power characterizes the photodetector and is
designated the NEP (Noise Equivalent Power). The NEP represents
the smallest optical power which can be detected for a given noise
equivalent bandwidth associated with the electronic system. Based on
the definition of the response coefficient, we then have, for linear
photodetectors and a unit bandwidth:
R =
I
ph
P
opt
=
I
obs
NEP
[7.44]
For a given noise equivalent bandwidth, an identical equation can
be written in terms of rms values. Thus, we can directly express the
signal-to-noise ratio in the form:
S
B
(
(
)
)
dB
=10log
P
opt
rms
( )
NEP
[7.45]
We can also define another characteristic parameter of
photodetectors, the detectivity, which is the inverse of the NEP:
D =
1
NEP
W
1
( )
In the case of a photoreceiver, we will keep in mind when defining
the NEP that the basic noise used as a reference consists of the basic
noise of the photodetector and the basic noise of the amplification
system, brought down to the level of the photodetector. For a
bandwidth 'f, we therefore have:
NEP =
S
i
f
( )
total
'f
R
[7.46]
where S
i
(f )
total
= S
i
(f )
photodetector
+ S
i
(f )
amplifier.
7.5.2. PIN photodiodes
These are diodes, of mesa or planar structure, in which the n and p
regions are separated by an intrinsic semiconductor region where the
photocarriers are generated and separated (see Figure 7.4).
Figure 7.4. Schematic cross-sections of PIN photodiodes using
(a) mesa and (b) planar technology
The most widely used semiconductor materials are shown in Table
7.2 along with their main characteristics.
Table 7.2. Main characteristics of the most widely used semiconductor materials
A theoretical approach to these devices must take into account the
band structure (direct or indirect) and optimize its geometric
dimensions as a function of the bandwidths, operating temperature,
dark currents, carrier lifetimes, leakage currents, series resistances, etc.
Figure 7.5 shows the ideal I = f (V) characteristics of a photodiode
at different incident light powers.
Figure 7.5. Typical characteristics of a PIN photodiode under different illuminations
Under reverse-bias, which is the normal mode of operation of a
PIN photodiode, the conduction consists of two components: the dark
current I
dark
and the photocurrent I
ph
= RP
opt
. Two shot noise
generators are associated with these two currents, and taking into
account the series resistance R
s
which is the source of the thermal
noise, the equivalent noise circuit is shown in Figure 7.6.
Figure 7.6. Equivalent noise circuit for a reverse-biased
PIN photodiode under illumination
Starting from this equivalent circuit, we can calculate the intrinsic
NEP of the device from equation [7.46]. For the large bandgap
materials used to detect short wavelengths, the dark current is
negligible, but technical issues are associated with the metal-
semiconductor contacts due to the very high resistivity of these
materials. For small bandgap materials, which are used for infrared
detection, the dark current will be the main limitation on the operation
of the device at ambient temperature.
7.5.3. Avalanche photodiodes
Avalanche photodiodes are PIN photodiodes which make use of
the physics of impact ionization to achieve multiplication through the
effect of the electric field on the charge carriers, with a mean gain M.
This process generates noise as discussed in section 7.2.3. Figure 7.7
shows a cross-section through two avalanche photodiodes.
Figure 7.7. (a) Structure of a n
+
pp
+
silicon avalanche photodiode for detection
at O = 0.9 m. (b) Heterostructure avalanche photodiode with separate absorption
and multiplication regions, for detection at O = 1.6 m
The equivalent noise circuit is the same as for a PIN photodiode,
but the photocurrent is written as:
I
ph
= MRP
opt
A
( )
For the shot noise, the spectral density of the noise current can at
first approximation be written as:
S
i
f
( )
=2q RP
opt
I
dark

M
2
F M
( )
A
2
H
2
( )
[7.47]
In general, the I
dark
component is multiplied by a factor M, which
is different from that for the photocurrent because we do not have
control over the injection into the multiplication region of the carriers
originating from the dark current.
7.6. Noise optimization of photodetectors
The main aim when designing a photoreceiver is to minimize the
optical power that needs to be detected while maintaining a specified
signal-to-noise ratio. Achieving this optimal sensitivity depends on the
choice of photodetector and the amplification sequence. Specifically,
at the level of the detector, the signal-to-noise ratio can be written:
S B
( )
detector
=
i
signal
2
Si f
( )
detector
df
'f
[7.48]
which in the presence of a preamplifier becomes:
S B
( )
receiver
=
i
signal
2
Si f
( )
detector
Si f
( )
preamp

df
'f
[7.49]
Thus, the signal-to-noise ratios of the receiver and the detector are
connected by the relationship:
S B
( )
receiver
=
1
F
S B
( )
detector
[7.50]
where F is the noise factor of the preamplifier, whose input load is the
photodetector. In order to achieve the minimum degradation in the
signal-to-noise ratio of the photoreceiver, the amplifier must be
matched to the detector in order to minimize F.
7.6.1. Formulation of the problem
Bearing in mind section 7.4, Figure 7.8 shows the equivalent noise
circuit of the detector-preamplifier system.
The photodetector is represented by its admittance Y
s
in parallel
with a noise current generator I
s
. The preamplifier noise is represented
at its input by a voltage generator e
n
and a current generator i
n
. These
two generators are correlated, and we write i
n
in the form of one part
which is correlated with e
n
(which we will call i
c
), and another
uncorrelated part (i
nc
):
i
n
f
( )
= i
c
f
( )
i
nc
f
( )
[7.51]
Figure 7.8. Equivalent noise circuit for a photoreceiver,
positioned at the system input
Since the correlated part is proportional to e
n
at a given frequency,
we will write it in terms of a correlation admittance Y
cor
:
i
c
f
( )
=Y
cor
f
( )
e
n
f
( )
[7.52]
The equivalent circuit in Figure 7.8b can then be drawn in the form
shown in Figure 7.9.
Figure 7.9. Equivalent circuit used to calculate the spectral density
of the total noise current of the photoreceiver
Bearing in mind our frequency-based approach to random
phenomena, we have in terms of spectral densities:
S
i
6
f
( )
= S
i
s
f
( )
Y
s
Y
cor
2
S
en
f
( )
S
i
nc
f
( )
[7.53]
The different spectral densities can be expressed as a function of
their equivalent noise conductances (see section 7.4):
S
i
s
f
( )
=4kTG
s
f
( )
S
en
f
( )
=4kTR
n
f
( )
= 4kTG
n
1
f
( )
S
i
nc
f
( )
=4kTG
nc
f
( )
S
i
n
f
( )
=4kTG
i
n
f
( )
'
1
1
1
1
1
1
1
1
1
+
1
1
1
1
1
1
1
1
1
[7.54]
To avoid encumbering the notation, from hereon in we will not
bother indicating that the conductances are a function of frequency.
The noise factor of the photoreceiver now be written:
F =
S
i
6
f
( )
S
i
s
f
( )
=1
1
G
s
G
nc
R
n
G
s
G
cor
( )
2
B
s
B
cor
( )
2
l
l
l
l
'
|
1
1
+
1
1
'
1
1
+
1
1
[7.55]
Bearing in mind equations [7.51] and [7.52], we have:
G
nc
=G
in
R
n
Y
cor
2
[7.56]
and the noise factor can be written as:
F =1
1
G
s
G
in
R
n
G
s
2
B
s
2
2 G
s
G
cor
B
s
B
cor
( )

[7.57]
In order to minimize the noise factor, we set to zero the derivative
of F with respect to the variables G
s
and B
s
. We find:
B
sopt
=B
cor
[7.58]
then:
G
sopt
= G
cor
2
G
n
G
nc
= G
in
G
n
B
cor
2
[7.59]
Under these conditions, the minimum noise factor can be written:
F
min
=12
G
sopt
G
cor
G
n
[7.60]
It follows that:
F = F
min

1
G
s
G
n
G
s
G
sopt
( )
2
B
s
B
sopt
( )
2 '
|
1
1
+
1
1
'
1
1
+
1
1
[7.61]
Hence, if we constrain F to be greater than or equal to F
min
by
modifying the source admittance Y
s
= G
s
+ jB
s
, we find we must solve
the following equation in terms of G
s
and B
s
:
G
s
2
B
s
2
2G
s
G
sopt
F F
min
( )
G
n

2B
s
B
sopt
G
sopt
2
B
sopt
2
=0 [7.62]
which in the basis G
s
, B
s
represents a circle with center:
G
so
=G
sopt
F F
min
( )
G
n
B
so
= B
sopt
'
|
1
1
1
+
1
1
1
1
[7.63]
and radius:
R = F F
min
( )
G
n
F F
min
( )
G
n
2G
sopt

[7.64]
Figure 7.10. (a) Circles of constant noise factor in the basis G
s
, B
s
.
(b) Plot of the surface defined by the noise factor
We note that for F = F
min
the circle shrinks to its center such that:
G
so
=G
sopt
B
so
=G
sopt
The results obtained are shown in Figures 7.10a and 7.10b.
It must be remembered that the series of circles and their resultant
surface is only representative of a single frequency. The matching of
the detector and preamplifier must therefore be as independent as
possible of this parameter.
7.6.2. Concepts for photodetector-transistor matching
A conceptual diagram for the matching is shown in Figure 7.11.
Figure 7.11. Conceptual diagram for matching a photodetector to its
preamplifier in order to minimize the noise factor
Y
s
is the admittance presented by the matching quadripole at the
input of the preamplifier. In order to have F = F
min
, it is necessary that
Y
s
= Y
opt
. Matching is achieved using a quadripole or matching
transformer and/or through a matching admittance Y
A
. Y
d
is the
admittance of the photodetector. For example, a photodiode with a
capacitance C
d
of 0.1 pF and a series resistance R
s
of 20 : presents an
admittance of Y
d
= 0.03 mS + j 1.2 mS at 2 GHz and Y
d
= 0.78 mS + j
6.3 mS at 10 GHz.
Similarly, for a field effect transistor with a gate of width 200 m
we have, at 2 GHz: Y
sopt
= 2.5 mS j 2.1 mS, and at 10 GHz: Y
opt
=
12.8 mS j 10.8 mS. Thus, in the complex plane of admittances, the
matching of the photodetector to its transistor move from the point Y
D
to the point Y
opt
(see Figure 7.12a).
Figure 7.12. (a) Representation in the admittance plane of the photodetectors
Y
d
and of Y
opt
which leads to a minimum of the noise factor, (b) movement due
to the matching admittance Y
a
= G
a
jB
a
, (c) matching by lossless admittance
and reduction in the width W of the gate, (d) matching by use of a transformer
In order to do this, we can imagine the following points:
a) connect the photodetector directly to the transistor. The noise
factor will not be minimal. However, we will be close enough to the
axis G of the conductances that C
d
will be weak;
b) place an admittance Y
a
= G
a
+ jb
a
in parallel with Y
d
so that Y
a
=
Y
opt
+ Y
d
(see Figure 7.12b). The noise factor will be minimized, but at
the expense of increasing the source noise. It follows that the signal-
to-noise ratio is degraded: increasing the real part of the admittance
seen by the transistor increases the noise and so is not a worthwhile
solution;
c) reduce the width W of the gate of the field effect transistor in
order to have R
e
{Y
opt
} = R
e
{Y
d
} [I think that R
e
should read Re [i.e.
the real part of]] (Figure 7.12c). The signal-to-noise ratio of the
preamplifier is optimized. It should however be noted that the
reduction in W leads to a reduction in the transconductance of the gain.
This may then not be large enough for it to be possible to neglect the
noise of the subsequent stage;
d) use an impedance transformer (Figure 7.12d). Y
d
becomes
K
e
{Y
d
+ jB
a
} = Y
opt
. In this case, it is also necessary to match the
imaginary part by using a lossless admittance jB
a
. The real part of Y
s
has been amplified along with the signal, and the signal-to-noise ratio
of the source remains unchanged. In practice, this requires being able
to build a lossless (and hence noiseless) matching circuit with a
significant K factor. It also requires that the lossless load Y
a
has a
negative susceptance (inductance). Finally, the system must operate in
the useful frequency range, bearing in mind that the admittance of an
inductance decreases with frequency when I
m
{Y
opt
} grows. This
makes this option challenging to achieve in practice.
7.7. Calculation of the noise of a photoreceiver
7.7.1. Basic equations
We need to express the equivalent noise circuit in Figure 7.8 in
terms of the basic noise associated with the transistor which forms the
preamplifier. The behavior of the transistor will be described here in
terms of its admittance matrix Y
ija
, and its background noise will be
represented by two correlated current generators i
b1
and i
b2
at the input
and output of the device respectively (section 7.5.2). This leads to the
arrangement depicted in Figure 7.13.
Figure 7.13. Equivalent circuit for the preamplifier,
highlighting the different sources of noise
The calculation of the noise present at the input is made by
considering the short-circuit output, and this does not detract from the
generality of the problem since the current gain will also be calculated
under these same conditions.
We set:
Y
11
=Y
s
Y
11a
[7.65]
and for V
2
= 0 we have:
i
2
=i
b2
Y
21a
V
1
V
1
=
1
Y
11
i
s
i
b1
( )
'
1
1
1
1
1
+
1
1
1
1
1
[7.66]
given that the output short-circuit current gain can be written:
G
i
cc
)
v
2 0
=
y
21a
Y
11
[7.67]
we find that the spectral density of the noise current associated with
current i
2
is:
S
i
2
f
( )
= S
i
b2
f
( )
G
icc
2
S
is
f
( )
S
ib1
f
( )
2 Re G
icc
S
i
b1
i
b2
f
( )

[7.68]
and so the total noise at the input to the transistor can be written:
S
i
e
f
( )
= S
i
s
f
( )
S
i
b1
f
( )
1
G
i
cc
2
S
i
b2
f
( )
2 Re
1
G
i
cc
*
S
i
b1
i
b2
f
( )
'
|
1
1
1
+
1
1
1
'
1
1
1
+
1
1
1
[7.69]
Setting Y
s
=G
s
jB
s
and S
i
b1
i
b2
f
( )
Y
21
a
=G
cor
jB
cor
, we have:
S
ie
f
( )
= S
is
f
( )
S
i
b1
f
( )
G
11a
G
s
( )
2
B
11a
B
s
( )
2
Y
21a
2
S
i
b2
f
( )
[7.70]
2 G
cor
G
s
G
11a
( )
B
cor
B
s
B
11a
( )
Insofar as S
is
(f) can be identified as the thermal noise of G
s
, S
ie
(f)
takes the form:
S
ie
f
( )
=C GG
s
BB
s
A G
s
2
B
s
2
( )
[7.71]
which represents the basic equation for the input noise of a
photoreceiver, with:
C = S
i
b1
f
( )
S
i
b2
f
( )
Y
11a
Y
21a
2
2 G
cor
G
11a
B
cor
B
11a

G =2
G
11a
Y
21a
2
S
i
b2
f
( )
G
cor
'
|
1
1
1
1
+
1
1
1
1
'
1
1
1
1
+
1
1
1
1
4kT
B =2
B
11a
Y
21a
2
S
i
b2
f
( )
B
cor
'
|
1
1
1
1
+
1
1
1
1
'
1
1
1
1
+
1
1
1
1
A= S
i
b2
f
( )
Y
21a
2
'
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
+
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
[7.72]
For a given value of S
ie
(f), equation [7.71] generates a circle in the
admittance plane, similar to that which we saw previously.
7.7.2. Models of transistor noise
In this section, we will elucidate the two noise generators i
b1
and
i
b2
as well as their correlation.
7.7.2.1. Case of a field effect transistor
i
b1
represents the noise associated with the gate and i
b2
the noise
from the channel (common-source configuration of the transistor). In
general terms, we have:
S
i
b1
f
( )
=
A
g
f
2qIg 16S
2
kT
C
gs
2
f
2
g
m
R [7.73]
There is a 1/f noise component, a shot noise component associated
with the gate current, and a component produced by noise in the
channel through capacitative coupling. Here R is a dimensionless
parameter which depends on the bias point of the device:
S
i
b2
f
( )
=
A
c
f
4kTg
m
P [7.74]
Here we have a 1/f noise contribution and a component associated
with the diffusion noise (thermal noise) of the channel. P is a
parameter which also depends on the bias point. A
g
, along with A
c
, has
units of A
2
and represents the value of the spectral density of the 1/f
noise at 1 Hz. Working from the definition of the correlation
coefficient C
r
between two noise sources (here the channel and gate
noises):
C
r
=C
re
jC
im
=
S
i
b1
i
b2
f
( )
S
i
b1
f
( )
S
i
b2
f
( )
[7.75]
we have from equations [7.73] and [7.74], where it is assumed that
only the thermal noise contribution leads to correlation:
S
i
b1
i
b2
f
( )
= 4kTCgsZ RP C
re
jC
im

[7.76]
The small-signal equivalent circuit for a field effect transistor,
along with the two noise generators i
b1
and i
b2
, is shown in Figure
7.14. The access points R
g
, R
s
and R
D
may be noisy and contribute
thermal noise in the form of 1/f noise.
Figure 7.14. Equivalent circuit of a field effect transistor,
with noise current generators at its input and output
7.7.2.2. Case of a bipolar transistor
The two noise sources model the shot noise associated with the
base current:
S
i
b1
f
( )
=2qI
B
= 2q
Ic
E
[7.77]
and the collector current:
S
i
b2
f
( )
=2qI
C
[7.78]
On top of the shot noise, it is useful at low frequencies to add the
1/f noise sources (at least at the level of the base) and maybe account
for the thermal noise of the terminal resistances (particularly for r
bb
).
At high frequencies, bearing in mind that, from the point of view of
the detector, the transistor is equivalent to its input resistance r
in
parallel with a capacitance C
T
, the total input noise to the device can
be written:
S
i
T
f
( )
=
2qI
C
E

1
r
S
2
C
T
2
Z
2
'
|
1
1
1
+
1
1
1
'
1
1
1
+
1
1
1
2qI
C
g
m
2
[7.79]
Given that r
S
= E
kT
qI
c
and g
m
r
S
= E , we find:
S
i
T
f
( )
=2qI
c
1
E

1
E
2
'
|
1
1
+
1
1
'
1
1
+
1
1
2
k
2
T
2
qI
c
C
T
2
Z
2
[7.80]
An increase in the capacitance can be counteracted by an increase
in the current I
C
. The minimum S
iT
(f) is obtained for the value of the
current I
C
which cancels out the derivative of S
iT
(f), so that:
I
c
=
kT
q
C
T
Z
1
E
1
E
2
[7.81]
The input spectral noise density becomes:
S
i
opt
f
( )
=4kTC
T
Z
1E
1
E
[7.82]
showing that this optimum varies linearly with frequency. The small-
signal equivalent circuit for the bipolar transistor, along with the two
noise generators i
b1
and i
b2
are shown in Figure 7.15.
Figure 7.15. Equivalent circuit for a bipolar transistor, with
noise generators at its input and output
7.7.3. Example calculation: a PIN-FET photoreceiver
Figure 7.16 shows the structure of a PIN-FET photoreceiver for
direct detection of the high impedance type (without R
f
) or the
transimpedance type (with R
f
).
Figure 7.16. Circuit diagram for a PIN-FET photoreceiver
From this circuit diagram, the elements of the admittance matrix of
the preamplifier can be written (ignoring parasitic elements):
Y
11a
=
1
R
f
jZ C
gs
C
gd
( )
Y
12a
=
1
R
f
jZC
gd
Y
21a
= g
m

1
R
f
jZC
gd
= g
m

1
R
f
Y
22a
=
1
R
f
jZC
gd
'
1
1
1
1
1
1
1
1
1
1
1
1
1
1
+
1
1
1
1
1
1
1
1
1
1
1
1
1
1
[7.83]
In addition, combining the PIN diode with its capacitance C
d
, the
source admittance takes the form:
Y
s
=
1
R
pol
jC
d
Z [7.84]
Designating the equivalent resistance of R
pol
and R
f
in parallel as
R
T
, and the equivalent capacitance of C
d
, C
gs
and C
gd
in parallel as C
T
(there may also be parasitic capacitances C
p
to take into account), the
equation Y
11
= Y
11a
+ Y
s
can then be written:
Y
11
=
1
R
T
jC
T
Z [7.85]
The photodetector noise is accounted for in the shot noise that
stems from the dark current I
dark
and the signal current I
signal
, with a
further contribution from thermal noise in the series resistance R
series
of the photodiode, which we can correct by taking account of the
access resistance R
g
at the gate of the transistor:
S
i
s
f
( )
=2q I
obs
I
signal
( )
4kT R
series
R
g
( )
Cd
2
Z
2
[7.86]
The total noise entering the input of the system can be written (see
equations [7.67] and [7.69]):
S
i
6
f
( )
= S
i
s
f
( )
S
i
b1
f
( )
Y
11
Y
21a
2
S
i
b2
f
( )
2 Re
Y
11
*
Y
21a
*
S
i
b1
i
b2
f
( )
'
|
1
1
1
+
1
1
1
'
1
1
1
+
1
1
1
[7.87]
Thus, we have:
S
i
6
f
( )
=2q I
dark
I
signal
I
g

A
g
f

A
c
f
1
R
T
2
C
T
2
Z
2
g
m
2
C
gd
2
Z
2
4kT
1
R
T
R
series
R
g
( )
C
d
2
Z
2
Cg
s
2
Z
2
g
m
Rg
m
P
1 R
T
2
C
T
2
Z
2
g
m
2
C
gd
2
Z
2
'
|
1
1
1
+
1
1
1
2C
gs
Z RPRe C
re
jC
im
( )
1 R
T
jC
T
Z
g
m
jC
gd
Z
l
l
l
l
l
'
1
1
1
+
1
1
1
[7.88]
At this point, we can reorganize this equation to make clear the
frequency behavior of the noise at the input to the preamplifier. For
instance, where g
m
>> Cg
d
Z, we have:
S
i
6
f
( )
=
1
f
Ag
Ac
R
T
2
g
m
2
'
|
1
1
1
+
1
1
1
'
1
1
1
+
1
1
1
2q Ig = I
D

4kT
1
R
T

P
R
T
2
g
m
'
|
1
1
1
+
1
1
1
'
1
1
1
+
1
1
1
f 4 f S
2
Ac
C
T
2
g
m
2
16S
kT
R
T
RP
C
gs
g
m
'
|
1
1
1
+
1
1
1
'
1
1
1
+
1
1
1
f
2
16S
2
kT
g
m
C
T
2
PC
gs
2
R2C
im
C
T
RPC
gs
( )
'
|
1
1
+
1
1
'
1
1
+
1
1
[7.89]
As an example, Figure 7.17 shows the variation of pA Hz
in S
i
6
f
( )
as a function of frequency, for a system whose
characteristics are listed in the caption.
Figure 7.17. Calculation of S
i
6
f
( )
as a function of frequency, using the
appropriate values for the characteristic parameters of the photoreceiver.
Dotted lines indicate the contributions from 1/f noise, thermal
and shot noise, as well as those of the f and f
2
contributions
C
d
C
p
=0.2pF, C
gs
=0.2pF
I
g
=150 nA, I
Diode
=20 nA
R
p
=30 k:, g
m
=12 mS; P =1.9, R =0.4
C
re
=0.2, C
im
=0.4
A
c
=10
11
pA
2
, A
g
=10
3
pA
2
'
|
1
1
1
1
1
1
1
1
1
1
+
1
1
1
1
1
1
1
1
1
1
At high frequencies, in the case where Cg
d
Z remains smaller than
g
m
, equation [7.88] can be written:
S
i
6
f
( )
=4kT
1
R
T
R
series
Rg
( )
Cd
2
Z
2
C
gs
2
Z
2
g
m
R
'
|
1
1
1
+
1
1
1
P
g
m
1
R
T
2
C
T
2
Z
2
(
(
)
)
2C
gs
Z RP Cre
1
R
T
g
m
C
im
C
T
Z
g
m
(
(
)
)
'
1
1
+
1
1
[7.90]
Starting from equation [7.90], it can be seen how the width W of
the transistor gate can be calculated to match it to the photodetector,
setting:
g
m
= g
mo
W
C
gs
=C
gso
W
C
gd
=C
gdo
W and Co =C
gso
C
gdo
We then have:
S
i
6
f
( )
=4kT
1
R
T
R
series
Rg
( )
Cd
2
Z
2
Cgs
o
2
g
m
0
WZ
2
R
P
g
m
0
W
1
R
T
2
'
|
1
1
1
+
1
1
1
P
C
o
2
WZ
2
g
m
0
P
Cd
2
Z
2
g
m
0
W
2P
CdCo
g
m
0
Z
2
2Cgso
g
m
0
RPC
im
CoWZ
2
2Cgso
g
m
0
RPC
im
CdZ
2
2
Cgso
g
m
0
Z RP
Cre
R
T
'
1
1
1
+
1
1
1
[7.91]
In terms of the variable W, this last equation can be rearranged to
obtain:
S
i
6
f
( )
=W
4kT
g
m
0
Cgso
2
R PCo
2
2Cgso RPC
im
Co

Z
2
4kT
1
R
T
2
P
g
m
0
CdCoZ
2
2
Cgso
g
m
0
RP C
im
CdZ
2
Cre
R
T
Z
(
(
)
)
'
|
1
1
1
+
1
1
1
'
1
1
1
+
1
1
1
1
W
4kT
1
g
m
0
P
1
R
T
2
Cd
2
Z
2
(
(
)
)
R
s rie
Rg
( )
Cd
2
Z
2
Wg
m
0
'
|
1
1
1
+
1
1
1
'
1
1
1
+
1
1
1
[7.92]
Since the gate access resistance is inversely proportional to:
W Rg =
Rgo
W
(
(
)
)
it is reasonable to make the following approximation when integrating

the photoreceiver:
R
s rie
Rg
( )
Cd
2
Z
2
Wg
m
0
= R
s rie
0
Rg
o
( )
Cd
2
Z
2
g
m
0
[7.93]
Additionally, the term representing the noise induced by the channel
in the gate, 4kT
Cgs
2
Z
2
g
m
R, can be written as:
4kT
Co
2
Z
2
g
m
R' with R' =
Cgs
2
Co
2
R [7.94]
Equation [7.92] then takes the final form:
S
i
6
f
( )
= 4kT W
Co
2
g
m
0
R' P2 R' PC
im
( )
'
|
1
1
1
+
1
1
1
1
R
T
2
P
g
m
0
CdCoZ
2
2
Co
g
m
0
R' P C
im
Cd Z
CreZ
R
T
(
(
)
)
1
W
1
g
m
0
P
1
R
T
2
Cd
2
Z
2
(
(
)
)
Cd
2
Z
2
R
series
0
Rg
o
( )
g
m
0
l
l
l
l
l
'
1
1
1
+
1
1
1
[7.95]
The derivative of this final result, with respect to W, leads us to an
optimal value of W which minimizes the noise, such that:
W
opt
=
Cd
Co
P 1
1
R
T
2
Cd
2
Z
2
(
(
)
)
g
m
0
R
series
Rg
o
( )
R' P2 RPC
im
[7.96]
We then have:
Si
6
f
( )
= 4kT
1
R
T
2
CoCdZ
2
g
m
0
P R' P C
im

Cre
R
T
CdZ
(
(
)
)
l
l
l
l
l
'
|
1
1
1
+
1
1
1
P 1
1
R
T
2
Cd
2
Z
2
(
(
)
)
g
m
0
R
series
Rgo
( )
'
|
1
1
1
+
1
1
1
'
1
1
1
+
1
1
1
R' P2 R' PC
im
( )
[7.97]
In the case where the value of R
T
dominates, we have:
Si
6
f
( )
=8kT
Co
g
m
0
CdZ
2
P PR'C
im
Pg
m
0
R
series
Rg
o
( )
R' P2 R' PC
im
[7.98]
which can be written as:
Si
6
f
( )
min
=16kT
Co
g
m
0
CdZ
2
* [7.99]
with:
* =
1
2
P PR'C
im
P R
series
Rg
o
( )
g
m
0

R' P2 R' PC
im

'
|
1
1
+
1
1
'
1
1
+
1
1
We can consider * as an equivalent noise parameter for the
channel which allows us to treat the system as if the gate noise is zero
(R = C
r
= 0). Equation [7.99] accounts for the high-frequency noise of
the receiver. Now returning to the sources of noise that we have
ignored up to now, the total input noise after the gate width has been
optimized is:
Si
6
f
( )
min
=2q I
dark
I
signal
I
g
( )
A
g
f
A
c
4S
2
C
T
2
g
m
2
f
64S
2
kT
Co
g
m
0
Cd*f
2
[7.100]
This equation gives the frequency dependence of the optimized
noise.
7.8. Comments and conclusions
Several comments can be made based on the example that we have
worked though. There are four sources of noise that are a function of
frequency:
a 1/f contribution which dominates at low frequencies and is
connected with the manufacturing technology of the devices;
a white noise component which represents the minimum noise
level of the photoreceiver;
a component proportional to frequency, associated with the high
frequency performance of the transistor and the capacitative coupling
between the channel and the gate, which brings into play the low
frequency noise of the channel;
a component proportional to the frequency squared, associated
with the * coefficient defined above.
The noise contribution from each element in the photoreceiver
circuit is clear:
the intrinsic characteristics of the PIN photodiode (I
obs
, C
d
and
R
series
) contribute to the white noise and the high frequency noise as a
result of its matching to the amplifier;
the bias resistance of the photodiode has an effect on the white
noise and low frequency noise of the system;
the transistor characteristics (I
g
, g
m
, C
gs
, C
gd
, A
c
, A
g
and *) affect
the noise performance of the photoreceiver across the whole
frequency range of its operation.
Finally, at the level of the whole photoreceiver, and to summarize
all the concepts presented in this chapter, we calculate the signal-to-
noise ratio for an incident optical power P
opt
and a photodiode
response coefficient R through the equation:
S
B
=
RP
opt
( )
2
Si
6
f
( )
df
B
en
[7.101]
where B
en
represents the equivalent noise bandwidth.
This type of calculation can be applied to different photoreceivers
involving bipolar or field effect transistors connected to PIN or
avalanche photodiodes.
7.9. Bibliography
[BER 93] BERTHIER P., Transistors effet de champ AlInAs/(Al)GaInAs(P) pour
photodtection intgre 1,3-1,5 m, PhD thesis, Montpellier University,
December 1993.
[BER 94] BERTHIER P., GIRAUDET L., SCAVENEC A., RIGAUD.D, VALENZA M., DAVIES
J.I., BLAND S.W., InGaAsP channel HFETs on InP for OEIC applications,
Journal of Light Wave Technology, vol. 12, n 12, December 1994.
[BIQ 92] BIQUARD M.F., Signaux, systmes linaires et bruit en lectronique,
Ellipses, 1992.
[DAS 95] DAS M.B., CHEN J.W., JOHN E., Designing optoelectronic integrated
circuit (OEIC) receivers for high sensitivity and maximally flat frequency
response, Journal of Light Wave Technology, vol. 13, n 9, September 1995.
[FIS 93] FISH P.J., Electronics noise and low noise design, Macmillan New
Electronics series, 1993.
[HEN 89] HENTSCHEL C., Fiber Optics Handbook, Hewlett-Packard, 1989.
[HOO 76] HOOGE F.N., 1/f noise, Physica, vol. B83, 1976.
[JOI 96] JOINDOT I.M., Les tlcommunications par fibres optiques, Dunod, 1996.
[KIM 97] KIM M.J., KIM D.K., KIM S.J., DAS M.B., Determination of bit-rate and
sensitivity limits of an optimized p-i-n/HBT OEIC receiver using SPICE
simulations, IEEE Transactions on Electron Devices, vol. 40, n 4, April 1997.
[KRE 93] KRESSEL H., Semiconductors Devices for Optical Communication,
Springer-Verlag, Topics in Applied Physics, vol. 39, 1987.
[LEG 93] LEGROS E., Photorcepteur intgr pour transmission cohrente sur fibre
optique, University Thesis, Orsay, December 1993.
[PAR 93] PARK M.S., MINASIAN R.A., Ultralow noise 10Gb/s p-i-n-HEMT optical
receiver, IEEE Photonics Technology Letters, vol. 5, n 2, February 1993.
[MCW 57] MC WHORTER A.L., Semiconductor Surface Physics, University of
Pennsylvania Press, Philadelphia, 1957.
List of Authors
Robert ALABEDRA
CEM2
University of Montpellier II
France
Baudoin DE CREMOUX
Central research laboratory
Thomson-CST
France
Didier DECOSTER
IEMN
Lille University of Science and
Technology
France
Carmen GONZALEZ
Alcatel Thales III-V Lab
Marcoussis
France
Joseph HARARI
IEMN
Technology
France
Vincent MAGNIN
IEMN
Technology
France
Antoine MARTY
LAAS-CNRS
Toulouse
France
Eva MONROY
INAC
Nanophysics and
Semiconductors Laboratory
Grenoble
France
Franck OMNES
Neel Institute
Grenoble
France
Dominique RIGAUD
CEM2
University of Montpellier II
France
Grard RIPOCHE
Alcatel Research Centre
Marcoussis
France
Index
1/f noise, 14, 206, 207, 231, 232,
233, 234, 256, 257, 262
II-VI compound, 2, 101
III-V compound, 2, 5, 51, 59, 94,
100, 101, 111, 112, 114, 177, 186,
195, 201, 202
IV-IV compound, 2
A
absorption, 3, 10, 16, 19, 28, 33, 36,
39, 42, 43, 44, 45, 46, 47, 50,
51, 52, 54, 57, 58, 74, 80, 82,
83, 87, 88, 91, 93, 95, 96, 97,
99, 102, 104, 105, 111, 112,
113, 114, 115, 116, 119, 164,
165, 166, 169, 170, 182, 184,
187, 189, 191, 193, 197, 198,
204, 205, 211, 240, 245
coefficient, 39, 43, 45, 46, 74, 88,
91, 93, 111, 113, 164, 165, 166,
197, 198, 211
acceptors, 2, 5
AlGaN, 187, 196, 198, 199, 200, 202,
205, 206, 208, 209, 210, 212, 214,
215
autocorrelation function, 225, 226,
227, 233
avalanche, 7, 10, 23, 27, 32, 49, 57,
59, 60, 62, 63, 64, 65, 66, 67,
68, 69, 71, 73, 75, 76, 78, 80,
90, 95, 97, 99, 100, 103, 106,
171, 175, 176, 214, 223, 229,
230, 231, 232, 244, 245, 267
region, 62, 63, 64, 69, 75, 76, 78
B
bandgap, 1, 2, 3, 14, 19, 20, 21, 28,
29, 31, 36, 43, 45, 46, 47, 48, 49,
50, 52, 61, 65, 74, 80, 88, 90, 95,
96, 99, 101, 102, 112, 113, 122,
123, 163, 164, 169, 171, 183, 186,
187, 188, 189, 191, 198, 203, 204,
205, 208, 209, 211, 216, 232, 240,
244
bandwidth, 5, 11, 12, 13, 14, 37, 40,
53, 58, 71, 72, 75, 76, 82, 88, 98,
99, 100, 102, 104, 105, 115, 148,
166, 202, 206, 207, 215, 217, 227,
241, 267
bipolar, 111, 112, 114, 115, 118, 119,
121, 122, 123, 125, 126, 127,
131, 133, 144, 145, 146, 150,
152, 214, 238, 257, 258, 259,
267
transistor, 119, 121, 123, 125,
126, 127, 131, 146, 214, 257,
258, 259
breakdown, 7, 23, 27, 31, 32, 33, 41,
57, 59, 60, 64, 65, 66, 67, 68, 76,
77, 78, 79, 81, 83, 84, 86, 88, 94,
104, 159, 160, 161, 173, 176
C
cancers, 185
capacitance, 7, 18, 33, 37, 74, 78, 99,
123, 128, 172, 173, 175, 177, 202,
206, 212, 251, 257, 258, 260
circuit, 16, 23, 25, 26, 27, 28, 34, 37,
39, 40, 41, 43, 47, 53, 71, 88, 118,
127, 128, 132, 138, 141, 143, 144,
145, 146, 148, 167, 172, 173, 174,
177, 196, 224, 227, 228, 235, 244,
245, 246, 247, 248, 253, 254, 257,
258, 259, 267
cutoff frequency, 13, 35, 38, 39, 48,
55, 93, 94, 137, 140, 157, 171,
172, 173, 174
D
dark current, 21, 26, 27, 29, 30, 31,
44, 45, 47, 48, 49, 51, 60, 64, 66,
67, 68, 78, 81, 86, 88, 90, 91, 92,
94, 95, 97, 98, 100, 102, 103, 104,
105, 156, 158, 159, 161, 162, 163,
176, 177, 194, 205, 206, 207, 209,
213, 228, 240, 243, 244, 245, 260
depleted region, 28, 52, 54, 63, 97
diffusion
current, 25, 90
length, 6, 25, 28, 122, 124, 194,
203, 205, 211
dislocations, 80, 103, 192, 195, 197,
198, 205, 207, 213
donors, 2, 5, 8
E
electric field, 3, 4, 5, 6, 7, 9, 10, 16,
17, 18, 23, 24, 25, 26, 28, 31, 32,
33, 34, 36, 57, 60, 61, 62, 63, 64,
67, 69, 76, 77, 79, 82, 83, 84, 85,
86, 87, 92, 95, 96, 97, 102, 104,
114, 119, 120, 156, 157, 159, 160,
161, 162, 166, 167, 173, 191, 193,
214, 215, 229, 240, 244
electrode, 156, 158, 159, 160, 162,
166, 168, 169, 171, 173, 174, 175,
177
electrode, 9, 10, 15, 141, 143
epitaxial growth, 59, 77, 95, 97, 106,
171, 218
F, G
fabrication, 45, 48, 49, 50, 54, 80, 81,
85, 86, 87, 92, 99, 100, 104, 112,
113, 117, 162, 163, 188, 209, 214,
216, 217, 234
filter, 51, 133, 184, 187, 191, 208,
216
GaAlAs, 43, 45
GaAs, 2, 43, 45, 46, 59, 70, 88, 89,
100, 105, 112, 113, 114, 115, 123,
144, 149, 162, 165, 166, 171, 177,
190, 208
gain, 5, 7, 10, 11, 12, 23, 27, 32, 40,
49, 53, 55, 57, 58, 59, 66, 67, 68,
71, 72, 73, 76, 82, 84, 86, 88, 94,
98, 100, 102, 104, 105, 111, 118,
121, 123, 124, 125, 126, 127, 129,
133, 134, 135, 137, 140, 144, 146,
147, 149, 150, 171, 176, 196, 197,
209, 215, 216, 217, 229, 239, 244,
252, 254
GaInAs, 38, 39, 40, 46, 47, 48, 49,
53, 99, 164, 166, 169
gain-bandwidth product, 76, 100,
102, 104, 105
GaN, 2, 187, 188, 195, 196, 202, 204,
207, 209, 210, 211, 212, 213, 214,
215
GaSb, 49, 102
generation-recombination noise, 232
germanium, 2, 46, 47, 48, 59, 60, 61,
62, 72, 90, 93, 94, 95
Index 273
guard ring, 75, 77, 78, 81, 85, 87, 88,
91, 93, 94, 97, 102, 103
H, I, J
heteroepitaxial growth, 207
implantation, 78, 81, 84, 85, 86, 87,
92, 93, 94, 97, 103
InP, 38, 39, 40, 46, 47, 48, 49, 53, 59,
61, 62, 70, 95, 96, 97, 98, 99, 100,
101, 102, 103, 105, 106, 112, 113,
115, 123, 145, 146, 147, 148, 149,
150, 169, 177
junction, 4, 5, 6, 7, 8, 9, 10, 15, 17,
18, 30, 31, 32, 33, 34, 37, 48, 58,
60, 62, 64, 65, 66, 68, 69, 73, 74,
76, 77, 78, 79, 80, 83, 84, 85, 92,
94, 97, 99, 103, 108, 114, 115,
118, 119, 120, 121, 123, 128, 130,
134, 135, 139, 147, 156, 158, 161,
162, 163, 164, 186, 189, 191, 192,
193, 194, 204, 210, 211, 212, 213,
214, 217
L, M
large signal, 126
mixer, 131, 132, 135, 142, 143, 146,
149, 150
MSM photodiode, 156, 174, 188,
209, 217
multiplication factor, 57, 62, 63, 64,
65, 66, 94, 229
N
noise, 11, 13, 14, 49, 58, 59, 61, 66,
68, 69, 70, 71, 72, 73, 76, 79, 80,
81, 84, 86, 87, 88, 89, 91, 92, 93,
94, 97, 98, 100, 101, 102, 103,
104, 105, 118, 138, 139, 140, 146,
155, 157, 175, 176, 177, 183, 202,
206, 207, 210, 213, 214, 217, 220,
223, 224, 226, 227, 228, 229, 230,
231, 232, 233, 234, 235, 236, 237,
238, 239, 240, 241, 243, 244, 245,
246, 247, 248, 249, 250, 251, 252,
253, 254, 255, 256, 257, 258, 259,
260, 261, 262, 264, 265, 266, 267
noise equivalent power (NEP), 11,
13, 207, 210, 241, 244
noise factor, 69, 70, 71, 76, 80, 84,
86, 87, 88, 90, 91, 93, 94, 97, 98,
100, 101, 102, 103, 104, 105, 176,
230, 231, 238, 239, 246, 248, 249,
250, 251, 252
O, P
ohmic contact, 4, 194, 210
optical
fiber, 44, 47, 59, 80, 93, 106, 111,
112, 113, 116, 117, 146, 149,
169
links, 91, 106
photoreceiver, 106, 138, 148, 176,
223, 234, 235, 239, 241, 245, 246,
247, 248, 253, 255, 259, 262, 264,
266, 267
phototransistor, 111, 112, 113, 114,
115, 116, 117, 118, 119, 121, 122,
123, 124, 125, 126, 127, 128, 129,
131, 133, 134, 135, 137, 138, 139,
140, 141, 142, 143, 144, 145, 146,
147, 148, 149, 150, 214, 215, 217
PIN photodiode, 15, 16, 18, 25, 28,
29, 32, 34, 35, 36, 38, 39, 40, 42,
48, 49, 50, 66, 72, 74, 75, 115,
118, 146, 169, 173, 177, 214, 228,
242, 243, 244, 245, 267
Q, R
quantum efficiency, 11, 12, 26, 27,
28, 29, 36, 39, 44, 45, 48, 50, 51,
52, 53, 54, 74, 82, 90, 94, 104,
117, 124, 126, 164, 166, 169, 170,
187, 192, 193, 194, 197, 198, 203,
204, 211, 240
reliability, 58, 79, 91, 97, 98, 102,
106, 172, 186, 188, 216, 217
response coefficient, 11, 12, 13, 14
S
Schottky contact, 9, 156, 164, 205,
206, 209
semiconductor, 1, 3, 4, 5, 7, 8, 9, 10,
11, 12, 15, 23, 42, 43, 46, 49, 50,
58, 60, 74, 75, 80, 88, 90, 112,
155, 156, 158, 159, 160, 161, 162,
163, 164, 167, 168, 169, 171, 175,
183, 186, 189, 217, 229, 230, 240,
242, 243, 244
shot noise, 13, 68, 69, 71, 139, 175,
176, 213, 229, 232, 234, 236, 243,
245, 257, 260
SiC, 187, 188, 190, 194, 195, 216
silicon, 2, 21, 23, 43, 44, 45, 46, 59,
60, 61, 62, 70, 72, 75, 87, 90, 96,
100, 111, 112, 115, 116, 141, 144,
145, 150, 165, 166, 186, 187, 188,
191, 196, 214, 231, 245
small signal, 174
submarine links, 59
T
telecommunications, 39, 44, 45, 46,
47, 48, 49, 59, 103, 106, 112, 145,
150, 184
temporal response, 12, 13, 198, 199,
200, 210, 212, 215, 217
thermal noise, 13, 71, 73, 139, 175,
232, 234, 238, 244, 256, 257, 260
transit time, 34, 35, 36, 40, 44, 50,
52, 53, 55, 73, 74, 75, 94, 99, 104,
105, 123, 124, 128, 129, 164, 167,
170, 171, 172, 173, 174, 175, 177,
210, 212, 228, 229
transparent contacts, 189
transport, 15, 16, 54, 83, 96, 101,
121, 122, 123, 125, 127, 146, 166,
167, 169, 171, 175, 213
tunnel effect, 26
U, W
ultraviolet (UV), 10, 11, 44, 87, 181,
216
UV-visible contrast, 188, 189, 209
waveguide photodiodes, 52, 53
wavelength, 4, 12, 20, 27, 36, 39, 43,
45, 51, 52, 53, 58, 80, 81, 88, 112,
113, 142, 146, 163, 167, 168, 169,
181, 184, 186, 187, 189, 190, 192,
193, 194, 195, 196, 203, 204, 208,
211, 217, 240

Op To Electronic Sensors

Uploaded by

Copyright:

Available Formats

Op To Electronic Sensors

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

Op To Electronic Sensors

Uploaded by

Copyright:

Available Formats

What are some of the different types of photodetectors discussed in the document?

What factors affect the performance of photodetectors?

This page intentionally left blank

, known as the primary photocurrent, is established

originates from a current

was introduced. The

to the received optical power P

. Several authors have shown

induced in the base-collector photodiode.

placed between the base B and the collector C. The output

, which defines the dynamic current gain

generated by the illumination at the level of the base-collector junction,

in the context of this assumption, suggests that the

and the electrical bias

, to the bias current of the phase

it is reasonable to make the following approximation when integrating

You might also like