Introduction To Terahertz Electronics (2021)
Introduction To Terahertz Electronics (2021)
Introduction To Terahertz Electronics (2021)
Rieh
Introduction
to Terahertz
Electronics
Introduction to Terahertz Electronics
Jae-Sung Rieh
Introduction to Terahertz
Electronics
Jae-Sung Rieh
School of Electrical Engineering
Korea University
Seoul, Korea (Republic of)
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
To my family: Sohee 소희, Gyujin 규진, and Woojin 우진
Preface
vii
viii Preface
professionals in the RF fields with electronics background, who are newly entering
this emerging arena of THz electronics, will also find this book useful. I also hope
this book will serve as a brief guide toward THz electronics for the researchers who
are involved in THz vacuum electronic devices as well as those in the field of optics-
based THz.
There was a question that hovered around me when I was about to embark on the
task of writing, “Do we still need books as means to transfer knowledge?” We are
living in a world where all the information is virtually a click away. We increasingly
rely on online information with a higher level of graphic presentation, a tendency
more obvious with young generations. So the question appeared legitimate from
many accounts. Nevertheless, I managed to come up with justifications for me to
move ahead. First, the value of information is greatly enhanced when they are
structured. Structured information guides the readers through the relation between
pieces of information. A book would serve this critical purpose around a focused
topic. The readers may further develop a more detailed structure based on what is
provided by the book. Second, the interpretation of information depends on who the
author is. An analogy can be found in music. A piece of music can sound vastly
different depending on who plays it, a consequence of different interpretations. The
readers will perceive knowledges through the unique frame of interpretation pro-
vided by the author of the book. These couple of points encouraged me to drop the
hesitation and move forward to writing.
It turned out that writing a book needed more devotion than I expected, requiring
great patience and pain. The difficulty doubled when it was written in a language that
is not my mother tongue. At the same time, however, the process provided me with
lots of pleasure. It was a wonderful opportunity for me to accumulate a greater
amount of knowledge and structure them in a systematic way. In a sense, I myself
may be the first beneficiary of this book. With the book now released to public space,
I wish it will benefit a wide range of students and researchers surrounding the topic
of THz electronics.
This book would not have been completed in this form without the precious helps from
so many intellectual and kind people around me. I first would like to thank the critical
reviews on the early manuscripts provided by Prof. EunMi Choi (Ulsan National
Institute of Science and Technology (UNIST)), Prof. Sanggeun Jeon (Korea Univer-
sity), Prof. Jinho Jeong (Sogang University), Prof. Byung-Sung Kim (Sungkyunkwan
University), Prof. Moonil Kim (Korea University), Dr. Jung-Won Lee (Korea Astron-
omy and Space Science Institute (KASI)), Prof. Munkyo Seo (Sungkyunkwan Uni-
versity), Prof. Dongha Shim (Seoul National University of Science & Technology
(SeoulTech)), Prof. Joo-Hiuk Son (University of Seoul), Prof. Ho-Jin Song (Pohang
University of Science and Technology (POSTECH)), and Prof. Daekeun Yoon
(National Chiao Tung University). The highly valuable comments by them definitely
helped to improve the contents. I greatly appreciate my diligent students, Minje Cho,
Doyoon Kim, Jungsoo Kim, Gihyun Lim, Heekang Son, and Junghwan Yoo, who
provided great helps on the figures. I would also like to extend my gratitude to all my
current and former students for extensive literature surveys reflected in the trend charts
in the book, as well as the technical discussions in and out of the lab, which I enjoyed a
lot. Further, I would like to take this opportunity to thank Prof. M. C. Frank Chang for
having me stay at UCLA for my 1-year sabbatical, which allowed me a perfect
environment to focus on writing, including an access to the engineering library located
at the Boelter Hall. I remember that the peculiar scent emitted from the books sitting on
the Boelter stacks somehow promoted the spirit of writing. About two-thirds of this
book was written during my stay at the Westwood campus. Certainly, I appreciate
Korea University for providing a comfortable yet highly stimulating environment for
research. Finally, special thanks should go to my family: my wife Sohee and two kids,
Gyujin and Woojin. They bore with me with great patience when I spent so much of my
time at home writing, which otherwise would have been shared with them. Their
understanding and encouragement have made this lengthy task successful in the end.
ix
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Terahertz Band . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Properties of THz Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Approaches to THz Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.1 Approaches to THz Signal Generation . . . . . . . . . . . . . . . . 11
1.3.2 Approaches to THz Signal Detection . . . . . . . . . . . . . . . . . 13
1.4 THz Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2 THz Sources and Related Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1 Vacuum Device Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1.1 Klystrons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.1.2 Traveling-Wave Tubes (TWTs) . . . . . . . . . . . . . . . . . . . . 23
2.1.3 Backward Wave Oscillators (BWOs) . . . . . . . . . . . . . . . . 26
2.1.4 Gyrotrons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.5 Free Electron Lasers (FELs) . . . . . . . . . . . . . . . . . . . . . . . 29
2.1.6 Magnetrons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2 Diode Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2.1 Gunn Diodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.2 IMPATT Diodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.3 Resonant Tunneling Diodes (RTDs) . . . . . . . . . . . . . . . . . 38
2.3 Transistor Circuit Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.3.1 Oscillator Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.3.2 LC Cross-coupled Oscillators . . . . . . . . . . . . . . . . . . . . . . 42
2.3.3 Colpitts Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.3.4 Ring Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.3.5 Voltage-Controlled Oscillators (VCOs) . . . . . . . . . . . . . . . 50
2.3.6 Phase-Locked Loops (PLLs) . . . . . . . . . . . . . . . . . . . . . . . 56
2.4 Frequency Enhancement Techniques . . . . . . . . . . . . . . . . . . . . . . 61
2.4.1 n-Push Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.4.2 Frequency Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
xi
xii Contents
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
About the Author
Jae-Sung Rieh received the B.S. degree in electronics engineering from Seoul
National University, Seoul, Korea, in 1991. He spent one and a half years in Korean
Army for the mandatory military service before coming back to school for a graduate
study in 1993. In 1995, he received the M.S. degree at Seoul National University,
with a thesis entitled, “Fabrication of devices for HEMT MMIC’s and design of low
noise amplifiers.” In the summer of 1995, he moved to the United States for a
doctoral study, which resulted in his Ph.D. degree in electrical engineering from the
University of Michigan, Ann Arbor, MI, USA, awarded in 1999. The dissertation
title was “Development of SiGe/Si HBT’s and their applications to microwave and
optoelectronic monolithic integrated circuits.”
In 1999, he joined IBM Microelectronics, Essex Junction, VT, USA, where he
contributed to the development of 0.18 μm SiGe BiCMOS technology. In 2000, he
moved down to IBM Semiconductor R&D Center, Hopewell Junction, NY, USA,
where he was involved in the research and development activities for 0.13 μm SiGe
BiCMOS technologies. In particular, he led the company efforts to develop the next-
generation technology, which resulted in a 350-GHz SiGe HBT, the first Si-based
transistor operating in the sub-millimeter wave band. It helped to push Si-based
technologies toward THz electronics.
In 2004, he came back to Korea to pursue his long-desired career, teaching at
university. He joined the School of Electrical Engineering, Korea University, located
in Seoul, Korea, where he has been teaching and supervising students for more than
15 years. His major research interest lies in the mm-wave and terahertz devices and
circuits. In 2012, he spent his first sabbatical at Submillimeter Wave Advanced
Technology (SWAT) team in JPL, Pasadena, USA, which greatly promoted his
interest in the THz research. For another sabbatical year spread over the 2018–2019
period, he was with High Speed Electronics Lab (HSEL) in UCLA, Los Angeles,
USA, during which he started to work on his book Introduction to THz Electronics.
Prof. Rieh was a recipient of IBM Faculty Award (2004) and two-time co-recip-
ients of IEEE EDS George E. Smith Awards (2002 and 2006). He was also
co-awarded IEEE Microwave and Wireless Component Letters Tatsuo Itoh Best
xv
xvi About the Author
Paper Award (2013). He has served as an Associate Editor of the IEEE Microwave
and Wireless Components Letters (2006–2009) and the IEEE Transactions on
Microwave Theory and Techniques (2010–2013). He is a Fellow of IEEE.
Chapter 1
Introduction
In this opening chapter, the background and basic concepts of terahertz electronics
will be presented along with a brief overview of the topics to be covered throughout
this book. A discussion on the definition of the terahertz (THz) band is first provided,
followed by a review of various unique properties of the THz band that have
attracted many researchers for scientific interests and practical applications. In the
next section, a preview on the technical approaches to the generation and detection
of THz waves is presented, which will serve as a quick reference to the readers
before detailed descriptions are presented in the following chapters. In addition, a
brief overview of the application areas of the THz band will be provided, which will
be more detailed in Chap. 6 of this book.
A proper starting point of the discussion on the terahertz electronics would be the
review of the definition of the “terahertz (THz) band.” A clear definition of the THz
band, however, does not exist as of today, and it varies over different authors and
literatures. The lower boundary of the band falls on either 100 or 300 GHz, while the
upper boundary hovers around 3 THz, 10 THz, and sometimes 30 THz. Although it
is supposed to be a technical term, a proper definition may follow the rule of general
languages in accepting terms, say, the popular usage by the users. A scan over the
existing literature with a standard search engine reveals that “100 GHz–10 THz” is
the most widely used definition with a reasonably large margin, followed by
“100 GHz–3 THz” and “300 GHz–10 THz.” With this usage pattern identified,
“100 GHz–10 THz” may serve as a reasonable definition of the THz band, unless
there arises a strong shift in the usage preference in the future. Hence, this definition
will be used in this book. Another related term occasionally found in the literature is
the sub-terahertz band. Its definition is as vague as that of the THz band, popular two
versions being “100 GHz–1 THz” and “100 GHz–300 GHz.” Again, based on the
usage counts in the literature, the former can be accepted as the definition of the
sub-terahertz band, literally indicating the band below 1 THz.
The discussion on the band definition naturally brings us to the definitions of
other frequency bands surrounding the THz band. An overview of the bands outside
the THz band may provide an informative backdrop, against which the relative
position of the THz band can be better pictured from a macroscopic point of view.
There are various frequency band notations used by radio frequency (RF) commu-
nity. The three most prevailing ones are the ITU (International Telecommunication
Union) frequency bands, IEEE (Institute of Electrical and Electronics Engineers)
radar bands, and the waveguide bands.
The ITU band notation, shown in Table 1.1, divides the spectrum from 0.03 Hz
up to 3 THz by a decade span and assigns band numbers and symbols [1]. Some of
the symbols, such as VHF (very high frequency) and UHF (ultra high frequency), are
familiar to the general public as those band names were adopted for the carriers of
television broadcasts. Although some may feel that the symbol expressions are too
generous (e.g., “ultra high” frequency represents a band of merely 300 MHz to
3 GHz), the nomenclature structure is highly systematic and covers the entire
frequency range that would be of practical use from electronics point of view. It is
also noteworthy that the origins of the widely adopted band definitions, millimeter-
wave (30–300 GHz) and sub-millimeter-wave (300 GHz–3 THz), can be traced back
to this ITU notation, corresponding to “millimetric waves” (Band 11) and
“decimillimetric waves” (Band 12), with a slight modification in the names. Related
to this, it can be pointed out that the definition of the THz band as discussed above
ranges across the millimeter-wave and the sub-millimeter-wave bands.
IEEE band nomenclature, officially called “IEEE Standard Letter Designations
for Radar-Frequency Bands” [2], is shown in Table 1.2. The band designations for
frequencies lower than 1 GHz in this nomenclature were apparently borrowed from
the ITU notation with a small change (the range for UHF band is truncated), while
those for beyond 1 GHz inherit the notations from the radar community. The letter
assignments for the bands, such as L, S, C, X, Ka, K, Ku, V, and W, do not appear to
follow any obvious rule or pattern and may look arbitrary at first glance. As such,
getting familiar with them has been a kind of ritual process that new microwave and
radar engineers have to go through. In fact, however, they are not totally arbitrary,
and a brief reference to their historical background would help readers to understand
(and remember) the designation as introduced in [3]: The wavelength used for the
early radars was 23 cm, while a shorter wavelength of 10 cm was later adopted, each
leading to L (long) and S (short) symbols for the corresponding frequency bands. A
fire control radar was later introduced, for which letter X indicates a spot marking,
leading to X-band. C is for an intermediate band located in between, thus
‘compromising’ the advantages of S-band and X-band. K-band took the initial
from a German word ‘kurtz’ (meaning ‘short’) indicating a higher frequency,
while Ku and Ka represent K-under and K-above, respectively. V indicates “very”
Table 1.2 IEEE standard letter designations for radar frequency bands
Band designation Frequency range Comments
HF 3–30 MHz
VHF 30–300 MHz
UHF 300–1000 MHz
L 1–2 GHz L for “long” wave
S 2–4 GHz S for “short” wave
C 4–8 GHz C for “compromise” between S and X band
X 8–12 GHz X for “cross” (or spot markings)
Ku 12–18 GHz Ku for “kurz-under”
K 18–27 GHz German “kurz” means short
Ka 27–40 GHz Ka for “kurz-above”
V 40–75 GHz V for “very” high frequency band
W 75–110 GHz W follows V in the alphabet
mm 110–300 GHz
4 1 Introduction
high frequency and W is a letter that simply follows V, both of which were newly
added in the revision of the IEEE standard made in 1984. With the current version of
the IEEE standard, revised on 2002 [2], the frequency band beyond 110 GHz is
designated simply as “mm,” indicating the mm-wave. Hopefully, new designations
may appear in the next revision and assign additional letter designations to expand
the frequency range covered by this standard.
There are other letter-based band definitions widely used among microwave
engineers, including Q-band, U-band, E-band, F-band, D-band, and G-band. They
are not part of IEEE band standard described above. Instead, they are related to the
notation system based on the rectangular waveguides, with one-to-one mapping on
to various bands defined by the waveguide standard as shown in Table 1.3. The
waveguide standard notation starts with “WR” (Waveguide Rectangular), followed
by a number that indicates the broad-side width of the inner dimension of a
rectangular waveguide, expressed in terms of hundredths of an inch (¼10 mils).
As will be described in detail in Sect. 4.3.1, the inner dimension of a metallic
rectangular waveguide determines the operation frequency range of the waveguides
as shown in Table 1.3 and thus appears in the band notation to indicate the range.
One may notice that V-band and W-band, a part of IEEE band standard, are also
included in this list, although the range for V-band in this notation (50–75 GHz)
slightly differs from the V-band defined in the IEEE standard (40–75 GHz). There
are a few other band designations beyond 220 GHz sparsely used by some such as
H-, Y-, or J-bands, although their usage has been limited and not fully accepted yet.
Other definitions have also been used. Frequency bands that are frequently found
in relation to AM broadcasting or amateur radio communication are short-wave
(SW), medium-wave (MW), and long-wave (LW), as described in Table 1.4. These
bands are roughly distributed over HF, MF, and LF of the ITU band notation. Note
that most of the AM broadcasting systems are based on medium-wave (MW) from
this definition, while some AM broadcasts are based on short-wave and long-wave.
The term “microwave” is widely used as a general term indicating higher parts of
radio frequencies, but the official definition of microwave band is the frequency
range of 300 MHz–300 GHz [4]. It comprises UHF, SHF, and EHF of the ITU
notation.
In the above, the frequency band definitions have been reviewed for radio
frequencies. If we move our attention to the other side of the THz band, there are
band definitions in the optical spectrum. One major difference encountered as we
enter the domain of optics is that the electromagnetic bands are now described in
terms of wavelength instead of frequency as was the case for radio frequencies. This
probably arises from the fact that most optical phenomena are better described in
terms of wavelength than frequency. In the discussion below, we will present the
optical bands mainly in terms of wavelength to follow the convention, but the
corresponding frequency ranges will also be provided as an effort to view the optics
from radio frequency perspective. The optical band with the lowest frequency
(longest wavelength) is the infrared band, which is sequentially followed by visible
light, ultraviolet, X-ray, and γ-ray bands. The boundaries of each optical band,
however, are not clearly defined and vary over the sources. You will notice that
even for visible light band, the boundaries slightly vary depending on where you find
the information. Table 1.5 lists the boundaries of the optical bands in terms of the
frequency and wavelength following a standard college textbook [5]. The infrared
band can be divided into smaller sub-bands, typically into the far-infrared (FIR),
mid-infrared (MIR), and near-infrared (NIR) bands, although there are other ways of
dividing the band. FIR is the region where optics and the traditional radio engineer-
ing encounter each other. It is often defined as 15 μm–1 mm (300 GHz–20 THz),
while the upper boundary (15 μm) is sometimes set differently. By most of the
definitions, FIR contains the sub-millimeter-wave band (300 GHz–3 THz) and in
great part overlaps with the THz band (100 GHz–10 THz) as discussed earlier.
The band definitions introduced so far are based on either frequency or wave-
length. However, electromagnetic waves can be described by other physical quan-
tities as well. In some occasions, wavenumber, defined as an inverse of the
wavelength, is used. The readers will notice that most of the spectroscopy data are
presented in terms of wavenumber. Photon energy is another way to represent
electromagnetic waves, which assumes that the energy of electromagnetic waves is
quantized into photons. Photon energy hν can be converted to temperature with a
relation of kT ¼ hν, where k and h are Boltzmann constant and Plank constant,
respectively. In this case, the temperature T roughly corresponds to the temperature
Table 1.5 Optical frequency Band designation Frequency range Wavelength range
bands
Infrared 300 GHz–385 THz 1 mm–780 nm
Visible light 385 THz–770 THz 780–390 nm
Ultraviolet 770 THz–34 PHz 390–9 nm
X-ray 24 PHz–50 EHz 12.5–0.006 nm
γ-ray >2.4 EHz <0.125 nm
6 1 Introduction
Table 1.6 Conversion between frequency and various physical parameters: (a) Shown in the
decades of frequency, (b) Shown in the decades of wavelength
Frequency Time period Wavelength Wavenumber Photon energy Temperature
(a)
1 kHz 1 ms 300 km 0.0033 km1 4.14 peV 48 nK
10 kHz 100 μs 30 km 0.033 km1 41.4 peV 480 nK
100 kHz 10 μs 3 km 0.33 km1 414 peV 4.8 μK
1 MHz 1 μs 300 m 0.00003 cm1 4.14 neV 48 μK
10 MHz 100 ns 30 m 0.0003 cm1 41.4 neV 480 μK
100 MHz 10 ns 3m 0.003 cm1 414 neV 4.8 mK
1 GHz 1 ns 30 cm 0.03 cm1 4.14 μeV 48 mK
10 GHz 100 ps 3 cm 0.33 cm1 41.4 μeV 480 mK
100 GHz 10 ps 3 mm 3.33 cm1 414 μeV 4.8 K
1 THz 1 ps 300 μm 33.3 cm1 4.14 meV 48 K
10 THz 100 fs 30 μm 333 cm1 41.4 meV 480 K
100 THz 10 fs 3 μm 3333 cm1 414 meV 4800 K
1 PHz 1 fs 300 nm 33,333 cm1 4.14 eV 48,000 K
(b)
3 kHz 333 μs 100 km 0.01 km1 12.4 peV 144 nK
30 kHz 33.3 μs 10 km 0.1 km1 124 peV 1.44 μK
300 kHz 3.33 μs 1 km 1 km1 1.24 neV 14.4 μK
3 MHz 333 ns 100 m 0.0001 cm1 12.4 neV 144 μK
30 MHz 33.3 ns 10 m 0.001 cm1 124 neV 1.44 mK
300 MHz 3.33 ns 1m 0.01 cm1 1.24 μeV 14.4 mK
3 GHz 333 ps 10 cm 0.1 cm1 12.4 μeV 144 mK
30 GHz 33.3 ps 1 cm 1 cm1 124 μeV 1.44 K
300 GHz 3.33 ps 1 mm 10 cm1 1.24 meV 14.4 K
3 THz 333 fs 100 μm 100 cm1 12.4 meV 144 K
30 THz 33.3 fs 10 μm 1000 cm1 124 meV 1440 K
300 THz 3.33 fs 1 μm 10,000 cm1 1.24 eV 14,400 K
of a blackbody that would emit photons with an energy around hν (to be more
precise, the most probable photon energy from a blackbody with a temperature of
T is 2.8 kT rather than kT). The conversions between these parameters are summa-
rized in Table 1.6, in terms of the decades of frequency and wavelength separately.
For a quick reference, 1 THz corresponds to a wavelength of 300 μm, wavenumber
of 33.3 cm1, photon energy of 4.14 meV, and temperature of 48 K.
One of the well-known properties of THz waves is the high absorption in water. The
absorption spectra and the causes of the absorption vary over the state of water.
Hence, the discussion needs to be made separately for three phases of water: water
1.2 Properties of THz Waves 7
vapor, liquid water, and ice. For water vapor, the absorption results mostly from the
transitions related to the rotational modes, the vibrational modes, and the electron
states of the water molecules. The energy associated with the transitions between
rotational modes is relatively small and corresponds to the THz band, leading to the
absorption of THz waves, whereas vibration-mode and electron-state transitions
affect shorter wavelength (higher frequency) regions. For liquid water, the hydrogen
bonding between water molecules, which results from the shorter distances between
water molecules than in vapor state, suppresses the rotations of the water molecules.
Instead, the vibration modes are affected by the hydrogen bonding, and the relevant
energy is shifted into the THz region, dominating the absorption in the THz band.
For ice, the situation is similar to the liquid water case, with a stronger inter-
molecular interaction by the hydrogen bonding. The water absorption spectra for
all three water phases are shown in Fig. 1.1 [6]. It is interesting to note that the
absorption profile for liquid water (and ice) does not show sharp peaks as shown for
water vapor absorption, which is a result of the broadened peaks due to inter-
molecular interaction in condensed state. Overall, it is obvious from the figure that
the water absorption in the THz band by water, regardless its phase, is much stronger
than in the adjacent bands, such as visible light and lower-frequency radio bands.
Internal to the THz band, the absorption tends to increase with frequency. For
105
Water vapor
Liquid water
Absorption (cm–1)
103
101 Water
vapor
10-1
Ice T = –1°C
T = –5°C
10–3 T = –20°C
T = –60°C
10–5
10–2 10–1 100 101 102 103 104 105 106 107
Wavelength (µm)
Fig. 1.1 Water absorption spectra for the three phases of water: water vapor, liquid water, and ice.
(Adapted by permission from Springer Nature: [6] © 2007)
8 1 Introduction
example, the absorption coefficient for liquid water at 10 THz is around 1000 cm1,
while it is below 100 cm1 at 100 GHz. The absorption levels by water vapor and ice
inside the THz band show even a shaper increase with frequency as indicated by the
profiles.
Another notable property is that various optically opaque dielectric materials,
such as paper, clothe, leather, wood, and most of plastics and ceramics, are trans-
parent to THz waves. One may be tempted to see what is behind such transparency
observed with the listed materials. Obviously, the transparency of dielectric mate-
rials is closely linked to the absorption level of electromagnetic waves in the given
material as represented by the absorption coefficient, although it is also partly related
to the reflection and scattering. The frequency dependence of the absorption coeffi-
cient is related to so many macroscopic and microscopic properties of the material
and cannot be discussed in a general way. The readers may recall the basic electro-
magnetic theory introduced in college classes, in which the absorption coefficient (α)
of electromagnetic waves in a dielectric material with a finite loss is given as:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u "rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
h i2
#
uμε σ
α¼ω t 1þ 1 , ð1:1Þ
2 ωε
where μ, ε, and σ are the permeability, permittivity, and conductivity of the material,
respectively, and ω is the angular frequency. It shows a general tendency for α that
increases with increasing frequency. However, the reality is not that simple and
affected by lots of other factors as well. For example, μ, ε, and σ in the equation,
which are rather macroscopic parameters, have their own frequency dependences,
which will also affect the trend. More importantly, in the microscopic level, the
transitions between energy states inside atoms, molecules, and lattices constituting
the material would lead to a great number of absorption lines that are specific to the
incident photon energy or electromagnetic wave frequency. This will also strongly
affect the frequency dependence of the absorption. In fact, the frequency dependence
of the water absorption shown in Fig. 1.1 is mostly dominated by this factor. With all
these observations given, there seems no single theory that explains the behavior of
THz waves in dielectric materials. Instead, all the factors mentioned above rather
collectively affect the transparency observed with the materials listed above. As will
be discussed in further detail later, the transparency of those materials is surely a
highly favored property for various practical applications including imaging. It is fair
to note, though, that such transparency may apply to the lower radio frequency bands
as well, even with less absorption. However, the THz waves additionally benefit
from the smaller wavelength, which is favored for higher resolution desired in
imaging applications.
Reflection on metallic surfaces is also often cited as a major property of THz
waves. The high reflectivity is useful to obtain a high contrast between metallic and
non-metallic surfaces in imaging applications, among other advantages. As a matter
of fact, the reflection on metallic surfaces, which results from the high free-electron
density in metals, is a fundamental property of electromagnetic waves in general, not
1.3 Approaches to THz Systems 9
limited to the THz band. At this point, however, it would be worthwhile to briefly
review the frequency-dependent behavior of the reflection near the metallic surface.
The reflection of electromagnetic waves on metallic surfaces is, in fact, not a total
reflection unless the conductivity of the metal ffiis infinite. With a very large conduc-
pffiffiffiffiffiffiffiffiffiffiffiffiffi
tivity, α in Eq. (1.1) is reduced to ωμσ=2 , which is basically the absorptionpffiffiffiffiffiffiffiffiffiffiffiffiffiffi
coefficient of a conducting material. The inverse of this parameter, or 2=ωμσ ,
has a dimension of length and is called the skin depth, which indicates the average
depth of electromagnetic wave penetration below the metallic surface. To be more
precise, it is the depth at which the field strength is reduced down to 36.8% (1/e) of
the original value at the surface. It shows that the extent of the finite penetration of
electromagnetic waves into metal depends on the frequency, which decreases with
increasing frequency. Hence, it can be said that a more ideal reflection on metallic
surfaces can be obtained for THz waves when compared with lower-frequency
waves. We will come back to the issue of the skin depth in Sect. 4.3.2 with more
practical situations.
An outstanding advantage of THz waves for imaging applications, particularly
against the prevailing X-ray imaging, is that THz radiation is non-ionizing radiation,
imposing little hazardous effects on human body when exposed. Ionizing radiation is
a kind of radiation that carries photon energy large enough to detach electrons from
atoms or molecules, leading to their ionization. An exposure of a human body to a
high dose of ionizing radiation may cause damage to DNA, leading to increased
possibility of developing cancer afterwards. When the energy is larger than ~10 eV,
we say the radiation is ionizing. Typical X-ray imaging systems for medical purpose
involves photon energies close to 100 keV. Hence, accumulated exposures to a high-
dose X-ray radiation may be harmful for humans. In contrast, the energy
corresponding to THz waves falls on the order of tens of meV at most, remote
from the energy range of concerns. This clearly indicates that THz imaging is a safe
imaging modality, especially when compared with X-ray imaging.
The THz band is situated between the spectra covered by the traditional electronics
and optics, both of which have been heavily explored for various science and
engineering purposes. In contrast, the THz spectrum has been relatively less
exploited so far. For this reason, the THz band is often referred to as the “THz
gap.” The relatively slow development of the THz band for applications can be
attributed to at least a couple of major challenges. First, the attenuation level of THz
waves in the Earth’s atmosphere is high. As shown in Fig. 1.2 [7], the largest
attenuation occurs for a range of 1–10 THz, the core portion of the THz band,
which arises mostly from the strong absorption by water molecules in the atmo-
sphere. The high attenuation will limit the long-distance propagation of the THz
waves in the air. Second, we lack devices to properly manipulate THz waves,
including generation and detection. This is symbolically revealed by Fig. 1.3 [8],
10 1 Introduction
Fig. 1.2 Attenuation of electromagnetic waves in the atmosphere of Earth in various conditions.
The largest attenuation occurs for 1–10 THz frequency range. (© 2007 IEEE [7])
0.1
0.01
0.001
10 100 1,000 10,000 100,000
Frequency (GHz)
which shows the output power from solid-state signal sources as a function of
frequency. When approached from low frequencies based on electronic devices,
the output power decreases with increasing frequency, while the power decreases
with decreasing frequency when approached from high frequencies with optical
devices. As a result, there exists a region where neither approach provides a
sufficient output, which is around the THz band as obviously indicated in the figure.
While the first challenge is not something we can engineer, the second one can be
overcome with technological advances. In fact, a great deal of advances are being
1.3 Approaches to THz Systems 11
Fig. 1.4 Various techniques for THz systems. (a) Signal generation (b) Signal detection
made in device technologies, both electrical and optical, to narrow down the THz
gap. Although the details will be discussed in Chaps. 2 and 3 (electrical methods) as
well as in Chap. 5 (optical methods), it would be helpful to take an overview of the
approaches to build up THz generation and detection systems. This section will
guide the readers through various techniques to generate and detect THz signals,
which are highlighted in Fig. 1.4.
The signal generation has long been a major bottleneck for exploiting the THz band,
and, as such, extensive efforts have been made to develop THz sources with
sufficient output power. Those efforts have been recently rewarded by some prom-
ising results, from both electrical and optical fronts, as will be briefly reviewed
below.
12 1 Introduction
Let us first start with the devices with the electrical approach. There are two
distinct electrical approaches to generate THz signals, one based on vacuum devices
and the other based on solid-state devices. THz signal generation with vacuum
devices is basically an extension of the efforts to generate microwave signals with
vacuum devices, which can be traced back to the early twentieth century. Signal
generation with vacuum devices is basically a process of converting modulated
electron current into electromagnetic wave radiation. The electron current modula-
tion, which happens inside a vacuum tube, can be either longitudinal or transverse.
The devices with longitudinal electron current modulation, a process that necessarily
involves bunching of electron beam, include klystrons, traveling-wave tubes
(TWTs), and backward wave oscillators (BWOs). Categorized as the devices with
transverse modulation are gyrotrons and free electron lasers (FELs), the former with
helical motion and the latter with wiggling motion of electrons. Inside these devices
with transverse modulation, the phase velocity of the electromagnetic wave exceeds
the speed of light, and thus they are also called fast-wave devices. On the other hand,
the devices described earlier (klystrons, TWTs, BWOs) are called slow-wave
devices as the wave phase velocity is smaller than the speed of light. Another
vacuum device that is based on a quite different operation principle from those
listed above is a magnetron, in which electrons travel in the radial direction. It can be
categorized as a slow-wave device, too.
Solid-state electrical signal sources are typically based on semiconductors, and
they can be grouped into diode-based sources and transistor-based sources. Widely
used diode-based sources are resonant tunneling diodes (RTDs), Gunn diodes, and
IMPATT (IMPAct ionization Transit Time) diodes. They are all based on different
physical mechanisms as will be described in Chap. 2 in detail. There are sources
based on more conventional diodes, such as p-n junction diodes and Schottky
diodes, which are in fact oscillator circuits employing a resonant structure realized
with the diodes. A more preferred approach for oscillator circuits, though, is to
employ transistors as the core circuit components. A variety of circuit topologies
including transistors can create oscillating current, which can be radiated through an
antenna or can be delivered to neighboring circuit components. Realizing transistor-
based oscillators operating at the THz band is challenging and has been limited due
to the transistor performance that is not sufficient for such a high-frequency band.
However, the barrier is getting lowered with the recent progress in semiconductor
process technologies. Figure 1.5 shows the transistor operation speeds reported by
various technologies based on fT (cutoff frequency, where the current gain becomes
unity) and fmax (maximum oscillation frequency, where the power gain becomes
unity), which indicates that transistors are now capable of operating at several-
hundred GHz or even beyond 1 THz depending on technologies. With the help of
this advancement in semiconductor technologies, a growing number of transistor-
based signal sources are now being reported based on various technologies, as will
be discussed in Chap. 2.
Optical methods are a more traditional mode for THz signal generation, which is a
major reason for the dominating position of the optics-based approach for THz
research till today. While the signal generation with electronic devices is mostly
continuous wave (CW), a more prevailing form of THz signals generated with
1.3 Approaches to THz Systems 13
Fig. 1.5 Transistor operation speeds reported by various semiconductor process technologies
represented in terms of fT (cutoff frequency) and fmax (maximum oscillation frequency). (From
various sources)
THz detectors are another key component to complete THz systems together with
THz sources. There have been various types of THz detectors developed, for which
14 1 Introduction
high responsivity and low noise level are desired. They can be categorized into
thermal detectors, square-law detectors, and heterodyne detectors. Also, THz pulses
can be detected with optical methods.
For thermal detectors, the temperature rise due to THz wave absorption is
exploited to estimate the incident power level, for which various techniques can be
employed. Bolometers convert the temperature rise into resistance change of the
constituting material, which is in turn converted into a measurable electrical quantity
such as current or voltage. Golay cell is another popular THz detector, in which the
temperature rise of the gas contained in a cell due to THz wave illumination causes
inflation of the gas cell, which alters the reflection path of a light incident on the cell
surface because of the deformed surface by the inflation. In pyroelectric detectors,
the temperature change due to illumination causes change in the polarization prop-
erty of pyroelectric materials, which can be converted to voltage change across a slab
of such materials and then measured. Thermopile makes use of the temperature
difference established across a material by illuminating THz waves on one side of
the material, which is converted into a voltage difference by Seebeck effect. In a
sense, these thermal detectors can be categorized as detectors based on electrical
approach, although the details vary over the device types.
Square law detectors exploit the nonlinear response property of devices or
circuits, which converts incoming CW sinusoidal signal into DC signal (as well as
other harmonic frequency components). By detecting the amplitude of the generated
DC signal, the strength of the incident signal can be determined. Since the measured
DC voltage amplitude is proportional to the square of the input signal voltage, they
are called square law detectors. In fact, because the input power is proportional to the
square of the input signal voltage for a given load, the measured DC voltage is
proportional to the input power for normal operation. For square law detectors, in
principle, any nonlinear device can be used, while diodes and transistors are widely
adopted for compatibility with other electronic parts.
For heterodyne detectors, the input signal is first frequency down-converted to a
lower frequency then detected at intermediate frequency (IF). For the frequency
down-conversion, a mixer is a key component, and thus those mixers operating at the
THz band have long been considered as effective THz detectors. Traditional THz
mixers are based on diode structures, which include Hot Electron Bolometers
(HEBs), Semiconductor-Insulator-Semiconductor (SIS) mixers, and Schottky Bar-
rier Diode (SBD) mixers. With the advent of various high-frequency semiconductor
technologies, heterodyne receivers based on transistor circuits have also reached
sufficiently high operation frequencies to be used in the THz band. With this
approach, a high-level integration is possible, which would comprise a mixer,
local oscillator, IF components, and sometimes a low-noise amplifier at the front,
all on a single chip. The detection based on the electrical methods will be detailed in
Chap. 3.
Optical methods can also be used for THz detection, mostly for the recovery of
incoming THz pulses. THz signals applied on photoconductive (PC) material can be
detected when an ultrashort optical pulse from a femtosecond laser is simultaneously
applied, as the pulse triggers a current with photo-generated carriers. With the
1.4 THz Applications 15
sampling process (also called gating process) based on the controlled time delay of
the optical pulses arriving at the detector in synchronization with THz pulses, the
profile of the incoming THz pulses can be recovered. Electro-optic (EO) material can
also be used to recover THz pulses. Linearly polarized optical beam incident on an
electro-optic material will go through a polarization-dependent phase delay with the
presence of THz field applied on the material. The phase difference between the two
polarized beams emerging from the EO material will determine the strength of the
incoming THz signal. In combination with a proper sampling process, it will enable
the recovery of the THz pulse shapes, similar to the case with PC material. Details of
the optical detection will be described in Chap. 5.
It should be fair to say that the interests on the THz band have been so far more
inclined toward the research purposes than the practical ones, especially when
compared with the neighboring spectra heavily explored by electronics and optics
for many useful practical applications. However, there are limited but important
application fields that have long relied on the THz spectrum. Also, on a more recent
but promising side, there are rapidly emerging practical THz applications that are
expected to be more closely linked to our daily lives. Some of these THz application
areas are introduced below, which will be described in more detail in Chap. 6.
Spectroscopy is the analysis of the interaction between matter and electromag-
netic waves as a function of frequency or wavelength, usually based on the emission
or absorption properties of the matter. The response of a material to electromagnetic
waves provides a great amount of physical and chemical information of the material,
and the spectroscopy is a useful tool to extract such information from the given
material. The readers should be aware that the key clues for understanding the
internal structure of atoms came from the spectral lines emitted by the atoms, as
spectroscopic methods were successfully applied to the atomic study in the late
nineteenth century. Different ranges of spectrum carry different information. THz
spectrum, for example, contains critical information on the properties of gas mole-
cules, since the photon energy of the THz band corresponds to the transition energies
in the rotational and vibrational modes of the molecules as mentioned earlier. This
makes THz spectroscopy a precious tool for the study of the behavior of gas
molecules. Traditionally, THz spectroscopy was mainly based on the frequency-
dependent reflection behavior of gratings, but Fourier-transform spectrometer that
emerged later has extended the capability of THz spectroscopy. This was followed
by the THz time-domain spectroscopy (TDS), which became a mainstream of
modern THz spectroscopy till today. See Sect. 6.1 for more details of these methods.
Imaging refers to a process of mapping the locational property of an object to
another space, mostly 2D but sometimes 3D, based on the spatial distribution of the
radiation from the object that comes from reflection, transmission, or self-generation.
When the radiated electromagnetic wave belongs to the THz band, we call it THz
16 1 Introduction
imaging. While there are lots of imaging modalities including the most conventional
one based on visible light, THz imaging has its own attractive features. The major
unique properties of THz waves described in Sect. 1.2 can be mostly converted to the
advantages of THz imaging. The high water absorption level of THz waves results in
high-contrast images with materials with a high water content. The transparency of
optically opaque materials to THz waves offers opportunities for security imaging,
especially useful when it is applied to humans owing to the non-ionizing character-
istics of THz waves. When compared with microwave imaging, THz imaging pro-
vides an enhanced resolution with the shorter wavelength. As for the actual methods
to practice THz imaging, there are many different options. There are active imaging
and passive imaging, while active imaging can be obtained with either transmission
or reflection mode. It may be based on THz pulse or CW THz wave. THz imaging
can also be obtained with either single-pixel detectors or array detectors. The details
of these various modalities are described in Sect. 6.2.
Wireless communication defines our modern lives in various ways in the form of
cellular networks as well as local connectivity, such as WiFi, Bluetooth, zigbee, and
so forth. We are always hungry for higher data rate for quick downloads and uploads
with these networks. One of the most efficient ways to achieve such a high data rate
is to increase the bandwidth, which in turn can be realized with a higher carrier
frequency. From this perspective, the THz band is a highly attractive and preferred
candidate when compared with the currently prevailing frequency band of 1–2 GHz
band as well as to the emerging mm-wave band. Of course, there are great challenges
for the THz waves to be successfully applied to the wireless communication, which
arise from the transmitter/receiver performance limit as well as the attenuation due to
Earth’s atmosphere and high free-space path loss (FSPL). However, there are
intensive efforts growing recently to overcome these challenges and pave the way
for THz communication systems in various application scenes, which will be more
detailed in Sect. 6.3.
Radar (RAdio Detection And Ranging) is a system that determines the range
(distance) and velocity of a target object based on the radio wave detection. While
its application has long been driven by military needs, its commercial application is
attracting growing interests recently, one good example being the car radars based
on 77–79 GHz frequency range. While the application of the THz band to radars is
still on its infant stage, there are obvious advantages expected for THz-based
radars. The shorter wavelength of THz waves than usually used for current radars
will provide an improved resolution for the ranging data acquired. Also, for
imaging radars, which are gaining increasing attention, the image resolution will
be improved, too, with enhanced angular accuracy. Although limited by the same
set of challenges that affect the other applications, the expected advantages will
push the development of THz radars for various practical applications. The details
will be discussed in Sect. 6.4.
References 17
References
In this chapter, a review on various electrical methods to generate THz signals and
related techniques will be presented. Traditional THz sources based on vacuum
devices and diodes will be first discussed, followed by oscillator circuits with
transistor technology operating at the THz band. Discussed next will be two effective
ways to boost the oscillation frequency of THz signal sources, the n-push technique
and frequency multiplication. This will be followed by the description of two widely
used power-boosting techniques, amplification and power combining. Lastly,
beamforming, which is a special case of power-combining technique that addition-
ally enables the steering of beams with improved directivity, will be treated as a
separate topic.
THz signal generation with vacuum devices is an extension of the efforts to generate
microwave signals based on vacuum devices, which can be traced back to the early
twentieth century. Signal generation with vacuum devices is basically a process of
converting a modulated electron current into electromagnetic wave radiation. The
signal output power obtained from vacuum sources is typically much larger than the
output power from solid-state devices. This is mainly because the electron current in
vacuum devices, from which the radiation is extracted, is a convection current
through the vacuum, whereas the electron current in solid-state devices is basically
a conduction current that travels through a medium that necessarily involves carrier
scattering. The absence of scattering inside the vacuum devices leads to a more
efficient transport as well as low heating effect, a big advantage of vacuum device-
based signal generation.
Inside a vacuum device, or tube, electron current modulation is applied for signal
generation. The electron current modulation can be either longitudinal or transverse.
The devices with longitudinal electron current modulation, a process that typically
involves bunching of the electron beam along the longitudinal direction, include
klystrons, traveling-wave tubes (TWTs), and backward wave oscillators (BWOs).
Categorized as the devices with transverse modulation are gyrotrons and free
electron lasers (FELs), the former with helical motion and the latter with wiggling
motion of electrons. Internal to this type of devices, the phase velocity of the
generated electromagnetic wave is larger than the speed of light, so they are often
called fast-wave devices. In contrast, the devices with longitudinal modulation, as
mentioned earlier, are called slow-wave devices, with the phase velocity smaller than
the speed of light. Yet another vacuum device, which is based on a quite different
operation principle than those mentioned above, is the magnetron. It does not
involve electron modulation but still can be categorized as a slow-wave device if
the aforementioned definition is applied.
In this section, a brief snapshot of some selected vacuum devices is provided. For
a more detailed description, the authors are encouraged to refer to other sources
focusing on vacuum devices [1–4].
2.1.1 Klystrons
The klystron is essentially an amplifying device based on a vacuum tube, which can
be converted to an oscillator for signal generation when needed. It was invented by
the Varian brothers at Stanford back in 1939 [5]. It modulates the electron beam
emitted by an electron gun inside a cavity resonator as triggered by the input signal,
generating a stream of electron bunches along the tube. The bunched electrons excite
electromagnetic waves at another cavity resonator located on the opposite side of the
tube, and the generated electromagnetic waves are extracted as the device output
signal. The power level of the output signal generated in this way is much larger than
that of the input signal. Hence, amplification is attained with the device.
The operation can be described in more detail with the help of Fig. 2.1a, which
shows the typical structure of a klystron. Electrons are emitted from the cathode that
is heated for thermal emission and then accelerated by the electric field established
by the anode at a more positive potential. The unit composed of the cathode and the
anode is called the electron gun. The electron beam emerging from the gun maintains
a steady flow with a uniform density along the traveling (longitudinal) axis until it
reaches the first cavity resonator, which is called the “buncher” cavity. This cavity
with a circular cross section serves as a cylindrical resonant cavity. That is, when it is
excited by the AC input signal supplied through a coaxial cable or waveguide, an
alternating electric field is induced across a pair of grids that are located at the input
and output boundaries of the cavity. The alternating electric field, which is of the
same frequency as the input signal, periodically accelerates and decelerates the
electrons that pass through the cavity. In consequence, as the modulated electrons
stream out of the cavity and travel through the space along the tube between the two
cavities (called the drift space), accelerated electrons tend to catch up the
2.1 Vacuum Device Sources 21
Fig. 2.1 (a) Schematic of a typical klystron. (b) Space–time diagram of electron beam showing the
formation of electron bunches
22 2 THz Sources and Related Topics
decelerated electrons that move ahead. As a result, a series of electron bunches are
formed along the axis by the time they arrive at the second cavity.
This bunch-forming process can be better understood with Fig. 2.1b, which plots
the distances traveled by the velocity-modulated electrons as a function of time
[6]. Electron bunching induced by the periodic modulation is the most pronounced at
location L. This is the location where the second cavity is located. Note that this
cavity, called the “catcher” cavity, has the same dimension and thus the same
resonant frequency as the buncher cavity. The electrons passing through the catcher
cavity excites electromagnetic waves inside the cavity with the same frequency as
the input signal. The generated electromagnetic waves are extracted as the output
signal at the output port. It shows a much larger power level than the input signal
power, as it has gained energy from the electron beam passing through the cavity. To
be more precise, the induced electrical field across a pair of grids at the catcher cavity
acts to decelerate the electrons and reduce their kinetic energy, and the energy
“stolen” from the electrons is transferred to electromagnetic waves being generated.
In fact, the induced electric field at the catcher cavity, triggered by the electron
bunches, is an alternating field, and it will not only decelerate but also partly
accelerate the electrons passing through the cavity as was the case for the buncher
cavity. There is a difference, though. The alternating field at the catcher cavity acts
on bunched electrons, whereas the field at the buncher cavity acts on a uniform flow
of electrons. The deceleration cycle at the catcher cavity coincides with the peak of
the bunches, while the acceleration cycle falls on the valleys between the peaks.
Hence, there are far more electrons decelerated than those accelerated, leading to a
net reduction in the energy of the electron beam. The electrons that have lost kinetic
energy to the catcher cavity will eventually arrive at the collector located at the end
of the tube and exit the device.
So far, the standard klystron that contains two cavities has been considered to
explain the operation. However, additional cavities can be added to the device to
improve the performance, such as gain or bandwidth. This type of klystron is called
the multi-cavity klystron. There are other variants as well: the reflex klystron with a
single cavity, the extended interaction klystron (EIK) with multi-gap cavities, the
optical klystron operating at optical frequencies, and so forth. Although the basic
function of klystrons is the amplification as described above, it can always be
modified to work as an oscillator by placing a proper feedback path between the
buncher and catcher cavities, as suggested by the original paper by Varian brothers
[5]. In this case, any electrical noise in the system will be amplified repeatedly
around the feedback loop to establish a sustaining oscillation with a sufficient output
power, in which the oscillation frequency will be the same as the resonant frequency
of the cavities. Also, klystrons may be operated in the pulsed mode instead of the
standard continuous wave (CW) mode when needed. Modern high-frequency kly-
strons are known to operate up to a few hundreds of GHz and the typical output
power can reach beyond 10 W (for CW operation) near 100 GHz [7, 8]. The main
challenge for THz operation of klystron is to reduce the electron beam diameter to a
level much smaller than the operating wavelength. The heat management and the
2.1 Vacuum Device Sources 23
tight assembly accuracy, which become more challenging with increasing fre-
quency, also need to be addressed for THz applications [9].
The traveling-wave tube (TWT) is another type of vacuum device used for the
amplification of electromagnetic waves applied as a signal. Although many
researchers have contributed to developing the concept of the early TWT, Kompfner
is widely credited for the invention of TWT in the early 1940s [10]. The overall
operation principle of TWTs [11] is not much different from that of klystrons.
Electrons are emitted from an electron gun, and electron bunches are formed as
they travel down to the collector along the tube. The electron bunches, created by the
input signal, will end up generating the output signal with a much larger power than
the input signal by transferring a part of their kinetic energy to electromagnetic
waves. The main difference, however, lies in the manner the electron bunches are
created and its energy is delivered to the output signal. This will be described below
with its conceptual diagram provided in Fig. 2.3a.
Remember the electron bunches in klystrons are created by the electric field of the
standing wave localized inside a cavity. On the other hand, the electron bunches in
TWTs are created by the electric field of the wave traveling along the tube, which is
the origin of the device name. For the electrons emitted from the electron gun to
effectively interact with the traveling wave injected at the input port, the electron
movement needs to be synchronized with the phase of the electromagnetic wave.
That is, the electron velocity should be same as the phase velocity of the electro-
magnetic wave. In this way, the electrons can remain stationary against the electric
field profile of the traveling wave, when viewed from a reference frame moving at
the same speed as the electrons and the wave. This is a preferred situation for a
consistent mutual interaction between the electrons and the wave. One issue in this
process is that the electron velocity established by the electron gun is typically much
smaller than the wave velocity, which is in fact the speed of light in vacuum. Hence,
a mechanism to slow down the traveling wave is required, which is typically
achieved by employing a slow-wave structure. There are various forms of slow-
wave structures. A traditional one that has been widely used for TWTs (and also for
backward wave oscillators to be described later) is the helix, or a coiled wire, as is
assumed in the device depicted in Fig. 2.2a. In this case, the phase velocity of
electromagnetic waves guided by the helix will be reduced by a factor of sinψ when
the wave travels along the longitudinal axis, where ψ is the angle between the tilted
coil and the vertical (transverse) axis (see Fig. 2.2a).
With the slow-wave structure in act and the velocities in synchronization, some
electrons will experience the accelerating phase of the field while some will expe-
rience the retarding phase of the field as they travel along the tube, depending on
their relative location against the periodic field profile. As a result, electrons stream-
ing down the path will be squeezed into a series of bunches as is described in
24 2 THz Sources and Related Topics
Helix Coil
Cathode Anode Electron Beam Collector
Electron Focusing
Gun magnet
RF in RF out
(a)
E E
Electron Bunch
distance distance
Electron Bunch
(b) (c)
Fig. 2.2 (a) Schematic of a typical TWT. (b) Relative position of the longitudinal electric field and
electrons to generate electron bunches. (c) Relative position of the longitudinal electric field and
electrons to transfer electron kinetic energy to electromagnetic wave by electron deceleration. (The
direction of the electric field is shown with reference to electrons for convenience. The actual field is
in the opposite direction)
Fig. 2.2b. After the bunch formation, the next step is for the electron bunches to lose
some of their kinetic energy, so that it can be used to build up the output signal
amplitude. For this to happen, the electron bunches need to be decelerated as they
travel toward the collector. With the configuration given in Fig. 2.2b, however, no
deceleration of electron bunches will be obtained as the electrons are located at zero-
field positions. But if we assume the electrons were initially moving slightly faster
than the phase of the field, electrons would gradually catch up the regarding phase of
the field, leading to the configuration depicted in Fig. 2.2c. In this situation, the
electrons will begin to lose the kinetic energy, which will be transferred to the output
signal that grows until the electron reach the collector, completing the amplification
action. This energy transfer mechanism is sometimes interpreted as a Cerenkov
(Cherenkov) radiation, in which any charged particle traveling at a speed larger
than the phase velocity of electromagnetic waves traveling in that medium will
radiate electromagnetic waves. This can be compared to another type of mechanism,
2.1 Vacuum Device Sources 25
Focusing
Cathode Helix Coil Collector
Anode magnet
Electron Beam
d
Electron vg vph
Gun
Matched RF
RF out terminaon
(a)
n=1 n=0 n = -1
Forward Backward
interaction
Frequency (w )
interaction
Beam lines
-1 -0.5 0 0.5 1
Wavenumber (normalized to 2p /d)
(b)
Fig. 2.3 (a) Schematic of a typical BWO. (b) Dispersion relation of electromagnetic wave inside a
BWO showing the group velocity and phase velocity. (© 2011 IEEE [7])
which is called the transition radiation. In this latter case, electrons will radiate when
they pass through a boundary between two media with different refraction indices or,
in a more general term, travel in a medium with non-homogeneities. In fact, the
energy transfer at the catcher cavity of klystrons as described earlier can be viewed
as an example of the transition radiation. There is yet another type of transition,
bremsstrahlung radiation, which will be discussed shortly.
In the operation of TWTs described above, a helix was assumed to be employed
as the slow-wave structure. As the dispersion (speed distribution over frequency)
26 2 THz Sources and Related Topics
with a helix is known to be very small, TWTs with a helix typically show excellent
broadband characteristics. Sometimes, however, other slow-wave structures are
employed, such as the cavity-coupled structures that exhibit narrowband but high
output power. It is also noted that typical TWTs employ a focusing magnet to tightly
confine the electron beam around the center with a magnetic field formed along the
longitudinal axis. For this purpose, either electromagnet with a solenoid or perma-
nent magnets arranged in a periodic alternating configuration can be used.
The operation frequency of TWTs can reach up to several hundreds of GHz, as
indicated by a recent report that demonstrated a TWT amplifier working beyond
600 GHz [12]. Typical average output power can reach well beyond 100 W near
100 GHz [13]. Although the operation frequency of TWTs has already reached deep
into the THz band, there are still challenges to further increase the operation
frequency. It is known that the main issues mostly involve the circuit size reduction
for increased frequency, which also requires a precise beam alignment. With these
issues addressed, the operation will be extended to cover higher frequency bands
around 1 THz.
The backward wave oscillator (BWO) is an oscillator based on vacuum tubes, the
invention of which is traced back to the early 1950s [14, 15]. As opposed to
klystrons and TWTs described above, which are basically amplifiers, BWO is
intrinsically an oscillator. It can be regarded as a TWT modified for oscillator
function. Hence, BWOs share the same main structure with TWTs such as the
electron gun and the slow-wave structure. They also share the same mechanism
for electron bunch formation. Instead of applying an external feedback loop to
achieve oscillation, which is a typical method of converting an amplifier to an
oscillator as was the case for the klystron oscillator described above, BWOs are
elegantly designed to excite electromagnetic waves with a negative group velocity to
meet the oscillation condition inside the device. As a result, the electromagnetic
waves established in the tube show a group velocity and a phase velocity with the
opposite directions, the former traveling backward while the latter traveling forward.
This is how the device name bears “backward wave.” Operation principle is
described below with the help of Fig. 2.3a that depicts a typical BWO with a helix
as the slow-wave structure.
In order to understand the opposite direction (sign) for the group velocity and the
phase velocity obtained in BWOs, the readers are referred to the definition of the two
velocities, which are shown below for the sake of the readers’ convenience:
∂ω ω
vg ¼ , vph ¼ , ð2:1Þ
∂k k
2.1 Vacuum Device Sources 27
where vg and vph are the group velocity and phase velocity of the wave, respectively,
ω is the angular frequency, and k is the wave vector. As in the dispersion relation for
the electromagnetic waves inside a BWO as shown in Fig. 2.3b, there exist regions
where the group velocity and the phase velocity become opposite. The example
point indicated in the plot belongs to the first backward spatial harmonic, for which
the group velocity is negative and the phase velocity is positive. Note that, in
principle, there may exist a huge number of harmonics due to the periodicity of
the slow-wave structure. In BWO operation, the phase velocity is aligned to the
electron beam velocity for bunch formation, in the forward direction toward the
collector, while the group velocity is formed backward toward the electron gun. As
the wave energy is carried by the group velocity, the kinetic energy taken by the
electromagnetic waves from the electron beam is transported backward, and thus, the
generated electromagnetic waves grow along the backward direction. Hence, when a
wave completes a round trip along the signal path inside the tube, its magnitude
increases, and when it repeats, oscillation is achieved. The output signal is extracted
from the output port, typically located near the electron gun side of the tube.
As was the case for TWTs, various types of slow-wave structures may be applied
although a helix was assumed in Fig. 2.3a. Also, BWOs can be categorized into
either O-type or M-type (true for TWTs as well). The types are distinguished by the
direction of the static magnetic field introduced in the device. For O-type devices, the
direction is longitudinal, or along the direction of electron flow, whereas for M-type
devices, the direction is transverse, or perpendicular to the direction of electron flow.
Figure 2.3a describes an O-type BWO structure. The operation frequency of BWOs
of today has exceeded 1 THz [16]. The maximum output power obtained with
BWOs near 100 GHz is around 1 W. Variants of BWO have also been developed
for improved output power, which include clinotron [17] and orotron [18], adopting
a tilted tube and Fabry–Pérot cavity for structural variation, respectively. Around
100 GHz, for example, they show improved power by a factor of up to 10 compared
to conventional BWOs [7].
2.1.4 Gyrotrons
The gyrotron is a high-power oscillator based on a vacuum tube. Its first demon-
stration was made in 1964 by a group of Russian researchers [19]. In a sense,
gyrotrons are similar to other vacuum devices described above such as klystrons,
TWTs, and BWOs, in that they are based on a tube along which electron beam
travels and generates electromagnetic waves. However, gyrotrons contrast with
those devices in at least three main aspects. First, they do not employ slow-wave
structures that are often used in other devices to slow down the phase velocity of
electromagnetic waves along the longitudinal direction. Second, electrons travel at a
very high speed, close to the speed of light, so that relativistic effects need to be
considered. Third, the electrons travel with a helical motion, or they gyrate as they
travel along the tube. These differences end up with a significantly higher output
28 2 THz Sources and Related Topics
Superconducting mirrors
main magnet Generated
Electron gun wave
cavity
anode
collector
cathode
window
Superconducting
gun coil Electron beam
Radiation
qB
ω¼ , ð2:2Þ
m
where B is the magnitude of the longitudinal magnetic field along the tube, q is the
elementary charge, m is the electron mass (distinguished from the rest mass of
electron m0), which increases with the electron speed (ve) by Lorentz factor γ ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1= 1 ðve =cÞ2 . The increase in m is negligible at lower speed but becomes
2.1 Vacuum Device Sources 29
significant for electrons traveling at a relativistic speed, which is the case for
electrons inside gyrotrons. The electrons in a helical motion are in random phase
when they are first emitted and the radiated electromagnetic waves are yet incoher-
ent. However, as they travel into the resonator cavity, the electrons that lose kinetic
energy to the electromagnetic waves are slowed down and become lighter, leading to
increased angular frequency and thus attain a phase lead. On the other hand,
electrons that gain kinetic energy speed up and become heavier, leading to a
decreased angular frequency and a phase lag. As a result, phase-bunching is
achieved for the electrons with aligned phase and the radiated electromagnetic
waves becomes coherent. The electrons that have completed the travel across the
resonant cavity arrive at the collector and exit the device. The generated electro-
magnetic waves are extracted through a window located on the side (as is the case for
the structure shown in Fig. 2.4) or on the end of the cavity.
In fact, fast-wave vacuum devices generating coherent radiation with gyrating
motion of electrons, or gyro-devices, can be realized based on other types of devices
described earlier such as klystrons, TWTs, and BWOs by modifying the basic
structures. In those cases, they are called gyro-klystrons, gyro-TWTs, and gyro-
BWOs, respectively, operating as either oscillator or amplifier according to the
function they serve. To distinguish from these other types of gyro-devices, the
device described in Fig. 2.4 is sometimes called the gyro-monotron, a name derived
from the fact that it is based on a single-cavity structure. In this case, the term
gyrotron can be reserved for a more general use, referring to all the types of gyro-
devices, also known as electron cyclotron masers (ECMs) or cyclotron resonance
masers (CRMs).
One advantage of gyrotrons for THz operation is that increasing oscillation
frequency can be achieved with an increased magnetic field as indicated by
Eq. (2.2), not necessarily requiring a scaling of the device structure dimension as
was the case for the slow-wave devices described above. For oscillation frequencies
higher than readily available with a reasonable strength of the magnetic field,
harmonics are typically utilized. The operation frequency of gyrotrons has continu-
ously increased since its invention, leading to oscillation frequencies now exceeding
1 THz [20]. Around 100 GHz, output powers higher than 1 MW have been reported
for CW operation, while pulse-mode operation can be employed for even higher
achievable peak powers [21].
The free electron laser (FEL) is a device that generates coherent radiation on a very
wide range of spectrum, from microwave to X-ray. Oscillation with an FEL was first
demonstrated in 1977 [22], although its basic concept can be traced back to as early
as 1951 [23]. Being a fast-wave device, it shares more common features with
gyrotrons than other devices, mostly on the basic fact that electromagnetic waves
are generated with bremsstrahlung radiation by electrons traveling at a relativistic
30 2 THz Sources and Related Topics
0 5 0 5 0 5 0
5 0 5 0 5 0 5
Fig. 2.5 Schematic of a typical FEL. (Adapted by permission from Springer Nature: [24] © 2009)
speed inside a cavity. However, there are major differences that characterize the FEL
as follows: electron speed is heavily relativistic (very close to the speed of light) in
FELs while it is only moderately relativistic (comparable to the speed of light) in
gyrotrons; electrons travel in a wiggling motion in most FELs while they travel in a
helical motion in gyrotrons; there are mirrors located on both ends of the cavity in
FELs, which are absent in typical gyrotrons. A sketch of a typical FEL is presented
in Fig. 2.5 [24].
Since electrons moving at very high speed are desired in FELs, the electron-
emitting element is a bit different from other vacuum devices. Rather than a
conventional electron gun composed of a cathode and an anode, more sophisticated
and powerful electron accelerators are often adopted for electron emission. The
emitted electrons travel through a path flanked on both sides by an array of magnets
with alternating polarity placed along the longitudinal direction, which is called the
undulator. Due to the alternating periodic lateral magnetic field, the electron beam
follows a zigzag or wiggling path. With a periodic change of the velocity direction,
the electrons radiate, which is initially incoherent. However, as the generated
electromagnetic waves are reflected back and forth by the pair of mirrors at the
cavity ends and spatially confined within the cavity with proper modes established,
electron bunches are formed, with its period aligned to the wavelength of the
electromagnetic waves. Accordingly, the radiation from the bunched electrons
becomes coherent. With this process, a high-power coherent radiation is obtained,
which is extracted at the output port through a coupler.
In fact, FELs are not exactly a laser by a strict definition. As indicated by the full
name, Light Amplification by Stimulated Emission of Radiation, a laser involves a
stimulated emission. FELs are based on spontaneous emission from the accelerated
electrons (see Sect. 5.2.3 for a brief description of emission types and laser opera-
tion). Nevertheless, FELs share a key feature with the conventional lasers in that they
both employ a cavity bounded by a pair of mirrors that help to confine the radiation
for high density and generate coherent radiation. This is how the original term is still
maintained, which was first suggested in 1973 [25].
The operation frequency of FELs is mainly determined by the spatial period of the
magnet arrays and Lorentz factor, with an additional affect by the magnetic field
strength. Hence, FELs can cover a wide range of spectrum with a proper undulator
design and control over the electron acceleration, including the THz band. It is noted
2.1 Vacuum Device Sources 31
that a typical FEL is not a type of device that can be hand-carried or installed on a
laboratory table top, but rather a bulky facility, shared by researchers from a wide
range of locations. The operation frequency range and achievable output power vary
over different facilities. The Novosibirsk FEL facility in Russia, for example,
operates over a frequency range of 1–3 THz with an average power and a peak
power up to 0.5 kW and 1 MW, respectively [26].
2.1.6 Magnetrons
The magnetron is another type of vacuum tube-based oscillators, which has found its
way in a wide range of applications. While it is best known for its popular adoption
for microwave ovens ubiquitous over the world, it was a key component in military
applications in its early days. The magnetron was developed in the first half of the
twentieth century based on cumulative contributions by a number of investigators
around the world. A pioneering work by Hull in 1921 [27] that detailed the key
operation principles of magnetrons triggered some earlier versions of magnetron
development, while more practical versions with a higher output power based on a
structure similar to modern cavity magnetrons were demonstrated in 1940 by
Randall and Boot [28]. The structure and operation of magnetrons are substantially
different from the other vacuum devices described so far, mainly in that the electrons
in magnetrons travel in the transverse direction in perpendicular to the magnetic
field, a key feature of M-type devices.
The basic structure of a typical magnetron is shown in Fig. 2.6. As can be seen
from the cross section of the magnetron, the cathode, a cylindrical rod, is located at
Fig. 2.6 Cross sectional view of a typical magnetron, which shows the trace of traveling electrons
and the direction of the magnetic field. Induced charge at the cavities are shown as + and . Also
shown in the zoomed box are the equivalent circuit components corresponding to the cavity
resonator
32 2 THz Sources and Related Topics
the center along the tube axis and is concentrically surrounded by the anode, which
contains multiple circular cavity resonators along the perimeter. A magnetic field is
applied by an external magnet along the longitudinal (axial) direction, while a
transverse (radial) electric field is established between the cathode and the anode.
Electrons emitted from the cathode travel initially toward the anode in the radial
direction as affected by the electric field. Soon after the emission, the electrons
experience a force in the azimuthal (rotational) direction due the magnetic field,
which tends to prevent electrons from landing on the anode surface, causing
circulation of the electrons through the space between the anode and the cathode.
When they eventually arrive at the anode, the electrons are localized near the
opening gap of the cavities on one side, inducing an electrical field across the gap
of each cavity. This opening gap acts as a capacitor while the path around the cavity
in the anode body behaves as an inductor, as described by the zoomed box in
Fig. 2.6, creating a resonator for each cavity. As a result, electromagnetic waves
are excited by the cavities, which will interact with the circulating electron beam,
causing bunching of the electrons. The generated electromagnetic waves are
extracted by an antenna tapped into one of the cavities or by an open hole leading
to a waveguide.
From the viewpoint of circulating electrons in the magnetron, they keep passing
by a series of cavity openings, as if they travel along a very long multiple-cavity
klystron [29]. In this view, electrons are “recycled” instead of reaching the collector
after a single trip, which can explain the higher DC-to-RF efficiency typically
exhibited by magnetrons compared to other vacuum devices. The oscillation fre-
quency of magnetrons, which is determined by the cavity dimension and electron
emission energy, is relatively lower than that of other devices described above.
However, there have been efforts to push magnetrons toward THz operation,
including a recent demonstration of a 210-GHz magnetron that utilizes a spatial
harmonic [30]. Around 100 GHz, an output power larger than 10 kW can be
achieved with magnetrons [31]. It is interesting to note that the magnetron is
sometimes mentioned as a supporting example for the assertion that technical
strength may significantly affect the outcome of military conflicts. During World
War II, there were two types of radars competing, one based on magnetrons and the
other on klystrons. There exists a notion that the superior performance of the radars
based on the magnetrons critically affected the outcome in favor of the allies that had
adopted magnetron-based radars. Ironically, afterwards, magnetrons have gradually
faded out from radars mainly due to their persistent frequency instability issues,
giving way to klystrons as well as TWTs as preferred signal sources for radars.
Up to this point, THz sources based on vacuum electronic devices have been
reviewed. Another main branch of electronic sources for THz operation is the
solid-state devices. As mentioned earlier, the main advantages of solid-state THz
2.2 Diode Sources 33
sources are their compact sizes as well as the low DC power dissipation, although
they generate much lower output power than the vacuum sources. Solid-state THz
sources can be based on either diodes or transistors. Diode-based sources will be
discussed in this section, while transistor-based ones will be treated in the following
section. Widely used diode-based sources are Gunn diodes, IMPATT (IMPAct
ionization Transit Time) diodes, and resonant tunneling diodes (RTDs). They
share a common main feature that the oscillation is attained based on the negative
differential resistance (NDR). If NDR is established across a diode, I–V relation will
show a region with a negative slope as is conceptually shown in Fig. 2.7. Then
oscillation can be achieved with a proper termination of the device with a resonant
circuit when biased correctly. The details of the physical principle to obtain NDR
will differ for each diode, though, as will be described in this section.
The Gunn diode is a two-terminal device with NDR originated from the Gunn effect.
Its operation as an oscillator was first reported by Gunn in 1963 [32]. When the
oscillation was first observed, the physical mechanism of the oscillation was not
exactly understood, but Kroemer soon pointed out [33] that the behavior is consis-
tent with a theory that had been described earlier by Ridley and Watkins [34] as well
as Hilsum [35]. Formally known as Ridley-Watkins-Hilsum (RWH) theory, it
describes how the NDR can be induced by the transfer of electrons between two
adjacent energy valleys in the conduction band of III–V semiconductors, as briefly
explained below.
Many III–V compound semiconductors widely employed for electronics, such as
GaAs, InP, and GaN, share a following feature in the band structure. The main valley
in the conduction band is located at the center of the E–k space (Γ point), and is
neighbored by another valley in proximity. This neighboring valley is called the
satellite valley. Its energy level is slightly higher than that of the main valley, as can be
easily seen on the E–k diagram depicted in Fig. 2.8. The satellite valley usually shows
a shallower curvature (larger radius of curvature) near the minimum point than the
34 2 THz Sources and Related Topics
main valley. This indicates a larger effective mass and thus lower mobility for
electrons residing in that valley. When the semiconductor is in the equilibrium,
most electrons remain in the main valley. When it is excited, typically by an external
electric field from the bias voltage, the electrons will gain energy. If the acquired
energy is large enough to exceed the energy separation between the two valleys (ΔE),
most electrons will be transferred to the satellite valley. After the transfer is com-
pleted, the average electron mobility in the semiconductor is degraded, since the
electrons in the satellite valley will have a larger effective mass and thus a lower
mobility than in the main valley. For this whole process to take place, the energy band
structure of a given semiconductor should satisfy following three conditions [35]:
(1) ΔE should be significantly larger than the thermal energy at room temperature kT
(otherwise electrons will enter the satellite valley even without an external electric
field). (2) ΔE should be smaller than the band gap (otherwise the semiconductor will
become highly conductive before the transfer occurs). (3) The satellite valley should
show a lower mobility, larger effective mass, and higher density of states (otherwise
the transfer will not cause mobility degradation). The semiconductors mentioned
above, GaAs, InP, and GaN, satisfy all these conditions, while Si and Ge do not.
Some III–V semiconductors, such as InAs, GaP, and InSb, do not qualify, either [36].
Now let us see what happens if a diode is made of such semiconductors that
satisfy the conditions listed above. When the electric field established inside the
device exceeds a certain critical value, the electron transfer will occur as electrons
will obtain an energy larger than ΔE and the average electron mobility will be
degraded. For an observer external to the device, this will appear as a sudden drop
in the current flowing through the device when the bias across the device is increased
beyond a certain voltage. This means that I–V curve will show a negative slope
beyond this critical bias voltage, indicating that NDR is achieved across the device.
By properly terminating the two electrodes of the device, typically with a resonant
circuit, oscillation is attained. The intrinsic factors such as device dimension and
doping profile, together with the resonant circuit, affect the oscillation frequency,
which can reach a few hundred of GHz or higher.
Figure 2.9a shows the performance of selected Gunn diodes reported in the past
years, in terms of the output power and the oscillation frequency for various
materials and different harmonic outputs (The data points in this figure and other
figures throughout this book, which show the recent performance trend, were
2.2 Diode Sources 35
40 40
Si Fundamental
Gunn Diode Si 2nd Harmonic
20 20 GaAs Fundamental
GaAs 2nd Harmonic
0 0
-20 -20
0 0
-20 -20
-40 -40
-60 -60
TUNNETT Diode RTD
-80 -80
0 100 200 300 400 500 600 700 800 0 400 800 1200 1600 2000
Frequency (GHz) Frequency (GHz)
(c) (d)
Fig. 2.9 Performance of recently reported diode-based signal sources: (a) Gunn diode, (b)
IMPATT diode, (c) TUNNETT diode, (d) RTD
obtained directly from the original sources. Due to the large number, they are not
individually referenced.). It shows that the output power of around 20 dBm can be
achieved for a frequency range of 100–200 GHz range, and the oscillation frequency
reaches nearly 500 GHz with output power of 10 to 15 dBm [37]. As of today,
Gunn diodes are widely adopted for commercial purposes, and packaged products
are available for various power and frequency ranges.
IMPATT (IMPAct ionization Transit Time) diodes attain NDR from time delays
caused by the avalanche multiplication and carrier transit across the device. Unlike
36 2 THz Sources and Related Topics
Fig. 2.10 (a) Structure of a typical IMPATT diode. (b) Applied voltage waveform and current
profiles after avalanche breakdown and externally measured
chain reaction of impact ionization, there will be a finite time delay before a large
number of carriers are generated. Besides, there will be an additional temporal delay
for the generated carriers to transit across the drift region and reach the terminal.
Hence, the total time delay will be composed of the two delay components.
Description above assumes a steady bias applied across the diode. Now let us
consider a more practical case, in which a sinusoidal AC voltage signal is superposed
on top of a DC bias that is very close to the breakdown voltage. During the positive
cycle of the AC voltage, therefore, avalanche multiplication will be triggered as the
total voltage will higher than the threshold for the avalanche process, leading to a
sudden surge in the local hole current. The avalanche current, however, will quickly
decay during the negative cycle that follows. Hence, the local hole current level near
the junction will fluctuate with the same cycle as the applied AC signal, but with a
finite phase lag. The generated holes will subsequently travel through the drift region
before exiting the device at the p+ contact region. The delay due to the drift will
cause an additional phase delay. With a proper design of the drift region, the total
phase delay between the diode current and the applied AC voltage can be tailored to
match 180 , as shown in Fig. 2.10b [41]. In this case, NDR is induced across the
diode terminals, which will open a possibility for oscillation with the diode. It is
noted that the avalanche delay is a material property and thus strongly affected by the
choice of the semiconductor employed for the diode. With a smaller avalanche
delay, the hole generation can respond to a quicker transition of the AC signal,
leading to a higher oscillation frequency. Similarly, a shorter drift region will help to
increase the frequency. It is interesting to note that the avalanche delay is smaller for
Si than GaAs, so that Si IMPATT diodes can operate faster than GaAs IMPATT
diodes.
Instead of attempting to achieve a smaller avalanche delay for faster device
operation, one may improve oscillation frequency by inducing a breakdown from
a totally different mechanism that causes faster carrier generation. This is the idea
behind the TUNNETT (TUNNEling Transit Time) diode, which is a variant of
IMPATT diode [42]. It is well known that there are two mechanisms to cause a
breakdown in a p-n junction: avalanche multiplication and tunneling. For a lower
doping concentration near the junction, the former mechanism dominates, while the
latter becomes dominant when the doping concentration is sufficiently high. For the
generation of carriers to be injected into the drift region, TUNNETT diodes rely on
the tunneling, as opposed to the avalanche multiplication for IMPATT diodes.
Developing a tunneling current takes a shorter time than for the avalanche current,
which results in a faster operation achieved with TUNNETT diodes. It is further
helped by the suppressed dispersion effect that is available with the tunneling
mechanism. Consequently, TUNNETT diodes tend to show a higher operation
frequency and lower noise level than IMPATT diodes.
Figure 2.9b shows the reported performance of IMPATT diodes in terms of the
output power and the oscillation frequency. It shows the output power exceeding
20 dBm for 100–200 GHz range, and the maximum frequency goes up close to
400 GHz [43]. On the other hand, a GaAs TUNNETT diode operating beyond
700 GHz with fundamental mode has been reported as Fig. 2.9c shows [44].
38 2 THz Sources and Related Topics
The output power of TUNNETT falls short of IMPATT diode output power, though.
Also, it rapidly decreases with increasing frequency. Although experimental results
are yet to appear, IMPATT diodes with an oscillation frequency exceeding 1 THz
appears feasible from calculation [45].
RTDs achieve NDR by exploiting discrete quantized energy levels allowed for
electrons confined in a quantum well. While Gunn diodes and IMPATT diodes are
based on a single type of semiconductor, with either bulk or p-n junction, RTDs
involve heterojunctions between semiconductors with different energy gaps to create
quantum wells and barriers. The NDR due to the resonant tunneling through
quantum wells in semiconductors was first observed by researchers in IBM in
1974 [46], soon after a prediction by calculation from the same group [47]. The
operation principle is as follows.
When a diode is composed of a stack of semiconductor layers with different
energy bandgaps, quantum wells will be formed in the conduction band. This is
shown in Fig. 2.11a with a case of a single quantum well bounded by two energy
barriers. Inside the quantum well, discrete energy levels for electrons will be created
as predicted by the quantum theory. As the bias applied across the diode changes, the
energy levels inside the quantum well will shift against the energy levels in the
adjacent semiconductor regions across the barriers. When any of the discrete energy
levels inside the quantum well aligns with the Fermi level of the outside region, a
large current will flow across the barrier due to tunneling. On the other hand, when it
is misaligned, the current will be small. Therefore, as the bias voltage is swept from
low to high, there may exist a local reduction of the current level, which will happen
when the increase in bias tends to further misalign the energy levels in that range.
This will cause a negative differential resistance across the diode, which can be
exploited to generate oscillation if proper bias voltage is applied. The oscillation
frequency is affected by the electron dwell time in the quantum well and the transit
time toward the terminal, while some external components, such as device parasitics,
also affect the device speed. The reduction of these delay times by device optimi-
zation is the key path to RTDs with higher oscillation frequency.
Figure 2.9d shows the accumulated data points recently reported for RTDs. The
oscillation frequency of RTD is known to be the highest among the diode-based
oscillators, penetrating deep into the THz band. The highest oscillation frequency of
RTD reported so far is 1.92 THz [48], while it is expected to further grow in view
of the extensive recent research efforts, notably by the group in Tokyo Institute of
Technology led by Asada. The output power of RTDs, on the other hand, is
relatively smaller than those of Gunn or IMPATT diodes at low frequency range.
However, there is a growing effort to increase the output power with device structure
optimization as well as power combining with multiple RTDs.
2.2 Diode Sources 39
Fig. 2.11 (a) Conceptual energy band diagram of a typical RTD. In the upper case, the current does
not pass through due to misalignment of the energy level in the quantum well. In the bottom case,
current passes through owing to the energy level alignment. (b) Schematic of a typical modern RTD
oscillator: semiconductor layer structure (left) and the entire device structure (right) [48]. (Copyright
(2016) The Japan Society of Applied Physics)
The actual structure of modern THz RTD, which is integrated with a slot antenna
for radiation, is depicted in Fig. 2.11b [48]. It is composed of multiple layers
employing InGaAs and AlAs layers for narrow- and wide-band semiconductors,
respectively, deposited on top of a semi-insulating InP wafer by a molecular beam
epitaxy (MBE). The semiconductor layers are sandwiched by a pair of electrodes,
which directly leads to the antenna. The dwell time and the transit time, which are the
intrinsic delays, can be lowered by reducing the layer width of the RTD quantum
well and the collector space layer shown in the figure, respectively. However,
capacitance increase with the thinned layer may counteract the improvement with
the consequent RC delay increase, which requires a careful optimization of the layer
thicknesses. The extrinsic delay that arises from the parasitic components such as
those from the link to the antenna needs also be reduced for increased oscillation
frequency, which can be achieved through the optimization of device exterior as well
as fabrication.
40 2 THz Sources and Related Topics
The signal sources based on diodes as described in the previous section benefit from
an advantage over the signal sources based on vacuum device in that they are solid-
state components and thus can be realized on a semiconductor substrate. Signal
sources based on transistor circuits, as will be described in this section, provide
further advantages. Most prominently, it allows possibility of integration with other
circuits with different functions, digital circuits inclusive, enabling embedded signal
sources in a monolithically integrated system. The generated signals can be conve-
niently delivered to other neighboring on-chip circuit components. Also, being an
active device, transistors provide much more options for circuit implementation than
passive devices would do, notably the availability of gain, which can help to further
enhance output power after signal generation. The highly developed circuit design
environment also helps efficient design processes. Realizing transistor-based signal
source operating at the THz band, however, is certainly challenging and has long
been limited due to the transistor performance not sufficient for such a high-
frequency band operation. Nevertheless, the barrier is being lowered with the recent
progress in semiconductor process technology for high-frequency operation, leading
to a growing number of transistor-based THz signal sources being reported based on
both III–V semiconductor technology and Si-based technology. This section will
start with a brief introduction to the basics of oscillators, followed by the description
on popular oscillator topologies widely adopted for THz application, such as LC
cross-coupled oscillators, Colpitts oscillators, and ring oscillators. Also, additional
topics of voltage-controlled oscillators (VCOs) and phase-locked loops (PLLs) will
be covered with a focus on their THz operations.
It will be useful to review the basics of oscillator principles from the circuit
perspective before providing the details of the various oscillator topologies. There
can be two views of oscillators: one-port and two-port views. For the one-port view,
as is described in Fig. 2.12a, we assume a passive resonant circuit with a finite loss.
Although oscillation may start inside the stand-alone resonator, it will not be
sustainable as it gradually decays with the finite internal resistance. However,
when the resonator output port is terminated with a network with a negative
resistance, say R1 (with a negative value), the loss of the resonant circuit will be
compensated by the negative resistance if R1 ¼ R2, where R2 is the resistance
looking into the resonator. This will lead to a sustaining oscillation. One way to
obtain the negative resistance is to employ a diode with a special structure as was
discussed in the previous section. However, when transistors are available, the
negative resistance can be obtained by circuit techniques, for which conventional
transistors can be used with no special device-level modification. This is the primary
2.3 Transistor Circuit Sources 41
R1 R2
Network
with Resonator X(s) H(s) Y(s)
negative R
R1 = R2
(a) (b)
Fig. 2.12 (a) One-port view and (b) Two-port view of an oscillator
Y ðsÞ H ðsÞ
¼ : ð2:3Þ
X ðsÞ 1 H ðsÞ
Hence, when H(s) becomes unity at a certain frequency, ωo, the gain across the
closed-loop network will be infinity. This means that any noise component inside the
network that happens to have a frequency of ωo can be picked up, amplified, and
emerge at the output node with a very large amplitude, even without the input signal
externally applied. Hence, oscillation is achieved. Of course, the output amplitude
will not be infinite, as is suppressed by the nonlinearity of the network. As H(s) is a
complex number in general, the condition of H(s) ¼ 1 can be divided into the
magnitude and phase terms, leading to a pair of conditions:
1:jH ðsÞj ¼ 1
ð2:4Þ
2:∠H ðsÞ ¼ 0:
These are called the Barkhausen’s criteria for oscillation. The split of the original
condition into the two conditions provides a useful view of the oscillation.
According to these two conditions, the input signal (which may have originated
from noise) applied to the two-port will appear at the output node without phase and
magnitude change, and then be added to the incoming input signal after passing
through the feedback. The modified input signal, now twice in magnitude, will go
42 2 THz Sources and Related Topics
through the same path and be added to the incoming input signal again, leading to a
tripled magnitude. This will go on and on, and the output signal will keep growing in
its magnitude, until it reaches a stable magnitude as is limited by the circuit
condition. Hence, oscillation is achieved. With this view in mind, consider a slight
deviation from the Barkhausen’s criteria, say, |H(s)| being larger than unity, while
the phase condition remains unchanged. With this modified condition, the two-port
network will still develop an oscillation, presumably at a faster rate. Although this
modified condition is not exactly the Barkhausen’s criteria making Eq. (2.3) infinity,
it will still lead to an oscillation. In fact, this is a preferred condition in most practical
oscillator designs, as it provides a design margin over any uncertainty to guarantee
oscillation start-up.
Now let us consider oscillator topologies widely adopted for high-frequency
applications. Three most popular topologies for this purpose are LC cross-coupled
oscillators, Colpitts oscillators, and ring oscillators.
LC Tank
RLC Tank
X Y Rcore
+ vx -
ix v+ v- ix
M1 M2
ISS M1 M2
ISS
(a) (b)
2.3 Transistor Circuit Sources 43
As depicted in Fig. 2.13b, let us divide the oscillator into two parts: the core and
the LC resonator. According to the one-port view descried earlier, a negative
resistance should be seen when looking into the core from the LC resonator side.
This can be proved with a simple circuit analysis, which is presented here to
emphasize how a simple active network can generate a negative resistance. To
obtain the resistance looking into the oscillator core, Rcore, let us apply a test voltage
vx across the drain nodes of the differential pair. Then, vx can be expressed as
The current induced by the test voltage will be equal to the drain current of M1 and
the negative of the drain current of M2. That is:
Assuming gm1 ¼ gm2 (¼gm), which is valid for most integrated differential pairs, the
test voltage vx can be related to the induced current ix as:
ix i 2i
vx ¼ x ¼ x: ð2:7Þ
gm2 gm1 gm
vx 2
Rcore ¼ ¼ : ð2:8Þ
ix gm
This is a good example of how a negative resistance can be readily achieved with
standard transistors with a simple circuit topology, not relying on novel device-level
structures or phenomena. Note that the analysis given here is a small-signal analysis,
and thus the obtained negative resistance is the small-signal negative resistance,
which is basically equal to the negative differential resistance (NDR) mentioned
earlier by nature. The operation of LC cross-coupled oscillators can be analyzed with
the two-port view as well, treating the circuits as a cascaded two-stage amplifiers
with reactive loads and a feedback loop, which is not elaborated here.
LC cross-coupled oscillators are widely employed for various RF applications
owing to its simple topology and frequency stability especially when compared with
ring oscillators. Based on a differential topology, they are inherently robust against
common-mode noise and well suited for differential systems. If connection with
single-ended systems is needed, however, they may require a balun that converts the
differential output signal to a single-ended signal. Alternatively, the signal from one
output node of the two oscillator differential output nodes can be taken while the
other output node is terminated with a proper load, but with a risk of breaking the
symmetry in the circuit. One downside of LC cross-coupled oscillators (or any LC
oscillators that comprise inductors and capacitors) is the large area occupied by the
on-chip inductors, especially for designs intended for relatively low-frequency
44 2 THz Sources and Related Topics
10 10
Fundamental Si CMOS
2nd Harmonic SiGe HBT
0 3rd Harmonic 0 III/V HBT
4th Harmonic
Output Power (dBm)
-20 -20
-30 -30
-40 -40
-50 -50
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Frequency (GHz) Frequency (GHz)
(a) (b)
Fig. 2.14 Performance of recently reported LC cross-coupled oscillators operating beyond
100 GHz: (a) Plotted for various signal harmonics, (b) Plotted for different technologies
bands. A relief for high-frequency applications, including THz ones, is that the area
issue becomes less stringent with increasing frequency, since the required induc-
tance and thus the inductor size become small. High-frequency operation also allows
chances for transmission lines to replace inductors, which may also help to reduce
the circuit area with the reduced wavelength.
There are still challenges for developing LC cross-coupled oscillators operating at
the THz band. One is the precise control of the small capacitance and inductance
values employed for the LC tank to increase its resonance frequency. The required
smaller component sizes are favored for chip size reduction, but, on the other hand,
even a tiny change in the device dimension may result in a great impact on the
component values. Also, the effect of the parasitic components becomes increasingly
significant as the frequency rises. The limited availability of sufficient gm of tran-
sistors at high frequencies, which is needed to obtain enough negative resistance,
also deters the effort to push the oscillation frequency up to the THz band, although
the recent development of high-frequency semiconductor technologies is gradually
relaxing this constraint. A popular approach of taking the harmonic signals as the
output (will be discussed in Sect. 2.4.1) for elevated frequency will inevitably lower
the output power, which is also one of the challenges for THz operation of the LC
cross-coupled oscillators.
Despite the challenges, there have been a great number of reported LC cross-
coupled oscillators operating in the THz band. Figure 2.14a shows the performance
of recently reported LC cross-coupled oscillators operating near or beyond 100 GHz,
in terms of the output power and the oscillation frequency. A great portion of the
reported oscillators employ harmonic techniques to boost the oscillation frequency
(will be discussed in Sect. 2.4). Also, some data points are for the circuits with power
combining techniques (will be discussed in Sect. 2.5.2). Figure 2.14b presents the
same set of data again, now categorized by the technology they are based on for
2.3 Transistor Circuit Sources 45
C1 X ix
L Y R L L R
i1 i2
C2 VG
X C2
Zin Y C2 C2
M1 ix
C1 C1 C1
effectively further increase the voltage at X as the voltage drop across the load
diminishes, resulting in a positive feedback. This will cause oscillation. One may
argue that a positive feedback can be achieved even without the capacitive division,
questioning the need for the capacitive division. The truth is that, without the
capacitive division (meaning without C2), the node X will see a resistance equivalent
to 1/gm, which is not a very large value in typical cases. This will degrade the Q-
factor of the LC load. Such a situation is obviously not favored and can be avoided
by the adoption of the capacitive division that will transform 1/gm to a larger value,
or ~(1 + C1/C2)2/gm for the configuration in Fig. 2.15a [50]. This is the main idea
behind Colpitts oscillators.
As an alternative way to understand the operation of Colpitts oscillators, let us
take a quantitative view on the oscillator, in which we will obtain the impedance of
the oscillator core seen from the load inductor. For this, the oscillator is redrawn as
Fig. 2.15b, and the impedance is calculated as the ratio of vx and ix. The current i is
the sum of i1 and i2, which are the drain current of M1 and the current flowing across
C2, respectively. Noting that i1 is driven by vgs of M1, which is given by the voltage
across C1 with the sign switched, following relations can be obtained for the current
components:
i
i1 ¼ gm vgs ¼ gm x , ð2:9Þ
C1 s
i
i2 ¼ vx þ vgs C 2 s ¼ vx x C 2 s, ð2:10Þ
C1 s
i i
ix ¼ i1 þ i2 ¼ gm x þ vx x C 2 s: ð2:11Þ
C1 s C1 s
By rearranging the terms and assuming a sinusoidal signal so that s ¼ jω, the
impedance Zin can be obtained:
g C
ix 1þ m þ 2 ¼ jωvx C2 , ð2:12Þ
jωC1 C 1
vx g 1 1 1
Z in ¼ ¼ 2 m j þ : ð2:13Þ
ix ω C1 C2 ω C1 C2
The result clearly shows that a negative resistance is obtained, which is connected in
series with C1 and C2. Hence, by terminating the input with an inductor, which is
actually the case for Colpitts oscillators, the reactive and the (negative) resistive
components will be compensated by the inductance and the parasitic resistance of
the inductor, respectively. This will trigger oscillation.
The circuit considered above is based on the common-gate (or common-base for
bipolar version) topology. However, the common-drain (common-collector) topol-
ogy can also be adopted, for which the capacitive division is made between the gate
2.3 Transistor Circuit Sources 47
(base) and the ground. Also, for each of these topologies, differential versions can be
implemented. Figure 2.15c shows a differential Colpitts topology of the common-
gate case. For the differential cases, a phase difference of 180 needs to be
maintained between the two transistors that form the differential pair, which is
typically achieved through a capacitive or inductive coupling across the two
branches. It is noted that for some Colpitts oscillators, the capacitive divider is
realized by parasitic capacitance of transistors, without employing separate devoted
capacitors. This is especially true for oscillators operating at high-frequency ranges,
where small device capacitance value is sufficient to generate reactance required for
the capacitive division. It is to be mentioned that a variant of the Colpitts oscillator,
in which another capacitance is inserted in series with (sometimes in parallel with)
the load inductor in the basic Colpitts topology, bears a different name, Clapp
oscillator. As proposed by Clapp in 1948, this version of Colpitts oscillator is
known to provide a much better frequency stability [51] with an additional degree
of freedom for design obtained from the newly added capacitance component, and
often adopted for high-frequency applications.
For the designers of high-frequency LC oscillators, one of the first choices they
need to make is the topology of the oscillator, which typically narrows down to
either LC cross-coupled oscillators or Colpitts oscillators. Although there are various
performance aspects to be considered for the decision, the phase noise is usually at
the top of the list. To make the choice difficult, there is no obvious winner in terms of
the phase noise performance. Based on the analysis in [52], LC cross-coupled
oscillators show a superior phase noise performance than differential Colpitts oscil-
lators. This observation has been confirmed by some works that followed, but the
comparison was limited for oscillators based on CMOS technologies only. Because
the way the device-level noise affects the phase noise of the overall oscillator is
different for LC cross-coupled oscillators and Colpitts oscillator (Colpitts with more
noise from device but less conversion to phase noise; LC cross-coupled with less
noise from device but more conversion to phase noise [52]), the preference may turn
out different for different device types employed.
For example, bipolar transistors exhibit much better 1/f noise characteristics than
MOSFETs, mainly because the main current path in (vertical) bipolar transistors
stays away from the interface usually full of traps, while (lateral) MOSFET current
runs just beneath the oxide–semiconductor interface, making it vulnerable to carrier
trap/detrap. Also, bipolar transistors show a much larger transconductance gm than
MOSFETs, a direct result of the exponential relation between the current and base
bias. These device properties certainly make difference in phase noise generation in
oscillators. In fact, in another comparison study for the two types of oscillators,
which was made in simulation based on bipolar transistor technology, it was
reported that differential Colpitts oscillators show better phase noise performance
than LC cross-coupled oscillators [53]. This result is consistent with the fact that
Colpitts oscillators are particularly popular with bipolar transistor technologies. The
relative performance may depend on the oscillation frequency range as well. The
comparison studies mentioned above are for frequencies below 10 GHz. For oper-
ations in high-frequency bands, including the mm-wave and THz bands, there is a
48 2 THz Sources and Related Topics
10 10
Fundamental Si CMOS
5 2nd Harmonic 5 SiGe HBT
3rd Harmonic III/V HBT
0 4th Harmonic 0 III/V HEMT
-5 -5
-10 -10
-15 -15
-20 -20
-25 -25
-30 -30
-35 -35
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
widely spread notion that Colpitts oscillators exhibit better performance in terms of
the phase noise. Despite this apparent complication in the comparison, it is com-
monly accepted that the start-up condition is more stringent for Colpitts oscillators,
partly due to the higher gm needed for Colpitts than LC cross-coupled oscillators.
This has prompted some techniques developed to mitigate this issue [54].
The challenges for realizing THz Colpitts oscillators are similar to those for LC
cross-coupled oscillators as discussed above, since they are both based on LC resona-
tors. Again, these challenges are gradually being addressed, leading to a growing
number of Colpitts oscillators operating beyond 100 GHz. Figure 2.16 shows the
performance of recently reported high-frequency Colpitts oscillators, again on the
output power—oscillation frequency space. The highest oscillation frequency reported
till today is 1 THz, which is from a 42-component Colpitts oscillator array based on
SiGe HBT with a combined output power of 10.9 dBm [55].
The third type of oscillator to be discussed is the ring oscillator. Ring oscillators do
not necessarily contain an LC resonator, a clear contrast to the two oscillator types
discussed so far. Instead, the delay inherent to inverter stages is exploited to generate
the phase shift that is required for the phase inversion in the two-port view of the
oscillator as was discussed above. A typical configuration of a ring oscillator is
shown in Fig. 2.17a. It is composed of a series of inverters cascaded, where the
output of the last stage is fed back to the input of the first, thus forming a loop or a
ring. The intrinsic delay of a single inverter is π (or 180 ). If there is an extra time
2.3 Transistor Circuit Sources 49
+– +– +– +–
–+ –+ –+ –+
(b)
delay of τ per each inverter stage, which is basically a gate delay due to the finite RC
time constant, the total delay around the loop will be equal to n(π + ϕ), where n is the
total number of the inverter stages around the loop and ϕ is the phase shift
corresponding to the time delay τ. If n is an odd integer and nϕ ¼ π, the total
phase shift around the loop will be an integer multiple of 2π. Hence, for every round
trip of a signal around the loop, the signal will be added in-phase and the signal
strength will grow, leading to an oscillation. Based on the relation between ϕ and τ
as follows,
τ
ϕ ¼ 2π , ð2:14Þ
T
where T is the period of the signal, the oscillation frequency is simply given as:
1 ϕ 1 1 π
f0 ¼ ¼ ¼ since ϕ¼ : ð2:15Þ
T 2π τ 2nτ n
-40
-50
0 200 400 600 800 1000 1200
Frequency (GHz)
oscillator, an even number of stages will satisfy the oscillation condition described
above.
Ring oscillators benefit from the small area owing to the absence of the bulky
on-chip inductors that are typically needed for LC oscillators. However, for ring
oscillators, the oscillation frequency is not very precisely predictable (due to the
uncertainty in the gate delay, etc.), and the phase noise is much worse than the
oscillators with LC resonators. When it is injection-locked with a much more stable
signal source, however, those issues are largely resolved, since the oscillation
frequency and the phase noise are now dictated by the signal source that locks the
ring oscillator rather than the ring oscillator itself. For this reason, ring oscillators are
often employed inside a locked circuit, such as phase-locked loops (PLLs) that will
be discussed in Sect. 2.3.6.
Figure 2.18 shows recently obtained performance of ring oscillators along with
those of LC cross-coupled and Colpitts oscillators in comparison. As the plot shows,
the performance of ring oscillators appears to fall slightly behind those of LC
resonator-based oscillators based on the given data points. Remember, though,
they have their own attractive features, such as the small area as just mentioned
above. Figure 2.18 also provides a direct comparison between LC cross-coupled
oscillators and Colpitts oscillators (the data points are based on Figs. 2.14 and 2.16.)
The two types of oscillators are almost on par in their output power performances,
while higher operation frequencies have been achieved with Colpitts oscillators.
Many applications of oscillators benefit from the tuning of the oscillation frequency.
For communication applications, for example, the channel or band selection would
require the tuning capability. For imaging applications, frequency tuning would
2.3 Transistor Circuit Sources 51
allow imaging over different frequencies, which is known as the spectral imaging.
There is a need from a highly practical point of view as well: when the fabricated
oscillator shows an oscillation frequency off the target, it can be still recovered if
frequency tuning is available. Finally, frequency tunability is indispensable for an
oscillator to be employed for PLLs, which will be discussed in more detail in the next
subsection.
Frequency tuning can be realized with various approaches. For oscillators based
on LC resonators, such as LC cross-coupled oscillators or Colpitts oscillators as
introduced earlier, the oscillation
pffiffiffiffiffiffi frequency is determined by the resonant frequency
of the resonator ( f osc / 1= LC ). Hence, the tuning of inductance or capacitance
will lead to frequency tuning. For oscillators driven by the gate delay, such as in ring
oscillators, it is the RC time constant that dictates the oscillation frequency ( fosc / 1/
RC). Therefore, the control over resistance or capacitance values will enable the
frequency tuning. Despite these various options, the tuning of the oscillation fre-
quency of most oscillators is predominantly achieved by capacitance variation.
Hence, a capacitor with variable capacitance, or varactor, is a core component for
oscillators with frequency turnability. Because the varactor tuning is accomplished
by voltage tuning in most cases, such frequency-tunable oscillators are called
voltage-controlled oscillators (VCOs). VCOs can be implemented with most types
of oscillators, including the three topologies discussed previously.
Varactors can be realized by various methods with semiconductor technologies.
The most straightforward approach is to make use of a p-n junction, which is readily
available in conventional semiconductor technologies. It is well known that the
capacitance (per unit area) of a p-n junction is given by ε0εs/xd, where ε0 is the
permittivity of vacuum and εs is the relative permittivity or the dielectric constant of
the semiconductor, and xd is the depletion width of the p-n junction. For one-sided
step junction (where doping concentration difference between the n- and p-region of
the p-n junction is so large that the depletion region is formed mostly in one side of
the junction only), which is the case for many practical p-n junctions inside semi-
conductor devices, the depletion width is given as:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2ε0 εs ðV bi þ V r Þ
xd ¼ , ð2:16Þ
qN
where q is the elementary charge, N is the doping concentration (of the lowly doped
side of the p-n junction), Vbi is the built-in potential of the p-n junction, and Vr is the
reverse bias across the junction. Hence, the capacitance is given as:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ε0 εs qN
C¼ : ð2:17Þ
2ðV bi þ V r Þ
Thus, capacitance varies over Vr and the junction can be used as a varactor. This is
a popular approach to realize a varactor, especially for bipolar transistor technolo-
gies, in which the base–collector junction is typically utilized for this purpose.
52 2 THz Sources and Related Topics
However, a downside of this type of varactor is the small tuning range. Vbi is already
close to 1 V for typical p-n junctions in Si technology and may exceed 1 V for GaAs
or InP technologies, whereas Vr is limited by a maximum voltage allowed for a given
technology. Assuming Vbi ¼ 800 mV and the maximum allowed Vr of 2 V, a simple
calculation reveals that the capacitance tuning range is around 87%. This is the ideal
case and the actual tuning range will be smaller, typically down to around 50%, due
to fixed capacitance components of the device interior and nonuniform doping
profile. Besides, when it is embedded in a circuit, fixed capacitance from other
neighboring circuit elements will additionally limit the circuit tuning range. Also, the
oscillation frequency has a square-root relation with the capacitance for LC oscilla-
pffiffiffiffiffiffi
tors (1= LC ), further suppressing the available frequency tuning of the oscillators.
There is a variant of the p-n junction varactor, which is devised to overcome the
issue and boost the tuning range. Called the hyperabrupt junction varactor [57], it
employs a doping profile tailored for a larger capacitance variation for a given
voltage tuning. To be more specific, it has a retrograded doping concentration profile
(lower doping when farther away from the junction), which causes a faster change in
depletion width (and thus capacitance) with voltage change than the uniform doping
case. As a result, capacitance tuning can be significantly improved, by a factor of 2 or
more when compared to the conventional junction varactors. However, this
approach would require an additional ion implantation step dedicated to the varactor
formation, adding to the total fabrication cost. This is what you need to pay in
exchange of performance improvement. Note that the conventional junction
varactors are basically “free,” as they make use of existing junctions formed for
other purposes.
Widely adopted in Si CMOS technologies is the accumulation-mode MOS
varactor [58, 59]. The capacitance–voltage (C–V ) curve of MOSFETs shows a
wide capacitance variation over the voltage across the gate and the source/drain
(Fig. 2.19a). This is because the width of the depletion region, formed just beneath
the oxide layer, changes with the gate voltage Vg, leading to a variation in the
capacitance. This depletion capacitance (Cd) is in series connection with the oxide
capacitance (Cox), resulting in the total gate capacitance (Ctot ¼ Cd || Cox). As Cd
decreases with increasing Vg, so does Ctot as shown in Fig. 2.19a with a lower curve
(denoted as high frequency). With this C–V curve given, any MOSFET can serve as
a varactor in principle. However, the MOSFET C–V relation shows a frequency
dependence, exhibiting a “snap-back” behavior if the voltage changes slowly or at
low-frequency operation. In this case, the capacitance returns back to the “high” state
(¼ Cox), although the MOSFET is still in the inversion mode with a sufficiently large
Vg. This is due to the junction structure of the conventional MOSFETs: n-type
source/drain regions formed on p-type substrate for NMOSFETs (or p-type source/
drain regions formed on n-type substrate for PMOSFETs). When the voltage change
is slow with this type of device in the inversion mode, the depletion width (and thus
the total space charge in the depletion region) does not respond to the change,
because a sufficient time is allowed for the inversion charges to be generated in
the source/drain region and supplied to the channel, which absorbs the effect of the
2.3 Transistor Circuit Sources 53
Ctot
Vg Cox
GND Cox
Cd(Vg, f) Low frequency
n+ n+ p+
High frequency
p-type substrate
Vg
(a)
Ctot
Vg Cox
GND Cox
Cd(Vg)
p+ p+
p-type substrate
Vg
(b)
Vg Ctot
Cox
GND Cd (Vg) Cox
n+ n+
n-type substrate
Vg
(c)
Fig. 2.19 (a) Structure of a conventional n-type MOSFET and its C–V curve. Note in the C–V
curve that the frequency-dependent split between low- and high-frequency states is bimodal,
without intermediate states. (b) Structure of a p-type accumulation-mode MOS varactor and its
C–V curve. (c) Structure of an n-type accumulation-mode MOS varactor and its C–V curve
voltage change. In view of the definition of capacitance, the ratio of charge variation
over voltage variation, there is no capacitance associated with the depletion region in
this case as the depletion charge is fixed at this low-frequency condition. Hence, Cd
does not contribute to Ctot at low-frequency operation, which leaves Cox the only
capacitance component of Ctot, leading to the high state of the capacitance. In
contrast, in high-frequency operation, the fast change of voltage would not allow
this process, and thus both Cd and Cox contribute to the total capacitance Ctot,
maintaining the low state of capacitance throughout the inversion mode. It is noted
in the C–V curve that the frequency-dependent split between low- and high-
frequency states is bimodal, without intermediate states in between.
54 2 THz Sources and Related Topics
C RS L RS
1 wL 1 1 1
Qvar = Qind = = +
w CRS RS Qtot Qvar Qind
Fig. 2.20 Simple circuit model and Q-factor of a varactor, inductor, and LC tank
40
Lg = 0.3 µm, W = 0.5 µm, Nf = 12
Gate finger
S/D
contact Lg = 0.3 µm, W = 1.0 µm, Nf = 6
Q-factor
Conventional 20
10
S/D
Gate
contact
island
0
10 100
Frequency (GHz)
IGV
(a) (b)
Fig. 2.21 Comparison of the island-gate varactor (IGV) and conventional MOS varactor: (a)
Layout (only the unit device is shown), (b) Measured performance (IGV is for 20 0.3 μm 0.3 μm
device). (© 2009 IEEE [60])
reduced series resistance and significantly improved Q-factor [60, 61]. The study
was extended to build and compare VCOs based on IGVs and the conventional MOS
varactors [62], which showed improved phase noise for the VCO employing IGV, as
expected.
There have been efforts to build VCOs without varactor as well. Various tech-
niques have been reported for such varactor-less VCOs, which are in many cases
based on variable inductors in principle although the details may vary. In an
approach named magnetic tuning [63], a transformer-capacitor tank is employed,
in which one of the two windings of the transformer forms a loop in series with a
capacitor. For this resonator structure, the resonant frequency is determined by the
self and mutual inductances of the transformer and the capacitance of the capacitor,
as well as the ratio of the currents in the two windings of the transformer. Hence, by
controlling the current ratio, a frequency tuning is achieved when it is employed in
an oscillator. The tunable oscillator built in this way is in fact a current controlled
oscillator, or ICO, as opposed to VCO. Another example of the current controlled
oscillator was reported in [64]. In this work, a tunable inductor is realized by
terminating the base of an emitter follower with an RL degeneration network. With
this structure, an effective inductance is obtained when looking into the emitter node
of the transistor, the value of which is controlled by the current level of the emitter
follower. Hence, the block works as a current-tuned inductor. By combining the
tunable inductor with a transformer in the differential topology, an oscillator oper-
ating at 200 GHz was realized with a frequency tuning range of 3.5%. Yet another
study reports a frequency tuning realized by tuning the transconductance. By
properly arranging RLC network with a dependent current source driven by a
transconductance, an inductive impedance can be established across the current
source. The effective inductance depends on the transconductance, and thus an
inductance tuning is achieved by controlling the transconductance, which can be
readily obtained by controlling a transistor bias current. A tunable oscillator was
realized with the tuning component showing a 23.6% tuning near 26 GHz [65].
f ref f0
Charge Loop
PFD
Pump Filter
Frequency
Divider chain
f0
f div = (÷N)
N
phase detector (PD) instead of PFD.) A detailed description of PLLs can be found
elsewhere, and we will here focus on the issues of realizing high-frequency PLLs,
which relates the high-frequency components in PLLs, say VCO and the front stages
of the frequency divider chain.
The key principle of PLL operation is simple: if there occurs a drift in frequency
(and thus the phase) with the embedded oscillator, it is immediately detected by the
PFD and a restoration process is triggered along the loop, which would bring the
frequency back to the original frequency. For this to happen, the oscillator needs to
have a tuning mechanism installed, preferably by voltage adjustment, which requires
the oscillator to be a VCO. Since the output of the VCO is the output of the PLL, the
issues of developing PLLs operating at higher frequency bands, including the THz
band, inherit those for high-frequency VCOs. These issues related to VCOs have
already been discussed above and will not be repeated here. Instead, a review of
frequency dividers operating at high frequencies will be presented below.
Frequency dividers (also called prescalers) operating at low frequencies can be
readily realized with digital logic circuits, for which CMOS latches with current-
mode logic (CML) are widely adopted. The digital CML frequency dividers gener-
ally show a wide locking range, and its operation frequency can be stretched up to
quite high frequencies, nearly 100 GHz as indicated by a recent result [66]. However,
such a frequency extension with digital frequency dividers typically accompanies a
rapid increase in the DC power dissipation. For example, the 94-GHz frequency
divider reported in [66], which is basically a flip-flop made of two CML latches,
consumes 65 mW. This is quite large and comparable to the power consumed by a
VCO in PLLs. The combined overall power would lead to an excessive PLL power
dissipation. Thus, from a practical viewpoint, the frequency dividers for high-
frequency PLLs, especially those located at the front-end of a divider chain, need
to take a different topology suited for higher frequency operation.
A dynamic divider is one of the favored high-frequency solutions. Also called
Miller divider after the name of the inventor [67] (sometimes it is called the
regenerative divider as well, reflecting the regenerative nature of the operation), it
is basically a mixer followed by a low pass filter, the output of which is fed back to
the input (see Fig. 2.23). With this configuration, a signal with a half frequency of the
58 2 THz Sources and Related Topics
M1 M2
B
Vinj M inj
input ( f ¼ fin/2) can be sustained at the output, while another output frequency of the
mixer ( f ¼ 3fin/2) is filtered out by the LPF. Although a higher division ratio can be
attained with an extra component inserted in the loop, a division ratio of 2 is
generally obtained from a dynamic divider as is achieved with the standard structure
shown in Fig. 2.23. The operation frequency of dynamic dividers easily enters the
THz band. With a SiGe HBT technology, a dynamic divider with an operation
frequency beyond 200 GHz has been reported [68], which showed a wide locking
range as well that exceeds 150 GHz. With an InP HBT technology, a dynamic
divider has reached beyond 500 GHz [69]. Although dynamic dividers are preferred
over the digital dividers (sometimes called static dividers) for high-frequency oper-
ation, it still suffers from the power dissipation issue. The 217 GHz operation in [68]
required a DC power dissipation of around 200 mW including the buffer circuits.
From a power dissipation perspective, injection-locked frequency dividers
(ILFDs) are the most attractive divider option for high-frequency applications
[70]. As can be seen from an example circuit in Fig. 2.24 [71], it is an oscillator
with an injection port, through which the input locking signal can be injected. Hence,
it is basically an injection-locked oscillator. When the injection signal frequency is a
multiple harmonic of the free-running frequency of the oscillator, the circuit
2.3 Transistor Circuit Sources 59
ω0 2 I inj
ωL / , ð2:18Þ
2Q π I osc
where ωL is the locking range, ω0 is the oscillation frequency of the ILFD core, Iinj is
the injection current, and Iosc is the oscillation current. Another approach to over-
come the narrow locking range of ILFDs and ensure the locking with the adjacent
VCO in a PLL is to employ the identical topology for the ILFD and the VCO
[75]. With this strategy, the two circuits will be affected in a similar way with
external variations, which may arise from device model, circuit simulation, and
fabrication issues, and so forth, possibly retaining the frequency alignment between
the ILFD and the VCO.
Up to this point, we have observed a variety of issues present in the high-
frequency components to be included in PLLs, i.e. VCOs and frequency dividers.
Despite all these issues and related challenges for these components, researchers
60 2 THz Sources and Related Topics
Fig. 2.25 (a) Block diagram of a CMOS 283-GHz PLL. (b) Chip photo of the fabricated PLL.
(© 2018 IEEE [77])
have found ways to operate PLLs at elevated frequencies including the THz band. A
growing number of PLLs working at the THz band are being reported, including the
one that operates at 560 GHz based on a Si CMOS technology [76]. Another THz
PLL to be introduced here as an example is shown in Fig. 2.25 [77], which is a Si
CMOS PLL operating around 280 GHz. For the VCO, a triple-push topology based
on ring oscillators is employed. The frequency divider chain converts 283 GHz
down to 5.8 MHz with a division ratio of 16,384, in which the first two dividers are
realized with ILFDs while CML static dividers complete the remaining division.
Owing to the availability of the fundamental signal from the embedded triple-push
VCO ( f0), which is one third of the PLL output frequency (3f0), the operation
frequency and the total division ratio for the divider chain is relaxed by taking f0
2.4 Frequency Enhancement Techniques 61
as the input signal frequency to the chain. With a total power dissipation of 114 mW,
about a half coming from VCO and one-fifth from the two ILFDs, the PLL exhibits
an output power of 22.5 dBm.
The technique to extract the nth harmonic signal internal to the oscillator is widely
referred to as n-push technique [78, 79]. The oscillators employing the technique are
called the n-push oscillator. The most widely adopted are the push-push (2-push)
[80, 81] and the triple-push (3-push) techniques [82, 83]. There have been reports on
oscillators with higher order harmonics as well [84]. A brief operation principle of
these techniques is introduced below based on Fig. 2.26. A push-push oscillator is
composed of two oscillator cores as shown in Fig. 2.26a, which are locked with a
180 phase difference. At the symmetric nodes, therefore, the fundamental-mode
signal will be canceled out due to the anti-phase relation, leading to a negligible
fundamental output signal when the output is taken at one of the symmetric nodes. In
contrast, for the second harmonic signals, the phase difference between the two cores
is doubled and becomes 360 , as indicated by the phase diagram in the figure. This
leads to the addition of the two signals at the output of the push-push oscillator. A
similar principle applies to triple-push oscillators as well. They are composed of
three oscillator cores as shown in Fig. 2.26b, each locked with a phase difference of
120 against the adjacent core. At the symmetric nodes, the fundamental mode and
the second harmonic signals vanish by cancelation, while the third harmonic signal
adds up, which is taken as the output signal of the oscillator. For the general case of
n-push oscillators, all the signals with harmonic number less than n (including the
fundamental signal) are eliminated at the symmetric node, while the superposed nth
harmonic signal is taken as the output, as indicated by Fig. 2.26c. In theory, n can be
as large as desired. In practice, however, n cannot be too large, since the signal power
62 2 THz Sources and Related Topics
push-push
fundamental 2nd harmonic
°
°
° °
(a)
triple-push
fundamental 2nd harmonic 3rd harmonic
° °
°
°
°
° °
° °
(b)
n-push
fundamental nth harmonic
°
×
°
°
°
°
° ( − )×
° °
×( − ) ×( − )
(c)
Fig. 2.26 The operation of n-push oscillators. The conceptual diagram is shown together with the
phase of the fundamental mode and the higher-order modes: (a) Push-push oscillator (n ¼ 2), (b)
Triple-push oscillator (n ¼ 3), (c) n-push oscillator
drastically reduces as the harmonic number grows. This is why most n-push oscil-
lators are realized with n ¼ 2 or 3 as mentioned earlier.
For the design of n-push oscillators, it is essential to maintain the symmetry in
both schematic and layout, since any broken symmetry may result in incomplete
suppression of the undesired harmonics. Also critical is to maintain the required
phase difference between the adjacent oscillator cores. For push-push oscillators, the
required phase difference can be achieved with a relative ease, because a differential
configuration naturally shows a phase difference of 180 , as is employed in the
example circuit presented in Fig. 2.27a. In this SiGe HBT push-push oscillator
oscillating at 320 GHz, Q1 and Q2 of the differential pair function as the two
oscillators cores [85]. At any symmetric point of the differential circuit, the funda-
mental signal vanishes and the second harmonic signal adds up. The collector
2.4 Frequency Enhancement Techniques 63
Fig. 2.27 Examples of n-push oscillators: (a) 320-GHz SiGe HBT push-push oscillator (© 2015
IEEE [85]), (b) 280-GHz CMOS triple-push oscillator (© 2015 IEEE [87])
(or drain) node is a popular choice for output signal extraction as is the case for this
example circuit. For triple-push oscillators, a three-stage ring oscillator is widely
employed to generate the third harmonic signal, because a phase difference of 120
between adjacent stages is enforced in this topology [86]. In Fig. 2.27b, an example
of triple-push oscillators based on Colpitts topology is introduced [87]. In this
oscillator operating at 280 GHz, the third harmonic output is taken at the common
drain node of the stages, which is supposed to be symmetric. It can be seen in the
schematic that the fundamental signal, with one-third of the output signal frequency,
is also extracted from the oscillator in addition to the third harmonic signal. It flows
into the frequency divider chain that follows, which would save a divide-by-three
frequency divider. While the fundamental signal exits, the circuit through the buffer
of the third stage, dummy buffers are attached to the first and second stages as well,
to retain the symmetry that is critical for the n-push operation.
64 2 THz Sources and Related Topics
For most of electronic systems, linearity is the highly preferred characteristic as most
of the nonlinear effects are malignant from a general system point of view. There are
some components of exception, however, which benefit from the nonlinearity and
exploit it. Mixers and frequency multipliers are well-known examples of these types
of circuits. The former operates on two inputs and the latter on a single input.
Frequency multipliers are discussed in this section, while mixers will be one of the
topics in Chap. 3.
Probably the simplest nonlinear device is the diode, as its output response is given
as a nonlinear function of the input, typically an exponential function. Hence, if
followed by an appropriate bandpass filter that suppresses the fundamental and
unwanted harmonics, a diode can operate as a frequency multiplier as its nonlinearity
inherently leads to the generation of harmonic signals. In fact, diodes indeed have
long been used as useful solid-state frequency multipliers. In particular, Schottky
barrier diodes (SBDs) have been widely adopted for high-frequency applications,
preferred over p-n diodes, owing to its faster switching time [88–90]. The faster
operation of an SBD can be attributed to the fact that it is a majority-carrier device.
Being a majority-carrier device (typically based on n-type semiconductor with
electrons as the majority carrier), it does not suffer from the long (dis)charging
time that degrades the speed as is found in minority-carrier devices such as p-n
diodes. Although SBDs are not included in the lineup of standard devices offered by
most commercial semiconductor technologies, they can be realized with a mix-and-
match of various layers available from those technologies. This requires customized
layout and in-house device modeling [91], but there have been reports on THz
frequency multipliers based on SBDs built in this way based on both CMOS [92]
and SiGe BiCMOS technologies [93]. They usually employ a single SBD or a pair of
SBDs with an anti-parallel connection as the nonlinear component, followed by a
filter that suppresses the fundamental frequency f0, often based on a λ0/4 open stub.
However, the conversion gain of diode-based multipliers is typically lower than
what is obtained with transistors due to the lack of the device gain. Also, circuit
topology options are rather limited with diodes. This brings our attention back to
transistor circuits for frequency multipliers.
Transistors are also nonlinear devices by nature and thus can be employed for
frequency multipliers. The detailed characteristics may vary depending on whether
they are bipolar or FET type, because the response of the former is characterized by
the exponential relation while that of the latter is generally given by the quadratic
relation. Nonetheless, the nonlinearity will be observed in both cases. It should be
noted, however, that the nonlinearity alone will not lead to frequency multipliers
with acceptable performance. The desired properties include high output power,
small conversion loss, and the suppression of undesired harmonics, which can be
achieved with properly tailored and optimized circuits. The availability of the device
gain and various circuit techniques with transistors will certainly help, leading to
richer options for the realization of frequency multipliers.
2.4 Frequency Enhancement Techniques 65
VDD
VDD
frequency
select frequency
circuit select
circuit
Vout
Vout+ Vout-
Vin+ M1 M2 Vin-
Vin+ Vin-
(a) (b)
In AMCs, the last stage is typically made of a frequency multiplier, often preceded
by a driving amplifier, enabling the final frequency boost beyond the maximum
operation frequency of the amplifiers available. A single-stage higher-order multi-
plier is also possible [94], but typically with a compromise in the performance.
While the examples in Fig. 2.28 takes on the differential configuration, single-ended
multipliers are also a popular option [95]. It is interesting to note that such single-
ended topologies are often found with III–V technologies, while the differential
approaches are more popular with Si-based technologies. This tendency is not
specific to frequency multipliers only but applies to most high-frequency circuit
components in general, partly due to the traditional individual component-based
design practices among III–V community. It can also be attributed to the larger
(lossy) substrate noise of Si circuits that calls for greater needs for common-mode
noise rejection.
There are various other ways to implement frequency multipliers. When a
wideband operation is required, traveling-wave frequency multiplier can be
employed [96, 97]. The structure is based on the artificial transmission line
constructed with transistor parasitic capacitance and line inductance (sometimes
with additional reactive components), which is a hallmark feature of the traveling-
wave amplifier (also known as the distributed amplifier). This enables a wideband
characteristic for frequency multipliers as is the case for traveling-wave amplifiers.
There are various ways to apply the artificial transmission line for the frequency
multipliers. Input signal can be injected at both ends of the line while the output is
extracted from each transistor stage and then combined [96], or input signal is
applied at one end while the output is extracted from the other end [97]. Another
way to implement a frequency multiplier is to simply employ a mixer and apply the
input signal to both RF and LO nodes. With this configuration, the second harmonic
signal will appear at the mixer output, rendering the circuit to work as a frequency
doubler. While the operation is straightforward, the saturation by the DC offset may
affect the performance with this approach. When frequency multiplication at high
power level is required, a grid frequency multiplier based on the quasi-optical
method (to be discussed in the next section) can be used. A frequency doubler
with a grid structure operating at 1 THz was reported with an input power of 47 W at
500 GHz, generating an output power of 24 mW in a pulse mode operation [98].
There are also frequency multiplier types that do not rely on the nonlinearity or
mixing, one popular example being the injection-locked frequency multiplier
(ILFM) [99]. Its operation is similar to the injection-locked frequency dividers
(ILFDs) discussed above. The only difference is that, for ILFMs, the output fre-
quency is the integer-multiple of the injected input frequency (sub-harmonic injec-
tion), whereas, for ILFDs, the output frequency is the integer-division of the input
frequency (super-harmonic injection). Therefore, in ILFMs, subharmonic injection
signal triggers oscillation near (but not necessarily exact to) the free-running fre-
quency of the core oscillator, which is taken as the output signal. For this purpose,
the n-push topology is nicely suited as the core oscillator, with the inject signal
corresponding to the fundamental signal and the output signal to its nth harmonic.
For ILFMs, the output power is determined by the bias condition of the circuit rather
2.5 Power Enhancement Techniques 67
3 6
-20 3 4
27
10
-30
0 200 400 600 800 1000 1200 1400
Frequency (GHz)
that the input power level as is often the case for conventional frequency multipliers,
which raises chances for high output signal power. Also, a better harmonic suppres-
sion is expected for ILFMs, since a small injection signal can trigger the oscillation
in the circuit.
A major challenge for THz frequency multiplier is to ensure sufficient output
power, which in turn requires both low conversion loss and the availability of high
input power level for most multipliers (note that input frequency is also high for THz
multipliers). Filtering undesired harmonics at the high frequency will be also more
difficult than with lower frequency cases. Figure 2.29 shows the reported perfor-
mance of frequency multipliers operating beyond 100 GHz in terms of the output
power and frequency with different technologies. The numbers beside each symbol
indicate the frequency multiplication factor (data points without numbers are for
frequency doublers (2)). As can be seen from the plot, operation beyond 1.3 THz
has been reached, which was obtained by a cascade of a 5 and 2 Si CMOS
multipliers [100]. Output power over 1 mW (0 dBm) has also been obtained beyond
600 GHz with a 36 InP HEMT multiplier chain [101]. As for the noise, it is
generally known that frequency multipliers will degrade (increase) the phase noise
by 20logN dB for the ideal cases, where N is the multiplication factor. The reported
phase noise degradations with frequency multipliers operating near the THz band are
quite close the ideal value [102, 103], indicating the dominance of the input phase
noise in usual cases.
Although THz signal generation with electrical means became feasible owing to
various techniques developed as discussed earlier, the output power level from a
stand-alone signal source is not large enough. As many of potential THz applications
68 2 THz Sources and Related Topics
prefer high output power, the lack of high-power sources is one of the major
bottlenecks for the THz applications. From this viewpoint, the development of
effective power enhancement techniques for THz signals is as critical as that of
individual signal sources. Two power enhancement techniques can be considered:
amplification and power combining. Each of these will be discussed in this section.
Amplification is the most effective way to boost the signal power and amplifiers are
an indispensable component in most electrical systems. The amplifier type that is
used to amplify and maximize the output power of a system is loosely called the
power amplifier (PA). At low-frequency bands, various types of amplifiers have
been developed to meet the key performance specs of power amplifiers such as
output power, efficiency, as well as gain. In particular, achieving a high efficiency is
critical for the increasingly popular mobile applications, for which the envelop
tracking technique or Doherty amplifiers are currently widely employed. In the
THz regime, however, amplifiers barely obtain a meaningful gain, since the opera-
tion frequency is close to the maximum operation frequency of the devices
employed, typically dictated by the maximum oscillation frequency fmax, or the
frequency on which device power gain reduces to unity. As such, the general
strategy for THz amplifier development is to ensure the required gain first, and
then meet the required output power, optionally with power combining if needed.
Efficiency is of less priority for amplifiers operating at the THz band yet, and a high-
efficiency THz amplifier would be regarded “luxury” at this stage of development.
For the same reason, THz amplifiers reported so far are exclusively linear amplifiers.
So how to ensure a gain at THz? Obviously, the best strategy is to have access to
high performance process technology. Probably the biggest factor that delayed the
birth of THz amplifiers was the lack of a proper semiconductor technology that
would support the gain required at the THz band. This rather fundamental issue was
first addressed by III–V compound semiconductor technologies such as HEMT
(high-electron mobility transistor) or HBT (heterojunction bipolar transistor) tech-
nologies, now both boasting fmax in excess of 1 THz [104, 105], which allows a
sufficient gain deep into the THz regime. There have been impressive improvements
in Si technologies as well, with CMOS and SiGe HBT technologies exhibiting fmax
of close to 500 GHz [106] and over 700 GHz [107], respectively. Owing to these
developments in the technology side, the designers now have a privilege to choose
their favorite from the lineup of technologies available for the THz amplifier design
depending on the application needs.
When the technology selection is done (hopefully affordable to the designers), the
next is to choose the amplifier topology. Three basic options are widely used for THz
amplifiers: common-emitter (CE) (or common-source (CS)), common-base (CB)
(or common-gate (CG)), and cascode. While each topology has pros and cons from
various aspects and needs, the focus will be maintained on the gain below. For
2.5 Power Enhancement Techniques 69
60 60
VGS = 0.7 V VBE = 0.9 V
50 VDS = 1 V 50 VCE = 1.2 V
40 40
(d B )
B)
MSG/MAG (dB)
dB
G ((d
30 30
MAG/MSG
MSAG
20 20
G//M
MASG
10 10
M
0 0
-10 -10
10 100 1000 10 100 1000
Frequency (GHz) Frequency (GHz)
(a) (b)
60 60
VGS = -0.2 V VBE = 0.8 V
50 VDS = 0.9 V 50 VCE = 1.5 V
40 40
SG/MAG (d(dB)
/MASGG(d(dBB) )
B)
30 30
MMAG/MSG
20 20
MMSAGG/M
10 10
0 0
-10 -10
10 100 1000 10 100 1000
Frequency (GHz) Frequency (GHz)
(c) (d)
Fig. 2.30 MAG/MSG of various configurations for four different technologies based on device
model: (a) 28-nm CMOS (AG ¼ 28 nm 1 μm 15), (b) 130-nm SiGe HBT
(AE ¼ 130 nm 1 μm 3), (c) 80-nm InP HEMT (AG ¼ 80 nm 20 μm 2), (d) 250-nm InP
HBT (AE ¼ 250 nm 1 μm 5)
largest gain, which is well expected considering its two-stage configuration. For
bipolar transistors, CB tends to show a larger gain than CE, while the opposite is
exhibited by FETs (CS shows a larger gain than CG). Still, most of the devices show
a gain well beyond 100 GHz, enabling their application to THz amplifiers. It would
be worthwhile to note that there is a stability issue with bipolar transistors in CB
configuration, partly related to its high gain. The issue is aggravated by the uncer-
tainty in the base inductance value in the device model, which has emerged as a
major concern and prompted some efforts to mitigate the effect [108, 109].
Can we get a gain higher than MAG/MSG? Yes, we can, but with an extra effort
in the circuit design. Consider the unilateral power gain U that is higher than
MAG/MSG below fmax, which offers an opportunity to attain a gain higher than
MAG/MSG. The readers are referred to the definition of U, which is the maximum
gain available for a two-port when the two-port is unilateral or S12 ¼ 0. Hence, if an
amplifier is modified to be unilateral, or “unilateralized” (sometimes called “neu-
tralized”), the gain may reach U for an ideal case (no loss, perfect input/output
match), which exceeds MAG/MSG. For transistor amplifiers, this can be effectively
achieved by canceling out the gate-drain capacitance Cgd for FETs or base-collector
capacitance Cbc for bipolar transistors by adding extra components, as those capac-
itances are the main bilateral path between input and output. However, in most cases,
the difference between U and MAG/MSG is not significant near fmax, as U and
MAG/MSG eventually merges at fmax (this merge at zero gain point allows fmax to be
defined based on either MAG/MSG or U ). Therefore, the extra loss caused by the
extra network may negate the extra gain attained with the unilateralization near fmax.
It is interesting to note, however, that U is the maximum gain that can be obtained
from a given two-port network assuming internal extra components can be added to
the network besides the input/output matching (note that MAG/MSG is the maxi-
mum gain achieved with optimum input/output matching only that are external to the
network). An extension of this idea was proposed in [110]. By embedding an
amplifier inside a carefully tailored linear lossless passive network, the amplifier
can achieve a gain higher than U. This new upper limit, dubbed the maximum
achievable gain or Gmax, can be as high as four times the value of U in the optimal
case, leading to a potential gain boost up to 6 dB. This is a highly attractive option for
THz amplifiers that are always hungry for gain and has been successfully applied to
amplifiers operating beyond 200 GHz [111, 112].
There are other factors to be discussed for THz amplifier designs. For each of the
three basic topologies mentioned earlier, both single-ended and differential config-
urations can be implemented. Besides the well-known advantage of common-mode
noise rejection, differential topology also provides a highly valuable virtual ground.
This is especially critical for THz applications, since the path to the external ground
always causes extra inductance, which becomes increasingly problematic with
increasing frequency as the reactance linearly grows with frequency (X ¼ jωL).
The availability of the virtual ground also obviates the need for bypass capacitors,
which are necessary components for single-ended circuits but with known issues at
high frequencies. The large capacitance required for bypassing typically tends to
involve a low self-resonance frequency (SRF). For example, bypass capacitors of
2.5 Power Enhancement Techniques 71
around 1 pF, a typical value for bypassing in high-frequency circuits, would result in
SRFs lower than 100 GHz with conventional Si technologies [113]. This implies that
circuits operating beyond this frequency will not have suitable RF bypassing
because the capacitor converts to an inductor at the operation frequency. This calls
for smaller bypass capacitors with higher SRF, which in turn leads to insufficient
bypassing. For these reasons, differential amplifiers are highly favored in many
applications, especially with Si technologies as briefly mentioned earlier. On the
other hand, single-ended structures have their own strengths, too. For example, they
benefit from its simpler layout and small size as well as lower power dissipation.
Also, the compatibility with single-ended on-chip antennas and test interfaces helps
in many practical situations, for which differential amplifiers would need extra
baluns.
Up to this point, the issues for THz amplifiers were discussed mostly in relation to
the circuit topologies. There are other aspects that need to be addressed for THz
amplifiers. One major challenge for successful development of amplifiers operating
at this high frequency is the availability of the accurate device model, both active and
passive. The accuracy of the device models provided by the foundry is not
guaranteed for the high frequencies that belong to the THz band. For this reason,
in-house modeling is often carried out to reduce the risk caused by possible model
mismatch. A typical RF model for active devices is composed of two parts: core
model for the intrinsic and parasitic components for the extrinsic part of the device.
While the core model is seldom touched by the circuit designers, improvements can
be still made by the users based on simple DC measurements. Modification is often
made for the parasitic components that are typically composed of series resistors for
electrodes (with occasional inclusion of series inductors), inter-electrode capacitors,
and a substrate network to account for the intra-device substrate coupling. The
component values can be modified, with an optional change in the parasitic network
topology itself, in a way to minimize the model-measurement discrepancy over a
wide range of bias, which is typically based on repeated RF measurements. If the
scalability of the original model is retained in such modification, it will be even more
useful in the actual design.
The accuracy of the passive device model is as critical as the active device model.
One relief for passive device modeling (and modification) comes from the availabil-
ity of EM (electromagnetic) simulators. The results from EM simulation can be used
to modify or develop passive device models. Alternatively, it can be directly used in
circuit simulation bypassing the modeling step. To assure accurate EM simulation,
however, it is critical to have precise dimension and material parameters of the
technology. They are routinely provided by the foundry, but it is still desirable to
check its accuracy with actual measurements with calibration. It is also a good
practice to cross-check the results from multiple EM simulators for further verifica-
tion. These issues and solutions related to the device models are in fact generic and
they can be applied not only to amplifier design but any type of THz circuit designs.
Figure 2.31 shows the recently reported performance of THz amplifiers in terms
of the gain versus frequency. As is indicated by the plot, an amplifier operating
beyond 1 THz has been recently achieved based on a 25-nm InP HEMT technology,
72 2 THz Sources and Related Topics
20
Gain (dB)
15
10
0
0 200 400 600 800 1000 1200
Frequency (GHz)
Fig. 2.32 One-THz InP HEMT amplifier: (a) Chip photo, (b) S-parameters. (© 2015 IEEE [105])
-60
0 200 400 600 800 1000 1200 1400
Frequency (GHz)
-50
-75
0 400 800 1200 1600 2000
Frequency (GHz)
A comparison is made in Fig. 2.33 for the output power obtained from THz
amplifiers and that of other signal sources based on transistor circuits such as
oscillators and frequency multipliers. As is obvious from the plot, the output
power with the amplifiers surpasses those of other circuits, which supports the
idea of employing power amplifiers as a power booster as is typically the case for
lower-frequency operations. Figure 2.34 provides a further extended comparison, in
which the output power of transistor circuit-based signal sources and that of diode-
based sources. It indicates a similar level of output power and frequency range for
both types of signal sources, but the transistor-based approach still benefits from the
essential advantage of being based on standard semiconductor technologies and able
to be integrated with other parts of electronics. Finally, a grand global comparison
between vacuum devices and solid-state devices, including both diodes and transis-
tors, is made in Fig. 2.35. It is clear that the vacuum devices provide much greater
74 2 THz Sources and Related Topics
9
120 10
Gyrotrons FELs
7
100 10
5
80 10
3
60
Transistors
1
40 Klystrons
Diodes 10
Vaccum devices
-1
20 10
TWT
BWO -3
0 10
-5
–20 10
-7
–40 10
Transistors
Diodes
-9
–60 10
-11
–80 10
100 1000 10000
Frequency (GHz)
Fig. 2.35 Performance comparison of vacuum devices and solid-state signal sources, including
both diode- and transistor-based ones. The data points include all the harmonic numbers, circuit
types, and process technologies (Si CMOS, SiGe HBT, III–V HEMT, III–V HBT) for transistor
circuits. The data points for vacuum devices are from [7]
output power than what is available with solid-state signal sources, while the
compact form factor and lower DC power dissipation are still attractive features
for solid-state sources. It should be stated that these two types of signal sources are
not competing for performance, but rather they have their own optimal application
fields.
There has been a strong thrust to develop various types of power combiners aiming
at achieving output power larger than the saturation power of individual (power)
amplifiers. At the same time, another drive to develop power combiners has been
efforts to achieve output power larger than the power of individual oscillators that
operate at frequencies beyond the operation frequency of amplifiers. From the
implementation point of view, there is little difference between the two needs, and
the discussion in this subsection may apply to both cases. Also, as most of the power
combiners are reciprocal, they can be used as dividers, too. This property is very
useful for amplifier power combining since the input power is initially divided into
multiple parallel amplifiers and then combined, for which symmetry is highly
2.5 Power Enhancement Techniques 75
desired. In this section, the discussion will be mostly made from the power com-
biners’ viewpoint, but it can be applied to power dividers as well.
There have been a large number of power-combining techniques developed for
microwave and millimeter ranges [120, 121], most of which can still be adopted to
the THz band. The techniques can be grouped into many categories, based on
various criteria. There are two-way (binary) combining and N-way combining,
depending on how many input channels are combined in a single step. When
preferred, multiple two-way combiners can be further combined to realize a large
number of combining, in which case two configurations can be considered: corporate
(tree) type and chain (serial) type. In the corporate type combining, two channels are
added in a repeated fashion, whereas, in the chain type, the power from each channel
is sequentially added to the main path of the outgoing power. The combining
techniques can also be grouped into resonant type vs. non-resonant type. In the
resonant type, the power combining takes place in a resonant cavity, where the input
ports are aligned with the resonant modes of the cavity to maximize the combining
efficiency, while the non-resonant type does not involve such a cavity structure. Yet
another way of grouping is to divide them into planar power combining and spatial
power combining, depending on where the combining physically happens. Discus-
sion below will follow this categorization.
Planar power combining literally refers to the techniques in which powers are
combined on a planar structure, typically on the surface of a semiconductor chip
(widely called “on-chip power combining”) or other types of substrate. There are
several different implementation approaches for planar combining [122–125], which
will be reviewed below.
Direct shunt power combining is the simplest way of combining powers, which
can be applied to either two-way or N-way combining cases (Fig. 2.36a). Since the
current from each channel is combined, it is also called direct current combining.
Although easily implemented, it suffers from a poor isolation between channels.
Also, the matching will be increasingly difficult as the number of combined channels
increases. This is because the parallel combination of the channels would require a
larger output impedance per each channel to match the fixed load impedance, which
contrasts with the low output impedance preferred for power amplifiers to boost the
output power.
Another widely employed planar-type power combiner, particularly for Si-based
high-frequency circuits, is transformer-based power combining (Fig. 2.36b). A
conventional transformer is composed of two interleaved solenoid coils, which
would require a three-dimensional structure. In this sense, realizing a planar com-
biner based on transformers would be a challenge. However, as the planar on-chip
inductors became a standard component for RF circuits with extensive development,
planar on-chip transformers also became readily available and widely adopted
[126]. Transformer-based power combiners overall show good isolation and
76 2 THz Sources and Related Topics
2-way N-way
(a)
1:n 1:n
RLoad vLoad
1:n 1:n
Voltage-combining Current-combining
(b)
Port 3 Port 1
∑
,
Port 1
D
Port 3 Port 4 Port 2
Port 2
(c) (d)
,
Port 1 Port 2
(0º) (270º) 0 j 1 0
–1 j 0 0 1
, , [S] =
2 1 0 0 j
0 1 j 0
Port 4 Port 3
(Isolated) (180º)
,
(e)
Fig. 2.36 Various power combining techniques: (a) Direct shunt power combining, (b) Trans-
former-based power combining, (c) Wilkinson power combiner, (d) Rat-race coupler, (e) Branch-
line coupler (shown with transfer matrix)
2.5 Power Enhancement Techniques 77
wideband characteristics. They also benefit from the inherent properties of trans-
formers, such as the impedance transformation and transition between single-ended
and differential nodes. There are two possible options for transformer-based power
combiners: voltage combiners and current combiners. The transformer-based volt-
age combiners add the voltage from each channel through a series connection of the
transformers of the input channels. In this case, the load impedance can be effec-
tively divided over the number of channels (assuming 1:1 turn ratio) for matching,
leading to a small required output impedance for each channel, a preferred situation
for power amplifiers. Of course, such impedance division can be further controlled
by employing different transformer turn ratio. For the transformer-based current
combiners, current from each channel is added through a parallel connection of the
transformers. This configuration may suffer from a similar issue encountered with
the direct shunt power combining as discussed above, a large output impedance
required per each channel. However, with the presence of transformers, this problem
can be elegantly mitigated by adopting a turn ratio larger than 1 (1:N turn ratio).
Further, the current-based combining benefits from the symmetry in the channels, a
highly preferred property to avoid amplitude and phase mismatch between the
channels [122]. Yet another type of transformer-based combiner is the distributed
active transformer (DAT) [127], which provides an alternative option for power
combining with high efficiency.
Wilkinson power combiner is also a widely adopted planar-type combiner. It can
be realized as both two-way and N-way configurations, although the two-way
version is by far more widely used. A two-way Wilkinson power combiner is
composed of two merged λ/4 transmission lines with an isolation resistor inserted
between the two lines (Fig. 2.36c).
pffiffiffi The matching is achieved when the characteristic
impedance of the λ/4 lines is 2Z 0 and the resistance of the bridging resistor is 2Z0,
with its input and output terminated with Z0. The advantages of Wilkinson power
combiner include the excellent isolation and ensured matching for all the channels,
the major reasons for its popularity. However, as the matching is achieved at a single
frequency, it is inherently a narrow band combiner. Also, its large area occupied by
the combiner is a downside for on-chip applications, as λ/4 line is the main
component constituting the entire structure. This dimension issue, though, is
expected to be gradually relaxed with increasing frequency, a favorable trend for
THz applications. For example, a λ/4 transmission line stretches over 380 μm at
100 GHz when it is immersed in silicon dioxide. This may be bulky, but the number
reduces down to 126 μm at 300 GHz and further down to 38 μm at 1 THz. We can
manage this range of dimension in most applications.
Other more traditional planar microwave structures can also be used as power
combiners, such as rat-race couplers and branch-line couplers. A rat-race coupler, as
shown in Fig. 2.36d, is composed of two input ports (Ports 2 and 3) and two output
ports (Ports 1 and 4). When two input signals with 180 phase difference enter the
coupler, the sum and difference can be taken at Port 1 and Port 4, respectively. This
is an ideal case for combining two differential signals if the output is taken at the sum
port. Also, this configuration provides an opportunity for the coupler to serve as a
78 2 THz Sources and Related Topics
balun, since the input is differential while the output is single-ended. A branch-line
coupler (also called quadrature hybrid coupler) is composed of two input ports
(Ports 2 and 3) and one output port (Port 1), the remaining port (Port 4) serving as the
isolation port (Fig. 2.36e). The sum of two inputs with 90 phase difference will
appear at the output node, leading to power combining. Of course, the opposite
operation is possible, making the coupler to be employed as a power divider (Port
1 as input and Ports 2 and 3 as output). In this case, the branch-line coupler will serve
as a quadrature signal generator, a valued component for systems that need quadra-
ture signals, such as quadrature modulation communication systems.
Both of these couplers were traditionally built with microstrip lines for off-chip
planar circuits. However, increased operation frequency has enabled the on-chip
integration of these structures, benefiting from the smaller size with higher fre-
quency. It is still under intense debate whether it is the best way to realize the
bulky passive components (including antenna) on the chip, as its “real-estate” value
is tremendous. The high area cost largely stems from the complex frontend-of-the-
line (FEOL) processes intended for the construction of active devices, which are not
much utilized for most (if not all) passive devices that only require backend-of-the-
line (BEOL) elements such as metal lines and dielectric layers. It is noted that there
are options for size reduction with these couplers. For rat-race couplers, a meander-
line structure can be employed with a significantly reduced area [128]. This approach
also lifts the main obstacle for rat-race couplers to be realized in commercial Si
processes, say, the design rule that prohibits circular shapes. For branch-line cou-
plers, the capacitive open-stub loading technique can be employed for size reduction,
which effectively shrinks the coupler dimension with the help of a series of open
stubs placed along the line edges [129].
While the techniques described so far can be considered rather classical, there
have been high-frequency signal sources operating at the THz band based on novel
techniques. One notable approach is the distributed active radiator (DAR)
[130, 131], which is briefly reviewed here. It is basically four cross-coupled differ-
ential pairs mutually injection-locked through a pair of concentric signal loops,
which separately link the drains of each pair of the core transistors (see Fig. 2.37).
The two loops are ingeniously cross-coupled, so that the second harmonic signals
traveling along the two loops are in-phase, while the fundamental signals traveling
along the same are out-of-phase. Hence, the two second harmonics add up and the
two fundamentals cancel each other as they radiate from the pair of loops, which
serve as the antenna. This allows an effective suppression of the fundamental signal
without resorting to a filter, which is typically lossy, as is practiced in many
conventional approaches. Additionally, an aperture opening was made for the
ground plane just beneath the loops, while retaining only radial paths between
differential pair source nodes and outer ground plane. This configuration prevents
possible radiation by the second harmonic return current on the ground that might
cancel out the main radiation because of the opposite direction of circulation if
exists. This approach, implemented in a Si CMOS technology as a 2 2 array,
resulted in a total output power of 80 μW around 290 GHz.
2.5 Power Enhancement Techniques 79
Radiating
Region
for 2fo
fo
fo h
r2
n at
tio p
ra rn
pa etu
@2fo
se R
@2fo
@fo
@2fo
The planar power combining techniques discussed so far combine powers in the
form of voltage or current on a planar surface. In contrast, the spatial power
combining techniques to be discussed below combine powers in the form of
electromagnetic waves in a guided or open space [132]. There are various advan-
tages expected with the spatial power combining as follows. Planar combiners
usually combine powers from multiple components (or channels) arranged in
one-dimension as the power propagates on the surface along the signal lines. On
the other hand, spatial combining works on a two-dimension array of components as
the power often (not always) propagates perpendicular to the surface after radiation.
This configuration allows spatial power combining to apply on a massive number of
components, counting up to tens or even hundreds. In this way, much higher output
power can be achieved than with planar combining. Also, as the powers from the
array components directly combine without any branched corporate structure that
involves accumulated loss from each stage (for corporate structures, widely
employed for multi-stage direct shunt power combining, the total loss is L log2N,
where N is the total number of devices combined and L is the loss per stage), a higher
combining efficiency can be expected with spatial combining. Additionally, a high
directivity can be obtained for the radiated power when the phase is well aligned
over the array components, a preferred property for many practical applications.
80 2 THz Sources and Related Topics
It is noted that EIRP rapidly increases with the number of elements by a factor N2,
which results from the fact that both total radiated power and directivity increase by a
factor of N. The definition of the antenna gain and a formal treatment of EIRP can be
found in Sect. 4.2. Finally, spatial power combining can be extended to beam
steering capability by properly incorporating phase shift components, which will
be detailed in the next section. On the downside, the bulky size of spatial power
combining structure may be an issue for some applications, as is imposed by the
antenna size and the minimum required spacing between components of around λ/2.
In a sense, however, this limitation may be an opportunity for THz applications, with
reduced dimension for increasing frequency.
There are various ways to implement the spatial power combining. They can be
largely divided into two different cases, depending on whether the power combining
occurs in the open space or inside metallic waveguide enclosures as depicted in
Fig. 2.38 (adapted from [132]). For the open space case, the input signal is incident
on the array of amplifying components directly or through a lens without any
confining structure. The output is radiated in a similar way. This approach is also
called the quasi-optic power combining. Although Fig. 2.38 assumes there exists an
incoming input power, there are cases where signals are generated from an on-chip
array and do not require input signal. In these cases, the array components will be
signal sources instead of amplifiers. Although the combining in the open space does
not require a complex packaging structure, it may suffer from focusing error,
diffraction loss, and, more importantly, poor heat sink. These issues can be
addressed by confining the power combining inside a metallic enclosure. In an
example shown in Fig. 2.38b, the input signal is fed through a waveguide and then
combined inside a metallic structure, which has a larger cross section than the
waveguide to accommodate the two-dimensional array of components. After com-
bined, the output power exits through the waveguide on the opposite side. The
widely used two-way or N-way waveguide power combiners, which do not need a
separate structure for power combining unlike the one described above, also belong
to this category of combiners. In fact, there exist many different configurations when
it comes to the details in actual realization [119, 133].
The THz application of the spatial combining has so far limited to arrays with a
relatively small number of components [94, 134]. However, a massive spatial
combining may appear with the advantage of small size required in the high-
frequency range.
2.6 Beamforming Techniques 81
(a)
Input Output
Waveguide Waveguide
(b)
As discussed so far, THz sources persistently suffer from the low output power level,
which would hinder them from many interesting THz applications. Once radiated, it
will be further suppressed by the free-space path loss (FSPL), which increases with
the square of frequency, as well as the large attenuation in Earth atmosphere near the
THz band. The two methods discussed in the previous section, amplification and
power combining, help to enhance the power level as we observed. Yet another way
to boost the THz power is the beamforming. In this case, the boost is directional and
applies to a specific direction of radiation with the total radiated power unchanged.
This is different from the two techniques described above and hence discussed
separately in this section.
Beamforming can be understood as a special case of the spatial power combining,
which effectively increases the power delivered to a certain direction by improving
the directivity in a controlled way. This definition, based on the viewpoint from the
signal generation, can be reciprocally applied to the signal detection for increased
directional sensitivity as well, but the discussion in this section will be made from the
signal generation perspective. We have already observed in the previous section that
a spatial power combining from an array of N elements with aligned phase will result
in N-factor increase in total power and N2-factor increase in the power delivered to a
certain direction compared to a single-element case. Hence, the “formation of a
beam” can be achieved with a simple spatial combining with an N-element array in
principle. However, the “beamforming” with a more conventional meaning implies
an additional important feature: the ability to control the direction of (or steer) the
beam by manipulating the phase delay for each element of the array. Typically
82 2 THz Sources and Related Topics
Beamforming can be understood with the help of Fig. 2.39 depicting various cases of
signal radiation. We can start with the single-element case, shown in Fig. 2.39a. Any
antenna has a certain degree of directivity although not very high. This is assumed in
the figure, showing a beam pattern gently pushed toward the direction of radiation.
Next, consider the case of an array of composed of N elements without any
individual phase control, leading to all elements radiating the signals at the same
phase. In this case, as is indicated in Fig. 2.39b, the beam pattern becomes sharper as
the directivity is increased by a factor of N. Hence, the beam direction will receive
much stronger power from the array than the single-element case. However, the
beam direction is fixed perpendicular to the array plane, and there is no control over
for the beam direction yet. Now consider the case shown in Fig. 2.39c, where a phase
control component, or phase shifter, is placed on every signal path so that an extra
phase can be added to each element. If the extra phase is designed so that it
progressively increases along the array of elements from one end to the other, the
direction of the beam pattern will not be perpendicular to the array front anymore,
but it will show some angle off the center. If the progressive extra phase increase is
more rapid, the angle will become larger. In this way, the beam direction can be
− N-1
= = = ·· · · = − − =
Fig. 2.39 Beam patterns from various source configurations: (a) Single source, (b) N-element array
without phase control, (c) N-element array with phase control
2.6 Beamforming Techniques 83
2Δ
controlled by precisely adjusting the phase shifts in the array elements. Such a
system is called the phased array.
An analytical description can be made with Fig. 2.40, which provides a more
detailed look into the phased array. In this N-element array, the initial distance
between the antenna and the single shared oscillator is assumed to be identical for
all the elements. Now suppose phase shift is applied to each channel along the path,
which equals kΔϕ, where k is the element index ranging from 1 (bottom) to N (top)
and Δϕ is the unit phase difference between adjacent channels. If Δϕ is zero, the
beam direction will be still perpendicular to the array front as mentioned earlier, and
thus the angle between the normal axis and the beam direction, defined as θ, will also
be zero. If Δϕ is not zero and has a finite value, then the radiation normal to the array
front will become smaller due to the destructive interference between the radiations
since the phase is not aligned over the elements anymore. Instead, a new phase
alignment will be established for a radiation that shows an angle θ with the normal
direction, which will be determined by Δϕ. This is because, for the radiation at this
angle, the phase added to each element (kΔϕ) exactly balances out the extra phase in
the free space along the path for each channel. Hence, a strong beam will be formed
in this direction due to a constructive interference. In a more quantitative term, the
relation between θ and Δϕ can be obtained as follows:
d sin θ Δϕ
¼ , ð2:20Þ
λ 2π
where d is the physical distance between the adjacent channels (more precisely, the
distance between adjacent antennas) and λ is the free-space wavelength of the beam.
Each parameter can be explicitly expressed in terms of the other:
2πd λ
Δϕ ¼ sin θ or θ ¼ sin1 Δϕ : ð2:21Þ
λ 2πd
84 2 THz Sources and Related Topics
When the preferred beam direction (θ) is decided, the required unit phase
difference (Δϕ) can be calculated for a given wavelength and inter-channel distance,
or vice versa. Although the analysis is made for one-dimensional array cases based
on Fig. 2.40 for a beam steering along a single axis, the analysis can be easily
extended to two-dimensional array cases that would allow a beam steering across a
biaxial plane. A similar analysis can be made for beamforming in the receiver side,
which will not be repeated here.
The expressions above include the wavelength as a parameter, and one may
wonder phased arrays are inherently a narrowband system since the given relations
are valid only for a given wavelength (or frequency). However, Eq. (2.20) can be
equivalently expressed in terms of time shift Δτ instead of phase shift Δϕ, as is
shown below:
d sin θ
¼ Δτ: ð2:22Þ
c
In this expression, the dependence on the wavelength disappeared, implying that the
relation is frequency-independent and thus is wideband. With this viewpoint, the
time delay, not the phase delay, imposed on each channel is balanced out with the
path difference in the free space after radiation. Hence, a wideband phased array can
be realized if a time delay component is employed in the channels, which is typically
achieved by controlling the signal path length. If explained in terms of the phase, for
a given time delay created with the time delay component, the corresponding phase
delay linearly increases with increasing frequency. At the same time, the phase
corresponding to the path difference of d sin θ also shows a linear increase with
frequency. Therefore, the balance between the two phases is retained independent of
the frequency, leading to a wideband behavior. In fact, a system configured in this
way may be better named timed-array instead of phased-array.
The properties of beamforming depend on the details of the phased-array design.
A larger N results in a stronger beam forming, increasing the power along the
direction of the main lobe, but a narrower beam pattern. It also leads to a larger
number of side lobes. A larger d improves the spatial resolution as the radiation angle
is more fine-tuned with a given Δϕ as indicated by Eq. (2.21). However, too large
d may cause the free-space path difference between adjacent channels (d sin θ) to
become larger than λ. If this happens, there exist multiple angles that satisfy
Eq. (2.20) because Δϕ is effectively equivalent to Δϕ + 2π. This results in multiple
main lobes, an unfavored situation. Hence, to avoid multiple main lobes for the range
0 < θ < 90 , d should be smaller than λ. If we consider the full angle range, or
90 < θ < 90 , the condition becomes more stringent, or d < λ/2 to avoid multiple
main lobes for this whole span of θ allowed in the ideal case. However, in reality, the
practical range of interest for θ will be narrower than this. In this case, the upper limit
of d will be slightly larger than λ/2, which justifies the popular choice of d ¼ λ/2 with
some margin.
The main advantage of beamforming is the ability to control the beam direction
without any mechanical rotation by means of electric beam steering. Another
2.6 Beamforming Techniques 85
There have been various types of phase shifters developed for high-frequency
applications, some of which have also been applied for THz band operation. Popular
among those are switched-line type, reflection type, and vector sum type shifters, as
depicted in Fig. 2.41, which are briefly reviewed below.
Switched-line phase shifters are composed of multiple phase shifters cascaded,
each of which has different shifting phase and can be turned on and off by an
embedded switch (Fig. 2.41a). Widely adopted for the individual phase shifters are
LC-based low-pass and high-pass filters, either T-type or Π-type, which provides a
phase shift that is dictated by the value of the included reactive components as well
as the frequency. The detailed relation between the desired phase shift and the
required reactive component values can be found in [135] for both low- and high-
pass filters with T- and Π-type configurations. The switches are digitally controlled
by a set of bits externally applied. For an n-bit phase shifter, n phase shifters are
cascaded, each with a shifting phase of 360 /2m where m is an integer satisfying
1 m n. This results in an overall phase resolution of 360 /2n.
Reflection-type phase shifters are based on a branch-line coupler, in which the
output is extracted from the isolation port with the input signal applied into the
regular input port (Fig. 2.41b). The through and coupled ports are terminated with
variable reactive components, for which a varactor or a higher-order LC network is
typically used. The applied input signal is reflected by the reactive termination
through the coupler and appear at the output port with a shifted phase, the amount
of which depends on the reactance of the termination jX. Hence, variable phase shift
can be achieved by controlling the terminal reactance. A reflection-type phase shifter
operating at 218–268 GHz range has been reported based on GaAs mHEMT
technology [136]. This work, which employs a variable gain amplifier and a pair
of couplers with reactive termination based on HEMT-based varactors, showed a
phase control range up to 247 with an average insertion loss of 1.6 dB.
86 2 THz Sources and Related Topics
Fig. 2.41 Various phase shifter schemes: (a) Switched-line phase shifter (n-bit), (b) Reflection-
type phase shifter, (c) Vector sum phase shifter
The two types of phase shifters discussed above are both passive phase shifters,
which benefit from the high linearity and the virtually zero power dissipation.
However, being passive, they usually show rather a large insertion loss. For appli-
cations where small loss is desired, active phase shifters can be employed. A popular
active phase shifter is the vector sum phase shifter. As shown in Fig. 2.41c, it is
composed of a pair of variable gain amplifiers connected in parallel, with a quadra-
ture signal applied to each amplifier through a quadrature generator. By adjusting the
individual gain of amplifiers, the phase of the sum of the two output signals can be
controlled. There has been a report on a vector sum phase shifter operating in the
range of 220–320 GHz [137]. Built on an InP HBT technology, this wideband phase
shifter employed a single Gilbert cell for each amplifier and exhibited continuous
phase shift up to 360 with a phase error and insertion loss smaller than 10.2 and
16 dB, respectively. In another vector sum phase shifter, based on a pair of high-gain
WR3.4 InP HBT amplifiers, a gain up to 10 dB was obtained although the phase
control range was limited up to 90 with a narrow band operation near
300 GHz [138].
Phased arrays require a higher level of integration than phase shifters, involving
more challenges, especially at a high-frequency range such as the THz band. While it
took some time, integrated THz phased arrays have become reality, leading to
exciting recent reports on integrated THz sources based on phased arrays. The
2 2 290-GHz source array based on DAR as described earlier [130], was extended
to facilitate phase control over each element, enabling a 2-D beamforming based on
4 4 DARs [131]. The required phase shift was achieved with a vector sum shifter
operating at the frequency of 47 GHz. The signal frequency was subsequently
boosted up to near 140 GHz by an injection-locked frequency tripler, then up to
References 87
280 GHz through the DARs that internally extract the second harmonic from the
incoming input signal. It is noted that the phase shift is obtained at a much lower
frequency than the THz output frequency, which relaxes the challenge for shifting
phases that grows more stringent for higher frequencies. The phased array, built on a
45-nm SOI CMOS technology, exhibited an EIRP of 9.4 dBm and a beam steering
range of over 80 in both axes with 820 mW of DC power dissipation.
A THz phased array based on a novel phase shifting technique was also reported
[139]. In this 2 2 phased array operating at 318–370 GHz range, fourth harmonic
signal is extracted from four unit cells, each composed of two differential pairs made
of SiGe HBTs. For beamforming, a new technique that controls base-emitter voltage
(VBE) of the transistors is employed. When a slightly larger VBE is applied to a unit
cell than that of an adjacent cell, the induced larger swing triggers a traveling wave
toward the adjacent cell, which effectively causes a phase difference between the two
cells for the generated signal. This rather intrinsic phase shift results in a
beamforming, which can be controlled over two axes due to the nature of the 2D
array. With this phase array, beam steering angles of 128 and 53 for E- and H-
plane, respectively, were obtained along with a peak radiation power of 6.8 dBm.
The phase shift is in fact obtained at the fundamental frequency of around
80–90 GHz, 1/4 of the output frequency, relaxing the challenges as was the case
for the previous circuit discussed.
References
12. K.E. Kreischer, J.C. Tucek, D.A. Gallagher, R.E. Mihailovich, Operation of a compact, 0.65
THz source, in 33rd International Conference on Infrared, Millimeter and Terahertz Waves,
(IEEE, New York, 2008), pp. 1–2
13. A.J. Theiss, C.J. Meadows, R. Freeman, R.B. True, J.M. Martin, K.L. Montgomery, High-
average-power W-band TWT development. IEEE Trans. Plasma Sci. 38(6), 1239–1243
(2010). https://doi.org/10.1109/TPS.2010.2041794
14. P. Guenard, O. Doehler, B. Epsztein, R. Warnecke, New UHF tubes with wide electronic
tuning range. C. R. Acad. Sci. (Paris) 235, 236 (1952)
15. R. Kompfner, N. Williams, Backward-wave tubes. Proc. IRE 41(11), 1602–1611 (1953)
16. A. Negirev, A. Fedorov, Wide-range miniature millimeter-range backward-wave tubes. Radio
Commun. Technol. 4(4), 43–46 (1999)
17. C.Y. Levin, A.Y. Kirichenko, A. Borodkin, S. Churilova, A.Y. Usikov, The clinotron, in 22nd
European Microwave Conference, vol. 1, (IEEE, New York, 1992), pp. 603–607
18. F. Rusin, G. Bogomolov, Orotron—An electronic oscillator with an open resonator and
reflecting grating. Proc. IEEE 57(4), 720–722 (1969)
19. G. Nusinovich, M. Thumm, M.I. Petelin, The Gyrotron at 50: Historical overview. J. Infrared
Millimeter Terahertz Waves 35(4), 325–381 (2014). https://doi.org/10.1007/s10762-014-
0050-7
20. V.L. Bratman et al., Gyrotron development for high power THz technologies at IAP
RAS. J. Infrared Millimeter Terahertz Waves 33(7), 715–723 (2012). https://doi.org/10.
1007/s10762-012-9898-6
21. K.L. Felch et al., Characteristics and applications of fast-wave gyrodevices. Proc. IEEE 87(5),
752–781 (1999). https://doi.org/10.1109/5.757254
22. D.A.G. Deacon, L.R. Elias, J.M.J. Madey, G.J. Ramian, H.A. Schwettman, T.I. Smith, First
operation of a free-electron laser. Phys. Rev. Lett. 38(16), 892–894 (1977). https://doi.org/10.
1103/PhysRevLett.38.892
23. H. Motz, Applications of the radiation from fast electron beams. J. Appl. Phys. 22(5), 527–535
(1951)
24. Y.-S. Lee, Principles of Terahertz Science and Technology (Springer, New York, 2009)
25. J. Madey, H. Schwettman, W. Fairbank, A free electron laser. IEEE Trans. Nucl. Sci. 20(3),
980–983 (1973)
26. O. Shevchenko et al., The Novosibirsk free electron laser–Unique source of terahertz and
infrared coherent radiation. Phys. Procedia 84, 13–18 (2016)
27. A.W. Hull, The effect of a uniform magnetic field on the motion of electrons between coaxial
cylinders. Phys. Rev. 18(1), 31–57 (1921). https://doi.org/10.1103/PhysRev.18.31
28. H. Boot, J. Randall, The cavity magnetron. J. Inst. Electric. Eng. IIIA Radiolocation 93(5),
928–938 (1946)
29. T.K. Ishii, Practical Microwave Electron Devices (Academic Press, San Diego, CA, 1990)
30. N. Avtomonov, V. Naumenko, D.M. Vavriv, K. Schünemann, A. Suvorov, V. Markov,
Toward terahertz magnetrons: 210-GHz spatial-harmonic magnetron with cold cathode.
IEEE Trans. Electron Devices 59(12), 3608 (2012)
31. J.-I. Kim, S.-G. Jeon, G. Park, Terahertz vacuum electronics, in Handbook of Terahertz
Technologies: Devices and Applications, ed. by H.-J. Song, T. Nagatsuma, (CRC Press,
Boca Raton, FL, 2015)
32. J.B. Gunn, Microwave oscillations of current in III–V semiconductors. Solid State Commun. 1
(4), 88–91 (1963)
33. H. Kroemer, Theory of the Gunn effect. Proc. IEEE 52(12), 1736–1736 (1964)
34. B. Ridley, T. Watkins, The possibility of negative resistance effects in semiconductors. Proc.
Phys. Soc. 78(2), 293 (1961)
35. C. Hilsum, Transferred electron amplifiers and oscillators. Proc. IRE 50(2), 185–189 (1962)
36. S.Y. Liao, Microwave Solid-State Devices (Prentice-Hall, Englewood Cliffs, NJ, 1985)
37. H. Eisele, 480 GHz oscillator with an InP Gunn device. Electron. Lett. 46(6), 422–423 (2010).
https://doi.org/10.1049/el.2010.3362
References 89
38. W.T. Read Jr., A proposed high-frequency, negative-resistance diode. Bell Syst. Tech. J. 37
(2), 401–446 (1958)
39. R.L. Johnston, B.C. DeLoach, B.G. Cohen, A silicon diode microwave oscillator. Bell Syst.
Tech. J. 44(2), 369–372 (1965). https://doi.org/10.1002/j.1538-7305.1965.tb01667.x
40. C. Lee, R. Batdorf, W. Wiegmann, G. Kaminsky, The read diode—An avalanching, transit-
time, negative-resistance oscillator. Appl. Phys. Lett. 6(5), 89–91 (1965)
41. S.M. Sze, Physics of Semiconductor Devices, 2nd edn. (John Wiley & Sons, New York, 1981)
42. J. Nishizawa, K. Motoya, Y. Okuno, GaAs TUNNETT diodes. IEEE Trans. Microw. Theory
Tech. 26(12), 1029–1035 (1978). https://doi.org/10.1109/tmtt.1978.1129540
43. M. Ino, T. Ishibashi, M. Ohmori, C.W. oscillation with p+-p-n+ silicon IMPATT diodes in
200 GHz and 300 GHz bands. Electron. Lett. 12(6), 148–149 (1976). https://doi.org/10.1049/
el:19760115
44. J. Nishizawa, P. Plotka, T. Kurabayashi, H. Makabe, 706-GHz GaAs CW fundamental-mode
TUNNETT diodes fabricated with molecular layer epitaxy. Phys. Status Solidi C 5(9),
2802–2804 (2008)
45. M. Mukherjee, N. Mazumder, S.K. Roy, Photosensitivity analysis of gallium nitride and
silicon carbide terahertz IMPATT oscillators: Comparison of theoretical reliability and study
on experimental feasibility. IEEE Trans. Device Mater. Reliab. 8(3), 608–620 (2008). https://
doi.org/10.1109/TDMR.2008.2002358
46. L.L. Chang, L. Esaki, R. Tsu, Resonant tunneling in semiconductor double barriers. Appl.
Phys. Lett. 24(12), 593–595 (1974). https://doi.org/10.1063/1.1655067
47. R. Tsu, L. Esaki, Tunneling in a finite superlattice. Appl. Phys. Lett. 22(11), 562–564 (1973)
48. T. Maekawa, H. Kanaya, S. Suzuki, M. Asada, Oscillation up to 1.92 THz in resonant
tunneling diode by reduced conduction loss. Appl. Phys. Express 9(2), 024101 (2016)
49. E.H. Colpitts, Oscillation generator, USA Patent 1,624,537 Patent Appl. 1,624,537 (1927)
50. B. Razavi, R. Behzad, RF Microelectronics (Prentice Hall, Upper Saddle River, NJ, 1998)
51. J.K. Clapp, An inductance-capacitance oscillator of unusual frequency stability. Proc. IRE 36
(3), 356–358 (1948)
52. P. Andreani, W. Xiaoyan, L. Vandi, A. Fard, A study of phase noise in colpitts and LC-tank
CMOS oscillators. IEEE J. Solid State Circuits 40(5), 1107–1118 (2005). https://doi.org/10.
1109/JSSC.2005.845991
53. V. Marotta, G. Macera, M.P. Kennedy, E. Napoli, Comparative analysis of differential colpitts
and cross-coupled VCOs in 180 nm Si-Ge HBT technology, in 2016 IEEE International
Symposium on Circuits and Systems (ISCAS), 22–25 May 2016 (2016), pp. 1650–1653.
https://doi.org/10.1109/ISCAS.2016.7538883
54. L. Li, P. Reynaert, M. Steyaert, A colpitts LC VCO with Miller-capacitance gm enhancing and
phase noise reduction techniques, in 2011 Proceedings of the ESSCIRC (ESSCIRC), 12–16
Sept 2011 (2011), pp. 491–494. https://doi.org/10.1109/ESSCIRC.2011.6045014
55. Z. Hu, M. Kaynak, R. Han, High-power radiation at 1 THz in silicon: A fully scalable array
using a multi-functional radiating mesh structure. IEEE J. Solid State Circuits 53(5),
1313–1327 (2018). https://doi.org/10.1109/JSSC.2017.2786682
56. B. Heinemann et al., SiGe HBT technology with fT/fmax of 300GHz/500GHz and 2.0 ps
CML gate delay, in 2010 International Electron Devices Meeting, 6–8 Dec 2010 (2010),
pp. 30.5.1–30.5.4, https://doi.org/10.1109/IEDM.2010.5703452
57. K. Lundien, R. Mattauch, J. Archer, R. Malik, Hyperabrupt junction varactor diodes for
millimeter-wavelength harmonic generators. IEEE Trans. Microw. Theory Tech. 31(2),
235–238 (1983)
58. P. Andreani, S. Mattisson, On the use of MOS varactors in RF VCOs. IEEE J. Solid State
Circuits 35(6), 905–910 (2000)
59. T. Soorapanth, C.P. Yue, D.K. Shaeffer, T.H. Lee, S.S. Wong, Analysis and optimization of
accumulation-mode varactor for RF ICs, in 1998 Symposium on VLSI Circuits, Citeseer
(1998), pp. 32–33
90 2 THz Sources and Related Topics
60. Y. Oh, S. Kim, S. Lee, J.-S. Rieh, The island-gate varactor—A high-Q MOS varactor for
millimeter-wave applications. IEEE Microw. Wirel. Compon. Lett. 19(4), 215–217 (2009)
61. Y. Oh, J.-S. Rieh, A comprehensive study of high-Q island-gate varactors (IGVs) for CMOS
millimeter-wave applications. IEEE Trans. Microw. Theory Tech. 59(6), 1520–1528 (2011).
https://doi.org/10.1109/tmtt.2011.2117437
62. N. Kim, Y. Oh, J.-S. Rieh, A 47 GHz LC cross-coupled VCO employing high-Q island-gate
varactor for phase noise reduction. IEEE Microw. Wirel. Compon. Lett. 20(2), 94–96 (2010)
63. G. Cusmai, M. Repossi, G. Albasini, A. Mazzanti, F. Svelto, A magnetically tuned quadrature
oscillator. IEEE J. Solid State Circuits 42(12), 2870–2877 (2007). https://doi.org/10.1109/
JSSC.2007.908727
64. P. Chiang, O. Momeni, P. Heydari, A 200-GHz inductively tuned VCO with 7-dBm output
power in 130-nm SiGe BiCMOS. IEEE Trans. Microw. Theory Tech. 61(10), 3666–3673
(2013). https://doi.org/10.1109/TMTT.2013.2279779
65. K. Kwok, J.R. Long, J.J. Pekarik, A 23-to-29 GHz differentially tuned varactorless VCO in
0.13 μm CMOS, in 2007 IEEE International Solid-State Circuits Conference, 11–15 Feb 2007
(2007), pp. 194–596. https://doi.org/10.1109/ISSCC.2007.373360
66. D.D. Kim, J. Kim, C. Cho, A 94GHz locking hysteresis-assisted and tunable CML static
divider in 65nm SOI CMOS, in 2008 IEEE International Solid-State Circuits Conference, 3–7
Feb 2008 (2008), pp. 460–628. https://doi.org/10.1109/ISSCC.2008.4523256
67. R. Miller, Fractional-frequency generators utilizing regenerative modulation. Proc. IRE 27(7),
446–457 (1939)
68. J. Al-Eryani et al., A 47–217 GHz dynamic frequency divider in SiGe technology, in Bipolar/
BiCMOS Circuits and Technology Meeting, (IEEE, New York, 2015), pp. 125–128
69. M. Seo, J. Hacker, M. Urteaga, A. Skalare, M. Rodwell, A 529 GHz dynamic frequency
divider in 130 nm InP HBT process. IEICE Electron. Express 12(3), 20141118 (2015)
70. H.R. Rategh, T.H. Lee, Superharmonic injection-locked frequency dividers. IEEE J. Solid
State Circuits 34(6), 813–821 (1999). https://doi.org/10.1109/4.766815
71. H. Seo, S. Seo, S. Jeon, J.-S. Rieh, A Q-band injection-locked frequency divider with
inductive feedback for a locking range enhancement. IEEE Microw. Wirel. Compon. Lett.
21(6), 317–319 (2011)
72. H. Tong, S. Cheng, A.L. Karsilayan, J.S. Martinez, An injection-locked frequency divider
with multiple highly nonlinear injection stages and large division ratios. IEEE Trans. Circuits
Syst. II Express Briefs 54(4), 313–317 (2007). https://doi.org/10.1109/TCSII.2007.895071
73. P.-H. Feng, S.-I. Liu, A 300 GHz divide-by-2 ILFD using frequency boosting technique. IEEE
Microw. Wirel. Compon. Lett. 23(11), 599–601 (2013)
74. B. Razavi, A study of injection locking and pulling in oscillators. IEEE J. Solid State Circuits
39(9), 1415–1424 (2004)
75. N. Kim, K. Song, J. Yun, J. Yoo, J.-S. Rieh, Two 122-GHz phase-locked loops in 65-nm
CMOS technology. IEEE Trans. Microw. Theory Tech. 64(8), 2623–2630 (2016). https://doi.
org/10.1109/TMTT.2016.2581816
76. Y. Zhao et al., A 0.56 THz phase-locked frequency synthesizer in 65 nm CMOS technology.
IEEE J. Solid State Circuits 51(12), 3005–3019 (2016)
77. J. Yoo, D. Kim, J. Kim, K. Song, J.-S. Rieh, A 283-GHz fully integrated phase-locked loop
based on 65-nm CMOS. IEEE Trans. Terahertz Sci. Technol. 8(6), 784–792 (2018)
78. S.-C. Yen, T.-H. Chu, An Nth-harmonic oscillator using an N-push coupled oscillator array
with voltage-clamping circuits, in 2003 IEEE MTT-S International Microwave Symposium
Digest, vol. 3, (IEEE, New York, 2003), pp. 2169–2172
79. U.L. Rohde, A.K. Poddar, J. Schoepf, R. Rebel, P. Patel, Low noise low cost ultra wideband
N-push VCO, in IEEE MTT-S International Microwave Symposium Digest (2005),
pp. 1171–1174. https://doi.org/10.1109/MWSYM.2005.1516884
80. R. Wanner, R. Lachner, G. Olbrich, A monolithically integrated 190-GHz SiGe push-push
oscillator. IEEE Microw. Wirel. Compon. Lett. 15(12), 862–864 (2005)
81. E. Seok et al., A 410GHz CMOS push-push oscillator with an on-chip patch antenna, in IEEE
International Solid-State Circuits Conference (ISSCC), (IEEE, New York, 2008), pp. 472–629
References 91
82. T. Yu-Lung, W. Huei, Triple-push oscillator approach: Theory and experiments. IEEE J. Solid
State Circuits 36(10), 1472–1479 (2001). https://doi.org/10.1109/4.953475
83. J. Grzyb, Y. Zhao, U. Pfeiffer, A 288-GHz lens-integrated balanced triple-push source in a
65-nm CMOS technology. IEEE J. Solid State Circuits 48(7), 1751–1761 (2013)
84. D. Shim, D. Koukis, D.J. Arenas, D.B. Tanner, K. Kenneth, 553-GHz signal generation in
CMOS using a quadruple-push oscillator, in 2011 Symposium on VLSI Circuits, (IEEE,
New York, 2011), pp. 154–155
85. J. Yun, D. Yoon, S. Jung, M. Kaynak, B. Tillack, J.-S. Rieh, Two 320 GHz signal sources
based on SiGe HBT technology. IEEE Microw. Wirel. Compon. Lett. 25(3), 178–180 (2015).
https://doi.org/10.1109/LMWC.2015.2391011
86. O. Momeni, E. Afshari, High power terahertz and millimeter-wave oscillator design: A
systematic approach. IEEE J. Solid State Circuits 46(3), 583–597 (2011)
87. J. Yoo, N. Kim, J. Yun, M.G. Seo, J.-S. Rieh, A CMOS triple-push 280-GHz VCO integrated
with 1/16,384 divider chain, in 2015 IEEE International Symposium on Radio-Frequency
Integration Technology (RFIT) (2015), pp. 58–60. https://doi.org/10.1109/RFIT.2015.
7377886
88. A.V. Raisanen, Frequency multipliers for millimeter and submillimeter wavelengths. Proc.
IEEE 80(11), 1842–1852 (1992)
89. G. Chattopadhyay et al., An all-solid-state broad-band frequency multiplier chain at 1500
GHz. IEEE Trans. Microw. Theory Tech. 52(5), 1538–1547 (2004)
90. A. Maestrini et al., Schottky diode-based terahertz frequency multipliers and mixers. C. R.
Phys. 11(7–8), 480–495 (2010)
91. V. Milanovic, M. Gaitan, J.C. Marshall, M.E. Zaghloul, CMOS foundry implementation of
Schottky diodes for RF detection. IEEE Trans. Electron Devices 43(12), 2210–2214 (1996).
https://doi.org/10.1109/16.544393
92. C. Mao, C.S. Nallani, S. Sankaran, E. Seok, K.O. Kenneth, 125-GHz diode frequency doubler
in 0.13-um CMOS. IEEE J. Solid State Circuits 44(5), 1531–1538 (2009)
93. U.R. Pfeiffer, C. Mishra, R.M. Rassel, S. Pinkett, S.K. Reynolds, Schottky barrier diode
circuits in silicon for future millimeter-wave and terahertz applications. IEEE Trans. Microw.
Theory Tech. 56(2), 364–371 (2008)
94. E. Ojefors, J. Grzyb, Y. Zhao, B. Heinemann, B. Tillack, U.R. Pfeiffer, A 820GHz SiGe
chipset for terahertz active imaging applications, in IEEE International Solid-State Circuits
Conference (2011), pp. 224–226
95. C.M. Grötsch, H. Mabler, A. Leuther, I. Kallfass, An active multiplier-by-six S-MMIC for
500 GHz, in 2018 43rd International Conference on Infrared, Millimeter, and Terahertz
Waves (IRMMW-THz), (IEEE, New York, 2018), pp. 1–2
96. O. Momeni, E. Afshari, A broadband mm-wave and terahertz traveling-wave frequency
multiplier on CMOS. IEEE J. Solid State Circuits 46(12), 2966–2976 (2011)
97. N. Mazor, E. Socher, A SiGe distributed millimeter-wave frequency Tripler. IEEE Microw.
Wirel. Compon. Lett. 24(12), 893–895 (2014). https://doi.org/10.1109/LMWC.2014.2361625
98. A. Moussessian et al., A terahertz grid frequency doubler. IEEE Trans. Microw. Theory Tech.
46(11), 1976–1981 (1998). https://doi.org/10.1109/22.734524
99. E. Monaco, M. Pozzoni, F. Svelto, A. Mazzanti, Injection-locked CMOS frequency doublers
for μ-wave and mm-wave applications. IEEE J. Solid State Circuits 45(8), 1565–1574 (2010)
100. Z. Ahmad, M. Lee, K.O. Kenneth, 1.4 THz, 13dBm-EIRP frequency multiplier chain using
symmetric-and asymmetric-CV varactors in 65nm CMOS, in 2016 IEEE International Solid-
State Circuits Conference (ISSCC), (IEEE, New York, 2016), pp. 350–351
101. A. Zamora et al., A high efficiency 670 GHz x36 InP HEMT multiplier chain, in 2017 IEEE
MTT-S International Microwave Symposium (IMS), (IEEE, New York, 2017), pp. 977–979
102. Z. Chen, P. Heydari, An 85–95.2 GHz transformer-based injection-locked frequency tripler in
65nm CMOS, in 2010 IEEE MTT-S International Microwave Symposium, (IEEE, New York,
2010), pp. 776–779
92 2 THz Sources and Related Topics
103. N. Mazor, E. Socher, X-band to W-band frequency multiplier in 65 nm CMOS process. IEEE
Microw. Wirel. Compon. Lett. 22(8), 424–426 (2012)
104. V. Jain et al., 1.0 THz fmax InP DHBTs in a refractory emitter and self-aligned base process for
reduced base access resistance, in Device Research Conference, 20–22 June 2011 (2011),
pp. 271–272. https://doi.org/10.1109/drc.2011.5994528
105. X. Mei et al., First demonstration of amplification at 1 THz using 25-nm InP high electron
mobility transistor process. IEEE Electron Device Lett. 36(4), 327–329 (2015). https://doi.org/
10.1109/LED.2015.2407193
106. H. Lee et al., Intel 22nm FinFET (22FFL) process technology for RF and mm wave applica-
tions and circuit design optimization for FinFET technology, in 2018 IEEE International
Electron Devices Meeting (IEDM) (2018), pp. 14.1.1–14.1.4. https://doi.org/10.1109/IEDM.
2018.8614490
107. B. Heinemann et al., SiGe HBT with fT/fmax of 505 GHz/720 GHz, in 2016 IEEE Interna-
tional Electron Devices Meeting (IEDM) (2016), pp. 3.1.1–3.1.4. https://doi.org/10.1109/
IEDM.2016.7838335
108. Y. Mao, K. Schmalz, J. Borngräber, J.C. Scheytt, A 245 GHz CB LNA in SiGe, in 2011 6th
European Microwave Integrated Circuit Conference, 10–11 Oct 2011 (2011), pp. 224–227
109. D. Yoon, M. Seo, K. Song, M. Kaynak, B. Tillack, J. Rieh, 260-GHz differential amplifier in
SiGe heterojunction bipolar transistor technology. Electron. Lett. 53(3), 194–196 (2017).
https://doi.org/10.1049/el.2016.3882
110. O. Momeni, E. Afshari, A high gain 107 GHz amplifier in 130 nm CMOS, in 2011 IEEE
Custom Integrated Circuits Conference (CICC) (2011), pp. 1–4. https://doi.org/10.1109/
CICC.2011.6055322
111. O. Momeni, A 260GHz amplifier with 9.2dB gain and 3.9dBm saturated power in 65nm
CMOS, in 2013 IEEE International Solid-State Circuits Conference (2013), pp. 140–141.
https://doi.org/10.1109/ISSCC.2013.6487672
112. D. Park, D.R. Utomo, J. Hong, S. Lee, A 230–260GHz wideband amplifier in 65nm CMOS
based on dual-peak Gmax-core, in 2017 Symposium on VLSI Circuits (2017), pp. C300–C301.
https://doi.org/10.23919/VLSIC.2017.8008516
113. D. Yoon, N. Kim, U. Pfeiffer, B. Heinemann, J. Rieh, A wide band 215–255 GHz CB
differential amplifier in a 0.25-μm SiGe HBT technology, in 2013 Asia-Pacific Microwave
Conference Proceedings (APMC) (2013), pp. 351–353. https://doi.org/10.1109/APMC.2013.
6695143
114. V. Radisic et al., InP HBT transferred substrate amplifiers operating to 600 GHz, in 2015 IEEE
MTT-S International Microwave Symposium (2015) pp. 1–3. https://doi.org/10.1109/
MWSYM.2015.7166750
115. J. Hacker, M. Urteaga, M. Seo, A. Skalare, R. Lin, InP HBT amplifier MMICs operating to
0.67 THz, in 2013 IEEE MTT-S International Microwave Symposium Digest (MTT) (2013),
pp. 1–3. https://doi.org/10.1109/MWSYM.2013.6697518
116. S. Yoon, I. Lee, M. Urteaga, M. Kim, S. Jeon, A fully-integrated 40–222 GHz InP HBT
distributed amplifier. IEEE Microw. Wirel. Compon. Lett. 24(7), 460–462 (2014). https://doi.
org/10.1109/LMWC.2014.2316223
117. V. Radisic, K.M. Leong, X. Mei, S. Sarkozy, W. Yoshida, W. Deal, Power amplification at
0.65 THz using InP HEMTs. IEEE Trans. Microw. Theory Tech. 60(3), 724–729 (2012)
118. J. Kim, S. Jeon, M. Kim, M. Urteaga, J. Jeong, H-band power amplifier integrated circuits
using 250-nm InP HBT technology. IEEE Trans. Terahertz Sci. Technol. 5(2), 215–222
(2015). https://doi.org/10.1109/TTHZ.2014.2387259
119. H. Kazemi et al., 350mW G-band medium power amplifier fabricated through a new method
of 3D-copper additive manufacturing, in 2015 IEEE MTT-S International Microwave Sympo-
sium (2015), pp. 1–3. https://doi.org/10.1109/MWSYM.2015.7167037
120. K.J. Russell, Microwave power combining techniques. IEEE Trans. Microw. Theory Tech. 27
(5), 472–478 (1979). https://doi.org/10.1109/TMTT.1979.1129651
References 93
The illumination of materials with electromagnetic waves generates heat and causes
a temperature increase inside the material. This temperature change induces various
secondary thermal effects, which can be exploited for the detection of the THz
V DC,out
ℜ¼ ½V=W: ð3:1Þ
Pin
As Eq. (3.1) indicates, the responsivity represents the effectiveness of the conversion
of the input power into the output voltage. NEP is defined as the input power that
leads to the output signal-to-noise ratio (SNR) of unity for a bandwidth of 1 Hz. The
average noise voltage of a system with a bandwidth of 1 Hz is often represented by
vn , and thus the condition for SNR ¼ 1 with 1-Hz bandwidth can be expressed as:
V DC,out ℜPin
SNRout ¼ ¼ ¼ 1 if Pin ¼ NEP: ð3:2Þ
vn vn
Hence,
h i
vn
NEP ¼ W=Hz1=2 : ð3:3Þ
ℜ
Let us consider the example case of NEP ¼ 1 pW/Hz1/2 (¼ 1012 W/Hz1/2) again. If
the bandwidth of the system is 10 kHz (¼ 104 Hz), for example, the minimum
detectible power is 100 pW (¼ 1010 W), as limited by the detection rate. It is useful
to remember that in some literatures NEP is expressed in the unit of [W] instead of
[W/Hz1/2] by omitting bandwidth normalization, but this is rare and not a general
practice. It is also noted that Eqs. (3.1)–(3.3) assume the detector output is given in
voltage, but the output may be current for some detectors such as photodiodes. In
those cases, the responsivity is given as IDC,out/Pin in the unit of [A/W] and NEP as
in =ℜ in the unit of [W/Hz1/2] (unchanged), where in is the average noise current in
the unit of [A/Hz1/2]. While a balanced performance assessment of detectors would
show both responsivity and NEP, it is fair to say that NEP is more intrinsic to the
detector. This is because the responsivity can be enhanced by placing an amplifier
following the detector, while NEP cannot be improved with such a simple add-on
scheme. Hence, in this chapter, NEP will be treated as the primary parameter in
comparing detectors.
As an additional note, the sensitivity of detectors is sometimes presented in terms
of another parameter, the specific detectivity D* (reads D-star) [1]. D* is defined as
the inverse of NEP normalized over the area, or
pffiffiffi h i
A
D ¼ cm Hz1=2 =W : ð3:4Þ
NEP
D* may provide a fair comparison over detectors with different active areas if NEP is
proportional to the square root of the area, which is often the case forpthe
ffiffiffiffiffiffibackground
limited cases. The readers may find some literatures define D* as AB=NEP with
bandwidth B included, but this is based on the unnormalized NEP in [W] as
mentioned above. The specific detectivity D* is often referred to as the detectivity
for simplicity, but there is a separate definition for detectivity D, which is simply the
inverse of NEP in [Hz1/2/W]. Still, D* is in far more popular use than D as many find
its area normalization useful.
3.1.1 Bolometers
Bolometers generally refer to a group of detectors that make use of the resistance
change due to the temperature rise caused by radiation. The invention of the
bolometer is credited to Langley, who devised a detector for astronomical observa-
tion in 1878 and named it “bolometer.” It was not until 1960s, however, that
bolometers began to be widely adopted with the emergence of the carbon resistance
bolometer [2], before which Golay cell was a popular choice. Since then, bolometers
have been widely selected for various applications in infrared and neighboring
bands, which was fueled by the introduction of germanium thermometer [3],
superconducting thermometer [4], and silicon monolithic thermometer [5] for
bolometers. As bolometers can be realized in many different ways, there are
98 3 THz Detectors and Related Topics
R(T)
V
-
Thermometer
Thermal
conductance
Thermal reservoir
(a) (b)
dΔT ðt Þ
P ¼ Gth ΔT ðt Þ þ C th , ð3:5Þ
dt
where ΔT(t) is the temperature rise induced by the illumination. (If the power
absorption from the ambient temperature is also to be considered, P will additionally
include the ambient term as well.) It is apparent from this equation that the thermal
response time will be given as Cth/Gth.
In most applications, a fast response is desired for bolometers, which would
require a small Cth and a large Gth. Small Cth demands the absorber (and the
supporting substrate) to have a low heat capacity, which in turn requires a small
volume and a low specific heat. The condition for a small volume, however,
contradicts the need for a large absorption area that will improve the bolometer
3.1 Thermal Detectors 99
area, allows a fast response, while a low NEP and a high responsivity can be
obtained with the help of the coupled antenna. There are reports on microbolometers
without antennas as well, suitable for cases that require high density for array
applications [9]. One may note that most of microbolometers are intended for
room temperature operation, unlike the mainstream bolometers, although some
cryogenic microbolometers have also been reported [10].
Bolometers operating at cryogenic temperatures, mostly near liquid helium
(He) temperature (4.2 K) or below, generally exhibit NEP in the range of 1017 to
1012 W/Hz1/2 depending on the bolometer type and the actual temperature. On the
extreme side, there have been efforts to meet the required NEP of in the order of
1020 W/Hz1/2 for spaceborne infrared telescopes being planned such as SPICA or
SAFIR, leading to some early results exhibiting NEP lower than 1019 W/Hz1/2
[11]. NEP of room temperature bolometers typically falls on the range of 1011 to
1010 W/Hz1/2, while lower NEP has also been reported that reaches below
1012 W/Hz1/2 at room temperature [12]. There is an interesting observation that
NEP of THz bolometers has been reduced by half every 2 years in average over the
past 70 years [13]. It is noteworthy to comment that the noise level of bolometers
does not show a strong dependence on the frequency. In fact, this wideband response
is an inherent property of any thermal detectors, including those to be discussed
below, as the radiation at any spectrum causes a temperature rise of the material it
illuminates. The response time of bolometers typically ranges over 1–100 ms, while
it can be reduced down below 1 μs depending on the structure and temperature [14].
polar axis
Support layer
the voltage along the load impedance (not shown), is the output of the detector,
which is usually magnified by an amplifier at the final stage. While pyroelectric
detectors are useful for the detection of radiation that changes over time as men-
tioned above, they can still be employed to detect continuous wave (CW) signal, if
the signal is properly modulated by a chopper or other similar modulating schemes.
As for the materials, TGS (triglycine sulfate) was widely used in the early stage of
pyroelectric detectors. Later on, various pyroelectric materials have appeared for
pyroelectric detectors, including SBN (strontium-barium niobite), LiTaO3 (lithium
tantalate), LiNbO3 (lithium niobate), BaTiO (barium titanate), PVF2 (polyvinylidene
fluoride), and so forth. Detailed comparisons between the materials can be found
elsewhere [17, 18]. It should be noted that pyroelectric materials are governed by
Curie temperature, beyond which the pyroelectric materials lose the unique magnetic
property and does not exhibit pyroelectricity anymore. Hence, the detector should be
operated below this temperature. In terms of the structure, there is a variance called
the compensated pyroelectric detector. As pyroelectric detectors respond to a tem-
perature variation, the ambient temperature variation can be falsely detected. To
suppress the effect, a dummy detector, which is screened from the external radiation,
is placed in proximity of the true detector. As the ambient temperature variation is
detected by both detectors while the radiation is incident only on the true detector,
the net effect of the radiation can be obtained by extracting the difference between
the two outputs. In fact, this is analogous to the case of the differential pair, which is
widely adopted for transistor circuits to suppress the common-mode noise. The
principle is basically the same.
Pyroelectric detectors show wideband characteristics like other thermal detectors
because of the reason mentioned earlier. Besides, they distinguish themselves from
other thermal detectors in that they exhibit a fast response below the thermal
relaxation time, since they do not need to wait until the thermal equilibrium is
established when there is a temperature change. This leads to the response time
easily below μs order. Pyroelectric detectors show fairly good NEP values, typical
values ranging around ~109 W/Hz1/2. The actual values vary over a quite wide
range depending on the vendors. It is true that this is several orders worse than that of
cryogenic bolometers and also falls short of NEPs with typical room-temperature
bolometers. However, pyroelectric detectors benefit from the simple structure, often
102 3 THz Detectors and Related Topics
housed into a low-cost transistor-type metal package with a window opening. This
makes it an affordable detector still capable of wide-range detection including the
THz band [19].
3.1.3 Thermopiles
ΔV ¼ SΔT, ð3:6Þ
where S is the Seebeck coefficient, which itself is a function of the temperature. The
minus sign is included here to follow the convention of the Seebeck coefficient, with
which the voltage is taken at the cold side with reference to the hot side. For
materials with a negative S (e.g., aluminum, platinum), the mean free path is longer
at higher temperatures, whereas for materials with a positive S (e.g., copper, gold),
the mean free path is shorter at higher temperatures. Note that the Seebeck effect is
the inverse process of the Peltier effect, which describes a temperature difference
induced across a metal rod when a voltage difference is applied across it.
The Seebeck effect can find various useful applications, one of which is the
thermocouple to be discussed here. Consider two metal rods with opposite signs of
the Seebeck coefficient, which are connected to form a junction as shown in
Fig. 3.3b. If the two sides are placed at different temperatures, voltage differences
will be induced for both metal rods, but with opposite polarities. This will lead to a
voltage difference established between the two metal tips on the other side of the
junction. A similar result will be obtained even for two metals with the same sign of
the Seebeck coefficient if the actual values are different, but the induced voltage
difference will not be as large as the case with two materials with opposite signs.
3.1 Thermal Detectors 103
Fig. 3.3 (a) Conceptual Material with negative S Material with positive S
view of Seebeck effect. For Thot Tcold Thot Tcold
materials with a negative
Seebeck coefficient S, the
mean free path of electrons
is larger at higher
temperature (lhot > lcold).
For materials with a positive lhot > lcold lhot < lcold
S, the opposite occurs nhot < ncold nhot > ncold
(lhot < lcold). As a result, the Vhot > Vcold Vhot < Vcold
induced electron (a)
concentration (n) and
voltage (V ) show the
opposite signs for the two T1 T2 Material with negative S
materials. (b) Structure of a
typical thermocouple, which ΔV
Material with positive S
is made of two materials
with the opposite signs of S. T1 > T2
(c) Structure of a typical of ΔV > 0
thermopile, which is (b)
basically multiple
thermopiles connected in
series T1 T2
T1 > T2
ΔV > 0
(c)
With the established voltage taken as the output, this structure can be used as a
temperature sensor, which is called the thermocouple. If the temperature difference
is induced by electromagnetic wave radiation, it can be employed as a radiation
detector. Now, suppose the configuration in which multiple thermocouples are
connected in series as depicted in Fig. 3.3c. In this case, the output voltage will be
even larger, by a factor of the number of thermocouples cascaded, leading to a larger
responsivity by the same factor (assuming the same power absorbed per unit
thermocouple). This is the thermopile.
Various materials and structures can be employed for thermopiles, including a
modern integrated thermopile based on CMOS technology [20, 22]. A typical NEP
of thermopile detectors falls on the order of 109 to 108 W/Hz1/2, an acceptable
value for many practical applications.
104 3 THz Detectors and Related Topics
Golay cells are another type of room-temperature uncooled thermal detector that can
be used for the detection of a wide range of frequency including the THz band. While
all the thermal detectors discussed so far are based on the solid-state phenomena that
respond to the temperature change, the Golay cell makes use of the expansion of the
gas-phase material by a temperature rise with the help of an optical method to detect
the expansion. It was first invented by Golay in 1947 [23], apparently named after
the inventor, and has been used as a very effective radiation detector until today.
The schematic of a typical Golay cell is illustrated in Fig. 3.4 [24]. The main body
is composed of a gas cell, an absorbing film, and a flexible mirror. When there is
incoming radiation through the cell window incident on the absorption film, the cell
temperature increases, resulting in the expansion of the gas confined in the pneu-
matic gas cell. The gas expansion causes the deformation of the flexible membrane
mirror located on the opposite side of the cell window. Hence, if the delicate
deformation on the mirror surface can be detected reliably, the temperature rise
and thus the radiation power can be estimated. This is performed by an optical
system composed of a light source, lenses, a line grid, and a photo detector as shown
in the right side of Fig. 3.4. The light from the light source reaches the surface of the
flexible mirror through the upper half of the pair of lenses and the line grid, and then
is reflected back through the lower half of the grid and the lenses, finally landing on
the photo detector. The details of the reflected grid image on the photo detector are
sensitive to the surface deformation of the mirror. Hence, by detecting the change in
the grid image pattern, the amount of the deformation is determined and so is the
radiation power absorbed by the cell. For this, in the Golay’s original design, the
lower half of the grid pattern was carefully designed so that it is slightly misaligned
with the incoming image of the upper-half grid pattern carried by the reflected light
beam when the mirror is totally flat. With this tactical scheme, the deformed mirror
Gas cell
Window
Grid
Radiation
Light source
Absorbing
film
Fig. 3.4 Structure of a typical Golay cell. (Adapted by permission from Springer Nature: [24]
© 2012)
3.2 Diode Detectors 105
surface with the gas expansion, in a concave or convex (in most cases) shape, will
either strengthen or weaken the light intensity falling on the photo detector. By
reading this change, the absorbed radiation power can be evaluated with a proper
calibration.
The desired property for the gas inside the cell is the low heat conductivity, for
which xenon (Xe) is typically adopted. The ambient temperature variation can lead
to the volume change of the gas cell, which needs to be avoided for a precise
radiation detection. A structural solution is to form a small leak between the gas
cell and the gas reservoir outside the cell, so that any change in the gas cell volume
due to the ambient temperature variation can be compensated by the leak. Only
constraint due to this compensation scheme in actual detection is that the detection
should be made before the leak takes place. For this reason, the incoming radiation is
often modulated by a chopper at a rate faster than the time constant of the leak,
typically at around 10–20 Hz with a detection response time of ~20 ms. This also
helps to suppress the 1/f noise of the detector.
The NEP values available from Golay cells are among the best for room-
temperature thermal detectors, ranging over 5 1011 to 109 W/Hz1/2 with a
wide spectrum coverage including the THz band. This explains the popular use of
this detector long after its invention without much change in the design. However,
there do exist modern versions of Golay cells developed for compact implementa-
tion. A miniature Golay cell based on planar micromachining technology was
reported [25], in which the deformation of a micro gas cell was detected by tunneling
displacement sensor instead of the conventional optical method. Built on a Si wafer
and intended for a room operation, it showed an NEP of 3 1010 W/Hz1/2. There
has also been an attempt to adopt a graphene film as the gas cell membrane for micro
Golay cell arrays, an interesting crossover of old and new technologies [26].
The term diode covers a wide range of devices, and its definition may vary
depending on specific application cases. Here, we will assume a rather broader
definition for the diode—an electronic device with two terminals. From this point
of view, there are various types of diodes that can serve as a THz detector. In this
section, three types of diode-based THz detectors will be discussed: SBD (Schottky
Barrier Diode), HEB (Hot Electron Bolometer), and SIS (Superconductor–
Insulator–Superconductor) mixer. In principle, these detectors operate as a mixer
based on the heterodyne technique, as opposed to the detectors described in the
previous section which are basically direct detectors. To help understand the differ-
ence between these two types of detection, a brief comparison will be first made
between direct and heterodyne detections before moving on to the description of
each type of the diode-based detectors.
106 3 THz Detectors and Related Topics
As just mentioned, there can be two types of THz signal detection: direct detection
and heterodyne detection. Sometimes they are referred to as incoherent detection and
coherent detection, respectively, as the latter may preserve the phase information of
the RF input signal while the former does not. Conceptual block diagrams for the
two cases are illustrated Fig. 3.5. For the direct detection case, detection is made at
RF (radio frequency) stage with an RF detector. For the heterodyne case, detection is
achieved at IF (intermediate frequency) stage with an IF detector after frequency
down-conversion with a mixer driven by an LO (local oscillator). It is true that
heterodyne systems can be employed in various receiver configurations, and thus an
IF detector does not need to be located at the end as shown in Fig. 3.5b. The
comparison with the direct detection, however, will be more straightforward by
including an IF detector, which will be assumed here. For both direct and heterodyne
cases, the detection sensitivity can be enhanced by employing a front-end amplifier,
which is a transistor-based circuit as will be discussed separately in Sect. 3.3.3. The
inclusion of an amplifier, however, is optional and will be practiced only when a
sufficient gain is available at the incident RF signal frequency, which is not
guaranteed for high-frequency regimes such as the THz band.
A comparison between the direct and heterodyne detections can be made in terms
of responsivity and noise performances [27]. Let us assume throughout the discus-
sion below that the front-end RF amplifier is not available and thus absent, which is
true for lots of THz detection cases. In this case, the responsivities for the direct and
heterodyne detections cases will be expressed as:
where ℜRF_detector and ℜIF_detector are the responsivities of the RF detector and the IF
detector, and Gmixer and GIF_amp are the conversion gain of the mixer and the gain of
the IF amplifier, respectively. Matched conditions are assumed for all the gain
parameters. There are a couple of factors that would favor the heterodyne detection
from the responsivity point of view. First, the responsivity of typical detectors tend
to show a larger value with lower frequencies, leading to ℜIF_detector typically larger
than ℜRF_detector. Second, the responsivity for the heterodyne detection can be
boosted with the gain factor, GmixerGIF_amp. The included mixer may exhibit a
conversion loss (Gmixer < 1) depending on the mixer type and the frequency
range. Even for such a case, a sufficiently large GIF_amp readily available from the
IF amplifier shall override the mixer loss, leading to a significantly large overall gain
factor. As a result, the heterodyne detection typically shows a higher responsivity.
The heterodyne detection is favored from the noise perspective as well. To
understand this, it will be helpful to exploit the Friis formula [28], which provides
a valuable insight on the noise properties of cascaded multi-stage systems such as
heterodyne systems. The difficulty here, however, is that the noise level of detectors
is conventionally presented in terms of NEP, not the noise factor F or the noise
temperature Tn that go along with the Friis formula. Hence, in principle, the Friis
formula cannot be directly applied to the noise analysis of detectors based on NEP.
The fact that the gain through these types of detectors is nonlinear also adds to the
difficulty in utilizing the formula [29]. However, the intuition provided by the Friis
formula is so useful, and thus, it can hardly be replaced by other analyses. Hence,
here, we will sacrifice rigor in favor of pragmatism to conceptually understand the
noise in detection systems by introducing the Friis formula and employing noise
factors to represent the detector noise level.
According to the Friis formula, as most readers should be aware, the total noise
factor of a cascaded multi-stage system cascaded is given by:
ð F 2 1Þ ð F 3 1Þ ð F 4 1Þ
F total ¼ F 1 þ þ þ , ð3:9Þ
G1 G1 G2 G1 G2 G3
where Fi and Gi are the noise factor and gain of the ith stage. The noise factor F is
defined as the ratio of the input SNR (signal-to-noise ratio) over the output SNR:
SNRin
F¼ : ð3:10Þ
SNRout
Here, the readers are advised not to mix up two parameters, noise factor F (in linear
scale) and noise figure NF (in dB scale). The latter is defined as NF ¼ 10logF (dB).
There is a close relation between F and Tn, and the two noise parameters can always
be converted to the other with simple relations:
108 3 THz Detectors and Related Topics
T0 þ Tn
F¼ and T n ¼ T 0 ðF 1Þ, ð3:11Þ
T0
where T0 is the ambient temperature, for which 290 K is typically assumed. Table 3.1
shows the conversion between Tn and F as well as NF for selected values. The noise
temperature Tn of a component can be interpreted as the temperature at which a
resistor would generate the same level of thermal noise as the component. The Friis
formula can be expressed in terms of the noise temperature as well, which appears
more compact:
T n,2 T T n,4
T n,total ¼ T n,1 þ þ n,3 þ , ð3:12Þ
G1 G1 G2 G1 G2 G3
On the other hand, the noise factor of the heterodyne detection case as depicted in
Fig. 3.5b will be given as follows:
F IF amp 1 F IF detector 1
F heterodyne ¼ F mixer þ þ , ð3:14Þ
Gmixer Gmixer GIF amp
In most practical cases, the gain of the IF amplifier can be made sufficiently large, so
that the noise contribution from the IF detector becomes negligibly small. How about
the noise contribution from the IF amplifier itself? It may or may not be suppressed
by the mixer depending on whether the mixer gain is larger or smaller than unity
(i.e., exhibits gain or loss). Even when the mixer shows a loss instead of gain,
however, the noise contribution from the IF amplifier will not be excessive as the
3.2 Diode Detectors 109
noise level of amplifiers is generally much lower than those of mixers and detectors.
With this reasoning, the noise comparison of the direct and heterodyne detections is
reduced to comparing the relative noise level of the RF detector and the mixer. A
direct comparison is hard because the noise level of detectors is typically given by
NEP and that of mixers is given by F. However, in a rather indirect way, it can be
shown that the required noise performance to obtain a certain level of the output
SNR is more relaxed for mixers in heterodyne detection than for an RF detectors in
direct detection [30], favoring the heterodyne detection for low-noise operation. This
is consistent with the analysis made in [31], where a direct comparison was made
between the direct detection and the heterodyne detection in terms of NEP. On the
other hand, heterodyne detection systems suffer from their own drawbacks. They
include more circuit components, notably a signal source as the LO if integrated,
leading to high DC power dissipation and large area occupation. This will be a
barrier particularly for array detector applications. Also, they tend to show a
narrower bandwidth, as limited by those of the mixer and the IF amplifier. The
narrow bandwidth will degrade the overall sensitivity for some applications, such as
passive imaging, as is often found in astronomy [32].
Here, although a bit of digression, a few brief comments on the unit of dB (as is
used for NF as mentioned above) are made, as they may benefit some readers. As the
readers should be aware, a ratio can be conveniently represented by its logarithmic
value when it varies over a wide range, such as orders of magnitude. Widely used in
engineering (and some science) is its representation in terms of dB (decibel), whose
value is obtained as 10log(x), where x is its linear value. The inclusion of the factor
10 is a result of using decibel as the unit, where “deci” is a Latin prefix meaning
one-tenth. A decibel is a quantity equal to 1/10 of a bel (after Alexander Graham
Bell). Hence, when a parameter is expressed in terms of decibel (dB), the values are
ten times as large as those expressed in terms of bel (B), for example, 10 dB is 1 B,
50 dB is 5 B, and so forth. The unit of B is rarely used despite its simpler definition
(i.e., log(x) when x is the linear value), presumably because the values can be more
accurately represented in terms of dB with one more significant digit available before
the decimal point. One confusion arises when the dB values are obtained with 20log
(x), instead of 10log(x), for certain parameters, such as current, voltage, or field,
while power is always given in 10log(x). Why are there these two different ways to
obtain dB values? In view of the definition of deci-bel, the expression based on 10log
(x) is fundamentally correct. The occasional adoption of 20log(x) is for the sake of
practical convenience in the conversion between those parameters (current, voltage,
field) and corresponding power values. For example, 20-dB voltage gain or 20-dB
current gain equals 20-dB power gain for a fixed load impedance with the adopted
convention. Finally, dB often comes with a suffix, which provides additional
information regarding the nature of the ratio being taken for the logarithmic conver-
sion. Some examples useful for THz researchers include: m for the ratio with
reference to 1 mW (dBm), V with reference to 1 V (dBV), c with reference to carrier
(dBc), i with reference to isotropic case as in antenna gain (dBi), and so forth. You
may invent a new dBx with your own preferred x.
110 3 THz Detectors and Related Topics
The Schottky barrier diode (SBD) is basically a metal-junction diode. Its history can
be traced back to 1874 when F. Braun first reported a rectifying behavior of a metal-
crystal junction [33], which precedes the birth of the p-n diode that occurred only in
the 1930s. Its much earlier appearance can be attributed to its simpler structure
available, a metal tip making a point contact to a crystal. Despite the simplicity,
SBDs came to offer various useful properties that are particularly important for high-
frequency applications, with its structure gradually evolving into much sophisticated
ones. As briefly mentioned earlier in Sect. 2.4.2, it shows a faster switching
operation than p-n junction diodes owing to the smaller junction charge/recharge
time, which is a main characteristic of majority carrier devices. It also benefits from a
smaller forward bias voltage drop. These advantages have led to a wide adoption of
SBDs for various high-frequency applications, which exploit the nonlinear behavior
of SBDs. One is the frequency multiplier (see Sect. 2.4.2), but a more popular
adoption has been with detectors. Interestingly, SBDs can be employed for both
direct detection and heterodyne detection, the former as a square-law detector and
the latter as a mixer. In this subsection, a brief overview of the device characteristics
of SBDs will be introduced, followed by the description of various SBD structures.
Its use cases as high-frequency detectors will be discussed separately for direct
detection and heterodyne detection afterwards.
Let us take a look at the characteristics of SBDs in terms of DC and RF behaviors,
starting with the former. An SBD consists of two regions: metal and semiconductor.
For the semiconductor side of SBDs, both p-type and n-type can be employed, while
n-type is almost exclusively used in favor of its superior speed owing to the higher
mobility of electrons. The energy band diagram for an n-type SBD is shown in
Fig. 3.6. The voltage barrier height ϕB, which is determined by the metal work
function and the semiconductor doping concentration, is a key parameter that
dictates the I–V characteristics. The I–V relation of SBDs shows an exponential
dependence on the bias voltage VD, as is the case for other junction diodes, which is
shown below:
xd
3.2 Diode Detectors 111
VD
ID ¼ IS exp 1 , ð3:15Þ
VT
where IS is the saturation current, VT is the thermal voltage given by kT/q with k for
Boltzmann constant, T for temperature, and q for elementary charge. It is useful to
remember VT ~ 26 meV at room temperature. The unique behavior of SBDs is
included in IS, which can be explicitly shown as:
ϕ
I S ¼ AA exp B , ð3:16Þ
VT
where A is the junction area and A* is the Richardson constant of the semiconductor
material employed for the diode. It is noted that IS has an exponential dependence on
ϕB, which governs the thermionic emission of electrons, making it sensitive to the
metal type and also the semiconductor doping concentration. The saturation current
density of SBDs is greater by a few orders of magnitude than that of p-n diodes in
typical cases. Such a high current level is the key property of SBDs that is derived
from its being a majority carrier device, also resulting in a smaller forward voltage
drop.
SBDs are widely known for their high-frequency operation. The RF or high-
frequency characteristics of SBDs can be represented by the diode cutoff frequency,
which is given by:
1
fT ¼ , ð3:17Þ
2πRS C 0
where C0 is the main junction capacitance component of the SBD measured at zero
bias (cold condition) and RS is the total series resistance. Equation (3.17) assumes a
simple SBD model consisting of a capacitance and a series resistance, although more
detailed SBD device models would involve additional components to account for the
contributions from various parts of the device [34]. fT is basically the inverse of RC
time constant and the high fT values typically found for SBDs are mainly due to a
small RS, which also explains the smaller charging/discharging time as mentioned
earlier. The cutoff frequency of SBDs reported so far has reached beyond 10 THz
with a GaAs SBD [35], while Si SBDs has provided a record fT around 4 THz
[36]. These values are approximately one order higher than the cutoff frequencies of
transistors based on similar material systems. However, the readers should note the
difference between the cutoff frequency of diodes and that of transistors. For
transistors, the cutoff frequency is defined as the frequency where the gain (current
gain for fT and power gain for fmax) reaches unity. Beyond this frequency, therefore,
the transistors cease to behave as an active device. For diodes, the cutoff frequency
does not point to a frequency of a clear physical significance. Instead, it can be
understood as a figure of merit to represent the high-frequency characteristics of
diodes, which simply corresponds to the inverse RC time constant.
112 3 THz Detectors and Related Topics
For the implementation of SBDs, GaAs and Si are the most popular choices for
the semiconductor side of the device. GaAs benefits from its higher electron mobil-
ity, which would further improve the high-frequency operation of SBDs. While
various metals have been employed to match GaAs to form a junction in SBDs,
aluminum (Al) has been the most widely adopted owing to its high thermal stability
as well as the stable contact formation on the surface of GaAs [37]. Gold (Au) is also
among the candidate metals that have been employed for GaAs SBDs. On the other
hand, Si-based SBDs are favored from a practical point of view in that they are
compatible with standard Si process technologies. Although they are yet to be
offered as a standard device from foundry Si processes with a supporting device
model, they can be easily realized by mix-and-matching various process options
available from the commercial technologies, which are widely practiced for Si
SBDs. While Al may serve as the metal region of Si SBDs, silicides are often
adopted owing to their great process compatibility. Figure 3.7a depicts the cross
section of a Si SBD built on the standard Si CMOS technology (adapted from [38]).
An n-well is used as the main body of the SBD to form a Schottky contact, while the
ohmic contact to the n-well is formed through the n+ source/drain diffusion regions.
Optionally, p-well could be used, in which case the ohmic contact will be made with
p+ source/drain regions. Cobalt silicide, widely employed to make ohmic contacts to
highly doped regions, is directly formed on top of the n-well region (lowly doped) to
serve as the metal side of the Schottky junction. The shallow trench isolations (STIs)
provide device isolation as well as a separation between the Schottky and ohmic
regions.
On a different approach, Si SBDs can be built on BiCMOS technologies as well.
In this case, the devices can benefit from process elements that are not offered from
the standard CMOS process. For example, with BiCMOS technologies, a highly
doped subcollector buried layer and the accompanying reach-through implant can be
used to provide contact to the cathode, which is the n-type semiconductor side of the
SBD. Also, the availability of the deep trench isolation (DTI) is another merit of
BiCMOS technology, which offers a better device isolation. An example of SBDs
based on a SiGe BiCMOS technology is shown in Fig. 3.7b (adapted from [39]). It
also relies on the cobalt silicide and n-well for the metal and semiconductor regions
of the SBD, respectively. For the cathode contact, in this particular example, a
dedicated deeper n+ layer was introduced together with an n-well tailored
ion-implanted layer, to support not only SBDs but also p-i-n diodes that were used
as a part of the circuit to be built.
With the device characteristics and structure given above, SBDs can be used for
THz detection, either as an RF detector for the direct detection or as an RF mixer for
the heterodyne detection. For the direct detection, SBDs basically operate as a
square-law detector, whose output voltage is proportional to the square of the
input voltage. Its square-law behavior can be easily derived from the exponential
I–V relation of SBDs mentioned above. With the help of Taylor expansion, the diode
current can be expressed as below, assuming the diode is under a forward bias and
the third and higher order terms are negligible:
3.2 Diode Detectors 113
n + Reach-through
n + Reach-through
STI n+ STI STI n+ STI
Deep Trench
Deep Trench
CoSi2-Si
n-Well
Schottky Barrier
Contact
n-Well Buried n+
(a) (b)
Fig. 3.7 Cross section of SBDs implemented on Si technologies: (a) Si CMOS technology (© 2005
IEEE [38]), (b) SiGe HBT technology (© 2008 IEEE [39])
VD V D V 2D
I D ¼ I S exp ’ IS 1 þ þ : ð3:18Þ
VT V T 2V 2T
With a sinusoidal input voltage of VD ¼ Vin cos ωt, the output current can be
expressed as:
V cos ωt V 2in cos 2 ωt
I D ¼ I S 1 þ in þ
VT 2V 2T
V cos ωt V 2in ð1 þ cos 2ωt Þ
¼ I S 1 þ in þ : ð3:19Þ
VT 4V 2T
If only the DC component is extracted with a proper low-pass filtering, the output
voltage induced across the load resistance RL will be given as:
V 2in I R
V out,DC ¼ I D,DC RL ¼ I S 1 þ 2 RL ’ S 2L V 2in : ð3:20Þ
4V T 4V T
Hence, the output voltage is proportional to the square of the input voltage magni-
tude and thus proportional to the input power. This can also be interpreted as a
rectifying behavior of SBDs, as the rectification process generally gives rise to a DC
component. The detailed analysis may vary depending on the specific operation
case, but the basic operation can be explained in this way. With this response
behavior and the high-frequency properties as discussed earlier, SBDs can be used
for the direct detection of the incoming THz signal power. It would be worthwhile to
mention that an SBD can detect not only the THz radiation incident on an antenna
linked to the SBD, but also the THz signals injected by an adjacent electrical device
integrated on the same chip. This can be compared to the thermal detectors described
earlier that respond only to THz radiation.
114 3 THz Detectors and Related Topics
For the heterodyne detection, the nonlinear property of SBDs used for detectors
can be now exploited for mixing. The approximated response shown in Eq. (3.18)
can be applied to the case where two sinusoidal input signals of different frequencies
are taken as an input. With VD ¼ V1 cos ω1t + V2 cos ω2t, it can be easily shown that
the output current for the mixed frequency components will be given as:
V 1 V 2 ð cos ðω1 þ ω2 Þt þ cos ðω1 ω2 Þt Þ
I D,IF ¼ I S : ð3:21Þ
4V 2T
With a proper low-pass filtering, the lower-side band output voltage induced across
the load RL will be:
I S RL
V out,IF ¼ I D,IF RL ¼ V 1 V 2 cos ðω1 ω2 Þt: ð3:22Þ
4V 2T
Thus, a mixing of the two input signals is achieved with an SBD. Hence, with the
help of the well-known high-frequency behavior of SBDs, a down-conversion can
be readily carried out for THz-band signals. Once the signal is converted down to the
IF spectrum, the resultant IF signal can be detected by a conventional detector that
does not particularly require a very high-frequency operation property. This will
complete the heterodyne detection of the signals at the THz band.
There have been a great number of reports on SBD-based THz detection with
various types of SBDs and for different applications. Most of the reported SBDs
operate at room temperature because the operation of SBDs in principle does not
require a cryogenic temperature, which is a major advantage of SBDs. There are,
though, some reports on cryogenic operation intended for further improved perfor-
mance as well. It is interesting to notice an obvious trend in the reports on SBDs.
Earlier, the reported THz SBDs were mostly based on GaAs and intended for use as a
mixer for heterodyne detection. More recent reports on THz SBDs, however, are
more likely based on standard Si technologies, targeting at the direct detection. This
is an intriguing shift in the points of interest regarding the THz application of SBDs.
As for the direct detection cases, SBD detectors operating in the range of several
hundred GHz routinely show NEPs in the range of 1011 W/Hz1/2. One SBD
detector that operates at 860 GHz showed an NEP of 4.2 1011 W/Hz1/2
[40]. NEP of this range is in fact lower than that of most Golay cells. It is in a
similar range as the better performances of room temperature bolometers, but it
should be noted that these SBDs are based on standard CMOS technology and thus
can be on-chip integrated with other circuits. This integration-friendly nature also
allows SBDs easily incorporated for array detectors.
To discuss the mixer application of SBDs, it should be noted that the noise
performance of SBD mixers is rarely represented by NEP unlike other direct
detectors considered so far. This is because a mixer is not the last device in a receiver
chain that generates a DC voltage (or current) as the final output of the detection
process. In this regard, the noise level of mixers is better expressed in terms of the
3.2 Diode Detectors 115
noise factor F, as the output is still in the form of AC signals. Or, alternatively, they
can be quoted in terms of the noise temperature Tn, as preferred by the traditional
THz community, which has a simple relation with F as shown earlier in Eq. (3.11)
and Table 3.1. In general, not only for SBDs, the noise of devices that have both
input and output as AC signals, such as mixers and amplifiers, will be represented by
F or Tn, while the noise of devices with AC input and DC output, such as detectors,
will be shown in terms of NEP, as mentioned earlier in this section.
The reported noise temperatures of SBD mixers operating at the THz band are
shown in Fig. 3.8. The plot includes only room temperature data for a fair compar-
ison, while there are reports on cryogenic temperatures as well. As the plot indicates,
they typically exhibit noise temperature in the order of 1000–100,000 K. It increases
rapidly with frequency (be aware that the plot is in logarithmic scale), which is a
primary distinct feature from the thermal detectors. The operation has reached
4.75 THz with a noise temperature (double-side band, DSB) measured as
70,000 K, which was achieved with a room-temperature GaAs SBD mixer that
was installed on an airborne observatory for astronomy researches, including the
observation of Orion Nebula as reported in [41].
although a vast majority of SIS structures are used as a mixer for the heterodyne
detection. SIS mixers are particularly popular for radio astronomy applications,
widely adopted for radio telescopes over the world operating below 1 THz. At
ALMA (Atacama Large Millimeter/submillimeter Array), for example, eight bands
(84–950 GHz) out of ten adopt SIS mixers for their heterodyne receiver front-
end [45].
Before the discussion of the operation of SIS mixers, the readers are first directed
to the basic properties of superconductors. When superconductors are cooled down
below a critical temperature TC, a pair of electrons form a Cooper pair. A Cooper pair
is a boson with an integer spin, because two half integers of electrons, which are
fermions, are combined. (In fact, a Cooper pair will have a zero spin because two
electrons with opposite spins will form a pair.) Hence, multiple Cooper pairs can
share an energy state, unlike the case of individual electrons that are governed by the
Pauli exclusion principle. When the Cooper pair gains enough energy from an
external cause, such as non-zero temperature, bias voltage, or external radiation,
that exceeds the binding energy, the pair is broken back to individual electrons
(to Bogoliubov quasiparticles, to be more accurate, which are often called just
quasiparticles for simplicity). The binding energy, typically represented by 2Δ, is
only a few meV. This explains why the superconductivity is observed only at low
temperatures below TC, because otherwise the thermal energy from the ambient
temperature can easily exceed the binding energy and break the Cooper pairs into
quasiparticles.
While Cooper pairs are the particles that are responsible for the famous zero-
resistance in the superconductor materials, it is the quasiparticles that play the key
role in SIS mixers. There used to be early attempts to use Cooper pairs for mixers,
but it gave way to quasiparticle-based approaches in favor of the better noise
performance. Hence, the operation of SIS mixers can be explained in terms of the
quasiparticles [46]. The behavior of quasiparticles can be described by first intro-
ducing the concept of the energy bandgap whose value is equal to the binding energy
of the Cooper pairs, or 2Δ. Unlike the electrons in metals, in which the energy states
are in continuum near the Fermi energy EF, there is a forbidden band for quasipar-
ticles in superconductors around EF. In a sense, this energy bandgap concept is
similar to that of semiconductors, although its bandgap, 2Δ, is much smaller (in the
order of meV) than those of typical semiconductors (in the order of eV). Another
notable difference lies in the profile of the density of states (DOS) near the bandgap
edge. For semiconductors, DOS for electrons decreases as it approaches the bandgap
edge, while DOS for quasiparticles in superconductors increases as it gets closer to
the edge.
Now consider the SIS structure described in Fig. 3.9a, which shows the energy
bandgap and DOS along the energy axis (vertical axis) at the equilibrium [46]. Note
in the plot that the horizontal axis is for DOS magnitude, not the distance as is the
case for the typical semiconductor energy band diagram. Still, the central region
indicates the insulator layer and the two sides represent the two superconductor
layers across the insulator. In this conceptual diagram, it is assumed for simplicity
that the same superconductor materials are used on both sides, which is often (but
3.2 Diode Detectors 117
S I S S I S
Empty
states
Bias
EF ∆ EF
Filled
states 2∆
=0 >
Density of states
(a)
Current (I)
Unpumped S I S
Pumped
EF
∆
hν /q
Density of states
(b) (c)
Fig. 3.9 (a) Energy band diagram of a typical SIS mixer before (left) and after (right) bias is
applied. (b) I–V relation of SIS mixers with and without LO pumping. (c) Photon-assisted tunneling
process at the presence of LO pumping. (Adapted by permission from Springer Nature: [24]
© 2012)
not always) the case. When a bias voltage is applied across the structure, the energy
profiles of the two superconductors is shifted as shown on the right side of Fig. 3.9a.
If the bias barely exceeds the bandgap voltage (2Δ/q), the lower band of one side and
upper band of the other side of the superconductor layers become slightly
overlapped. This will trigger a tunneling of quasiparticles across the thin insulator
barrier. This quasiparticle tunneling leads to an abrupt increase in the current across
the structure, which is by nature a tunneling current.
The abrupt current increase leads to a strong nonlinearity in the I–V curve of the
terminal current, which shows a nearly vertical increase near the critical (or turn-on)
voltage, as is depicted in Fig. 3.9b. On the other hand, when LO radiation (pumping)
is additionally applied, a photon-assisted tunneling will take place, the process of
118 3 THz Detectors and Related Topics
which is depicted in Fig. 3.9c. It will induce a tunneling at voltages lower than 2Δ/q
by an amount that equals the photon voltage hν/q or its multiples, where h is the
Planck constant and ν is the frequency of the photons. In this case, the I–V relation
will be modified, now showing a step-like increase of the current, with a slight
relaxation in the nonlinearity, which is shown in Fig. 3.9b as a solid line in
comparison with the case without LO pumping shown in a dotted line. When the
signal reaches an SIS mixer together with LO pumping, the nonlinearity will cause a
mixing between the incoming RF signal and LO, leading to a down-conversion of
the RF signal. The final detection will be made at IF stage as in the cases with usual
heterodyne detection systems.
The operation frequency of SIS mixers is limited by the bandgap energy of the
superconductors, since the incident radiation with a larger photon energy (and thus
higher frequency) than the bandgap (hν > 2Δ) will prevent the normal operation of
SIS mixers. This implies that the employment of superconducting materials with a
higher bandgap will help to boost the operation frequency. For this reason, NbN
(niobium nitride) or NbTiN (niobium titanium nitride), whose 2Δ are larger than
5 meV (corresponding gap frequency of ~1.2 THz), can be used instead of the rather
standard Nb (niobium) with 2Δ around 2.8 meV (gap frequency of ~680 GHz) if
high-frequency operation is desired. This will increase the fabrication difficulty, but
these high-bandgap materials are still often employed for one or both side(s) of the
junction for operation frequency extension beyond what is covered with Nb. The
critical temperature TC of these superconductors ranges over 9–18 K for bulk
materials. It is known that TC is significantly lowered for thin films, but liquid He
temperature (4 K) is usually low enough to ensure their superconductivity and thus is
widely employed.
For the insulator barrier, AlOx (aluminum oxide) is commonly used, while AlN
(aluminum nitride) is also employed in favor of the film quality in some cases. A
typical cross-sectional view of an SIS mixer is shown in Fig. 3.10 [13]. It is also
noted that SIS mixers are often integrated with a tuning circuit, which will tune out
the intrinsic capacitance from the parallel plate structure of the SIS mixers. As for the
signal feeding, the RF signal and LO pumping, typically coupled by a beam
combiner, are received by an antenna, which can be implemented either on-chip or
off-chip. In the case with an on-chip antenna (quasi-optic approach), the chip with
the SIS mixer is mounted on an elliptical or hyper-hemispherical lens that focuses
the incoming radiation onto the on-chip wideband antenna, such as a log-spiral or
log-periodic, integrated with the SIS mixer. In the case with an off-chip antenna
Substrate
Tunnel barrier
3.2 Diode Detectors 119
LB
LB
Au Au
Nb (normal) Nb (superconducting) Nb (normal)
superconducting materials for superconducting HEBs, Nb, NbN, and NbTiN, are
also shared with SIS mixers. When it comes to the structure, however, there is a clear
difference. While an SIS mixer involves a junction made of a thin insulator and two
superconductor layers, a superconducting HEB is, being a thermal detector, com-
posed of a single superconductor piece, typically in a form of a narrow wire bridge
(called microbridge) with contacts formed on both ends. Hence, one can expect that
the way of utilizing superconductivity for mixing is much different for the two cases.
The structure of a typical superconducting HEB is shown in Fig. 3.12 [50]. The
HEB under mixer operation is cooled down below the critical temperature TC of the
superconductor forming the microbridge. When it is biased with an external DC
source, the microbridge will be heated up by joule heating from the DC power
dissipation, leading to a temperature rise. Excessive heating, additionally promoted
by the radiation of the LO pumping beam, will push the device operation beyond TC,
which triggers a transition from the superconducting state into the normal
conducting state. The resultant sudden increase in resistance will induce a
nonlinearity in the I–V relation of the bolometer, which can be utilized for mixing
of the RF signal and the LO pumping. The IF signal obtained with this heterodyne
process can be detected by a low-frequency detector. As the operation is related to
the transition from the superconducting to the normal conducting states, it is also
called the transition-edge sensor [51]. A more detailed model [52] indicates that a hot
spot will be created in the middle of the microbridge, in which a high temperature
exceeding TC will be induced. This leads to a local normal-conducting region in the
center with the other regions remaining as superconducting. The formation of the
regional hot spot will affect the device current level as the total resistance will be
dominated by the highly resistive hot spot. Because the electrons are heated up faster
than the phonons in the device due to the different heat capacity, the device is called
“hot electron” bolometer. The way the RF signal and the LO pumping are coupled
into a HEB mixer is in most cases similar to the SIS mixer cases described above.
The coupling may rely on a quasi-optic technique for HEB mixers integrated with a
wideband on-chip antenna, or waveguide technique that involves a waveguide
3.2 Diode Detectors 121
transition with an external antenna. In some cases, though, a direct coupling through
an absorption region without antenna is still employed as is the case for conventional
bolometers [4].
The reported performance of superconducting HEBs are shown in Fig. 3.13. As
indicated by the plot, the operation frequencies of superconducting HEBs are
distributed roughly over 0.5–5 THz range. The HEB that marks the highest operation
frequency of 5.25 THz in the plot is based on NbN microbridge embedded in a
log-spiral antenna built on high-resistivity Si substrate, which shows a noise tem-
perature of 1150 K [53]. Figure 3.14a compares the performances of the three diode
detectors discussed so far, SBDs, SIS mixers, and superconducting HEBs, by
overlapping the data points from Figs. 3.8, 3.11, and 3.13. The same set of data
are repeated in Fig. 3.14b, now presented in terms of NF based on the conversion
with Eq. (3.11). As Fig. 3.14 clearly shows, the operation of HEB stretches up to
much higher frequencies than that of SIS mixers, but the noise level of HEB mixers
is still higher than that of SIS mixers at lower frequencies. For example, near 1 THz,
the noise temperatures of superconducting HEBs are reported around 1000 K, while
those of SIS mixers are mostly populated in the range of 200–500 K. Near 100 GHz,
the noise temperature as low as below 10 K is obtained with SIS mixers, which
translates to NF of only 0.15 dB (see Table 3.1 for conversion between noise
parameters). The maximum operation frequency ranges are similar for
superconducting HEBs and SBDs, while the noise level is much higher for SBDs.
This is obviously due to the fact that SBDs mostly operating at room temperature
(only room temperature data included in the plots for SBD), while the operation of
superconducting HEBs as well as that of SIS mixers are based on cryonic temper-
atures. Although rarely reported, SBDs operating at cryogenic temperatures show a
lower noise than room-temperature SBDs, but larger than typical HEBs.
122 3 THz Detectors and Related Topics
The transistor is a versatile device that can provide a vast variety of functions on
electronic systems with various topologies available for transistor circuits. That
versatility reaches the field of signal detection as well, and many different types of
detectors can be realized with transistors, for both direct and heterodyne detection.
The availability of gain, which is the most favored aspect of transistors over diodes,
may not be a necessity when it comes to signal detection. However, it still helps
various aspects related to the detection, and the enhancement of transistor speed, or
the frequency limit of the gain availability, contributes to the THz detection based on
transistors. In this section, the direct detection with transistors will be first
3.3 Transistor Circuit Detectors 123
introduced. Then, the key circuit blocks of the heterodyne detection, such as mixers
and low noise amplifiers (LNAs), and their implementation by transistor circuits will
be discussed. This will be followed by the description of integrated heterodyne
detectors and the examples. Although LO is also a core part of heterodyne architec-
ture, it will not be discussed in this section as it can be realized by the transistor-
based signal sources described in Sect. 2.3.
Direct detectors based on transistors can be largely divided into two groups: active
mode and passive mode. In the active-mode detectors, which can be implemented
with either FETs or BJTs, the detection is made exploiting the nonlinearity of the
transistor operation, typically realized with a pair of transistors in the differential
configuration. The topology of a typical active-mode direct detector is shown in
Fig. 3.15, for both FET and BJT cases, in the differential configuration (the figure
shows the input applied through the gate, but it can be applied through the source as
well). It is basically a differential pair with the two drain (or collector) nodes tied
together, which serves as the output node. One may notice that it shares almost the
same topology as the frequency multiplier with the even harmonics described in
Sect. 2.4. In fact, the two operations are very similar in their nature. Only difference
is that, for the output, the DC component is taken for direct detectors while the even
harmonic component is extracted for frequency multipliers, with a proper filtering
network. The distinct feature of the active-mode detection in contrast with the
passive-mode detection is that the transistor is fully biased typically with the gate
(base) and drain (collector) biases. For the passive-mode detection, as shown in
Fig. 3.15 Topology of a typical active-mode direct detector: (a) FET-based, (b) BJT-based
124 3 THz Detectors and Related Topics
Fig. 3.16 (a) Topology of a typical passive-mode direct detector based on FETs. (b) Simplified
single-ended picture of the passive-mode detector, showing Cgd for resistive self-mixing
Fig. 3.16a based on FETs, only gate is biased (not explicitly shown in the figure)
without drain bias.
A brief small-signal analysis is made below to help clarify the operation of the
active-mode detection with the circuit shown in Fig. 3.15a. Let iD1 and iD2 be the
small-signal drain currents through M1 and M2, respectively, and iD the sum of
the two drain currents at the output. At the presence of the transistor nonlinearity,
where it was assumed that vGS2 ¼ vGS1 with an ideal differential operation. Note
ðnÞ
that gm is given as:
1 ∂ I D
n
gðmnÞ ¼ : ð3:25Þ
n! ∂V nGS V ¼const
DS
Now assume the input signal is sinusoidal, so that vGS1(t) ¼ VRF cos ωt. Then,
which suggests that only DC and the second harmonic (and even harmonics if higher
order terms were included) terms survive at the common node. Odd harmonics will
vanish at this node as they are canceled out due to the differential operation. With a
proper low-pass filtering, a DC voltage will appear across the load resistance RL
(neglect RC or RD for now) as follows:
This clearly indicates the circuit serves as a square-law detector. For MOSFETs, the
ð2Þ
explicit form of gm can be obtained from the drain current expression assuming a
long channel device:
1 W
I D ¼ μCox ðV GS V th Þ2 , ð3:29Þ
2 L
where W and L represent the gate width and length, respectively, μ is the channel
electron mobility, and Cox is the gate oxide capacitance per unit area.
ð2Þ
With this current relation, gm is given as:
2
1 ∂ ID 1 W
gðm2Þ ¼ ¼ μC : ð3:30Þ
2 ∂V 2GS 2 ox L
Then the output voltage can be expressed in terms of the device parameters:
μC ox RL W 2
V out ¼ V : ð3:31Þ
2 L RF
The square-law relation to the magnitude of the input voltage is obvious from the
expression. The dependence on the device parameter is mostly as expected.
The same analysis can be applied to the BJT case shown in Fig. 3.15b. The output
voltage can be accordingly obtained by applying the expression for the collector
current in place of the drain current. By assuming a one-side step junction with
emitter doping much higher than base, which is often the case, the collector current
for an npn bipolar transistor is given as:
126 3 THz Detectors and Related Topics
V BE qADn n2i V BE
I C ¼ I S exp ¼ exp , ð3:32Þ
VT NBW B VT
where A is the emitter–base junction area, Dn is the electron diffusion constant in the
base, ni is the intrinsic carrier concentration in the base, NB is the base doping
concentration (assumed uniform or average value), and WB is the neutral base width.
ð2Þ
Then, based on Eq. (3.25) with VGS replaced by VBE, gm can be obtained as:
IC qADn n2i V BE
gðm2Þ ¼ 2 ¼ exp , ð3:33Þ
2V T 2N B W B V 2T VT
Again, the square-law relation is clearly shown. One notable difference from the
FET case is that the output voltage is proportional to the DC bias current, or the
collector current IC for BJT case.
To employ the resistive mixing for the passive-mode detection with FETs, the RF
signal for detection should be applied to both gate and drain nodes. For this to
happen, a strong AC coupling between the gate and the drain needs to be established
with the RF signal applied to the gate. The coupling can be achieved with the
parasitic gate-drain capacitance Cgd alone, or it can be further enhanced with an
addition of external capacitance across the gate and the drain. Note that this situation
is opposite to the normal mixing operation of resistive mixers where the coupling
should be suppressed, which may require LO and RF filters for isolation instead of
coupling. An analysis can be made to explain the detector operation based on the
self-mixing as presented in [55]. Consider the simplified circuit schematic shown in
Fig. 3.16b, in which Cgd may be purely parasitic or enhanced by extrinsic capaci-
tance. Assume that the RF signal injected to the gate is sinusoidal, so that the total
gate voltage is given by vGS(t) ¼ VRF cos ωt + Vg, where Vg is the DC gate bias. Due
to the capacitive coupling, the AC component of the gate voltage will appear at the
drain node, leading to vDS(t) ¼ VRF cos ωt with no drain DC bias applied. Since the
RF signal is small enough to keep the device to stay in the triode region in most
practical detector applications, the drain current will be given as a product of the
voltage and the conductance across the drain and the source, leading to:
Note that gDS(t), the conductance between the drain and the source, is basically the
channel conductance, which is approximated in strong inversion as [55]:
W v ðt Þ
gDS ðt Þ ¼ μC ox vGS ðt Þ V th DS : ð3:36Þ
L 2
With vGS(t) and vDS(t) as given earlier substituted to Eq. (3.36), one can obtain the
DC component of the drain current as:
μCox W 2
ID ¼ V : ð3:37Þ
4 L RF
If the output is terminated with a load resistor of RL, then the output voltage will be
simply:
μC ox RL W 2
V out ¼ V : ð3:38Þ
4 L RF
quasi-static operation that applies to the THz detection, in which the channel is
represented by distributed RC components. The result for non-quasi-static operation
as presented in [55] indicates that detection is still achieved, more effective with a
shorter channel length. For devices with a longer channel than needed for the
resistive mixing, the extra channel length will only add capacitance and series
resistance, degrading the noise and responsivity of the detector.
There is yet another mode of transistor-based direct detection, widely known as the
plasma wave resonant-mode detection. The principle behind this phenomenon, as
suggested by Dyakonov and Shur [56], is briefly introduced here. If the mean free
path for electron-electron scattering in the channel of an FET is shorter than those for
electron-impurity or electron-phonon scatterings, the total scattering is dominated by
electron-electron scattering. In this case, the electrons in the channel will behave as if
they are 2D fluid rather than 2D gas. This condition can be achieved with a channel
with a high electron mobility and high electron concentration. In this situation, the
electrons can be excited to generate waves inside the channel, which is called the
plasma wave. With the presence of the plasma wave, the channel can serve as a
resonator with a proper set of boundary conditions. Assuming the boundary condi-
tion of the source and the drain being AC-wise short and open, respectively, the
resonance will be obtained at the fundamental frequency f0 shown below and its
odd-harmonics:
s
f0 ¼ , ð3:39Þ
4L
where L is the channel length and s is the velocity of the plasma wave in the channel.
The velocity s can be expressed as a function of device parameters as:
rffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffi
qV 0 q2 n
s¼ ¼ , ð3:40Þ
m m C
where q is the elementary charge, V0 is the gate overdrive (Vg Vth), m* is the
effective mass of electron, n is the channel electron concentration per unit area, and
C is the gate capacitance per unit area. The equality on the right-hand side is based
on the assumption that the device is in the strong inversion region and n CV0/q
[57]. Since s is given as a function of device parameters as well as the bias condition,
its value and thus the resonant frequency can be tuned by the external gate bias. It is
noted that a typical value of s is in the order of 108 cm/s, larger than the electron
saturation velocity in most semiconductors by a factor of an order. This leads to the
fundamental resonant frequency in the order of THz, opening the possibility of THz
detection. For this resonance to show a sufficiently large Q-factor, the channel length
L should satisfy the following condition:
3.3 Transistor Circuit Detectors 129
106 106
102 102
0 10 20 30 0 2 4 6 8
0.05 μm
Responsiv ity (V/W )
104 103
0.8 m2 / Vs
103
102
102 0.25 μm
0.8 m2 / Vs
101 101
0 10 20 30 0 2 4 6 8
Frequency (THz) Frequency (THz)
Fig. 3.17 Calculated responsivity profiles for the resonant-mode detection of plasma-wave detec-
tor with various gate lengths and mobilities. (© 1996 IEEE [56])
sμm
L Lcr ¼ , ð3:41Þ
q
where Lcr is the critical channel length and μ is the electron mobility. Hence, for the
resonant-mode operation of the plasma-wave FET, a sufficiently small channel
length is also required. This is an additional condition that adds to the conditions
for the existence of the plasma wave, or high mobility and high carrier concentration
as mentioned above.
When all these conditions are satisfied, a plasma-wave resonant-mode detection
can be obtained at the resonant frequencies. The signature of the resonant-mode
detection is the periodic occurrence of sharp peaks in the responsivity with very large
values at the resonant frequencies ( f0 and its odd harmonics) along the frequency
axis. Figure 3.17 shows calculated responsivities with different channel length and
mobility conditions that clearly exhibit such peaks, indicating the onset of the plasma-
wave resonance [56]. The profiles in Fig. 3.17 implies that a highly sensitive narrow-
band THz detection can be achieved if the above-listed conditions for the plasma-
wave resonance are satisfied. Those conditions are quite stringent and hard to achieve,
limiting the early demonstration of the resonant-mode detection to FETs based on III–
V semiconductors [58]. However, the evidence of possible resonant-mode plasma
wave detection with Si MOSFETs has been subsequently reported with aggressively
scaled gate length [57]. While the resonant-mode detection is narrow-band by nature,
less sensitive broadband detection can still be made with a relaxed condition for the
130 3 THz Detectors and Related Topics
Fig. 3.18 Reported NEP of transistor-based direct detectors with various technologies, shown
together with selected thermal detectors for comparison (room temperature data only). For Golay
cells and pyroelectric detectors, only the noise range is presented assuming the responses are
frequency-independent
NEP for transistor-based detectors drops rapidly, reaching as low as below 1015 W/
Hz1/2 around 100 GHz (with the help of an LNA) [60], while microbolometers do
not show a significant drop. Hence, it is this region of frequency, roughly below
200 GHz, where transistor-based detectors exhibit a huge advantage over the room-
temperature thermal detectors. In fact, the NEP values exhibited by the transistor-
base detectors around 100 GHz at room temperature is comparable to some of
bolometers operating at cryogenic temperatures. Further, when integrated with a
low noise amplifier, the noise level will be further reduced. This will be discussed
shortly later in this section.
3.3.2 Mixers
It will be useful to take a brief review of the basic mixer operation principles before
moving into the details of various mixer circuit topologies, although we have already
partially started discussion on mixers in the previous sections. Mixing can be
obtained by either nonlinearity or time-varying response of a system. These two
mechanisms will be explained by taking RF and LO as the two inputs and IF as the
output, as is the convention for down-conversion mixers.
For the mixers based on a nonlinearity, the sum of RF and LO signals, possibly
through a power combiner located at the front, enter the mixer and go through the
nonlinear process. Assume a two-tone sinusoidal input signal of
Then, one can easily show that the resultant output will have the following frequency
components: ω0, ω1, 2ω0, 2ω1, ω0 + ω1, ω0 ω1, 3ω0, 3ω1, 2ω0 + ω1, 2ω0 ω1,
2ω1 + ω0, 2ω1 ω0. In general, at the presence of the higher order terms, the
following frequency components will be obtained:
where n and m are integers. The desired frequency component can be selected with a
filter at the output. Hence, any nonlinear system can serve as a mixer if two input
signals are applied together.
Now consider mixers based on a time-varying response. If we assume the system
is linear although time-variant, the response can be shown as below with an input of
x(t):
When the time variance of the system, α(t), is periodic with an angular frequency of
ω0, the response can be expressed with the help of Taylor expansion as:
X
1
yð t Þ ¼ αn ejnω0 t xðt Þ, ð3:46Þ
n¼1
X
1
A1 X
1
yðt Þ ¼ αn ejnω0 t A1 cos ω1 t ¼ αn ejnω0 t e jω1 t
þ ejω1 t : ð3:48Þ
n¼1
2 n¼1
Again, a filtering at the output will provide only the desired frequency compo-
nent. Hence, a linear time-varying system can serve as a mixer if one of the input
signals, typically LO, induces a periodic time-variance in the system response. A
nonlinear time-varying system can also serve as a mixer, of course, in which case the
output will include frequency components of |nω0 mω1| with an additional integer
of m. If the response variation over time is extreme so that the system periodically
turns on and off, then we can say the mixing is due to switching.
3.3 Transistor Circuit Detectors 133
Transistors are versatile devices, and they can be assembled in many different ways
to form circuits intended for certain functions. Transistor-based mixers are no
exception, and they can be realized based on various approaches. Depending on
the mode of operation, transistor mixers can be divided into passive mixers and
active mixers. For passive mixers, transistors are typically not fully biased and
operate in the absence of DC bias current. One example is the resistive mixer briefly
mentioned in the previous section [54], which belongs to the case where mixing is
based on a time-varying response. Another example is a circuit shown in Fig. 3.19,
in which a transistor acts as a switch driven by LO that controls the RF signal path,
probably the simplest implementation case of a switching mixer. In this configura-
tion, the transistor does not need the drain DC bias current for switching, and thus
involves no DC power dissipation, which is a major advantage of passive mixers.
Passive mixers also exhibit an excellent linearity, a signature of passive operation,
which stems from the absence of the saturation at high power levels as is generally
observed in active circuits. Noise level can be maintained low owing to the reduced
1/f noise that often plagues active devices. Let us take a brief digression here. There
may be a notion that using transistors for passive-mode circuits is a waste of the
precious functional resources that can be provided by the transistors employed in the
circuit, such as the gain. However, ironically, the transistor is the cheapest device
available on a chip when fabricated in a standard semiconductor technology, since it
occupies the smallest area. True expense for using transistors comes from the DC
power dissipation, rather than from the area, at least for sparsely integrated analog
and RF circuits. Hence, there is no particular economic loss in employing transistors
for passive circuits.
While passive mixers based on transistors are suited for some occasions, active
mixers are still largely favored for most applications whenever they are available.
The reason for this favor is the availability of the gain, the hallmark feature of
transistors, for this type of mixers. With a conversion gain (larger than unity), the
noise from the following stages will be suppressed by the mixer gain, as is indicated
by the Friis formula shown in Eq. (3.9). Note that, in passive mixers, the noise from
the following stages will be effectively magnified by the conversion loss, as opposed
to the active mixer case. Additionally, a conversion gain will contribute to the total
gain of the heterodyne detector, enhancing the responsivity of the entire detector. It
is still true that a conversion gain larger than unity cannot be always guaranteed for
active mixers, especially for very high-frequency operations due to the gain roll-off
Vg
(a)
(b)
3.3 Transistor Circuit Detectors 135
most cases, periodically alters the bias of the transistor up and down. This in turn
periodically changes the response of the transistor stage, or the transconductance of
the transistor, that works on the RF signal, resulting in mixing between the LO and
the RF signals. In this view, one can consider this mixing as resulting from the time-
variance of the transistor response as driven by the LO signal. Or, at the same time, it
can be interpreted as stemming from the nonlinearity of the transistor stage, which is
inherent in transistors operating in the normal operating condition (forward active
mode for FETs and saturation mode for BJTs). Dual-transistor mixer shown in
Fig. 3.20b have separate inputs for RF and LO. In this scheme, the drain current
that carries the RF signal amplified by M1 is switched by M2 driven by the LO signal,
leading to mixing between the two input signals. Hence, this is another example of
switching mixers operating in the active-mode.
Much more popular among the active mixers, especially with Si-based technol-
ogies, are differential mixers, examples being shown in Fig. 3.21. Differential
mixers, or balanced mixers as sometimes referred to, include one or more differential
pair(s), and generally grouped into two categories, single-balanced and double-
balanced. For a single-balanced mixer, as depicted in Fig. 3.21a, a single-ended
136 3 THz Detectors and Related Topics
RF signal enter the tail transistor M1 at the bottom (transconductance stage), while a
differential LO signal is applied through the differential pair M2/M3 at the top
(switching stage). Similar to the dual-transistor mixer described above, the LO
differential pair switches the RF signal riding on the drain current of M1. The
difference is that the differential LO signal alternatively turns on and off, switching
transistors M2 and M3, leading to a differential IF output signal. For the positive
cycle of LO, RF signal is delivered to the positive node of the IF output, while for the
negative cycle of LO, RF is delivered to the negative node of IF. For the double-
balanced mixer shown in Fig. 3.21b, which is commonly called Gilbert mixer, both
RF and LO are applied as differential signals. Hence, the transconductance stage that
accepts RF signal is formed by a differential pair (M1/M2), and the switching stage
driven by LO is composed of two sets of differential pairs (M3/M4 and M5/M6), each
of which controls the drain current of M1 and M2. The drain nodes of the two
switching differential pairs are cross-coupled to generate a differential IF signal. For
the positive and negative cycles of LO, RF signal is delivered to IF output in-phase
and out-of-phase, respectively, by a current-commutating action [62, 63]. Double-
balanced mixers benefit from the smaller LO-to-IF leakage, because the DC com-
ponent of the RF signal is absent unlike the single-balanced case, eliminating the
mechanism to carry LO to the output. They also show a less even-order distortion.
However, their noise level is higher due to more active components included and the
voltage budget is tighter with an additional stage stacked (if a tail transistor is
employed as a current source) when compared to the single-balanced mixers.
Up to this point, a rather general overview of RF mixers has been provided. For THz
mixers, the biggest challenge stems from the limitation in the power and frequency
of the LO signal. This issue can be addressed with the introduction of subharmonic
mixers. As Eqs. (3.44) and (3.49) indicate, a mixing can be made between the
harmonics of RF and LO signals, with either nonlinear or switching (or time-varying
response) mixers. This implies that IF signal frequency does not need to be the
difference between the actual frequencies of the LO and the RF signals as in the case
of the fundamental mixer. It can be difference between the integer multiples of RF
and/or LO signal frequencies, and such a mixer is called the subharmonic mixer. For
practical purposes, the case that takes the harmonics only from LO signal is almost
exclusively employed, such that ωIF ¼ |nωLO ωRF|. In this case, the requirement
for the LO frequency is relaxed by a factor of the harmonic number n, which is a
great relief for THz operations. Also, there is a good chance that a higher LO power
level will be obtained at the lower frequency. Its own challenge, however, lies in the
degradation in the conversion gain and the noise level. An analysis in [27] shows that
the conversion gain and the noise can be degraded by several dB when second-order
subharmonic mixing is introduced instead of fundamental mixing for a 300 GHz
mixer assuming a fixed LO power. Despite the expected performance degradation,
however, subharmonic mixing is a widely practiced solution in many THz-band
3.3 Transistor Circuit Detectors 137
VLO_Q-
VLO_I+
VLO_I-
VRF+ VRF-
Another challenge for mixers operating at the THz band is the low conversion gain.
Even with the active operation, the lowered transconductance of transistors at raised
frequencies affects the availability of sufficient conversion gains for THz mixers.
One frequently adopted technique to enhance the conversion gain of active mixers is
138 3 THz Detectors and Related Topics
the Gm-boosting technique, which effectively boosts the Gm of the mixer core. There
are various ways to boost Gm, one of which is the cross-coupling technique that can
be applied to the common-gate or common-base differential configurations. One
example is presented in Fig. 3.23, which shows a D-band SiGe HBT mixer with the
cross-coupling technique [71, 72]. The cross connection between the base of Q1 (Q2)
to the emitter of Q2 (Q1) in the differential pair enhances the VBE swing of the
transistors. This happens because, for the positive cycle applied to Q1 (Q2) base, Q2
(Q1) base is biased with the negative cycle, pulling down the potential at the Q1 (Q2)
emitter due to the coupling, which results in increased VBE1 (VBE2) for every cycle.
The enhanced VBE drive effectively boosts Gm of the differential pair and thus the
conversion gain of the mixer. In another view of the same mechanism, it can be
interpreted such that the cross-coupling provides a feedback with a gain of A,
which results in Gm boost by a factor of (1 + A). The boosted Gm will help to improve
the noise performance of the mixers in addition to the conversion gain enhancement.
There are other techniques to boost Gm in mixers, such as back-gate connection and
current reuse configuration adopted in [73]. It is also noted that the Gm boosting is a
general technique that can be widely adopted for various types of high-frequency
circuits other than mixers as well, including amplifiers [74] and detectors [27].
There have been quite a lot of mixers reported for operation beyond 100 GHz.
Figure 3.24 presents the measured noise figure of mixers shown as a function of
frequency based on various technologies (the plot includes the cases where addi-
tional circuits, such as a buffer amplifier, are included after the mixer, as they do not
affect the noise performance significantly and, in a sense, they can be considered as a
part of the mixer circuit). Depending on the mixer configuration, NF as low as
~10 dB has been reported around 200 GHz [75, 76], while it rapidly increases with
the frequency leading to ~50 dB near 800 GHz [77].
3.3 Transistor Circuit Detectors 139
Fig. 3.24 Reported noise figure of THz mixers based on various technologies
As for the amplifier topology, CE (CS) or cascode configurations have been widely
adopted for high-frequency LNAs owing to their favorable noise properties, which
trend also applies to THz LNAs. The adoption of CB (CG) for THz LNA is less
popular because its noise performance is not as good as those with the other topology
options, despite its known advantage of wideband performance that arises from the
absence of the Miller capacitance. One aspect that should not be overlooked,
however, is that the gain of CB (CG) is larger than that of CE (CS). This fact is
particularly relevant to the noise-gain balance just discussed above, since the high
gain of CB (CG) should help the suppression of the noise from the following stages.
In fact, there is an analysis indicating that a two-stage CB amplifier shows a lower
noise and higher gain than those of cascode, apparently owing to the higher gain of
CB than CE at the front stage [78].
The issues surrounding the differential vs. single-ended amplifiers for THz
operation were discussed in Sect. 2.5.1, and they are valid for LNA cases as well.
To repeat, the differential topology will be preferred from the design and perfor-
mance aspects, such as common-mode noise rejection and the availability of the
virtual ground, while the interface with single-ended antennas and test environments
may favor the single-ended topology although integrated baluns can always be
considered. Particularly for LNAs, their outputs often drive mixers that follow in
heterodyne structures, so it is affected by the topology of mixers widely employed.
From this aspect, it is not surprising to see that more of Si-based LNAs are designed
in differential topology while III–V LNAs are mostly in single-ended scheme,
consistent with the trend in mixer topologies. In this regard, it is clear that III–V
community is more influenced by the legacy from the hybrid designs based on
individually packaged components.
The noise characterization of LNAs (and receivers) at the THz band is challenging,
and a discussion on this topic will benefit the readers. The noise figure (or noise
factor), a widely adopted parameter that represents the noise level of two-port
networks including LNAs, is usually obtained with a measurement employing a
noise figure analyzer and a noise source. A mixer may be optionally included to
down-convert the frequency into the range covered by the noise figure analyzer
when the operating frequency is higher than the range. However, noise sources
operating at the THz band are unfortunately not readily available as a commercial
product, currently limited below 200 GHz. Without noise sources available, which
are usually compact and convenient for use, one needs to resort to a bulky physical
setup that produces noise at multiple temperatures required for the Y-factor method, a
standard method to extract the noise figure. A brief introduction to the Y-factor
method and the measurement setup is provided below [79, 80].
3.3 Transistor Circuit Detectors 141
The starting point of the Y-factor method is the expression for the noise power Pn
measured at the output of a system to be characterized in the absence of a signal,
which is given by:
Pn ¼ Gk T ref þ T sys B, ð3:50Þ
where Tref is the reference temperature faced by the input of the system, Tsys is the
noise temperature of the system, G is the system gain, and B is the system bandwidth.
k is, of course, Boltzmann constant. According to Eq. (3.50), Pn is given as a linear
function of the reference temperature, as is conceptually drawn in Fig. 3.25. Remem-
ber that the main object of the Y-factor method is to obtain Tsys, and this requires the
linear function to be determined first. Obviously, this can be achieved if we know the
noise power levels at two different reference temperature points such that:
Phot ¼ Gk T hot þ T sys B ð3:51Þ
Pcold ¼ Gk T cold þ T sys B, ð3:52Þ
where the subscripts hot and cold represent the two temperatures employed. Now,
the Y-factor is defined as the ratio between Phot and Pcold, thus leading to:
T hot YT cold
T sys ¼ : ð3:54Þ
Y 1
Based on Eq. (3.54), one can obtain Tsys with the knowledge of the Y-factor as
well as the two temperatures Thot and Tcold. A measurement is needed to determine
142 3 THz Detectors and Related Topics
the value of the Y-factor by evaluating the noise power at the system output with the
injection of external noise corresponding to the two reference temperatures. With the
obtained Y-factor and the temperature values, the noise temperature of the system
Tsys will be decided. If noise figure is preferred as a noise parameter instead of the
noise temperature, a simple conversion can be made with a relation Tn ¼ T0(F 1)
as mentioned earlier (Eq. 3.11). The elegance of the Y-factor method lies in the fact
that no information on system gain G or bandwidth B is needed, because it is the
ratio of the noise powers that matters, which is independent of G or B, as indicated
by Eq. (3.53). In fact, the system gain G is a parameter that can be simultaneously
determined from this procedure because the slope of the curve in Fig. 3.25, or GkB,
is proportional to the gain. One additional information needed to accurately deter-
mine G from the slope, though, is the value of B. To ensure a control over B, a filter
of well-defined bandwidth is typically inserted in the measurement setup. As for the
reference temperatures, the room temperature (around 300 K) and the liquid nitrogen
temperature (77 K) are typically employed for their easy availability, although a
larger temperature difference would result in a more reliable measurement with a less
fluctuation in the data.
A typical setup for noise measurement with the Y-factor method is illustrated in
Fig. 3.26a. A cryogenic container with liquid nitrogen (LN2) and a moving shutter
covered with an absorber are placed in front of the DUT (Device Under Test), which
is interfaced with a horn antenna at the input and a subharmonic mixer at the output.
This is followed by a low-frequency LNA and a band-pass filter. With this config-
uration, the DUT will face the room-temperature absorber when the shutter is closed
(the hot-load state) while it looks into LN2 through a parabolic mirror when the
shutter is open (the cold-load state). By measuring the noise power at the output with
these two states, the Y-factor can be obtained and thus Tsys can be determined. To be
more precise, the obtained Tsys with this process is not for the DUT alone but for a
structure that comprises the DUT as well as the probe and post-DUT components.
Hence, the extraction of Tsys for the isolated DUT will require an additional
calibration measurement that should be made in the absence of the DUT while
maintaining the rest based on otherwise the same setup. Also, when the noise power
difference between the two states are not large enough, which occurs when the noise
level of the device is relatively high, repeated measurements are necessary to extract
the reliable averaged value. It is noted that the conventional noise measurement
setup with a noise figure analyzer and a noise source is in fact based on the Y-factor
method, too, in which turning on and off the noise source emulates the hot and cold
states in the Y-factor method, respectively.
There is another method to extract the noise figure of LNAs that merits a brief
review here, too. Called the signal generator N-times power method, it makes use of
the ratio of the output power of the system with and without the input signal at a fixed
temperature, most likely the room temperature [80–82]. The measurement setup for
the signal generator N-times power method is depicted in Fig. 3.26b. It is similar to
the setup used for the Y-factor method, the main difference being the front cryogenic
part replaced with a signal generator. As it is performed at a single fixed temperature,
it alleviates the need for a low-temperature measurement and thus can be
3.3 Transistor Circuit Detectors 143
Moving
Computer
stage
Shutter
Parabolic
mirror Mixer
Power
DUT
Meter
Horn
BPF
antenna LNA
Freq.
Multiplier
Liquid
Nitrogen
Signal source
(a)
Mixer
Freq. Power
DUT
Multiplier Meter
Signal BPF
source LNA
Freq.
Multiplier
Signal source
(b)
Fig. 3.26 Setups for noise measurement: (a) Y-factor method with cryogenic temperature, (b)
Signal generator N-times power method
where Psig is the input signal power. From these relations, one can easily arrive at the
following expression for the system noise temperature:
144 3 THz Detectors and Related Topics
Psig
T sys ¼ T ref : ð3:57Þ
kBðPon =Poff 1Þ
Hence, with the values of input power Psig, two output powers Pon and Poff, and the
system bandwidth B, the noise temperature Tsys can be obtained. It is worthwhile to
note that Eqs. (3.54) and (3.57) are in fact equivalent if the following relation is
valid:
Psig
T hot ¼ þ T cold : ð3:58Þ
kB
The biggest advantage of using transistor-based circuits for detection is that all the
circuit components needed to build a detector circuit (especially that of heterodyne
detector) can be integrated and implemented in a single chip. Also, if desired, they
can be further integrated with other part of electronics, including digital parts, to
enhance the function of the chip. This integration feasibility is the most outstanding
feature of the transistor detector circuits when compared to other approaches based
on thermal detectors and diode detectors. The major limitation for the transistor-
146 3 THz Detectors and Related Topics
Fig. 3.28 Comparison of reported noise figure of diode-based detectors including SBDs, SIS
mixers, superconducting HEBs, and transistor circuits including mixers and LNAs. The data points
are taken from Figs. 3.14, 3.24, and 3.27. Data for SIS mixers and superconducting HEBs were
obtained at cryogenic temperatures. The others are all room-temperature data
based approach has been the upper frequency limit for active operation (with gain),
which is being gradually lifted with the advent of transistor technologies with a
device speed much faster than before. The availability of discrete circuit compo-
nents, though, does not guarantee the successful operation of integrated heterodyne
detectors as there exist additional design considerations in stitching them into one
piece meticulously. Besides the matching between the circuit blocks, the configura-
tion (differential or single-ended) needs to be aligned, which may require additional
balun circuits. The harmonic order of the mixer needs to be decided based on the
available power and frequency of the LO. Although rarely happens at the THz band,
the signal saturation also requires attention. Satisfying all these requirements is
challenging, but a growing number of integrated heterodyne detectors are being
reported.
It would be useful to review an example of integrated heterodyne detectors, as
presented Fig. 3.29 [27]. In this design, a 300-GHz SiGe HBT heterodyne detector is
composed of an on-chip antenna, a mixer, an oscillator as LO, an IF amplifier, and an
IF detector, all integrated together on a chip. LNA is not included since an amplifier
operating at 300 GHz was not readily available with the SiGe HBT technology used
in this work. The generation of the 300-GHz signal for LO, however, was feasible
with the help of the push-push technique employed to attain the second harmonic
signal based on a differential Colpitts oscillator. Owing to this 300-GHz LO
available on-chip, the mixer was designed to operate at the fundamental mode,
which would benefit from the lower noise and higher conversion gain compared to
the cases with subharmonic mixers. Since the oscillator output is single-ended as is
typical with the push-push configuration, the mixer adopted a single-balanced
topology that accepts differential RF and single-ended LO inputs. The two-stage
3.3 Transistor Circuit Detectors 147
Fig. 3.29 Schematic of the 300-GHz SiGe HBT heterodyne detector circuit
example including an LNA can be found in [79]. The 300-GHz heterodyne receiver,
based on a III–V HBT technology, integrates a five-stage differential common-base
amplifier with a double-balanced fundamental mixer driven by an on-chip 300-GHz
fundamental-mode oscillator for LO signal generation. The availability of the
300-GHz amplification and 300-GHz fundamental-mode signal generation can be
attributed to the superior performance of III–V technologies over Si-based technol-
ogies. The receiver showed a peak conversion gain of 26 dB and a noise figure of
12.0–16.3 dB around the operation frequency of 300 GHz.
In the previous sections, we have reviewed various detectors that can be used for
THz wave detection. Those detectors can be used as a stand-alone individual
detector, or employed as part of a group of multiple detectors operating collectively.
The former case is widely referred to as a single-pixel detector, a term apparently
originated from imaging applications, while the latter is called an array detector or a
detector array. The array detectors are a highly popular option for imaging applica-
tions, while spectroscopy and communication systems can benefit from them, too.
Let us briefly review the merits of array detectors. Essentially, array detectors can
improve the effectiveness in the detection process over the single-pixel counterpart,
by converting the serial process in the time domain into the parallel process in the
spatial domain in detection. For example, in imaging applications, single-pixel
detectors will need a scan of the object (or, alternatively, scan of the detector) to
acquire 2D images, but array detectors will remove the scan step or at least reduce
the number of scans, depending on the array size (pixel count). For spectroscopy,
array detectors (usually 1D array in this case) will allow the simultaneous detection
of a wide spectrum spread over a distance, alleviating the need for mechanical
scanning of the detector that would be needed with single-pixel detectors. In fact,
communication systems will also benefit from the array detectors. MIMO (multiple-
input and multiple-output) systems [93] include multiple receiver antennas, together
with multiple transmitter antennas, for improved channel capacity of the data link,
which can be considered as one kind of array detectors. The discussion below,
though, will assume array detectors largely intended for imaging applications. It
should be mentioned at this point that another advantage of the array detector is the
capability of steering the sensing direction (reverse of beam forming) by controlling
the phase of each pixel, or by phased array. This topic, however, will not be treated in
this section, as it was covered in Sect. 2.6 as a separate topic.
3.4 Array Detectors 149
Despite the apparent advantages briefly mentioned above, there are various issues in
realizing detector arrays. First, there is an issue in the power level received by each
pixel. For an array detector, the total power incident on the detector is spread over the
pixels, leading to a significant reduction in the received power per pixel and
degraded SNR, especially when the pixel count is large. Simply put, when there
are N pixels in the array, each pixel will receive 1/N of the total incident power. This
contrasts with the case with a single-pixel detector, where nearly the entire incident
power can be focused on the single detector in the ideal case. Additionally, in most
practical cases, the signal detected by each pixel is sequentially extracted from the
array. Although the situation may depend on the detection scheme details, it is likely
that the detector in each pixel will not be collecting the incident power for the entire
exposure time but only for a fractional time allocated per each pixel. Again, with a
pixel count of N, the allocated time per each pixel will be 1/N of the total exposure
time in the ideal case. In reality, it will be even smaller than this due to the overhead
such as the time required for transition between pixels. This will be more significant
with a larger video bandwidth (higher frame rate) if the array is employed for real-
time imaging, as the time allowed for the entire array scan will be smaller. As a
result, both spatial and temporal distribution of the power over the pixels will lead to
a significantly degraded SNR for each pixel, severely limiting the receiver perfor-
mance and image quality.
Second, there is an issue of nonuniformity. When an array is made of discrete
components, it typically suffers from serious nonuniformity due to the performance
variance over the components. If the array is made of a single chip with monolithic
approach, the situation will be much better with a tighter performance distribution,
but it cannot avoid the variation over the pixels. For example, Fig. 3.30 shows the
measured distribution of NEP of a Si CMOS 7 7 array implemented on a single-
Fig. 3.30 Nonuniformity in measured NEP over a single-chip Si CMOS 7 7 array: (a) NEP
distribution over the array chip, (b) Distribution of the NEP values. (Reprinted by permission from
Springer Nature: [94] © 2020)
150 3 THz Detectors and Related Topics
chip, which still shows a significant variation [94]. This expected nonuniformity
mandates a calibration process to be carried out after the image acquisition, which
can mitigate the issue to a certain extent. Yet another nonuniformity issue comes
from the variance in the THz beam intensity across the beam cross section. As will
be detailed in Sect. 4.1, THz beams will have a finite lateral dimension (even at focal
points) and will show a Gaussian profile in its intensity along the transverse
direction. Hence, when the beam from a THz source arrives at a 2D array detector,
the incident power will show variation over the pixels, showing the peak at the center
and gradually decreasing toward the edge. This nonuniform power distribution will
cause distortion in the acquired images and thus also require a calibration process for
correction after data acquisition.
Third, the effect of the readout circuit needs to be minimized. Array detectors
necessarily include a readout circuit to sequentially select a pixel at a time and
transfer the data to the output. The readout circuit will include switches for
row/column selection and physical signal paths between each pixel and the output
node. Although indispensable for array operation, the readout circuit will affect the
performance of the array detector. It will cause loss and add noise as the signal from
the selected pixel travels down to the output node. Hence, an additional design care
should be taken, and an amplifier, on-chip or off-chip, may be needed to compensate
for the losses. It is also noted that the loss caused by the readout circuit will be
different over the pixels depending on the location of each pixel. This is another
cause of nonuniformity in array detectors. For example, the schematic of a Si CMOS
7 7 array shown in Fig. 3.31 indicates that the signal from Pixel 0 should travel a
...
...
...
Signal out
Fig. 3.31 Chip-level schematic of a Si CMOS 7 7 array (4 4 mm2). (© 2020 K. Song [95])
3.4 Array Detectors 151
much longer path than that of Pixel 48 to reach the output node, suffering from a
larger loss [94, 95]. Fortunately, for this type of nonuniformity, the variation is not
random but rather systematic, which can be calibrated and canceled out with a
relative ease.
For the advantages and challenges of the array detectors discussed above, the direct
detection scheme was implicitly assumed. While it is true that array detectors are
mostly adopted with the direct detection scheme, arrays can be applied to heterodyne
systems as well. The above-mentioned advantages and issues of arrays are valid for
the arrays with the heterodyne scheme as well, plus there will be additional issues
pertinent to heterodyne arrays. Compared with the direct detector, the unit detector in
the heterodyne arrays will occupy a large area, as it contains more components,
leading to a larger pixel size. This also means a higher power dissipation per pixel,
which may accumulate to a massive level for the entire array. To avoid excessive
power dissipation, it is highly desired to turn on only the pixel that is selected for
signal detection at a given time. In fact, this is true also for the direct detection arrays
if power-hungry components are included in each pixel. A symmetric placement of
the pixels will also be hard for heterodyne arrays, especially for 2D arrays, because
of the elongated shape often found for heterodyne detector layout. This would make
1D arrays a better candidate for heterodyne arrays. LO implementation is another
main issue for heterodyne arrays. The central distribution scheme, which is based on
a single LO source located in the center of the array, will suffer from the low LO
power per pixel (or channel) as the power from the LO is shared by all the pixels.
Also, if the phase information is to be retained, which is the key advantage of the
heterodyne detection over the direct detection, the distance from the LO to the pixel
needs to be identical for all the pixels. This requirement is challenging particularly
for THz applications due to the small wavelength, as a small distance delta will be
significant in terms of the electrical length. An alternative LO scheme for heterodyne
arrays is the localized LOs, in which each pixel has its own LO source. In this case, a
reasonable LO power can be obtained for each pixel, but the phase synchronization
over the pixels will need a precise phase locking between adjacent oscillators. Also,
there may arise an issue of LO power uniformity over the pixels, which is difficult to
achieve when individual LO drives each pixel.
As for the array dimension, they can be either 1D or 2D. With 1D arrays, usually
in a linear configuration, the scan step cannot be completely eliminated to acquire 2D
images, since a scan along the direction perpendicular to the array axis is still needed.
Nonetheless, the total scan time can be reduced by a factor of the pixel count along
the array. In fact, 1D arrays are widely deployed for practical applications, including
the body scan at the airport security points as well as the product scan over a moving
conveyer belt for quality inspection. 2D arrays provide possibility of acquiring 2D
images without any scan step, in which case the 2D array can be considered as a
camera. In this case, 2D images can be obtained with a single exposure and real-time
152 3 THz Detectors and Related Topics
imaging becomes also available, if the pixel count is large enough to satisfy the
required image resolution. If the pixel count is not large enough, scan may be still
needed to meet the resolution target, but the total scan time will be significantly
reduced compared with the imaging with single-pixel detectors [94].
A couple of comments can be made related to 2D arrays for imaging. The 2D
detector arrays used for imaging applications are often referred to as the focal plane
array (FPA) [13, 55, 96]. The name, focal plane array, tends to imply that the array is
supposed to be located at the focal plane of the imaging system. However, this is not
always true. Technically, the focal plane is the plane that contains the focal point on
it, but not necessarily the plane on which images are focused. The focused images
will be obtained on the “image plane” instead, which is in general different from the
focal plane. Hence, to acquire high-quality images, the array should be located at the
imaging plane. There is a special case, though, when the focal plane will coincide
with the image plane. This happens when the object is located at the infinity, as in the
case with telescopes for astronomy including radio telescopes. It is no surprise that
the term, focal plane array, originated from the astronomy, in which all the objects
are virtually at the infinity. We will come back to this issue again in Sect. 6.2.
Another comment can be made on 2D array configuration as follows. In typical 2D
arrays, pixels are packed inside the given array dimension, arranged in rows and
columns. Sometimes, however, images of a similar quality can be obtained with
detectors sparsely scattered over the given area, not necessarily being packed, with
the help of a signal-processing technique called the compressed sensing [97]. In this
case, a large number of detectors can be saved, which will benefit especially the
arrays based on individual components rather than monolithic arrays. The com-
pressed sensing can also be used with single-pixel detectors, in which case raster
scan can be possibly eliminated [98].
responsivity and minimum NEP of the unit pixel detector at the optimum bias point
were measured to be 140 kV/W and 100 pW/Hz1/2, respectively, at a chopping speed
of 5 kHz. The completed chip is mounted on a hyper-hemispherical Si lens for an
efficient backside detection. The photo of the fabricated array detector is shown in
Fig. 3.32b. Note that this array detector functions as a THz camera, capable of
acquiring still images in a single shot and real-time images, which will be discussed
in Sect. 6.2.4.
For the detector chip described above operating near 1 THz, pixels counting up to
~1000 were able to be contained within a reasonable chip size. For lower frequen-
cies, however, the allowed spacing between adjacent pixels is larger as dictated by
154 3 THz Detectors and Related Topics
Fig. 3.33 (a) Photo of the CMOS multi-chip 2-D array with 28 28 pixels (16 16 mm2). (b) The
multi-chip array mounted on a package module
carries the integrated read-out circuits [9]. The array consists of 320 240 bolom-
eter pixels. A partial photo view of the array is given in Fig. 3.34b. With a very wide
operation bandwidth of 1.5–100 THz, the bolometer exhibited NEP of around
0.1–0.2 pW/Hz1/2 within the band. With the large pixel count, real-time images
were successfully obtained by employing quantum cascade laser (QCL) as a THz
source.
The array examples introduced so far are all microscopic, built on tiny chips
based on microelectronics. There are macroscopic large-scale THz array detectors as
well. One good example is Atacama Large Millimeter/submillimeter Array (ALMA)
[102], as briefly mentioned earlier, which consists of 66 radio telescopes with dish
reflector sizes of 12 or 7 m in diameter. Built on Chajnantor plateau in Chile at 5000-
m altitude for optimal observation condition, they operate in a wide frequency range
of 35–950 GHz, which are sub-divided into ten bands. The antennas are mobile and
can move around over a baseline ranging up to 16 km for a flexible arrangement of
the array configuration. The operation of ALMA is in fact based on the principle of
interferometry, which provides a resolution of an aperture size that is equal to the
maximum separation between antenna elements. Accordingly, ALMA operates with
an effective aperture size of over a 10-km range. Based on the same principle, radio
telescopes scattered over the Earth can be linked for co-operation, in which case the
Earth effectively becomes a single enormous antenna for space observation. This is
the basis for Very-Long-Baseline Interferometry (VLBI), which was utilized for the
Event Horizon Telescope (EHT) (see Fig. 3.35b) that has successfully taken the very
first (shadow) image of a black hole at the center of galaxy M87 at a wavelength of
1.3 mm ( f ¼ 230 GHz) [103].
Fig. 3.34 Microbolometer-based 320 240 array detector: (a) Unit bolometer structure, (b) Partial
image of the fabricated array. (Reproduced with permission from [9], © 2008 SPIE)
156 3 THz Detectors and Related Topics
Fig. 3.35 (a) Locations of the eight stations of the Event Horizon Telescope (EHT), which renders
the Earth an enormous virtual antenna. (b) The first (shadow) image of the black hole taken with the
EHT at 230 GHz. (Published in [103] under CC BY 3.0 license)
References
1. R.C. Jones, Phenomenological description of the response and detecting ability of radiation
detectors. Proc. IRE 47(9), 1495–1502 (1959)
2. W.S. Boyle, K.F. Rodgers, Performance characteristics of a new low-temperature
bolometer. J. Opt. Soc. Am. 49(1), 66–69 (1959). https://doi.org/10.1364/JOSA.49.000066
3. F. Low, Low-temperature germanium bolometer. J. Opt. Soc. Am. 51(11), 1300–1304 (1961)
4. J. Clarke, G. Hoffer, P. Richards, N.H. Yeh, Superconductive bolometers for submillimeter
wavelengths. J. Appl. Phys. 48(12), 4865–4879 (1977). https://doi.org/10.1063/1.323612
5. P.M. Downey et al., Monolithic silicon bolometers. Appl. Opt. 23(6), 910–914 (1984). https://
doi.org/10.1364/AO.23.000910
6. P. Richards, Bolometers for infrared and millimeter waves. J. Appl. Phys. 76(1), 1–24 (1994)
7. S.E. Schwarz, B.T. Ulrich, Antenna-coupled infrared detectors. J. Appl. Phys. 48(5),
1870–1873 (1977). https://doi.org/10.1063/1.323940
8. T.L. Hwang, S.E. Schwarz, D.B. Rutledge, Microbolometers for infrared detection. Appl.
Phys. Lett. 34(11), 773–776 (1979). https://doi.org/10.1063/1.90669
9. N. Oda et al., Detection of terahertz radiation from quantum cascade laser using vanadium
oxide microbolometer focal plane arrays, in Infrared Technology and Applications XXXIV,
vol. 6940, (International Society for Optics and Photonics, Bellingham, WA, 2008),
p. 69402Y
10. A. Luukanen, A.J. Miller, E.N. Grossman, Passive Hyperspectral Terahertz Imagery for
Security Screening Using a Cryogenic Microbolometer (Defense and Security) (SPIE, Bel-
lingham, WA, 2005)
11. P. Khosropanah et al., Ultra-low noise TES bolometer arrays for SAFARI instrument on
SPICA, in Millimeter, Submillimeter, and Far-Infrared Detectors and Instrumentation for
Astronomy VIII, vol. 9914, (International Society for Optics and Photonics, Bellingham, WA,
2016), p. 99140B
12. B. Guillet, S. Liu, A. Aryan, C. Adamo, D.G. Schlom, L. Mechin, Uncooled phonon noise
limited La0.7Sr0.3MnO3 suspended bolometers, in 41st International Conference on Infrared,
Millimeter, and Terahertz waves (IRMMW-THz) (2016), pp. 1–2. https://doi.org/10.1109/
IRMMW-THz.2016.7758405
References 157
13. A. Rogalski, F. Sizov, Terahertz detectors and focal plane arrays. Opto-Electron. Rev. 19(3),
346–404 (2011)
14. D. Rutledge, S. Schwarz, Planar multimode detector arrays for infrared and millimeter-wave
applications. IEEE J. Quantum Electron. 17(3), 407–414 (1981). https://doi.org/10.1109/JQE.
1981.1071097
15. A. Chynoweth, Dynamic method for measuring the pyroelectric effect with special reference
to barium titanate. J. Appl. Phys. 27(1), 78–84 (1956)
16. J. Cooper, A fast-response pyroelectric thermal detector. J. Sci. Instrum. 39(9), 467 (1962)
17. S.T. Liu, D. Long, Pyroelectric detectors and materials. Proc. IEEE 66(1), 14–26 (1978).
https://doi.org/10.1109/PROC.1978.10835
18. A. Hossain, M.H. Rashid, Pyroelectric detectors and their applications. IEEE Trans. Ind. Appl.
27(5), 824–829 (1991). https://doi.org/10.1109/28.90335
19. K.Y. Kim, A.J. Taylor, J.H. Glownia, G. Rodriguez, Coherent control of terahertz
supercontinuum generation in ultrafast laser–gas interactions. Nat. Photon. 2, 605 (2008).
https://doi.org/10.1038/nphoton.2008.153
20. I. Kasalynas et al., Design and performance of a room-temperature terahertz detection array for
real-time imaging. IEEE J. Sel. Top. Quant. Electron. 14(2), 363–369 (2008). https://doi.org/
10.1109/JSTQE.2007.912629
21. S.O. Kasap, Principles of Electronic Materials and Devices (McGraw-Hill, New York, 2006)
22. P. Sarro, H. Yashiro, A. Herwaarden, S. Middelhoek, An integrated thermal infrared sensing
array. Sensors Actuators 14(2), 191–201 (1988)
23. M.J. Golay, A pneumatic infra-red detector. Rev. Sci. Instrum. 18(5), 357–362 (1947)
24. E. Bründermann, H.-W. Hübers, M.F. Kimmitt, Terahertz Techniques (Springer, Berlin, 2012)
25. T. Kenny et al., Micromachined infrared sensors using tunneling displacement transducers.
Rev. Sci. Instrum. 67(1), 112–128 (1996)
26. E. Ledwosinska, T. Szkopek, A. Guermoune, M. Siaj, Application of Graphene Membrane in
Micro-Golay Cell Array (SPIE OPTO) (SPIE, Bellingham, WA, 2012)
27. D. Yoon, J. Kim, J. Yun, M. Kaynak, B. Tillack, J. Rieh, 300-GHz direct and heterodyne
active imagers based on 0.13-μm SiGe HBT technology. IEEE Trans. Terahertz Sci. Technol.
7(5), 536–545 (2017). https://doi.org/10.1109/TTHZ.2017.2715419
28. H.T. Friis, Noise figures of radio receivers. Proc. IRE 32(7), 419–422 (1944). https://doi.org/
10.1109/JRPROC.1944.232049
29. S. Gambini, N. Pletcher, J.M. Rabaey, Sensitivity analysis for AM detectors. Tech. Rep. UCB/
EECS-31, EECS Department, University of California, Berkeley, 2008
30. E. Ojefors, B. Heinemann, U.R. Pfeiffer, Subharmonic 220- and 320-GHz SiGe HBT receiver
front-ends. IEEE Trans. Microw. Theory Tech. 60(5), 1397–1404 (2012). https://doi.org/10.
1109/tmtt.2012.2190092
31. E. Brown, Fundamentals of terrestrial millimeter-wave and THz remote sensing. Int. J. High
Speed Electron. Syst. 13(04), 995–1097 (2003)
32. C.K. Walker, Terahertz Astronomy (CRC Press, Boca Raton, FL, 2015)
33. F. Braun, On the current conduction through metal sulphides. J Ann. Phys. Chem 153,
556 (1874)
34. K.S. Champlin, G. Eisenstein, Cutoff frequency of submillimeter Schottky-barrier diodes.
IEEE Trans. Microw. Theory Tech. 26(1), 31–34 (1978). https://doi.org/10.1109/TMTT.1978.
1129302
35. W.C.B. Peatman, T.W. Crowe, Design and fabrication of 0.5 micron GaAs Schottky barrier
diodes for low-noise terahertz receiver applications. Int. J. Infrared Millimeter Waves 11(3),
355–365 (1990). https://doi.org/10.1007/bf01010434
36. S.-M. Chen, Y.-K. Fang, F.R. Juang, W.-K. Yeh, C.-P. Chao, H.-C. Tseng, Terahertz Schottky
barrier diodes with various isolation designs for advanced radio frequency applications. Thin
Solid Films 519, 471–474 (2010)
37. B. Sharma, Metal-Semiconductor Schottky Barrier Junctions and their Applications (Springer
Science & Business Media, Berlin, 2013)
158 3 THz Detectors and Related Topics
38. S. Sankaran, K.O. Kenneth, Schottky barrier diodes for millimeter wave detection in a foundry
CMOS process. IEEE Electron Device Lett. 26(7), 492–494 (2005). https://doi.org/10.1109/
LED.2005.851127
39. U.R. Pfeiffer, C. Mishra, R.M. Rassel, S. Pinkett, S.K. Reynolds, Schottky barrier diode
circuits in silicon for future millimeter-wave and terahertz applications. IEEE Trans. Microw.
Theory Tech. 56(2), 364–371 (2008)
40. R. Han et al., Active terahertz imaging using Schottky diodes in CMOS: Array and 860-GHz
pixel. IEEE J. Solid State Circuits 48(10), 2296–2308 (2013). https://doi.org/10.1109/JSSC.
2013.2269856
41. A. Betz, R. Boreiko, A practical Schottky mixer for 5 THz (Part II), in Proc. of the 7th Int.
Symp. on Space THz Technology, vol. 503 (1996)
42. A.H. Dayem, R.J. Martin, Quantum interaction of microwave radiation with tunneling
between superconductors. Phys. Rev. Lett. 8(6), 246–248 (1962). https://doi.org/10.1103/
PhysRevLett.8.246
43. P.L. Richards, T.M. Shen, R.E. Harris, F.L. Lloyd, Quasiparticle heterodyne mixing in SIS
tunnel junctions. Appl. Phys. Lett. 34(5), 345–347 (1979). https://doi.org/10.1063/1.90782
44. G.J. Dolan, T.G. Phillips, D.P. Woody, Low-noise 115-GHz mixing in superconducting
oxide-barrier tunnel junctions. Appl. Phys. Lett. 34(5), 347–349 (1979). https://doi.org/10.
1063/1.90783
45. B.-K. Tan, Development of Coherent Detector Technologies for Sub-Millimetre Wave Astron-
omy Observations (Springer International Publishing, Cham, 2016)
46. M.J. Wengler, Submillimeter-wave detection with superconducting tunnel diodes. Proc. IEEE
80(11), 1810–1826 (1992)
47. S. Asayama, T. Noguchi, H. Ogawa, A fixed-tuned W-band waveguide SIS mixer with
4.0–7.5 GHz IF. Int. J. Infrared Millimeter Waves 24(7), 1091–1099 (2003)
48. A. Karpov, D. Miller, F. Rice, J. Stem, B. Bumble, Low noise NbTiN 1.25 THz SIS mixer for
Herschel Space Observatory, in Proc. 16th Int. Symp. Space Terahertz Technol. (2005), p. 450
49. D.H. Andrews, W.F. Brucksch Jr., W.T. Ziegler, E.R. Blanchard, Attenuated superconductors
I. For measuring infra-red radiation. Rev. Sci. Instrum. 13(7), 281–292 (1942). https://doi.org/
10.1063/1.1770037
50. D.W. Floet, J. Baselmans, T. Klapwijk, J. Gao, Resistive transition of niobium
superconducting hot-electron bolometer mixers. Appl. Phys. Lett. 73(19), 2826–2828 (1998)
51. K.D. Irwin, G.C. Hilton, Transition-edge sensors, in Cryogenic Particle Detection, (Springer,
Berlin, 2005), pp. 63–150
52. D.W. Floet, E. Miedema, T. Klapwijk, J. Gao, Hotspot mixing: A framework for heterodyne
mixing in superconducting hot-electron bolometers. Appl. Phys. Lett. 74(3), 433–435 (1999)
53. W. Zhang et al., Noise temperature and beam pattern of an NbN hot electron bolometer mixer
at 5.25 THz. J. Appl. Phys. 108(9), 093102 (2010). https://doi.org/10.1063/1.3503279
54. S.A. Maas, A GaAs MESFET mixer with very low intermodulation. IEEE Trans. Microw.
Theory Tech. 35(4), 425–429 (1987)
55. E. Ojefors, U.R. Pfeiffer, A. Lisauskas, H.G. Roskos, A 0.65 THz focal-plane array in a
quarter-micron CMOS process technology. IEEE J. Solid State Circuits 44(7), 1968–1976
(2009)
56. M.I. Dyakonov, M.S. Shur, Plasma wave electronics: Novel terahertz devices using two
dimensional electron fluid. IEEE Trans. Electron Devices 43(10), 1640–1645 (1996). https://
doi.org/10.1109/16.536809
57. W. Knap et al., Plasma wave detection of sub-terahertz and terahertz radiation by silicon field-
effect transistors. Appl. Phys. Lett. 85(4), 675–677 (2004). https://doi.org/10.1063/1.1775034
58. W. Knap et al., Resonant detection of subterahertz radiation by plasma waves in a submicron
field-effect transistor. Appl. Phys. Lett. 80(18), 3433–3435 (2002). https://doi.org/10.1063/1.
1473685
59. A. El Fatimy et al., Resonant and voltage-tunable terahertz detection in In GaAs/InP nanome-
ter transistors. Appl. Phys. Lett. 89(13), 131926 (2006)
References 159
60. F. Caster II, L. Gilreath, S. Pan, Z. Wang, F. Capolino, P. Heydari, Design and analysis of a
W-band 9-element imaging array receiver using spatial-overlapping super-pixels in silicon.
IEEE J. Solid State Circuits 49(6), 1317–1332 (2014)
61. S. Emami, C.H. Doan, A.M. Niknejad, R.W. Brodersen, A 60-GHz down-converting CMOS
single-gate mixer, in IEEE Radio Frequency Integrated Circuits (RFIC) Symposium. Digest of
Papers, (IEEE, New York, 2005), pp. 163–166
62. M.T. Terrovitis, R.G. Meyer, Noise in current-commutating CMOS mixers. IEEE J. Solid
State Circuits 34(6), 772–783 (1999). https://doi.org/10.1109/4.766811
63. M.T. Terrovitis, R.G. Meyer, Intermodulation distortion in current-commutating CMOS
mixers. IEEE J. Solid State Circuits 35(10), 1461–1473 (2000). https://doi.org/10.1109/4.
871323
64. K. Schmalz, W. Winkler, J. Borngräber, W. Debski, B. Heinemann, J.C. Scheytt, A
subharmonic receiver in SiGe technology for 122 GHz sensor applications. IEEE J. Solid
State Circuits 45(9), 1644–1656 (2010). https://doi.org/10.1109/jssc.2010.2051474
65. A. Muller, M. Thiel, H. Irion, H. Ruob, A 122 GHz SiGe active subharmonic mixer, in
European Gallium Arsenide and Other Semiconductor Application Symposium (2005),
pp. 57–60
66. E. Ojefors, U. Pfeiffer, A 94-GHz monolithic front-end for imaging arrays in SiGe:C technol-
ogy, in 2008 38th European Microwave Conference (2008), pp. 1449–1452. https://doi.org/
10.1109/EUMC.2008.4751739
67. R.M. Kodkani, L.E. Larson, An integrated 50-GHz SiGe sub-harmonic mixer/downconverter
with a quadrature ring VCO, in 2007 Topical Meeting on Silicon Monolithic Integrated
Circuits in RF Systems (2007), pp. 223–226. https://doi.org/10.1109/SMIC.2007.322825
68. K. Nimmagadda, G.M. Rebeiz, A 1.9 GHz double-balanced subharmonic mixer for direct
conversion receivers, in 2001 IEEE Radio Frequency Integrated Circuits (RFIC) Symposium,
(IEEE, New York, 2001), pp. 253–256
69. B.G. Perumana et al., A SiGe sub-harmonic mixer for millimeter-wave applications, in 2007
European Microwave Integrated Circuit Conference (2007), pp. 80–83. https://doi.org/10.
1109/EMICC.2007.4412652
70. D. Yoon, J. Kim, J.-S. Rieh, A 290-GHz CMOS heterodyne integrated imager, in 2016 IEEE
International Symposium on Radio-Frequency Integration Technology (RFIT), (IEEE,
New York, 2016), pp. 1–3
71. D.-H. Kim, J.-S. Rieh, A SiGe 140-GHz low power Gm-boosted down-conversion mixer, in
Asia-Pacific Microwave Conference Proceedings (2011), pp. 1126–1129
72. D.H. Kim, J.-S. Rieh, A 135 GHz differential active star mixer in SiGe BiCMOS technology.
IEEE Microw. Wirel. Compon. Lett. 22(8), 409–411 (2012). https://doi.org/10.1109/lmwc.
2012.2205909
73. N. Kim, V. Aparin, L.E. Larson, A resistively degenerated wide-band passive mixer with low
noise figure and +60dBm IIP2 in 0.18μm CMOS, in 2008 IEEE Radio Frequency Integrated
Circuits Symposium (2008), pp. 185–188. https://doi.org/10.1109/RFIC.2008.4561414
74. X. Li, S. Shekhar, D.J. Allstot, Gm-boosted common-gate LNA and differential colpitts
VCO/QVCO in 0.18-μm CMOS. IEEE J. Solid State Circuits 40(12), 2609–2619 (2005).
https://doi.org/10.1109/JSSC.2005.857426
75. I. Kallfass et al., 64 Gbit/s transmission over 850 m fixed wireless link at 240 GHz carrier
frequency. J. Infrared Millimeter Terahertz Waves 36(2), 221–233 (2015)
76. S. Bouvot et al., A D-band passive receiver with 10 dB noise figure for in-situ noise
characterization in BiCMOS 55 nm, in 2017 IEEE 17th Topical Meeting on Silicon Monolithic
Integrated Circuits in RF Systems (SiRF), (IEEE, New York, 2017), pp. 107–110
77. E. Ojefors, J. Grzyb, Y. Zhao, B. Heinemann, B. Tillack, U.R. Pfeiffer, A 820GHz SiGe
chipset for terahertz active imaging applications, in IEEE International Solid-State Circuits
Conference (2011), pp. 224–226
78. Y. Mao, K. Schmalz, J. Borngräber, J.C. Scheytt, A 245 GHz CB LNA in SiGe, in 2011 6th
European Microwave Integrated Circuit Conference, 10–11 Oct 2011 (2011), pp. 224–227
160 3 THz Detectors and Related Topics
79. S. Kim et al., 300 GHz integrated heterodyne receiver and transmitter with on-chip funda-
mental local oscillator and mixers. IEEE Trans. Terahertz Sci. Technol. 5(1), 92–101 (2015).
https://doi.org/10.1109/TTHZ.2014.2364454
80. D. Kim, S. Kim, K. Song, J. Kim, J. Yoo, J.-S. Rieh, Characterization of a CMOS 135-GHz
low noise amplifier with two different noise measurement methods. J. Semicond. Technol. Sci.
18(4), 536–540 (2018)
81. Y. Yan et al., Monolithically integrated 200-GHz double-slot antenna and resistive mixers in a
GaAs-mHEMT MMIC process. IEEE Trans. Microw. Theory Tech. 59(10), 2494–2503
(2011). https://doi.org/10.1109/TMTT.2011.2161326
82. D.H. Kim, D. Kim, J.-S. Rieh, A D-band CMOS amplifier with a new dual-frequency
interstage matching technique. IEEE Trans. Microw. Theory Tech. 65(5), 1580–1588
(2017). https://doi.org/10.1109/TMTT.2017.2655508
83. D. Yoon, K. Song, J. Kim, M. Kaynak, B. Tillack, J.-S. Rieh, A D-band active imager in a
SiGe HBT technology. J. Infrared Millimeter Terahertz Waves 36(4), 335–349 (2015)
84. R. Smith, F. Jones, R. Chasmar, The Detection and Measurement of Infra Red Radiation
(Oxford University Press, Oxford, UK, 1968)
85. K.M.K.H. Leong et al., A 0.85 THz low noise amplifier using InP HEMT transistors. IEEE
Microw. Wirel. Compon. Lett. 25(6), 397–399 (2015). https://doi.org/10.1109/LMWC.2015.
2421336
86. P. Kangaslahti et al., Low noise amplifier receivers for millimeter wave atmospheric remote
sensing, in 2012 IEEE/MTT-S International Microwave Symposium Digest, 17–22 June 2012
(2012), pp. 1–3. https://doi.org/10.1109/MWSYM.2012.6259468
87. G. Moschetti et al., A 183 GHz metamorphic HEMT low-noise amplifier with 3.5 dB noise
figure. IEEE Microw. Wirel. Compon. Lett. 25(9), 618–620 (2015). https://doi.org/10.1109/
LMWC.2015.2451355
88. Z. Wang, P. Chiang, P. Nazari, C. Wang, Z. Chen, P. Heydari, A CMOS 210-GHz funda-
mental transceiver with OOK modulation. IEEE J. Solid State Circuits 49(3), 564–580 (2014).
https://doi.org/10.1109/JSSC.2013.2297415
89. K. Schmalz, J. Borngraber, Y. Mao, H. Rucker, R. Weber, A 245 GHz LNA in SiGe
technology. IEEE Microw. Wirel. Compon. Lett. 22(10), 533–535 (2012). https://doi.org/10.
1109/LMWC.2012.2218097
90. B. Thomas, A. Maestrini, G. Beaudin, A low-noise fixed-tuned 300–360-GHz sub-harmonic
mixer using planar Schottky diodes. IEEE Microw. Wirel. Compon. Lett. 15(12), 865–867
(2005). https://doi.org/10.1109/LMWC.2005.859992
91. P.J. Sobis, N. Wadefalk, A. Emrich, J. Stake, A broadband, low noise, integrated 340 GHz
Schottky diode receiver. IEEE Microw. Wirel. Compon. Lett. 22(7), 366–368 (2012). https://
doi.org/10.1109/LMWC.2012.2202280
92. J. Treuttel et al., A 520–620-GHz Schottky receiver front-end for planetary science and remote
sensing with 1070 K–1500 K DSB noise temperature at room temperature. IEEE Trans.
Terahertz Sci. Technol. 6(1), 148–155 (2016). https://doi.org/10.1109/TTHZ.2015.2496421
93. J. Winters, On the capacity of radio communication systems with diversity in a Rayleigh
fading environment. IEEE J. Sel. Areas Commun. 5(5), 871–878 (1987)
94. K. Song, J. Kim, D. Kim, J. Yoo, J.-S. Rieh, A 300-GHz CMOS 7-by-7 detector array for
optics-less THz imaging with scan-less target object. J. Infrared Millimeter Terahertz Waves
41, 1–13 (2020)
95. K. Song, Development of THz Imaging Arrays Based on Si CMOS Technology, Ph.D.
Dissertation, Korea University, 2020
96. N. Oda, Uncooled bolometer-type terahertz focal plane array and camera for real-time
imaging. C. R. Phys. 11(7–8), 496–509 (2010)
97. D.L. Donoho, Compressed sensing. IEEE Trans. Inf. Theory 52(4), 1289–1306 (2006). https://
doi.org/10.1109/TIT.2006.871582
References 161
98. W.L. Chan, K. Charan, D. Takhar, K.F. Kelly, R.G. Baraniuk, D.M. Mittleman, A single-pixel
terahertz imaging system based on compressed sensing. Appl. Phys. Lett. 93(12), 121105
(2008)
99. H. Sherry et al., A 1kpixel CMOS camera chip for 25fps real-time terahertz imaging applica-
tions, in 2012 IEEE International Solid-State Circuits Conference, (IEEE, New York, 2012),
pp. 252–254
100. R. Al Hadi et al., A 1 k-pixel video camera for 0.7–1.1 terahertz imaging applications in 65-nm
CMOS. IEEE J. Solid State Circuits 47(12), 2999–3012 (2012)
101. N. Oda, A.W. Lee, T. Ishi, I. Hosako, Q. Hu, Proposal for real-time terahertz imaging system
with palm-size terahertz camera and compact quantum cascade laser, in Terahertz Physics,
Devices, and Systems VI: Advanced Applications in Industry and Defense, vol. 8363, (Inter-
national Society for Optics and Photonics, Bellingham, WA, 2012), p. 83630A
102. A. Wootten, A.R. Thompson, The Atacama large millimeter/submillimeter array. Proc. IEEE
97(8), 1463–1471 (2009)
103. K. Akiyama et al., First M87 event horizon telescope results. I. The shadow of the
supermassive black hole. Astrophys. J. Lett. 875, 1–17 (2019)
Chapter 4
THz Propagation and Related Topics
Probably the simplest optical principles can be provided by the geometrical optics, or
ray optics, in which the propagation of electromagnetic waves, or light as in most
optics domain, follows a straight-line path in the homogenous medium as well as
across the interface between different media that involves reflection and refraction.
Many basic optical behaviors of electromagnetic waves can be successfully
explained with geometrical optics. While it nicely applies to the cases where the
wavelength is much smaller than the dimension of interacting optical components, it
cannot explain some phenomena such as diffraction. When the size of the
For many practical applications, the final expression of Gaussian beam given as a
function of various propagation parameters will be certainly useful, while its deri-
vation may not for immediate engineering needs. Apart from the practical causes,
however, understanding of Gaussian beam should be greatly enhanced by reviewing
its derivation steps, which is presented below.
A formal derivation should start from the time-dependent Maxwell’s equations,
but a simpler approach may start with the Helmholtz equation, in which the time-
dependent term is eliminated assuming a sinusoidal waveform over the time. That is:
∇2 þ k 2 E ¼ 0, ð4:1Þ
4.1 Gaussian Beam 165
where k is the wave number (¼2π/λ), and E is the electric field with the time
dependent factor dropped out. Note that the expression will be also valid when the
electric field E is replaced by magnetic field H. However, we will stay with the
electric field E throughout the analysis and express the Gaussian beam as a function
of E. Equation (4.1) is general and valid for any type of sinusoidal waves, including
both plane wave and Gaussian beam. For plane waves, as is treated in college level
physics courses, E does not have x- or y-dependence assuming the wave propagates
along z-axis, and the equation is reduced to:
2
∂ E
þ k 2 E ¼ 0, ð4:2Þ
∂z2
which is basically a wave with a well-defined phase without variation over the
transverse direction. How about Gaussian beams? One of the major properties of
Gaussian beam that distinguishes itself from the plane wave is that E has x- and
y-dependence, or radial dependence, as well as z-dependence in addition to the
normal propagation factor, exp(jkz). Hence, E for can be expressed as:
With Eq. (4.4) substituted into the Helmholtz equation of Eq. (4.1), the following
equation is obtained for Gaussian beams where exp(jkz) dependence is factored
out:
2 2 2
∂ u ∂ u ∂ u ∂u
þ þ 2jk ¼ 0: ð4:5Þ
∂x2 ∂y2 ∂z2 ∂z
This equation is called the reduced wave equation, where both t-dependent and
z-dependence sinusoidal factors, ejωt and ejkz, are taken out. If we assume the
z-dependence, which is mostly due to the diffraction, is small, the third term in
Eq. (4.5) can be ignored. This is called the paraxial approximation, which leads to:
2 2
∂ u ∂ u ∂u
þ 2jk ¼ 0: ð4:6Þ
∂x2 ∂y2 ∂z
Equation (4.6) is called the paraxial wave equation and the derivation of Gaussian
beam usually starts from here. The paraxial approximation may not appear obvious
to many readers, but there are many different ways to prove its validity in practical
cases. Here, we will make yet another assumption that u has only radial dependence,
invariant over the angle change. In fact, this is not the general case, and the angular
166 4 THz Propagation and Related Topics
dependence should be considered for higher order modes. However, we will confine
the current analysis to this axially symmetric case only, which will correspond to
lower order modes including the fundamental mode. In this case, u will be governed
by the axially symmetric paraxial wave equation, which is better presented in
cylindrical coordinate such that:
2
∂ u 1 ∂u ∂u
þ 2jk ¼ 0, ð4:7Þ
∂r 2 r ∂r ∂z
where r is the radial distance from the axis of propagation (z-axis). Note that there is
no angular dependence in Eq. (4.7). A nice trial solution to the axially symmetric
paraxial equation can be written as follows:
r2
uðr, zÞ ¼ AðzÞ exp jk : ð4:8Þ
ð2qzÞ
A and q are parameters with complex values that have only z-dependence and their
actual expressions are to be obtained shortly after. The exponent in Eq. (4.8) should
contain the information on the radius of curvature of the beam and the decay of the
field magnitude in the radial direction. Hence, Eq. (4.8) can be expressed in terms of
parameters with more direct physical meaning such that:
r2 r2 πr 2 r2
uðr, zÞ ¼ AðzÞ exp jk ¼ AðzÞ exp j , ð4:9Þ
2RðzÞ ω2 ðzÞ λRðzÞ ω2 ðzÞ
1 1 λ
¼ j 2 , ð4:10Þ
qðzÞ RðzÞ πω ðzÞ
where R(z) is the radius of curvature (for the equiphase surface of the beam) and ω(z)
is the beam radius, which is defined as the radial distance where the electric field
magnitude drops to 1/e of the value on the propagation axis. The factor πr2/λR is the
off-axis phase shift observed at a radial distance of r, which is a result of the finite
radius of curvature of Gaussian beam. The detailed procedure to arrive at Eqs. (4.9)
and (4.10) will not be given here, but can be found elsewhere (for example in [1]).
The physical meaning of q may not appear as straightforward as those of R or ω to
the readers, but it does contain lots of related key information in it. It is called in
various ways such as the complex beam parameter, Gaussian beam parameter, or
simply beam parameter. Sometimes it is also called the complex radius of curvature,
indicating it is a complex generalization of R because it is reduced to R if purely real,
which corresponds to the spherical wave case. By the same token, it can be stated
that the Gaussian beam is a spherical wave with a complex radius of curvature
of q [2].
4.1 Gaussian Beam 167
Equation (4.9) already shows the r-dependence of u(r, z) as the Gaussian function
(i.e., including a factor of exp((x/x0)2), but still is not the final form and will be
developed further below. Before moving forward, however, a few practical discus-
sions will be useful. With the help from the boundary condition of q, R and ω can be
shown as an explicit function of z, such that:
2
1 πω20
RðzÞ ¼ z þ ð4:11Þ
z λ
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
λz
ωðzÞ ¼ ω0 1 þ , ð4:12Þ
πω20
where ω0 is the beam radius at z ¼ 0, which is also called the beam waist since it is
the minimum beam radius available as mentioned earlier. One may note that both
expressions include a common factor of πω20 =λ. This factor is called the confocal
distance zc with the following definition:
πω20
zc ¼ : ð4:13Þ
λ
z2c
RðzÞ ¼ z þ ð4:14Þ
z
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
z
ωðzÞ ¼ ω0 1 þ : ð4:15Þ
zc
Now let us get back to the expression for the Gaussian beam given as Eq. (4.9).
The only remaining undetermined is A, a z-dependent complex coefficient of u. From
the relation between A and q, which can be obtained by substituting the trial solution
(Eq. 4.8) into the paraxial equation (Eq. 4.6), one can show that the following
relation can be reached:
ω0
AðzÞ ¼ Að0Þ exp ð jϕ0 Þ, ð4:16Þ
ω
where ϕ0 is the Gaussian beam phase shift, or the Gaussian beam phase slippage,
which is given as:
168 4 THz Propagation and Related Topics
1 λz 1 z
ϕ0 ¼ tan ¼ tan : ð4:17Þ
πω20 zc
The positive value of ϕ0 implies that the phase of Gaussian beam on the propagation
axis slightly lags the plane waves. This on-axis phase shift should not be confused
with the off-axis phase shift, πr2/λR, as described above. With the expression for A as
obtained in Eq. (4.16), the final form of u is given as:
2
ω0 r jπr 2
uðr, zÞ ¼ Að0Þ exp þ jϕ : ð4:18Þ
ω ω2 0
λR
Equation (4.19) could be regarded as the final form of the Gaussian beam, but one
may wonder about A(0), the value of A at the origin, which still remains
undetermined. As the last step, a normalization can be made to finalize the form.
Here, the normalization condition suggested in [1] will be taken, which assumes that
the
R integration of the square of the electric field over the beam area equals unity, or
2πr|E|2dr ¼ 1. Although the unit is not exactly physical, the coefficient for the
electric field can be determined with this normalization, leading to:
pffiffiffiffiffiffiffiffi 2
2=π r jπr 2
E ðr, zÞ ¼ exp jkz þ jϕ0 : ð4:20Þ
ω ω2 λR
This is the final expression for the Gaussian beam in the fundamental mode. The
interpretation of this form will be discussed in the following section. Higher order
solutions to Gaussian beams can be found in other literatures that focus on this topic.
Understanding the behavior of Gaussian beams in free space (or, more precisely, in
homogeneous media) can start with inspecting the expression for Gaussian beam as
is given in Eq. (4.20). There are four terms in the exponent, and the beam behavior is
dictated by these four exponential factors. The first term relates the magnitude, and
the last three terms concern the phase of the Gaussian beam, each of which bears its
own physical meaning as follows:
2
• ωr 2 : A term indicating that the magnitude of the field decays with the Gaussian
function along the radial direction, from which the name of the beam is derived.
4.1 Gaussian Beam 169
where its z-dependence is not explicitly shown because the profile is governed by the
ratio r/ω regardless the position on the z-axis (as well as the actual value of ω). While
the electrical field is the main parameter we are interested in, the radial spread of the
beam may be better understood in terms of the power from a practical viewpoint. For
this purpose, the relative power normalized over the on-axis value is given in
Eq. (4.22). In fact, this ratio is called the edge taper Te, serving as a useful parameter
to present the radial spread profile of Gaussian beams.
Pðr Þ jE ðr Þj2 r 2
Te ¼ ¼ ¼ exp 2 : ð4:22Þ
Pð0Þ jEð0Þj 2 ω
One can easily notice that only difference from the normalized field is the factor of
2 introduced in the exponent, indicating a faster decay of the normalized power. The
edge taper is plotted in Fig. 4.2 as a function of the radial distance normalized to the
beam waist, or r/ω. An inspection of the plot shows that Te is reduced by half at r/
ω ¼ 0.589, a value that corresponds to HWHM (half-width-half-maximum) of
Gaussian beams. Te is further reduced down to 0.135 at r/ω ¼ 1, which implies that
a substantial power level is still maintained at the beam radius. But it will quickly fade
out and become negligibly small as r grows, as indicated by Te ¼ 0.0111 and
0.0003 at r/ω ¼ 1.5 and 2, respectively (outside the range of Fig. 4.2). The phase of
the field also varies over the radial distance, a clear difference from the plane wave, in
which case the same phase is retained over the any direction perpendicular to the
170 4 THz Propagation and Related Topics
propagation. As mentioned earlier, this phase shift, πr2/λR, arises from the finite
radius of curvature of Gaussian beam, which indicates the spherical-wave-like nature
that exists in the beam. A cartoon in Fig. 4.3 shows this phase shift visually: As the
wave front sets back as r increases, the phase at r will have an extra phase when
compared to the on-axis case with the same z position.
Now let us consider the axial dependence of the Gaussian beam along the z-axis.
The dominating feature in this consideration is the z-dependence of the radius of
curvature R and beam radius ω. Based on Eqs. (4.14) and (4.15), Fig. 4.4 shows how
R and ω develops as the beam propagates along z-direction. The radius of curvature,
which is infinite at z ¼ 0, exhibits an initial reduction down to a minimum value of
2zc at z ¼ zc, then snaps back upward and keeps increasing afterwards. Eventually,
the curve converges to a straight line of R ¼ z, as the second term in Eq. (4.14)
becomes negligible. This behavior reveals an important aspect of the beam: After a
long-distance propagation, a Gaussian beam can be approximated as a spherical
4.1 Gaussian Beam 171
R ω
ω0
Slope approaches 1 Slope approaches
zc
2zc
θ0
zc z z
(a) (b)
Fig. 4.4 Dependence of beam parameters on the propagation distance z: (a) Radius of curvature R,
(b) Beam waist ω
wave centered at the origin, as indicated by the radius of curvature that equals z. This
basically means that the source can be regarded as a point source for a very far-field
location along the axis, which makes good sense. The development of the beam
radius ω over the propagation would be of more practical interest to those readers
who work on optical setups to manipulate Gaussian beams. As shown in Fig. 4.4b,
the beam waist gradually spreads out with a monotonic increase, which is mainly a
diffraction effect caused by the source aperture size being comparable to the
wavelength. After a long propagation, ω eventually converges to a linear line with
a slope of ω0/zc, as the term z/zc in the square root in Eq. (4.15) becomes much larger
than 1. This indicates that, when far away from the source, the beam spreads in a
linear fashion, from which we can define a beam spread angle, or the far-field
divergence angle as is officially termed:
1 ω0 1 λ
θ0 tan ¼ tan : ð4:23Þ
zc πω0
One useful observation is that the angle will be larger if the aperture size of the
source is smaller in terms of the wavelength, assuming the minimum beam waist
ω0 is about the same size as the aperture. This also aligns with our common sense
because a stronger diffraction effect would be expected for a smaller aperture,
leading to a faster beam spread. The Gaussian beam phase shift, ϕ0 ¼ tan1(z/zc),
is also given as a function of z. Being the on-axis phase delay experienced by the
Gaussian beam against the plane wave, it grows with the beam propagation and
eventually converges to π/2 at very far-field locations. Hence, there would be a
maximum phase shift for Gaussian beams, which is a quarter-wavelength. A graph-
ical representation of the properties of Gaussian beam propagation based on the
discussion above is shown in Fig. 4.5.
172 4 THz Propagation and Related Topics
Fig. 4.5 Radius of curvature R and beam width ω of Gaussian beam along the propagation axis
r'
z0 Axis of propagation
With the understanding of Gaussian beam in the free space as reviewed in the
previous section, it will be useful to observe the way Gaussian beams are coupled
with optical or electromagnetic components. In particular, the coupling between
Gaussian beams and lenses is of high importance as it is widely practiced in practical
applications, and thus will be discussed in this section. The transformation of
Gaussian beams with lenses can be nicely described by introducing the ray transfer
matrices that originate from the geometrical optics. With this fact given, in this
subsection, the ray transfer matrix will be first introduced with various example
cases, and then its application to Gaussian beam transformation by lenses will be
discussed.
In the frame of geometrical optics, a ray can be described by two parameters, the
position from the propagation axis r and the angle against the axis r0 as depicted in
Fig. 4.6. Hence, naturally, the transformation of a ray by an optical component can
4.1 Gaussian Beam 173
where the subscripts “in” and “out” represent input and output, respectively. This
ABCD matrix configuration allows a convenient description of the commonly
encountered cases of the cascaded multiple optical components, which can be simply
represented by a progressive multiple product of the corresponding matrices. It is
noted that the determinant of the ray transfer matrix is unity, or ||ABCD|| ¼ 1, if the
starting and ending points are with the same refractive index, a property that will be
used later in the beam analysis. A few example cases relevant to practical cases with
lens transformation are introduced below.
The simplest case would be propagation through a homogeneous medium with a
length of L, as shown in Fig. 4.7a. In this case, one can see that very simple relations
will be held between input and output rays such that:
Note that this matrix is valid regardless the refractive index of the propagation
medium.
Another important case is the curved interface between media with different refrac-
tive indices, say, n1 and n2, as is described in Fig. 4.7b. The corresponding ray
transfer matrix Minterface can be obtained by considering the angles shown in the
figure. If we assume the incident angle is small, Snell’s law at the interface is
obtained as:
174 4 THz Propagation and Related Topics
n2 sin θ1 θ1
¼ or n1 θ 1 n2 θ 2 : ð4:29Þ
n1 sin θ2 θ2
r r
n1 sin 1 in tan 1 r 0in ¼ n2 sin 1 out tan 1 r 0out : ð4:30Þ
R R
Based on the equality rin ¼ rout at the interface and the assumption that the angle
components included in the expression are small, Eq. (4.30) is reduced to:
r in r n2 n1 1 n
n1 r 0in ¼ n2 in r 0out or r 0out ¼ r þ 1 r0 , ð4:31Þ
R R n2 R in n2 in
where the following approximations were used: sin1x tan1x x for small x.
Based on these relations, the transfer matrix becomes:
4.1 Gaussian Beam 175
" #
1 0
Minterface ¼ n 2 n1 1 n1 , ð4:32Þ
n2 R n2
where near-normal and near-parallel incident beam was assumed. For an arbitrary
incident angle, one can find information from other references such as [3]. Note that
the sign of the radius of curvature is defined so that it is positive for the case shown in
Fig. 4.7b (i.e., R > 0 if concave seen from left). Of course, this matrix is valid for a
negative radius of curvature as well (R < 0, convex seen from left), and is also true
for a flat interface where the radius of curvature is infinity (R ¼ 1).
With these two basic ray transfer matrices, we are now ready to work on matrices for
lenses. As a simpler case, we can first consider a thin lens where the propagation
interior to the lens can be neglected. With this simplification, a thin lens can be
considered as two interfaces with different radius of curvatures located in immediate
proximity. Accordingly, the ray transfer matrix for a thin lens can be obtained by
applying Eq. (4.32) twice with different radius of curvatures, say R1 and R2, in
sequence. Then the matrix becomes:
" #" #
1 01 0
Mthin lens ¼ n1 n2 1 n2 n1 1
n2 n1
n1 R 2 n2 R 1
n1 n2
2 3
1 0
¼ 4 n2 n1 1 1 5: ð4:33Þ
1
n1 R1 R2
Although a convex lens was assumed in Fig. 4.8a, Eq. (4.33) is general and can be
applied to other cases with arbitrary two radius of curvatures. Also note that when
multiple optical components are cascaded, the matrix for the first interacting com-
ponent will appear on the rightmost side of the matrix product since the matrix
application starts from the right. As we follow the convention of beam propagation
from the left, the readers need to be careful about the order of matrices in the matrix
product. It is also interesting to note that we can obtain a relation between the focal
length f and its radius of curvature and refractive index for thin lenses with the help
of Eq. (4.33), because it is known that the ray transfer matrix of a thin lens can also
be expressed as a function of f as follows:
176 4 THz Propagation and Related Topics
n2
(a) (b)
" #
1 0
Mthin lens ¼ 1 : ð4:34Þ
1
f
The procedure to arrive at Eq. (4.34) would be a good chance for the readers to be
reminded of the basic optical principles related to the focal length of a lens. The four
matrix elements for a thin lens can be decided if we consider two special cases shown
in Fig. 4.9, where the incident beam is parallel to the propagation axis (Fig. 4.9a) and
the exiting beam is parallel to the axis (Fig. 4.9b). In either case, the equality rin ¼ rout
is valid, leading to the first-row elements of 1 and 0. To determine the second row,
recall a basic principle that a parallel beam incident on a lens will converge at the
focal length f on the other side of the lens (if we assume a convex lens without loss of
generality). Then, as is clear from Fig. 4.9a, we obtain r 0in ¼ 0 and r 0out ¼ r in =f . A
reciprocal statement is that a parallel exiting beam will be obtained if a source is
located at the focal length on the opposite side. With the help of Fig. 4.9b, one can
infer r 0out ¼ 0 and r 0in ¼ r in =f . From these relations, one can obtain the second-row
elements of 1/f and 1, completing the matrix shown in Eq. (4.34). It is noted that
Eq. (4.34) is valid for an arbitrary incident angle although the two special cases
shown in Fig. 4.9 were used for the derivation above. Now, comparing Eqs. (4.33)
and (4.34), one can obtain the focal length f of a thin lens as a function of geometrical
parameters (R1 and R2) and material parameters (n1 and n2) as:
1 n2 n1 1 1
¼ : ð4:35Þ
f n1 R2 R1
For the analysis of thick lenses as is shown Fig. 4.8b, we need to consider the
finite thickness of the lens and propagation across it. It can be achieved by inserting
the length matrix of Eq. (4.28) between the two interface matrices that appear in
Eq. (4.33). Assuming the lens thickness as d, the matrix becomes:
4.1 Gaussian Beam 177
′
rout rin′
rin Axis of propagation rout
Axis of propagation
rout = rin
rout = rin
rin
′ = –
rin′ =0, rout rin
f rin′ = ′ =0
, rout
f f f
(a) (b)
Fig. 4.9 Two cases of a ray passing through a focal point of a thin lens: (a) Incident beam is parallel
to the propagation axis, (b) Exiting beam is parallel to the propagation axis
2 3 2 3
1 0 "1 d# 1 0
Mthick lens ¼ 4 n1 n2 1 n2 5 4n n 1 n 5
2 1 1
n1 R2 n1 0 1 n2 R1 n2
2 n n1 d 3
1þ 2
n1
d ð4:36Þ
6 n2 R1 n2 7
¼64 2
7:
5
1 ð n2 n1 Þ d n1 n2 d
1þ
f n1 n2 R1 R2 n2 R 2
One can see that Eq. (4.36) approaches Eq. (4.34) for small d. In view of the
complexity in Eq. (4.36), we will maintain the assumption of the thin lens for
remaining part of this discussion, which will provide reasonably accurate results.
Up to this point, we have walked through the ray transfer matrices for various
practical cases. Now, we will proceed to combine this matrix representation with
the Gaussian beam scheme, so that the transformation of Gaussian beam by an
optical component can be described by the corresponding ray transfer matrix. The
first task will be relating the input and output complex beam parameters of an optical
component, qin and qout, in terms of the matrix elements of the optical component,
A, B, C, and D. From the definition of the ray transfer matrix shown in Eq. (4.24), we
have r out ¼ Ar in þ Br 0in and r 0out ¼ Cr in þ Dr 0in . If we take the ratio of these two
expressions,
r out Ar in þ Br 0in
¼ : ð4:37Þ
r 0out Cr in þ Dr 0in
Now, from Fig. 4.10, we can see that the radius of curvature of Gaussian beam can
be simply given as r/r0 in the far field. Hence, Eq. (4.37) can be converted to an
equation relating Rin and Rout as follows:
178 4 THz Propagation and Related Topics
ARin þ B
Rout ¼ : ð4:38Þ
CRin þ D
Aqin þ B
qout ¼ : ð4:39Þ
Cqin þ D
Given Eq. (4.39), we are now ready to take on the transformation of Gaussian
beam by a thin lens. As is depicted in Fig. 4.11, let us assume the input beam waist,
ω0in, is located din away from the lens and the output beam waist, ω0out, is located at
dout. From a practical point of view, it is important to be able to express ω0out and dout
as a function of ω0in and din as well as the ray transfer matrix elements of the lens.
This is the short-term object of the following analysis. The ray transfer matrix of the
optical system described in Fig. 4.11 is:
" 1 0 #
1 d out 1 din
M¼ 1
0 1 1 0 1
f
2 3
d out din
6 1 d þ d 1
f in out
f 7
¼64
7:
5 ð4:40Þ
1 din
1
f f
If the matrix elements of Eq. (4.40) is substituted into Eq. (4.39), we obtain:
where the relation qin ¼ jzc is used, which is valid because R is infinite at the input
beam waist location (recall Eq. (4.10)). It would be worthwhile to note that
Eq. (4.41) is in fact valid for any point located right to the lens in Fig. 4.11 by
4.1 Gaussian Beam 179
substituting the distance from the lens for dout in the expression, although we will
confine our discussion to dout where the output beam waist is located.
By examining the real and imaginary parts of 1/qout, the inverse of Eq. (4.41), we
can retrieve two constraints that can be used to obtain dout and ω0out. First, at the
output beam waist, the radius of curvature R is again infinity, leaving the real part of
1/qout zero. Based on Eq. (4.41), one can derive:
dout d in =f 1
¼1þ : ð4:42Þ
f ðd in =f 1Þ2 þ ðzc =f Þ2
Hence, dout can be expressed as a function of din and zc as well as f, which will be
more intuitively understood if dout, din, and zc are shown normalized to f as is the
case in Eq. (4.42). Second, the imaginary part of 1/qout becomes –λ/πω0out2 at the
output beam waist (see Eq. 4.10). This relation can be equally expressed as
Im(1/qout) ¼ (ω0in/ω0out)2/zc since zc was defined with respect to the input beam
waist ω0in such that zc ¼ πω0in2/λ (see Eq. 4.13). With this relation, along with
||ABCD|| ¼ 1, which is valid as the refractive indices are the same in the regions din
and dout belong to, one can obtain the following:
ω0in
ω0out ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi: ð4:43Þ
2 2
ðd in =f 1Þ þ ðzc =f Þ
Therefore, we can express the output beam waist in terms of din and zc as well as f.
Occasionally, one will find a parameter called the “system magnification” useful,
which is defined as the ratio between output and input beam waist:
ω0out 1
M ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð4:44Þ
ω0in
ðdin =f 1Þ2 þ ðzc =f Þ2
One should not confuse this system magnification with the other more popular
definition of the magnification in optics, which is the ratio between the object and
image sizes.
180 4 THz Propagation and Related Topics
Now let us make some observations on the obtained output beam waist and its
location, starting with its location, dout. The relation between dout and din, as is
presented by Eq. (4.42), is depicted in Fig. 4.12a with zc as a parameter. This plot is
4.2 Antennas 181
extremely useful for those dealing with Gaussian beams, which they may want to
keep near at hand. The key observation from the plot is that dout ¼ f if din ¼ f, or the
output beam waist is located at the focal length of the lens if the input beam waist
(near the source position) is also located at the focal length on the other side, which is
also obvious from Eq. (4.42). If the input beam waist is pushed farther away from the
lens than f (i.e., din > f ), dout will also move farther away from the lens initially (i.e.,
dout > f ), then snap back and eventually approach the focal length for very large din/f.
If the input beam waist is brought closer to the lens than the focal length (i.e., din < f ),
the output beam waist will be also pulled closer to the lens (i.e., dout < f ). The
response of dout to the change in din will be more sensitive with smaller zc, or larger
wavelength and smaller input beam waist. One comment on Fig. 4.12a can be made
regarding the negative values of dout obtained for some small zc cases when din < f.
This implies that the beam will diverge immediately after passing through the lens
without forming a beam waist in those special cases. Next, let us make observation on
the output beam waist size. Figure 4.12b shows the ratio of the input and output beam
waist, or the system magnification, as a function of din normalized to f, following
Eq. (4.43). It is clear the output beam waist will be largest when din ¼ f, or when the
input beam waist is located at the focal length. As din is pushed away from f, toward
either direction from f, the output beam waist will be reduced. When din becomes very
large, the magnification converges zero, or the output beam waist will become
infinitesimally small, similar to what the geometrical optics would predict when the
point source is at infinity. The reaction of ω0out to the change in din is again more
sensitive when zc is smaller. A system magnification larger than unity will be obtained
only for zc smaller than f, because the maximum value of the system magnification
that is achieved when din ¼ f is f/zc, as indicated by Eq. (4.44).
4.2 Antennas
The antenna can be defined as a device that converts guided waves into electromag-
netic waves in open space or vice versa. For the propagation in the open media
discussed in the previous section, it was implied that antennas are employed for
radiation and detection. As many of the THz applications involve propagation in the
open space, it became clear that THz researchers are required to have some basic
knowledge of antennas. The topic of antenna is extensive, and an enormous amount
of related research results are being published in the journals and conferences
dedicated to the topic. A plenty of textbooks are also available that solely focus on
antennas, providing detailed description of the topic. However, for THz researchers
trying to get acquainted with the basics of antennas, those dedicated sources may
appear overwhelming. This section is intended to provide a very quick introduction
to the antenna basics, together with a few types of antennas that are available for
integration with planar THz circuits. Those readers who need a deeper look into
antennas are referred to more comprehensive coverages of this broad topic such as
[5, 6]. For the focused treatment of mm-wave and THz antennas, there is an excellent
review article as well [7].
182 4 THz Propagation and Related Topics
The radiation pattern, also called the antenna pattern, is a mathematical function or a
graphical representation of the radiation properties of an antenna, usually taken at the
far-field regime of the radiation [8]. While radiation patterns can be obtained for
various antenna properties, in most cases the patterns are presented for the magni-
tude of the field or power, in either linear or dB scale. An example would be useful to
describe the radiation pattern in more detail. Suppose an infinitesimal dipole located
at the origin of a coordinate system, which is basically a charge moving up and down
periodically along a small metallic wire aligned along the z-axis (Fig. 4.13). As many
of antennas can be modeled with a line source, this structure should be a nice
example to consider. With a simple theory but a rather lengthy derivation, one can
arrive at the expressions of the electric and magnetic fields at the far-field region as:
IΔzejβr
E ¼ jωμ sin θaθ , ð4:45Þ
4πr
IΔzejβr
H ¼ jβ sin θaϕ , ð4:46Þ
4πr
where ω is the angular frequency of the dipole oscillation and thus the radiated field,
μ is the permeability of the propagation medium, I and Δz are the current and the
4.2 Antennas 183
dipole size, β is the propagation constant, r is the distance from the origin, θ is the
angle from the z-axis in the spherical coordinate, aθ and aϕ are respectively the unit
vector for the θ and ϕ direction. It was assumed that the medium is lossless, in which
case the phases of E-field and H-field are the same, leading to the ratiop between
ffiffiffiffiffiffiffiffi the
field magnitudes being a real number. This ratio, given as η ¼ ωμ=β ¼ μ=ε with ε
as the permittivity of the medium, is known as the intrinsic impedance with a unit of
Ω (376.7 Ω in free space). Because the magnitudes of the E-field and H-field are the
same except for a constant factor (i.e., η), the field radiation patterns, typically
presented as normalized over the maximum value, are identical for the two fields.
Hence, the field radiation pattern can be created based on either E-field or H-field.
The 3D radiation pattern of the infinitesimal dipole is shown in Fig. 4.13b. It takes
on a doughnut shape and does not have ϕ dependence, thus is often referred to as
omnidirectional (different from isotropic, in which case the pattern is a perfect
sphere). While 3D patterns best depict the radiation pattern, 2D patterns that are
cut along reference planes are often preferred for practical purposes. In this case, it is
conventional to have 2D patterns obtained on the E-plane and H-plane, which are
Fig. 4.13 (a) A small metallic wire as an infinitesimal dipole. (b–e) Radiation patterns for the
infinitesimal dipole: (b) 3D radiation pattern for field, (c) E-plane field radiation pattern, (d) H-plane
field radiation pattern, (e) E-plane power radiation pattern
184 4 THz Propagation and Related Topics
called the E-plane radiation pattern and the H-plane radiation pattern, respectively.
These two radiation patterns are collectively called the principal plane patterns.
Here, the E-plane refers to the plane that contains the electric field vector of the
radiation. For instance, with the infinitesimal dipole depicted in Fig. 4.13, any plane
with a fixed value of ϕ (constant-ϕ plane) can be an E-plane as indicated in the
figure, because the electric field in this example does not have any ϕ-dependence.
Likewise, H-plane refers to the plane that contains the magnetic field vector of the
radiation. In the given example, any constant-z plane can be regarded as a H-plane,
but the patterns are typically taken for a plane that includes the maximum magnitude,
which is z ¼ 0 plane or the x–y plane in this example. 2D radiation patterns for field
magnitudes obtained on the E-plane and H-plane are shown in Fig. 4.13c, d,
respectively. A 2D radiation for power is also shown in Fig. 4.13e, on E-plane as
an example. The readers are cautioned not to make a naïve mistake of relating E-
plane and H-plane patterns to E-field and H-field patterns. Again, the (normalized)
field pattern should be the same for the two fields, and E-plane and H-plane simply
denote the reference planes on which the 2D radiation patterns are obtained.
Another example for a radiation pattern would be helpful to introduce a few more
definitions related to radiation patterns. Figure 4.14 shows a typical radiation pattern
from an antenna with a finite dimension, in which case the phase differences in
different parts of the antenna result in multiple lobes as shown in the pattern. The
pattern shows a main lobe (or major lobe) and multiple side lobes (or minor lobes).
The side lobe in the opposite direction of the main lobe is particularly called the back
lobe. (Sometimes a narrower definition of the “side lobe” may be used, in which a
side lobe indicates a minor lobe that is not the back lobe, or even indicates the lobes
immediately next to the main lobe. In many cases, however, side lobes and minor
lobes can be used interchangeably.) Since normalized to the maximum point value,
the main lobe tip corresponds to a value of 1 in terms of the magnitude.
pffiffiffi The angle
between the two directions where the field magnitude equals to 2, or 0.707 (which
corresponds to 0.5 in terms of power magnitude), is called the half-power beamwidth
(HPBW or HP). Another beamwidth is defined as the angle between the first nulls,
which is called the first null beamwidth (FNBW) (sometimes, but less often, called
the beamwidth between first nulls (BWFN)). There may be other definitions for
4.2 Antennas 185
Antennas usually have a preferred direction in which the radiation is stronger than
other directions, unless the antenna is perfectly isotropic. Such a property is quan-
tified with a parameter called the directivity D, which is defined as the ratio of the
radiation intensity in the given direction to the radiation intensity averaged over all
directions. Unless stated otherwise, it is implied that the directivity is made for the
direction of maximum radiation intensity, with the following expression:
Um
D¼ , ð4:47Þ
U avg
where Um is the maximum radiation intensity, and Uavg is the radiation intensity
averaged over all directions assuming the source is isotropic with the same total
radiation power. The definition will be better visualized with the help of Fig. 4.15,
which compares the actual radiation pattern and the isotropic equivalence with the
Um g
’ av
= U
G
Um
D
U
= U
U’avg
a
vg
Uavg
Averaged power
accepted by antenna Averaged power
at input (Pin) radiated by antenna
at output (P )
Fig. 4.15 Definition of directivity D and the gain G. (Adapted with permission from [6], © 2012
Wiley)
186 4 THz Propagation and Related Topics
same total radiation power. It is noted that the directivity is a unitless parameter,
often presented in dB.
When an antenna is used as a part of a system, not all of the power delivered from
the system to the antenna is radiated by the antenna. In realistic situations, there is
always a finite loss in the process of the power conversion due to the internal ohmic
loss, both conduction and dielectric. The radiation efficiency er is introduced to
account for this loss between the input and the output of antennas with the following
definition:
P
er ¼ , ð4:48Þ
Pin
where P is the radiation power and Pin is the power delivered to the antenna at the
input. It is noted that the power reflection due to the input mismatch is not considered
as part of the loss in the definition of the radiation efficiency, since Pin denotes the
power eventually accepted by the antenna. Another parameter, the total antenna
efficiency, less popular, accounts for the loss due to the reflection as well, leading to
a slightly smaller value than the radiation efficiency. Sometimes, it is useful to
express the radiation efficiency in terms of the radiation resistance Rr, which is
defined as Rr ¼ 2Pr/|IA|2 where Pr is the averaged radiation power and IA is the
current supplied to the antenna. For the antenna impedance given by ZA ¼ RA + jXA,
RA is composed of two components such that RA ¼ Rr + R0, where R0 represents the
resistance due to ohmic loss. In terms of these parameters, the radiation efficiency
can be also expressed as:
Rr
er ¼ : ð4:49Þ
Rr þ R0
With the definition of the radiation efficiency given, we can now discuss the
antenna gain. The antenna gain G is defined as the directivity multiplied by the
radiation efficiency, such that:
G ¼ er D: ð4:50Þ
In more formal terms, the antenna gain refers to the ratio of the radiation intensity in
the given direction to the radiation intensity that would be obtained if the power
accepted by the antenna were (entirely) radiated isotopically [8]. In a plain language,
it is basically a directivity that also takes into account the antenna loss. A visual
comparison of gain and directivity can be found in Fig. 4.15. As is the directivity, the
antenna gain is a unitless parameter and in many cases shown in dB scale, particu-
larly with a unit of dBi (i for isotropic). It should be remembered that there is no true
power gain obtained even if the antenna gain is larger than 1. An antenna is not an
amplifier, and, in fact, it is a passive device, which causes overall loss of power. The
antenna “gain” is achieved over the ideal isotropic radiation, not over the input
power.
4.2 Antennas 187
Now let us shift out attention to an antenna at the receiver side and consider the
relation between the power density incident on the antenna and the actual power
received by the antenna. By common sense, the received power will be the product
of the incident power density and the area. The issue here is how we should define
the area. Is this the physical area of the antenna? The answer is apparently no,
because there is not a clear definition of the physical area for many antenna types.
Even for those with a fairly well-defined physical area (such as aperture antennas), it
is not the entire power incident on the physical area that is accepted by the antenna.
For this purpose, we define the effective aperture, or the effective area, which is
defined with a relation:
PA ¼ SAe , ð4:51Þ
where PA is the available power for the receiving antenna, S is the incident power
density, and Ae is the effective aperture. Hence, Ae can be regarded as the virtual area
that relates the incident power density and the actual received power. The expression
given in Eq. (4.51) is based on the assumption that the antenna is oriented to the
maximum response, polarization-matched to the incoming wave, and impedance-
matched to the load. If we further assume that the internal ohmic loss of the antenna
is zero, we can introduce the maximum available received power Prm and the
maximum effective aperture Aem such that:
Here, Aem is related to Ae with Ae ¼ erAem (the relation between Aem and Ae may
appear differently for some literature [5], but we will follow that in [6]). Interest-
ingly, there is a simple relation between the directivity and the maximum effective
aperture as follows:
4π
D¼ Aem : ð4:53Þ
λ2
This relation is general and holds for any type of antenna. (The proof can be found in
general antenna textbooks.) If we recall the relation between the directivity and the
gain given in Eq. (4.50), we also have:
4π
G¼ Ae : ð4:54Þ
λ2
received power of a link between two antennas. Let us first consider the power
density S incident on the receiving antenna located at a distance R from the
transmitting antenna with an antenna gain of Gt and transmitting power Pt, which
is given as:
Gt Pt
S¼ : ð4:55Þ
4πR2
This relation can be understood intuitively, because an isotropic radiation would lead
to S ¼ Pt/4πR2 for a point at R, and a factor of Gt should be multiplied for a
directional radiation. With the given incident power density, the received power Pr
can be easily obtained with the help of Eq. (4.51):
Gt Pt
Pr ¼ Aer , ð4:56Þ
4πR2
where Aer is the effective aperture of the receiving antenna and the assumptions for
Eq. (4.51) still hold. Finally, by applying Eq. (4.54) to Eq. (4.56) with receiving
antenna gain of Gr, we obtain the final form of the Friis transmission equation:
2
λ
Pr ¼ Gt Gr Pt : ð4:57Þ
4πR
According to this equation, the received power reduces with R2 but can be improved
with antenna gain on either side. It is notable that the received power is proportional
to λ2, which stems from the fact that the effective aperture is proportional to λ2 for a
given antenna gain as can be seen from Eq. (4.54). Equivalently, it can be said that
the antenna gain is proportional to 1/λ2 for a given effective aperture. This is a natural
result from the diffraction viewpoint, because less diffraction will happen with a
smaller wavelength for a given aperture, leading to a higher directivity. However, it
will serve as one of the major challenges for THz communication systems, as the
radiated power decays increasingly faster with higher carrier frequency.
EIRP ¼ Pt Gt , ð4:58Þ
4.2 Antennas 189
where Pt is the transmission power and Gt is the transmitting antenna gain. This
relation can be explained again with the help of Fig. 4.15. If the total radiation power
Pt of a directive antenna is averaged over all directions to be isotropic, the power
level is down from the peak by Gt. If we want to recover the peak power level with an
isotropic antenna, the power needs to be increased by Gt, leading to Eq. (4.58). Here,
to be precise, Pt is the accepted power by the transmitting antenna from the system
attached at the input, which may be slightly larger than the radiated power consid-
ering the antenna loss.
Equation (4.58) can be substituted into Eq. (4.57), leading to:
2
λ
Pr ¼ Gr EIRP: ð4:59Þ
4πR
This relation implies that EIPR is a useful parameter to describe the transmitter when
the detected power at the receiver is available while the gain of the transmitting
antenna is unknown. For example, when characterizing a radiating on-chip THz
signal source, they often report EIRP instead of the source output power, since EIRP
can be directly obtained with received power from Eq. (4.59). Note that the receiving
antenna gain is usually well defined (such as the case of a commercial horn antenna)
unlike the gain of the transmitting antenna integrated with the source. In this case,
the actual source output power can be obtained from EIRP only with an estimated
gain of the on-chip transmitting antenna.
4.2.1.6 Polarization
The polarization is another main parameter that describes an antenna. Let us take a
review on the concept of polarization first. In a general sense, the polarization can be
understood as the time-varying behavior of the electric field vector of propagating
electromagnetic waves at a given location. Assume you are measuring the electric
field at a certain point inside the radiation field as a function of time. If the radiation
is polarized, there can be three scenarios for the behavior of the electric field vector.
First, the direction of the vector is fixed along a certain axis, but its magnitude
periodically changes (Fig. 4.16a). This case is called the linear polarization. Second,
the direction of the vector makes a circular rotation, while the magnitude remains
constant (Fig. 4.16b). This is called the circular polarization. Finally, the direction of
the vector makes a circular rotation simultaneously with a periodic change of the
magnitude, resulting in an elliptical trace of the point vector (Fig. 4.16c). This is a
general case positioned between the first and second extreme cases (i.e., linear and
circular cases), called the elliptical polarization. The three cases are illustrated in
Fig. 4.16 in terms of the trace of the electric field vector. The circular and elliptical
polarization can be sub-divided into left-handed and right-handed polarization,
which correspond to clockwise and counter clockwise rotation, respectively, when
propagation is toward the observer.
190 4 THz Propagation and Related Topics
Fig. 4.16 Three polarization cases in terms of the electric field vector: (a) Linear polarization, (b)
Circular polarization, (c) Elliptical polarization. Right-handed polarization is assumed for the
circular and elliptical polarizations with propagation toward +z direction
For the readers who are not fully comfortable with the concepts of the three types
of polarization with Fig. 4.16, it may help to compare the three cases in terms of the
relative phase of two orthogonal components as presented in Fig. 4.17. For the linear
polarization, let us first take the orientation of the two mutually perpendicular axes
on the wave front, so that the magnitude of the field is equally divided into the two
axes with 45 angle against both axes (x- and y-axis in the example shown in the
figure, with wave propagation along the z-axis). In this case, the phases of the x and
y components are perfectly aligned as depicted in Fig. 4.17a, which is shown along
with the vector diagram for reference. Also, the peak intensity will be the same for
the two components. If we relax the condition for the orientation selection of x- and
y-axis, so that the magnitude is not necessarily divided equally into the two axes,
which is in fact the general case, the waveform plot will now appear as shown in
Fig. 4.17b. For the circular polarization, there will be a phase difference of π/2 as is
described in Fig. 4.17c. It is noted that the waveform and phase difference will be
independent of the choice of x- and y-axis, which is a unique property of the perfect
circular polarization. As for the elliptical polarization, the phase difference will show
an intermediate value between 0 and π/2 in general, as is presented by Fig. 4.17d.
However, depending on the choice of polarization axis with respect to x- and y-axis,
the phase difference and relative intensity may appear differently. In the special case
when the polarization axis is aligned with either x- or y-axis, the phase difference will
be reduced to π/2, with the difference in the peak intensities maximized, which is
shown in Fig. 4.17e.
Although unpolarized electromagnetic waves do exist in the nature where the
multiple polarization states co-exist in a mixed manner, the radiation from an
antenna is always polarized with one of the polarization states. Accordingly, the
polarization of an antenna is defined as the polarization of the wave radiated by the
antenna. When the polarization varies with the radiation direction, which is often
the case, the polarization of the main beam is taken. The transmitting antenna and
receiving antenna should be matched with the same polarization states (in terms of
types as well as the axis orientation) for maximum transmission efficiency. If not
matched, it may result in a partial or even total loss of the transmission.
4.2 Antennas 191
Fig. 4.17 (a) Linear polarization with the field magnitude equally divided into x- and y-axis. (b)
Linear polarization with an arbitrary angle of the field with respect to x- and y-axis. (c) Circular
polarization. (d) Elliptical polarization with the field magnitude equally divided into x- and y-axis.
(e) Elliptical polarization with the polarization axis aligned with x-axis. Right-handed polarization is
assumed for the circular and elliptical polarizations
192 4 THz Propagation and Related Topics
4.2.1.7 Bandwidth
Bandwidth is also a key parameter of antennas. There are a few different ways to
define the bandwidth of antennas. The antenna bandwidth, simply denoted as BW, is
given as the difference between the upper boundary ( fU) and lower boundary ( fL) of
operation, or BW ¼ fU fL. However, a more practical measure of the frequency
range covered by an antenna is provided when the bandwidth is expressed as a ratio,
rather than in terms of the absolute difference. For narrow band antennas, the
fractional bandwidth is widely used, which is defined as ( fU fL)/fC where fC is
the center frequency. It is often presented in terms of percentage (%) as well. Here, fC
can be defined
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi by either the arithmetic average ( fU fL)/2 or the geometric average
f U = f L . The two averages are not much different for narrow-band antennas. For
broadband antennas, the ratio bandwidth is preferred, which is simply defined as fU/
fL. It is also common to present the ratio bandwidth in the form of fU/fL:1 (e.g., 10:1
or 5:1). Note that there is no standard way of determining fU and fL. They can be
generally regarded as boundaries for acceptable or satisfactory operation, for which
many different characteristics of antennas can be considered depending on the focus
of the performance. For example, the frequency boundaries can be determined based
on the input impedance, gain, side lobe level, beam width, polarization, beam
direction, and so forth [5].
If you look around, you will soon notice so many antennas are surrounding our daily
lives. They can be found in all kinds of the wireless mobile devices as well as base
stations, also with traditional devices receiving broadcast signals such as radios or
TVs. If you happen to work on experiments based on air radiation of electromagnetic
waves such as wireless links, lots of test antennas will be populating your laboratory.
Those antennas take on various sizes and shapes, and you can easily come up with a
long list of different types of antennas. Overviewing all those individual types of
antenna will be overwhelming, and we can instead group them into certain categories
based on common features and review each category. There can be many different
ways for categorization, while we will follow here the one suggested in [6], which
groups antennas into four different basic types: electrically small antennas, resonant
antennas, broadband antennas, and aperture antennas.
Electrically small antennas refer to those with a dimension much smaller than the
wavelength. Technically, they may include any antennas with a dimension that fit
into a sphere of radius a smaller than 1/k, where k is the wave number, or 2π/λ
[9]. However, for most practical electrically small antennas, a is much smaller than
1/k, by at least a factor of an order. As the antenna performance is generally degraded
if the dimension shrinks far smaller than the wavelength, one may want to avoid too
much scaling of the antenna sizes. Nevertheless, in many cases, there exist practical
4.2 Antennas 193
constraints that limit the antenna size, especially when the wavelength is larger than
the dimension that can be properly handled in a practical way. For example, AM
radio signal spans a wavelength range of a few hundred meters, but antennas with a
half of these wavelengths will not be a practical solution, in which case electrically
small antennas need to be used. They typically exhibit a low directivity, which is
somewhat expected from Eq. (4.53) as the effective aperture of small antennas is
much smaller than the wavelength. The low directivity is not necessarily a short-
coming and sometimes may be preferred. However, when it comes to the impedance
characteristics, the small antennas pay the price for the compactness as explained
below. Most of all, the radiation resistance is very small, which limits the radiation
power level with a relation Pr ¼ Rr|IA|2/2. For a short dipole antenna, which is one of
the typical examples of electrically small antennas that also include small loop
antennas (see Fig. 4.18), Rr is proportional to (Δz/λ)2 where Δz is the length of the
dipole. This leads to radiation power also proportional to (Δz/λ)2 if IA remains the
same. It is clear from this relation that the radiation power will drop significantly for
small antennas where Δz λ and so will the radiation efficiency. The radiation
efficiency of electrically small antennas may easily fall below 10%. The small
radiation resistance also results in a small real part of the antenna impedance (i.e.,
resistance), making the impedance match with the adjacent system, typically termi-
nated around 50 Ω, quite difficult. The large reactance component of electrically
small antennas also adds to the difficulty in the impedance matching. Another
downside of electrically small antennas is the small bandwidth as limited by the
large Q-factor, which rapidly increases with a smaller electrical dimension [10]. The
improvement of the bandwidth with reduced Q-factor needs a larger dimension or
enhanced structural optimization. As mentioned above, however, there are cases
where the adoption of small antennas cannot be avoided, especially for applications
with longer wavelengths.
Resonant antennas have dimensions that support the resonant modes of the
corresponding wavelengths. They typically involve a dimension of a half wave-
length or its integer multiples, which allows standing waves formed across the
antenna structure. Being based on the resonant mode, they naturally show narrow
bandwidth characteristics. On the other hand, resonant antennas in many cases
exhibit impedance characteristics that are favored for impedance matching, with
resistance not far from 50 Ω and small reactance values. The examples of the
194 4 THz Propagation and Related Topics
Fig. 4.19 Examples of resonant antennas: (a) Half-wave dipole antenna, (b) Half-wave folded
dipole antenna, (c) Patch antenna, (d) Yagi-Uda antenna
resonant antennas are shown in Fig. 4.19, which include half-wave dipole antennas,
patch antennas, and Yagi-Uda antennas. The first two antenna types will be
described in more detail separately in Sects. 4.2.3 and 4.2.4. The Yagi-Uda antenna
will be briefly described here. It is composed of a dipole-based driver and parasitic
elements, director and reflector, which are added to improve the directivity (see
Fig. 4.19d). The directors are typically composed of multiple conductor bars, placed
in the direction of the radiation. They significantly improve the directivity, which
increases with the number of the director elements, invoking a trade-off between size
and performance. The reflector, usually a single element bar, is located on the other
side of radiation, which reflects the power radiated to the opposite side of the main
lobe back to the main direction. The spacing between the reflector and the dipole
driver (SR) as well as between the director elements (SD) also affect the directivity,
the optimum value ranging around 0.15–0.25λ for SR and 0.2–0.35λ for SD. The
directors are usually slightly shorter than the dipole driver, while the reflector is
longer, typically around 0.5λ. Although not typically practiced, there are reports on
on-chip Yagi-Uda antenna for end-fire-type radiation as well [11, 12].
Broadband antennas literally indicate a group of antennas that show broadband
characteristics. Examples of broadband antennas include traveling-wave antenna,
helical antenna, biconical antenna, bow-tie antenna, log-periodic antenna, and spiral
antenna, as sketched in Fig. 4.20. When an extremely large bandwidth of larger than
10:1 is achieved, those antennas are particularly called frequency-independent
antennas. Spiral antennas and log-periodic antenna belong to this category. As spiral
antennas and log-periodic antenna are widely employed for planar THz circuits, they
will be described in more detail later in Sect. 4.2.5. Here a brief introduction to other
broadband antennas will be made. The traveling-wave antenna (Fig. 4.20a) is an
antenna that supports a traveling wave without reflection wave, which typically
covers a bandwidth larger than 2:1. A long conducting wire, longer than a wave-
length, terminated with a matched load may serve as a traveling-wave antenna. It
shows a main lobe formed with a certain angle off the longitudinal direction of the
wire with an azimuthal symmetry around it as depicted in the figure. There are
variants such as V antenna and rhombic antenna, designed to provide a main lobe
aligned toward the longitudinal direction of the wire. The helical antenna
4.2 Antennas 195
Fig. 4.20 Examples of broadband antennas: (a) Traveling-wave antenna (shown with the radiation
pattern), (b) Helical antennal, (c) Biconic antenna, (d) Bow-tie antenna, (e) Log-periodic antenna,
(f) Spiral antenna
(Fig. 4.20b) is basically a conductor wound to form a helix. When the radius of the
helix is small compared to the wavelength, it operates in the normal (broadside)
mode and the radiation is made toward the side of the main axis of the helix (normal
to the helix axis). When the helix diameter is comparable to or larger than the
wavelength, its operation is in the axial (end-fire) mode with the main lobe formed
toward the helix main axis. The latter is often considered more practical as a wider
bandwidth and higher efficiency can be achieved. The biconical antennas
(Fig. 4.20c) is a modified dipole antenna intended for a wider bandwidth. It is
composed of two cones (solid or shell) with a gap near the center, across which an
AC driving voltage is applied. It can be considered as a dipole antenna, in which
each half wire is tapered to obtain broadband characteristics. The radiation beam is
formed symmetrically on the side, similar to the case of dipole antennas. The bow-tie
antenna (Fig. 4.20d) is a planar version of the biconical antenna. Although bow-tie
antennas are more sensitive to frequency variation than biconical antennas, its
simpler structure and compatibility with integrated circuit have made it a popular
option for THz integrated circuits [13]. The log-periodic antenna (Fig. 4.20e) and
spiral antenna (Fig. 4.20f) will be described in more detail in Sect. 4.2.5.
Aperture antennas are characterized with the presence of a physical aperture,
through which the radiation or reception of electromagnetic waves are made. The
most outstanding property of aperture antennas is the high gain. It can easily exceed
196 4 THz Propagation and Related Topics
Reflector
(a) (b)
20 dBi and, if designed properly, it can reach a gain higher than ~50 dBi depending
on the actual type. Horn antennas and reflector antennas are two widely employed
examples of aperture antennas, which are sketched in Fig. 4.21. A horn antenna
(Fig. 4.21a) can be considered as an open-ended rectangular waveguide with a
tapered transition aperture attached at the end, which serves as a matching structure
to the open space, suppressing reflection and increasing the bandwidth. The aperture
is typically of rectangular or conical shapes in most practical cases. Horn antennas
show a moderate bandwidth of 50% or larger. Although they are usually built as
individual components, a horn antenna formed on a planar substrate for compatibil-
ity with integrated circuits has also been reported for THz operation [14]. Reflector
antennas (Fig. 4.21b) are composed of a reflector and a feed antenna located at the
focal point of the reflector. The reflector, typically with a parabolic surface, reflects
and focuses the incident plane wave into the feed antenna (for receiver), or radiate a
plane wave by reflecting the normally incident beam emitted from the feed antenna.
For the feed antenna, horn antennas or open-ended waveguides are frequently
employed, while dipole antennas can be also used at low frequency ranges such as
UHF. As the bandwidth of a reflector itself is large, limited by the reflector size
(lower boundary) and the surface distortion (upper boundary), the bandwidth of the
entire antenna is usually determined by that of the feed antenna. Often called dish
antennas, the reflector antennas are a kind of antenna that can achieve the highest
gain and are deployed to various parts of private and public applications that require
an extremely high gain including satellite and space communications.
In the following sections, a few antenna types widely employed for planar THz
systems will be reviewed, starting with dipole antennas. The operation of dipole
antennas can be better understood if we first consider an open-ended two-wire
transmission line as shown in Fig. 4.22. When a wave is guided by the open-
ended transmission line, there will be a reflection at the end of the transmission
line, establishing a standing wave with nodes formed at every λ/2 point from the end.
This will cause a periodic distribution of the current on the wire as shown on the top
4.2 Antennas 197
of Fig. 4.22. Also shown in the figure as arrows, the direction of the current on the
lines will alternate as it passes every node. Notably, the current direction is the
opposite for the two line sections facing each other, a characteristic of differential
signal. The electric and magnetic fields will be confined close to the wires. Now, if a
portion of the wires are bent outward as shown on the bottom of Fig. 4.22, the current
will oscillate up and down in the same direction for the two wires, and the alternating
fields will be exposed to the open space, resulting in radiation. This is the basic
structure of a dipole antenna. Assumed in the figure is that the bent was made at λ/4
point from the end, resulting in a dipole with a total length of λ/2. A dipole antenna
with this particular dimension is called the half-wave dipole antenna (also shown in
Fig. 4.19a). As the length exactly matches a single period of the standing wave, a
fundamental resonant mode is supported and the radiation efficiency will be maxi-
mized. This explains the popularity of the half-wave dipole antennas, and this
subsection will also focus mostly on this type of dipole antenna.
The current distribution on the half-wave dipole will be sinusoidal of a half cycle,
with a peak at the center and zeros on both ends. The corresponding voltage
distribution will be also a sinusoidal half cycle, but with a peak at the ends and a
zero at the center. This electrical configuration will result in a radiation pattern as
follows:
where θ is the angle from the axis of dipole. It shows only θ-dependence, as is
verified with the 3D radiation pattern shown in Fig. 4.23a. 2D E-plane radiation
pattern is shown in Fig. 4.23b, which is slightly elliptic with a HPBW of 78 .
Naturally, dipole antennas show a linear polarization aligned along the direction of
the dipole (called vertical linear polarization). H-plane radiation pattern is a simple
circle. Equation (4.60) can be compared with the case for a very small dipole (see
198 4 THz Propagation and Related Topics
Fig. 4.23 Radiation pattern of a half-wavelength dipole antenna: (a) 3D pattern, (b) 2D E-plane
pattern
Fig. 4.13), for which F(θ) ¼ sin θ. In this simple case, the E-plane radiation pattern
will be a pair of perfect circles in contact, leading to a HPBW of 90 . The directivity
of a half-wave dipole can be calculated from the total output power and the
maximum radiation intensity, leading to a value of 1.643 or 2.15 dBi. This is slightly
larger than that of a short dipole, which is 1.5. The directivity increases with the
dipole length, which can be expected from the increased effective aperture with
longer dipoles. Accordingly, the radiation pattern will be more elliptic for longer
dipoles, further deviating from the circle of short dipoles.
The impedance of a half-wave dipole can be also calculated, which is
73 + j42.5 Ω assuming the wire of the dipole is very thin. As the resistance (real
part of impedance) value is close to 50 Ω, half-wave dipoles are reasonably com-
patible with 50-Ω systems. For 75-Ω transmission lines, which are adopted for some
applications, half-wave dipoles provide an almost ideal resistance value. Both
resistance and reactance values show a strong dependence on the dipole length,
both increasing (more positive) at least until the length becomes close to λ. A slight
reduction of the dipole length leads to zero reactance, which is often practiced by
design. In fact, this is the actual resonance point of the dipole with a finite wire
thickness. Note that the characteristics and the numbers derived above are for the
ideal dipole antenna placed in free space. When the antenna is built on top of a
dielectric material, which is the case for the on-chip antennas that will be discussed
shortly, they will surely deviate from the ideal states. Being a resonant antenna, half-
wave dipole antennas exhibit relatively narrow bandwidth, roughly around in the
order of 10% in terms of input matching (VSWR < 2). One effective way to improve
the bandwidth is to increase the thickness of the dipole (or metal width of planar-type
dipole).
One variance of the dipole antenna is the folded dipole antenna (see Fig. 4.19b). It
can be understood as two dipole antennas placed in parallel and connected on both
4.2 Antennas 199
ends. The structure supports two modes depending on the current distribution, the
transmission mode and the antenna mode, the latter being used for antenna opera-
tion. The radiation pattern is similar to that of regular dipole antennas, but the
impedance is quadrupled, which is useful for some applications that involve high
impedance transmission lines such as 300-Ω lines. Another advantage of the folded
dipole antenna is the broader bandwidth, which makes them a popular choice as the
feeding antenna for other types of antenna including Yagi-Uda antennas.
The on-chip integration of dipole antennas is a feasible option as dipole antennas
can be configured compatible with integrated circuits based on standard semicon-
ductor fabrication processes [15, 16]. There are many advantages for antennas if they
are integrated with other circuit blocks on a single chip. For example, there will be no
need for off-chip connection schemes such as wire-bonding, which limits the
frequency response as well as the reliability of the whole system. Also, on-chip
integration allows the co-design of antenna and adjacent circuits, which will signif-
icantly relax the requirement for 50-Ω matching, lifting one major constraint for
antenna design and potentially removing some of matching circuits [15]. One
prerequisite for on-chip integration, however, is the reasonable size of antennas,
which has been a great barrier for low frequency operations. In this regard, THz
integrated circuits are favorably positioned, as the size of half-wavelength falls on an
acceptable range from a practical point of view for integration. Remember that λ/2
for 300 GHz is around 500 μm in free space, and is further reduced down by a factor
pffiffiffiffi
of εr if embedded in a dielectric, where εr is the dielectric constant of the
embedding dielectric material. For example, with the dielectric constant of ~3.9 of
silicon dioxide (SiO2), often used as inter-metal dielectric material for the standard Si
processes, the reduction factor will be around 2, resulting in λ/2 of around 250 μm at
300 GHz. Note that if the field lines around the dipole conductor is not fully
embedded in the dielectric material and partially exposed to the air (free space),
which is typically the case for on-chip implementation of dipole antennas, the
effective dielectric constant εeff should be used instead of εr. This will be discussed
in more detail in Sect. 4.3.2.
There have been lots of reports that successfully employed dipole antennas,
including folded variants, for THz integrated circuits including folded versions
[17–19]. When a dipole antenna is built on top of the BEOL (back-end of the line)
structure, the upper side of the antenna is (mostly) the air, while the back side is a
BEOL dielectric layer (usually several μm), which is stacked on top of the semicon-
ductor substrate (around 100–200 μm thick, depending on the extent of the wafer
thinning). With this configuration, only a fraction of the total radiation will radiate
upward, leading to a main beam formed toward the substrate due to the higher
dielectric constant of the material beneath the antenna structure. In theory, when an
antenna is placed on top of an infinitely large homogenous dielectric material filling
3=2
the bottom half of the space, the portion of the air radiation will be roughly 1=εr of
the total radiation, where εr is the dielectric constant of the dielectric material
[20]. Substitution of εr ¼ 11.7 of Si and 3.9 of SiO2 into the relation will lead to
2.5% and 13%, respectively. Although the substrates of practical uses have a finite
200 4 THz Propagation and Related Topics
dimension and thickness, still most of the radiation from the on-chip integrated
antennas without metal shield underneath will propagate downward through the
substrate, until it reaches the bottom of the substrate and emerge in the air. A major
issue here is that the substrate is naturally lossy and causes a degradation in the
radiation efficiency, which is more serious with Si substrate as it is more lossy than
other substrate options such as GaAs or InP. For higher frequency bands such as the
THz band, the situation is aggravated with the excitation of the substrate modes due
to the short wavelength that is comparable to the substrate thickness [21]. With the
onset of the substrate mode, the substrate wave will propagate inside the substrate
that serves as a waveguide retaining the wave, which will additionally degrade the
radiation efficiency and also affect the radiation pattern. There are techniques
proposed to suppress the substrate mode. The formation of randomly located cavities
or thru-substrate vias will help to mitigate the mode generation [22], while mounting
the chip on top of a silicon lens is also an effective and widely practiced technique
[19, 23].
As a supplementary note, it can be mentioned that slot antennas are also widely
employed as THz on-chip antennas [24, 25]. It is basically a conductor plane with a
hole, or slot, carved into it, for which a signal feed is made at the two slot edges
across the slot. In principle, a slot antenna can be considered as a complementary
structure of a dipole antenna with the dimension of the slot opening equivalent to that
of the dipole conductor. This mutual relation leads to the identical far-field radiation
patterns for the two types of antennas with the same complementary dimension,
except for the interchange of E-field and H-field. Also, there is a simple relation
between the input impedances of the two complementary antennas: the product of
the two impedances is η2/4, where η is the intrinsic impedance of the medium in
which the antennas are immersed. There are occasions where slot antennas are better
suited than dipole counterparts, such as when the antenna is integrated with other
slot-based on-chip planar structures [26]. For related details for slot antennas, the
readers may refer to [27].
Patch antennas are a kind of planar antennas based on the microstrip structures [28],
which is also called the microstrip antenna for this reason. It is composed of a thin
metal conductor placed on top of a dielectric layer with a ground plane on the bottom
as shown in Fig. 4.24a. The metal conductor, or patch, can be formed in various
shapes, including square, rectangular, and circular among the most popular. There
are also a few different options for signal feeding into the metal conductor, such as
the edge-feed that accepts signal from a microstrip line on the patch edge, or the
probe-feed in which signal is fed through a coaxial cable from the bottom. Also, the
proximity coupling provides a signal feed from a microstrip line located beneath
the patch without a direct contact, which helps to suppress spurious radiation and
improve bandwidth, but requires an extra metal layer. In the discussion in this
4.2 Antennas 201
Patch
L W z Vertical
x fields
d εr dielectric Ground
ground plane
(a) (b)
Fringing fields
y
Notches
x
(c) (d)
Fig. 4.24 (a) Structure of a typical patch antenna. (b) Side view showing the vertical components
of the electric field. (c) Top view showing the horizontal components of the fringing electric field.
(d) Patch antenna with notches. (Adapted with permission from [6], © 2012 Wiley)
upward. This is a clear difference from the case of on-chip dipole antennas that
radiate downward through the substrate. This upward radiation is favored in most
practical cases as the issues related to substrate loss are absent and mounting on Si
lenses is not needed. However, radiation through the narrow gap between the patch
and the ground will limit the operation as will be discussed shortly.
The radiation properties of a patch antenna are mostly determined by the length
L and width W of the patch, along with the thickness d and dielectric constant εr of
the dielectric layer underneath. The operation frequency is dictated by L, a natural
consequence of the standing wave being formed along the patch length direction. In
this case, one would expect a relation of L ¼ λ/2 is valid considering the half-wave
standing wave. However, the actual physical length L required will be slightly
smaller than this, because the fringing field protruding outside the patch also
contributes to the total effective length. To be more quantitative, the physical length
will be expressed as:
λ λ0
L ¼ 2ΔL ¼ pffiffiffiffiffiffi
ffi 2ΔL, ð4:61Þ
2 2 εeff
where λ0 is the free space wavelength, εeff is the effective dielectric constant, and ΔL
the fringing length that represents the effect of the fringing field on either side of the
patch length, which is approximated as [5, 6]:
As for εeff for microstrip structures including patches, the following approximate
expression is widely accepted [29]:
1=2
εr þ 1 εr 1 d
εeff ¼ þ 1 þ 12 , if W d: ð4:63Þ
2 2 W
Beware that this formula is based on a quasi-static analysis and does not explain the
frequency dependence of εeff. We will come back to this issue in Sect. 4.3.2. While
the patch length L dominantly influences the wavelength and thus the operation
frequency of the antenna, the width W is a parameter that can be used to tune the
input impedance of the antenna for a fixed L. For an edge-fed half-wave rectangular
patch antenna, the input impedance can be approximately expressed as [6, 30]:
ε2r L
2
Z A ¼ RA ¼ 90 ðΩÞ: ð4:64Þ
εr 1 W
Note that the reactance is zero as the resonant condition is assumed. As an additional
method to control the input impedance (and thus the antenna matching), the forma-
tion of a recess notch on both sides of the microstrip feedline is also widely adopted
4.2 Antennas 203
(see Fig. 4.24d). The input resistance decreases with the recess depth (along
x direction) until it reaches L/2 at which the resistance becomes zero, then it bounces
back eventually to the original value.
As briefly mentioned, the main lobe of the radiation is made on the broadside of
the patch antenna, a direction perpendicular to the patch plane. With the coordinate
denoted in Fig. 4.24a, the electric field that contributes to the far-field radiation is
immersed in the x–z plane. Hence, in terms of the spherical coordinate, the E-plane
corresponds to the ϕ ¼ 0 plane, while the H-plane is for ϕ ¼ 90 . The radiation
patterns for the two primary planes are:
βL
F E ðθÞ ¼ cos sin θ for E‐plane, ð4:65Þ
2
sin βW2 sin θ
F H ðθÞ ¼ cos θ βW for H‐plane, ð4:66Þ
2 sin θ
where β is the phase constant in free space. With the pattern, a linear polarization
along the feed direction (x-direction in Fig. 4.24a) is obtained for rectangular patch
antennas. The directivity of a rectangular patch antennas depends on the patch width
and shows a larger value than dipole antennas. If we assume d λ0/2π, the
directivity is around 6.6 or 8.2 dBi when W λ0, whereas it asymptotically
approaches 8(W/λ0) for W
λ0 [5].
One major downside of half-wave patch antennas is the narrow bandwidth, lower
than that of half-wave dipole antennas, typically falling below 10% for on-chip
cases. An empirical expression for the (fractional) bandwidth of rectangular patch
antennas is given as:
εr 1 W d
B ¼ 3:77 if d λ0 : ð4:67Þ
ε2r L λ0
It is apparent from the relation that large W/L and d/λ0 ratios help to improve the
bandwidth. The dependence on d, in particular, would favor higher frequency
operation, as it will correspond to a longer electrical dimension. For instance, for
on-chip patch antennas with typical parameter values of d ¼ 10 μm, εr ¼ 3.9 (SiO2),
and W/L ¼ 1, the estimated bandwidth for 100 GHz is 2.4%, but it will increase up to
7.2% at 300 GHz. The radiation efficiency also increases with dielectric thickness d,
an obvious result of the radiation barely making its way out through the gap (slot)
between the patch and the ground plane along the patch edge.
The inherent planar configuration of patch antennas has made them one of the
most popular choices for integrated on-chip antennas, especially for high-frequency
applications that allow a reasonable size for half-wave structures. With standard Si
technologies, the patch is typically formed with the top metal layer while the ground
plane is made of one of the bottom metal layers, between which is located the BEOL
dielectric layer. Hence, no extra layer or lithographic patterning step is required to
204 4 THz Propagation and Related Topics
build patch antennas, indicating the full compatibility of the antenna with the
standard fabrication process. One major challenge with this configuration arises
from the fact that the maximum available dielectric layer thickness is predefined
by the fabrication process, typically around 5–10 μm, limiting the performance,
especially efficiency and bandwidth. A report shows that the increase of dielectric
thickness from 3 to 6 μm improves the simulated efficiency from 12% to 43% near
400 GHz [31]. One may consider drastically increasing d by utilizing a backside
metal as the ground plane, in which case the Si substrate thickness will be added for
the total thickness. However, for standard Si technologies, backside metallization is
not offered as a standard process step. Even if in-house post-process backside
metallization were carried out, the resultant exposure to the lossy Si substrate will
degrade radiation properties. Regarding this issue, the situation for III–V technolo-
gies based on GaAs or InP substrates is much more favorable, since backside
metallization is often available and the substrate loss is significantly smaller than
in Si wafers, making it a viable option to use the backside metal as the ground for
patch antennas. In this case, however, the generation of the substrate mode may
degrade the efficiency, partially negating the advantage of increased dielectric
thickness.
Despite these issues, lots of application cases of on-chip patch antennas for THz
operation have been reported, many of them integrated with functional THz circuits
[32–36]. Various modifications have also been made for those on-chip THz patch
antennas to improve the performance or meet the specific applications, some of
which are briefly described below. Differential patch antennas, as opposed to the
standard patch antennas in the single-ended configuration, have been adopted for
some THz applications to facilitate a direct connection to the adjacent circuit block in
the differential configurations [32, 33]. In such a differential patch antenna, which is
driven by a pair of signal feeds carrying a differential signal, the orientation of the
standing wave is dictated by the location of the feed point, not necessarily along the
feed direction as is the case for regular single-ended patch antennas. It was reported
that stacking a dielectric resonator on top of a CMOS on-chip patch antenna resulted
in an antenna gain improvement by more than 5 dB around 340 GHz [37], which is a
similar approach to the antenna with superstrate that was demonstrated for on-chip
slot ring structures [38]. It was also reported that a segmented metallic resonator,
formed with the top metal layer, was stacked on top of a CMOS on-chip patch
antenna and resulted in a significantly improved efficiency and bandwidth around
300 GHz [39]. To mitigate the generation of the substrate mode in patch antennas
with backside ground metal plane as mentioned above, a cavity antenna operating
near 300 GHz has been developed in which via holes are formed around the patch,
effectively suppressing the substrate mode in the InP substrate [40].
4.2 Antennas 205
r ¼ r 0 eaϕ , ð4:68Þ
where ϕ is the angle from a reference axis, r0 is the initial radius at ϕ ¼ 0, a is the
flare rate that indicates the rate of radius increase. The actual shape of the curve is
depicted in Fig. 4.25a. The spiral antenna with shapes based on an equiangular spiral
Fig. 4.25 (a) Equiangular spiral curve with key structure parameters (Reproduced with permission
from [6], © 2012 Wiley). (b) The pattern of a typical of equiangular spiral antenna (log-spiral
antenna)
206 4 THz Propagation and Related Topics
curve is called the equiangular spiral antenna [43], which is the most popular type
of spiral antennas. It is also widely referred to as a log-spiral antenna. As shown in
Fig. 4.25b, it is composed of two symmetric interwound metallic patterns, the edges
of which are all based on an equiangular spiral curve, except for the outmost portion
that is truncated for a finite dimension. The inner minimum radius and the outer
maximum radius of the pattern determine the upper and the lower boundary of the
operation frequency band, respectively, which roughly corresponds to a quarter-
wavelength of the associated frequency. The signal is fed across the inner parts of the
two spiral arms. It can be seen from the figure that the antenna is self-
complementary, as the shape of the area covered by the metal pattern is basically
identical to the shape of the unpatterned area. Hence, in the given configuration, a
spiral dipole antenna and a complementary slot antenna coexist. Also, the pattern is
dictated by the angle as can be seen from Eq. (4.68), with the length being a
dependent parameter of the angle. With these features, the equiangular spiral antenna
is a good example of the frequency-independent antennas.
The radiation is made on the broadside of the antenna along vertical directions
perpendicular to the spiral plane, upwards and downwards, with an approximate
radiation pattern of cosθ, leading to a pair of symmetric main lobes if the antenna
structure is placed in the middle of a homogeneous medium (such as the air).
However, for the case of on-chip implementation, which is frequently found for
THz applications, the radiation will be mostly toward the substrate as was the case
for on-chip dipole antennas due to the higher dielectric constant of the substrate than
that of the air. The radiation is circularly polarized, with the sense (left- or right-
handed) opposite for the two beams upwards and downwards. The antenna imped-
ance is theoretically η/2 in the uniform medium (188.5 Ω in the air), but the actual
value is lower than this due to the finite arm length, non-zero thickness, etc. A slot
version of the equiangular spiral antenna is also available, in which the spiral arm
patterns are made of slots carved into a metal plane. There are other versions of spiral
antennas that are not based on Eq. (4.68). Archimedean spiral antennas maintain the
same metal width and gap as the spiral patterns flare out. A 3D version of the
equiangular spiral antenna is the conical spiral antenna.
The log-periodic antenna is another widely deployed antenna that belongs to the
class of frequency-dependent antennas [44]. In this type of antenna, the geometric
variation follows a periodic function governed by the logarithm of the distance. This
geometric feature is translated into the electrical characteristics of the antenna, such
as the impedance and radiation parameters, varying periodically over the logarithm
of the wavelength and thus the frequency. One well-known example is the TV
antennas that used to cover the roofs of many house buildings but are now fading
away, officially called the log-periodic dipole array (LPDA). For this antenna, an
array of dipole wire antennas is placed with the spacing varying exponentially with
distance (thus varying linearly with its logarithm). Here, we will focus on a different
4.2 Antennas 207
4.3 Waveguides
Rectangular waveguides have long been used from the early stage of microwave
engineering history, and their use has been recently extended for THz band appli-
cations. While planar waveguides, which will be discussed in Sect. 4.3.2, are
increasingly popular for high-frequency applications owing to their compatibility
with integrated circuits, rectangular waveguides are still widely employed in many
practical situations, such as packaging and inter-package links. A rectangular wave-
guide is basically a hollow pipe with a rectangular cross section. The structure is
typically made of metal, although they can also be built based on dielectric, silicon
for instance, if the interior wall surface is properly coated with a conducting film.
The inner space is usually left empty, or filled with the air, while in principle any
dielectric can occupy the space. Figure 4.27 shows a typical rectangular waveguide,
with a cross-section width and height of a and b along x- and y-axis, respectively.
The aspect ratio of the cross section for standard rectangular waveguides is 2:1, or
a ¼ 2b. The planes parallel to y–z plane (including the narrow-side walls) and x–z
plane (including the broad-side walls) are E-plane and H-plane, respectively, as they
contain the corresponding field in the basic operation mode, or the dominant mode as
will be described later.
The nomenclature of standard rectangular waveguides, as briefly mentioned in
Chap. 1, is WRx, where x is the width of the broad side (a in Fig. 4.27) shown in the
unit of 10 mils. “WR” stands for Waveguide Rectangular. The dimensions of
standard rectangular waveguides are summarized in Table 4.1, along with the
lower and upper boundaries of the operation frequency range that will be discussed
later in this section. One can easily notice that the dimensions of the rectangular
waveguides operating in the THz band are very tiny, mostly below millimeter range.
In addition to the raised precision level required for the fabrication, the small
dimension would make the connection between waveguide parts quite challenging,
as even a slight misalignment would lead to a significant loss. One unique feature of
rectangular waveguides is that the body (or the interior surface) is composed of one
piece of conductor. A main consequence of such a configuration is that TEM mode is
210 4 THz Propagation and Related Topics
not allowed inside the waveguide, leaving TE and TM modes only available states
for propagation. More detailed properties of rectangular waveguides will be
reviewed below mainly with TE modes, with a brief comparison with TM modes.
For TE modes, the magnetic field is composed of both longitudinal (Hz) and
transverse components (Hx and Hy), while the electric field has only the transverse
components (Ex and Ey). The non-zero field components can be obtained from the
wave equation, on which the boundary condition (the tangential components of E-
field are zero at the conductor inner surfaces) are applied, leading to the following
expressions [51]:
Fig. 4.28 (a) Components of a wave vector k assuming kc has only x component, which corre-
sponds to TEm0 modes. (b) Profile of E-field for TE10 mode
k2 ¼ k 2c þ β2 , ð4:74Þ
where k is the magnitude of k, also known as the wave number. A key point here is
that k and β are not identical, as opposed to the case with the plane waves, or TEM
mode propagation. This is a direct consequence of the fact that the wave propagation
inside a rectangular waveguide is not aligned with the waveguide axis along the z-
axis, or the longitudinal direction, due to the existence of the non-zero z-component
of H-field. In a sense, the wave is composed of the transverse and longitudinal
components, each corresponding to wave number of kc and β, respectively. We can
also define corresponding wavelengths, the cutoff wavelength λc and the guide
wavelength λg, each satisfying the relation kc ¼ 2π/λc and β ¼ 2π/λg. They can be
understood as effective wavelengths seen along the transverse and longitudinal
directions, respectively, while λ is the true wavelength that can be observed along
the propagation direction.
If we consider the two-dimensional cross-section of rectangular waveguides
normal to the longitudinal direction, the transverse direction can be further
sub-divided into two components, each corresponding to x and y direction, such that:
212 4 THz Propagation and Related Topics
where kcx and kcy, are x and y component of kc, respectively. Again, we can define
corresponding wavelengths for each component of kc, as denoted by λcx and λcy, so
that kcx ¼ 2π/λcx, and kcy ¼ 2π/λcy. Now, with this picture in mind, one can easily see
from Eqs. (4.70)–(4.73) that standing waves are effectively formed along the
transverse direction, both x and y directions. In this view, m and n represent the
number of nodes along x- and y-axis, respectively, satisfying the relations a ¼ mλcx/2
and b ¼ nλcy/2, or equivalently, kcx ¼ mπ/a and kcy ¼ nπ/b. For the longitudinal
direction, wave propagation is maintained as implied by the propagation factor ejβz.
This overall configuration is referred to as TEmn mode. It should be noted that the
solution for m ¼ 0 and n ¼ 0 is trivial as it corresponds to E ¼ H ¼ 0. That is, there is
no TE00 mode allowed inside a rectangular waveguide.
One may wonder why the transverse component kc is called the “cutoff” wave
number. To understand this, consider a modified expression of Eq. (4.74): β ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
k2 k2c . If k is smaller than kc, β will be imaginary. With an imaginary β, the
wave will decay exponentially and does not propagate any more. Hence, kc is the
smallest allowed value for k to maintain the wave propagation, leading to the name
“cutoff”. Now, one can easily infer that if there is a “cutoff wave number”, there
should be a “cutoff frequency” as well. Yes, this is certainly true, as manifested by
the simple one-to-one relation between k and f, or ck ¼ 2πf, where c is the speed of
light. To arrive at the expression for the cutoff frequency fc, first consider the
expression for kc with the boundary condition mentioned earlier:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
2ffi
mπ nπ
k c ¼ kcx þ kcy ¼
2 2
þ : ð4:76Þ
a b
This is the well-known cutoff frequency formula for TEmn modes, below which the
mode ceases to exist (beware that a and b are the waveguide dimensions, whereas
c is the speed of light in this equation and the description below). Each mode has its
own cutoff frequency, and the mode with the lowest fc is called the dominant mode.
For rectangular waveguides with a > b, the dominant mode is TE10. In this case, fc,10
is simply c/2a. The mode with the second lowest cutoff frequency will depend on the
4.3 Waveguides 213
Fig. 4.29 Decomposition of a wave vector k (top figures), shown along with the direction of the
corresponding wave propagation inside a rectangular waveguide (bottom figures) for various
frequency points: (a) At fc,10 ( fc of TE10), (b) Near fc,10, (c) Away from fc,10, (d) At fc,20 ( fc of TE20)
ratio between a and b. For standard rectangular waveguides with a ¼ 2b, this mode is
TE20, which will be excited if frequency is increased beyond fc,20, or c/a. Hence,
standard rectangular waveguides will maintain a single-mode operation for fre-
quency from c/2a up to c/a, beyond which the waveguide will be overmoded. The
operation frequency range of standard rectangular waveguides is typically given as
the range for the single-mode operation, or between c/2a and c/a, as listed in
Table 4.1, although wave propagation is allowed even in the overmoded condition
(but not allowed below fc as mentioned earlier). The “recommended operation
frequency” is narrower than this range to allow margins for practical reasons. In
fact, the frequency ranges for selected waveguide standards as shown in Table 1.3
are based on this recommended operation frequency.
It would be useful to observe what actually happens inside a standard rectangular
waveguide around the dominant TE10 mode including the behavior near the cutoff
frequencies. The readers should remember that, in this mode, E-field has only
y component and kc has only x component (i.e., kc ¼ kcx). Now, take a look at
Fig. 4.29, which illustrates the various cases for the wave propagation path inside the
waveguide with gradually increasing frequency from fc,10 up to fc,20. First consider
the case of f ¼ fc,10, in which k ¼ kc,10 and β ¼ 0 as shown in Fig. 4.29a. This
214 4 THz Propagation and Related Topics
condition indicates that there exists only the transverse component of the wave
vector and the wave does not make propagation along the longitudinal direction. If
f is slightly pushed beyond fc,10, as is shown in Fig. 4.29b, k becomes larger than
kc,10, and β takes on a non-zero real value. This triggers wave propagation, which
will progress along the waveguide following a zig-zagged path inside the waveguide
as depicted in the bottom of Fig. 4.29b. Note that, in reality, a pair of waves with
opposite transverse component will propagate in a symmetric fashion, although it is
not shown in the figure for simplicity. The angle θ between k and β, or the angle the
propagating wave forms against the waveguide wall, satisfies a relation of
sinθ ¼ kc,10/k ¼ fc,10/f ¼ λ/λc,10. With a further increase of f, k grows larger and so
does β (see Fig. 4.29c), which would lead to a more straightened path of the wave
propagation due to reduced θ, as described in the bottom of Fig. 4.29c. As
f continues to climb up and eventually reaches fc,20, k will become large enough to
excite TE20 mode with k ¼ kc,20 ¼ 2π/a. At this point (and beyond), the decompo-
sition of k into kc and β will have two scenarios depending on the mode, which is
depicted in Fig. 4.29d: First, if the wave stays with TE10 mode, β continues to absorb
the entire increase in k, while kc remains at kc,10 ¼ π/a. Second, if the wave enters
TE20 mode, kc will satisfy a new boundary condition of kc ¼ kc,20 ¼ 2π/a, which will
take up most of k, leading to an abrupt reduction of β. As a result, there will be two
modes available inside the waveguide, TE10 and TE20. With a further push of f, β for
TE20 mode will gradually recover the magnitude, although the (averaged) propaga-
tion speed along the waveguide axis will be much smaller for TE20 than TE10 mode.
The characteristic impedance of TE modes, ZTE, is given as the ratio of E-field and
H-field normal to each other, or
Ex E y ωμ k
Z TE ¼ ¼ ¼ ¼ η, ð4:78Þ
Hy Hx β β
where η is the intrinsic impedance of the dielectric inside the waveguide (in free
space, η ¼ η0 ¼ 376.7 Ω). This impedance can be compared with the TEM case
(although not allowed in rectangular waveguide), where ZTEM ¼ η. This is as
expected from the fact that k/β ¼ 1 for TEM, since there is no longitudinal
component for the field and thus no transverse component of the wave vector in
TEM mode. Equation (4.78) is valid for any TE mode, but the actual value will vary
over the mode since the ratio of k/β will be different for each mode. A better insight
can be obtained if the expression for ZTE is modified as follows based on the relation
β2 ¼ k2 kc2:
4.3 Waveguides 215
Fig. 4.30 Characteristic impedance ZTE of TE10 mode for various waveguide standards at the THz
band. It is assumed that the waveguide interior is filled with the air. Note that the values approach
the intrinsic impedance η0 (shown as dotted line) as frequency increases
kη kη η η
Z TE ¼ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , ð4:79Þ
β
k kc
2 2
1 ðkc =k Þ 2
1 ð f c =f Þ2
It is clear from Eq. (4.79) that ZTE is a function of frequency for each mode and will
show a sharp increase near the corresponding fc. Also, η will serve as the lower
boundary for ZTE, which will be approached with a very high f, provided the mode is
maintained. Examples are shown in Fig. 4.30 for various waveguide standards
operating at the THz band assuming the interior is filled with the air.
4.3.1.4 Loss
The loss of metallic rectangular waveguides predominantly arises from the finite
conductivity of the waveguide walls if dielectric loss can be neglected, which is true
for waveguides filled with the air. The corresponding attenuation due to the conduc-
tor loss for TE10 can be expressed in terms of the waveguide dimension as
follows [51]:
Rs
αc ¼ 2bπ 2 þ a3 k2 ðNp=mÞ, ð4:80Þ
a3 bβkη
where Rs is the sheet resistance of the waveguide wall. Rs can be obtained by 1/σδ,
pffiffiffiffiffiffiffiffiffiffi
where σ is the conductivity and δ is the skin depth given by 1= πf μσ , μ being the
216 4 THz Propagation and Related Topics
Fig. 4.31 Attenuation of TE10 mode for various waveguide standards at the THz band. It is
assumed that the interior is filled with the air and copper is used for the waveguide wall structure.
Also shown in dotted lines are fc for each waveguide standard
permeability of the wall conductor. Some readers may not be familiar with the unit
Np (Neper). It is simply a ratio that is shown in its natural logarithm. For a linear
ratio of x1/x0, for instance, its expression in Np is ln(x1/x0). Then, one can easily
show that the conversion from Np to dB can be made by multiplying a factor of
10log(e) (or 20log(e) if the ratio is for E- or H-field) to the value expressed in terms
of Np. Hence, the attenuation αc shown above can be expressed in terms of dB/m by
multiplying a factor of 20log(e) to the value given Eq. (4.80). The attenuation in
TE10 mode for various waveguide standards in the THz band is plotted in Fig. 4.31.
It is assumed that the interior is filled with the air and copper is used for the wall
structure. When compared with fc, also shown in the plot for reference, it is clear that
attenuation sharply increases as the frequency approaches fc.
Up to this point, the properties for TE modes have been discussed. For TM modes,
the fields are expressed as follows [51]:
where Bmn is another arbitrary coefficient. The fields given above may look similar to
the corresponding fields in TE modes at first glance. However, a careful inspection
reveals that all the field components for TM modes will be zero if either m or n is
zero, which is one outstanding difference from TE modes. This means that m and
n should be positive integers for TM modes for non-trivial solutions, not allowing
TM00, TM01, and TM10. Hence, TM11 is the lowest-order TM mode. Another
difference can be found in the characteristic impedance ZTM, which is given as
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Ex Ey β
Z TM ¼ ¼ ¼ η ¼ η 1 ð f c =f Þ2 : ð4:86Þ
Hy Hx k
Hence, η will serve as the upper boundary of ZTM, as opposed to the case of ZTE,
which will be approached with a very large f. Near fc, ZTM will show a value close
to zero.
Other than these, the properties of TMmn modes will be largely similar to those of
TEmn modes. Notably, the relation k2 ¼ kc2 + β2 still holds for TM modes, leading to
the identical expression for the cutoff frequency fc,mn for TM and TE modes. If we
consider a particular case of a ¼ 2b, which is valid for standard rectangular
waveguides, the expression for fc,mn will be reduced to the following for both TE
and TM modes:
c pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
f c,mn ¼ m2 þ 4n2 if a ¼ 2b: ð4:87Þ
2a
This expression predicts that, as the frequency is increased from the dominant mode
TE10, we will experience in sequence the onset of the modes TE20/TE01, TE11/TM11,
TE21/TM21, TE30, and so on. Several modes are listed in Table 4.2 along with the
cutoff frequency values for various waveguides at the THz band. In most practical
applications of rectangular waveguides, a single-mode operation is preferred. There-
fore, the cutoff frequencies of TE10 and TE20 serve as the lower and upper bound-
aries of operation frequency, respectively, for a given waveguide, which were
assumed in Table 4.1.
218 4 THz Propagation and Related Topics
Table 4.2 Cutoff frequency of the selected modes for various waveguides
Cutoff frequency fc (GHz)
Mode WR6.5 WR5.1 WR3.4 WR2.0 WR1.5 WR1.0
TE10 91 116 174 295 393 590
TE20 182 231 347 590 787 1180
TE11 203 259 388 660 880 1320
TM11 203 259 388 660 880 1320
TE21 257 327 491 835 1113 1669
TM21 257 327 491 835 1113 1669
TE30 272 347 521 885 1180 1770
TE31 327 417 626 1064 1419 2128
TM31 327 417 626 1064 1419 2128
The microstrip line is made of a single conductor strip placed on top of a dielectric
substrate with grounded backside. Figure 4.32 shows the structure of a typical
microstrip line and the field lines in normal operation. Its simple structure and
compatibility with interconnect lines of other electronic components explain its
popularity with a practical basis. As shown in Fig. 4.32b, the field lines are spread
over both air and dielectric regions, which complicates the propagation properties of
microstrip lines. If there were wave propagating in TEM mode along the microstrip
line with the given structure, the fields in the air and in the dielectric will show a
pffiffiffiffi
phase velocity difference by a factor of εr of the dielectric, leading to a phase
mismatch at the boundary. This would be inconsistent with the boundary condition
that enforces the continuity of the fields across the boundary. This expected incon-
sistency indicates that microstrip lines cannot support TEM mode, which would
have been supported if the line were fully immersed in a homogeneous medium.
However, the longitudinal components of E- and H-fields are relatively small,
leading to the wave behavior quite close to that of TEM mode. Hence, microstrip
4.3 Waveguides 219
Fig. 4.32 (a) Structure of a microstrip line. (b) Cross-sectional field lines
lines are often regarded as a quasi-TEM transmission line. In this case, the signal line
can be assumed to be immersed in a homogeneous medium with the effective
dielectric constant εeff, which was given earlier in Eq. (4.63). Again, the expression
in Eq. (4.63) is valid only at low frequencies and will increase and approach εr as
frequency increases as will be discussed shortly.
The characteristic impedance Z0 of microstrip lines can be approximated as
follows [53]:
60 8d W W
Z 0 ¼ pffiffiffiffiffiffiffi ln þ ð ΩÞ for 1
εeff W 4d d
120π W : ð4:88Þ
Z 0 ¼ pffiffiffiffiffiffiffih
i ð ΩÞ for 1
W W d
εeff þ 1:393 þ 0:667 ln þ 1:444
d d
According to these expressions, Z0 decreases with increasing W/d ratio. A larger εeff
also reduces the impedance. This behavior is somewhat expected if we recall the
pffiffiffiffiffiffiffiffi
ffi
characteristic impedance of the ideal transmission line, which is given as L=C ,
where L and C represent the inductance and capacitance per unit length, respectively.
A wider width will decrease L but increase C, while a smaller d will also contribute
to increase C. A larger εeff will raise C, of course, all consistent with the behavior
predicted by Eq. (4.88). The approximated expressions, however, are based on
quasi-static analysis. Hence, their accuracy is limited, and, more importantly, they
do not reflect the frequency dependence of the line characteristics that actually exist
as will be shortly discussed below. Nevertheless, these empirical formulas will be
still useful for the initial design of microstrip lines, although the final design should
be polished with the help of more rigorous analyses now readily available with
various EM simulation tools. A quick estimation with Eq. (4.88) shows that an
on-chip microstrip line with 50 Ω will require W/d of roughly 1.6 with εeff of 4, close
to that of SiO2. This translates to W of 16 μm, if d of 10 μm is assumed for a typical
Si technology. A design based on EM simulation suggests a slightly smaller W, but
not very far from the quick estimation.
The loss is another property of waveguides that affects the performance of the
circuit or system when employed [54]. Three types of losses can be considered in
microstrip lines: conductor loss, dielectric loss, and radiation loss [55]. The
220 4 THz Propagation and Related Topics
conductor loss arises from the finite conductivity of the metallic strip and the ground
plane, in conjunction with the skin effect and surface roughness. For a small
conductor loss, it certainly helps to employ metals with a lower resistivity, although
in many cases, especially when using standard semiconductor technologies, the
option is limited to copper or aluminum. As the field is confined approximately
within a skin depth δ below the surface, a larger skin depth will result in a lower
conductor loss. A concern for THz application is that thepskin ffiffiffiffiffiffiffiffiffiffidepth decreases with
increasing frequency as indicated by the relation (δ ¼ 1= πf μσ), leading to a higher
conductor loss. This is true to any metal-based waveguides and emerges as a major
challenge for their adoption to THz applications. For reference, the skin depths of
aluminum and copper calculated based on the bulk conductivity are shown in
Table 4.3 for various frequency points, along with the sheet resistance Rs (¼1/σδ)
calculated from these values. The surface roughness also affects the conductor loss,
as the effective current path length increases with the roughness. The dielectric loss
mainly stems from the lossy substrate of the microstrip lines. It shows a larger value
with a larger loss tangent of the dielectric, as expected, and also with a larger “filling
factor”, which indicates the extent of the field lines immersed inside the dielectric
instead of the air. The radiation loss is due to the microstrip line acting as an antenna
[56]. It is more pronounced when the substrate dielectric constant is small so that the
fields are only weakly coupled to the dielectric. Although it is the conductor loss that
dominates the total loss of microstrip lines in most cases, if lines are formed directly
on semiconductor substrates, especially on Si, the dielectric loss become significant
due to their lossy characteristics and may exceed the conductor loss. Rather fortu-
nately, most of on-chip microstrip lines of today are built on top of the BEOL
dielectric layer, which is electrically separated from the lossy semiconductor sub-
strate by a metal ground plane, leading to relatively small dielectric losses.
Conventional analyses on microstrip lines are quasi-static and assumes near-DC
operation condition. However, microstrip lines are dispersive and indeed show
frequency-dependent behaviors, partly due to the propagation that is not strictly
TEM. The effective dielectric constant εeff is one example, and its value will deviate
4.3 Waveguides 221
3.4
3.2 10 μ m
5 μm
3.0
2.8
0 200 400 600 800 1000
Frequency (GHz)
from the quasi-static expression given in Eq. (4.63) as frequency increases. The
frequency-dependent εeff can be approximated as follows [53]:
εr εeff ð0Þ
εeff ð f Þ ¼ εr 2 , ð4:89Þ
1 þ G f=fp
E
H
(a) (b)
Fig. 4.34 (a) Structure of a coplanar waveguide. (b) Cross-sectional field lines
As in the case of microstrip lines, the field lines in CPW are distributed over both
dielectric and air (see Fig. 4.34), leading to quasi-TEM mode propagation. As for the
effective dielectric constant that is employed to account for the given configuration,
the situation is simpler than the microstrip lines, since the field distribution is
symmetric across the air–dielectric boundary if an infinitely thick dielectric substrate
is assumed. With this assumption, the quasi-static effective dielectric constant εeff is
simply given as the average of the dielectric constant of the air and the substrate such
that:
εr þ 1
εeff ¼ : ð4:90Þ
2
With a finite dielectric thickness, which is more practical, εeff is reduced, apparently
due to the field lines exposed to the air in the space below the dielectric [59]. For
stacked dielectric layers, as in the case of on-chip CPWs formed on top of a BEOL
dielectric layer deposited on a semiconductor substrate, the dielectric constant of
both layers will affect εeff, while the effect of the top layer will be more dominant due
to its denser field lines [60].
The characteristic impedance of CPW depends on the properties of the dielectric
as well as the geometry of the conductor structures, the latter being represented by
three parameters, a, b, and d, as denoted in Fig. 4.34 (assuming the ground plane is
infinitely wide). The compact forms proposed in the literature for Z0 with CPW
mostly involve the elliptic integrals [52, 59] and are not introduced here. The general
trend predicted by the formulas is a smaller Z0 with a larger a/b and larger εeff, again
predictable from the dependence of Z0 on the unit length inductance and capacitance
as discussed above. d is a less critical parameter for CPW than in the case of
microstrip lines, as it does not intimately affect the field configuration in CPW,
except its effect on εeff. The loss of CPW comprises the conductor loss, dielectric
loss, and radiation loss, as was the case for microstrip lines. The conductor loss is
more affected by the lateral dimension than in microstrip lines. The a/b ratio of
around 0.5–0.7 will show the smallest conductor loss, while a/b close to 0 or 1 will
lead to a sharp increase in the conductor loss [52]. The dielectric loss of CPW will be
affected by the loss tangent and the filling factor in a similar way as in the microstrip
line case. As for the radiation loss, it is noted that the parasitic slotmode, often
excited near the line discontinuities, is particularly responsible for the radiation from
CPW structure.
The slot line is composed a conductor plane on a dielectric substrate with a gap, or
slot, that separates the conductor into two parts as shown in Fig. 4.35 [61]. As was
the case for CPW, slot lines can be built with a single-level metallization process, an
attractive feature for many planar applications. The slot width W serves as the main
geometrical parameter, which determines the overall line characteristics together
224 4 THz Propagation and Related Topics
z
z
(a) (b)
H
λeff /2
z
(c)
Fig. 4.35 (a) Structure of a slot line. (b) Field lines in the transverse section. (c) H-field lines in the
longitudinal section
with the dielectric constant of the substrate. The dielectric thickness d plays a minor
role as in CPW. The effective dielectric constant is identical to that of CPW, the
average of the dielectric constant of the air and the substrate, which is reduced if a
finite d is employed. As slot lines are formed on a large conductor plane with a
simple slot formation, it is possible for them to be built on the ground plane of
microstrip lines, allowing co-design with microstrip lines for various passive
components.
The propagation mode significantly deviates from TEM and is rather close to
TE, which results in the highly dispersive behavior of slot lines, limiting its
broadband applications [52, 62]. Also, the presence of a significant longitudinal
H-field component (see Fig. 4.35c) tends to promote radiation off the slot line
surface, leading to a significant radiation loss. In fact, slot line structures are
occasionally employed for antennas, resonant or non-resonant, to make use of
this tendency for radiation. To suppress the radiation, dielectric substrates with a
high εr are preferred, which helps to confine the field inside the dielectric. Or, its
radiating property can be exploited to use the slot line for a dual purpose of
waveguide and radiating element [26].
4.3 Waveguides 225
Fig. 4.36 Various dielectric waveguides: (a) Dielectric rod, (b) Dielectric slab waveguide, (c)
Image line, (d) Insulated image line, (e) Strip dielectric waveguide, (f) Inverted strip dielectric
waveguide
heat sink. However, the metallic plane may cause a conductor loss, the elimination of
which was the main motivation of adopting dielectric waveguides. A compromised
solution is the insulated image line shown in Fig. 4.36d, in which an insulating
dielectric layer is inserted below the guiding dielectric strip [64]. A main idea here is
to maintain the dielectric constant of the insulating layer (εr2) smaller than that of the
guiding strip (εr1), so that the field is mostly confined in the upper dielectric,
avoiding the conductor loss. The opposite choice for εr1 and εr2 (i.e., εr1 < εr2) is
also possible for the given structure, though, if one favors the reduction of the
radiation loss rather than the conduction loss. There have been other variants as
well, such as the strip dielectric waveguide (Fig. 4.36e) [65], in which both radiation
and conduction losses are suppressed with εr2 larger than both εr1 and εr3, and the
inverted strip dielectric waveguide with εr1 > εr2 (Fig. 4.36f), which inherits
the advantages of the strip dielectric waveguide while additionally eliminating the
dielectric loss from the bottom layer [66].
As for the properties of the wave propagating inside dielectric waveguides, it
should be noted that TEM mode is not supported by dielectric waveguides. They
support TE and TM modes, as well as the hybrid modes in which longitudinal field
components exist for both E-field and H-field. The hybrid modes are available only
for non-planar waveguides with two-dimensional field confinements, not available
for fully planar waveguides such as the slab waveguide shown in Fig. 4.36b. One
characteristic feature of the dielectric waveguides is that the fields in dielectric
waveguides are only loosely confined, and they can partially extend outside the
dielectric boundary of the waveguide. This is true even for the cases with the total
internal reflection mentioned above. The loose field confinement is likely to cause
radiation loss, particularly at the curved section, junction, and discontinuities. Also,
such a field configuration leads to the absence of analytic solutions to fully describe
the fields. Even for the simple rectangular dielectric waveguide structure shown in
4.3 Waveguides 227
Fig. 4.37 (a) Structure of a rectangular dielectric slab used for Eqs. (4.91)–(4.94). (b) Profile of Ey
along x-axis, (c) Profile of Ey along y-axis. The outside of the slab is the air
Fig. 4.36a, the exact solution is not available, which is in contrast to the metallic
rectangular waveguides where compact analytical solutions can be readily obtained
with the help of well-defined boundary conditions as discussed earlier.
Consider rectangular dielectric waveguides as an example for a little bit of details.
Inside rectangular dielectric waveguides, all the modes are hybrid as mentioned
above, and they can be grouped into two families, E xmn and E ymn [67]. For Exmn modes,
which are obtained with x-direction polarization, the largest field components are Ex
and Hy, while for the E ymn modes that are available with y-direction polarization, Ey
and Hx are the strongest. The dominant mode is obtained for m ¼ n ¼ 1 for both
cases. Let us take a brief look at the case of E y11 based on the structure depicted in
Fig. 4.37a, which is a simple dielectric slab surrounded by the air. Omitting the time
factor ejωt, approximate expressions for y-component of the E-field, Ey, are given as
follows for various regions [67, 68]:
jxj a, jyj b : E y ¼ E 0 cos ðk x xÞ cos ky y ejβz , ð4:91Þ
jxj > a, jyj b : E y ¼ E1 cos ðkx aÞ cos k y y ekx0 ðjxjaÞ ejβz , ð4:92Þ
jxj a, jyj > b : E y ¼ E2 cos ðkx xÞ cos k y b eky0 ðjyjbÞ ejβz , ð4:93Þ
jxj > a, jyj > b : Ey ¼ 0, ð4:94Þ
where β is the propagation constant, kx, ky and kx0, ky0 are the x- and y-direction
propagation constants inside and outside the waveguide, respectively. A quick
inspection of the expression reveals that a sinusoidal profile is obtained for the E-
field inside the waveguide, while it exponentially decays outside the boundary. We
have E1 ¼ E0 due to the continuity required for the tangential components of electric
field at the dielectric boundary. On the other hand, for the normal components,
E2 ¼ εrE0 as the normal component of the electric flux density (D) should maintain
continuity at the charge-free dielectric boundary. Based on these properties, the
actual field distribution will appear as depicted in Fig. 4.37b, c. One can clearly
observe a finite portion of the field extending outside the waveguide boundary as
228 4 THz Propagation and Related Topics
mentioned above. Consistent with this field extension, kx and ky are slightly smaller
than π/2a and π/2b, respectively, meaning that the half-wavelength is slightly larger
than the physical dimension of the given waveguide for both x and y directions. This
can be compared to the case of the metallic rectangular waveguide with the same
dimension, where kx and ky exactly match π/2a and π/2b, respectively, for the
corresponding mode. H-field components of the dielectric waveguide can be calcu-
lated based on the E-field given above. For E x11 mode, the roles of E-field and H-field
will be reversed. As we move up with increasing frequency beyond the dominant
mode, odd- and even-modes will be alternately show up for the higher modes, as are
characterized with sine and cosine profiles, respectively. The details of the fields and
modes inside dielectric waveguides can be found in other sources [67–69]. The loss
and dispersion characteristics will be dependent on the specific dielectric material
used in the guiding region.
There have been recent increasing efforts to apply the dielectric waveguides for
communication links based on millimeter-wave and THz carrier frequencies. In [70],
a rectangular dielectric waveguide based on polystyrene, with a cross-section
dimension of 8 mm 1.1 mm, was successfully employed for a link between
transmitter and receiver carrying a 12.5 Gbps AKS modulated signal at 57 and
80 GHz. A similar data link experiment through a dielectric waveguide was reported
in [71], for which a circular hollow polymer fiber tube was employed as a waveguide
for a 2.5 Gbps CPFSK (continuous-phase frequency shift keying) modulated signal
transmission at 120 GHz over a distance up to 7 m. The waveguide with an outer and
inner diameter of 2 mm and 1 mm, respectively, exhibited a loss of 2–4 dB/m for a
frequency range of 110–140 GHz.
Various types of waveguides have been so far described. The best-suited waveguide
will vary from application to application, and the optimum waveguide type will be
selected based on the details of the needs from the given application. Obviously,
there may arise situations where two different types of waveguides need to be
connected. The most common situation is the connection between a planar wave-
guide and a rectangular waveguide for packaging purposes, in which waveguide
transition structures need to be employed. There have been a variety of transition
structures developed for this purpose in microwave and mm-wave bands, many of
which can be adopted for the THz region as well. In this subsection, a brief overview
of those transition structures will be presented with focus on two cases: transition
between a microstrip line and a rectangular waveguide, and transition between a
CPW and a rectangular waveguide.
The challenge in the transition between a planar waveguide to a rectangular
waveguide stems from the fact that planar waveguides are composed of two con-
ductors, signal and ground, while rectangular waveguides are made of one single
4.3 Waveguides 229
conductor body. How can those two highly heterogeneous structures be linked
together, preferably with a low insertion loss and broadband characteristics? A
clue can be found if we recall the transverse electric field distribution in the
rectangular waveguides. When the dominant TE10 mode is assumed, Ey will be
given as a sine function of x with a peak located at the middle of the broad-side of the
rectangular cross section and zeros at both edges, as indicated by Eq. (4.73). Ex will
be non-existent. This implies that the largest potential difference is established
between the centers of the two broad-side walls of the waveguide. Consequently,
the best approach is to link the signal and ground conductors of the planar wave-
guides to the two broad-side centers of the rectangular waveguide. Two traditionally
popular transition structures, ridge waveguide transitions and fin-line transitions, are
both based on this approach. Another approach is to radiate (or detect) electromag-
netic waves with the signal line placed inside the rectangular waveguide, which is
the basis for the E-plane probe transition. These three types of transitions are
discussed below.
Another way to make transition is to employ the fin-line structure [74]. For the
transition with a microstrip line as shown in Fig. 4.39, the signal line and the ground
plane are patterned, so that they are gradually curved into the side edges in the
opposite direction, forming fin-shaped transitional regions on top and bottom sides
of the substrate, respectively. The patterned structure is inserted into the rectangular
waveguide on the E-plane, or with its main plane in parallel with the narrow-side
walls of the waveguide (see Fig. 4.39b). In the completed structure, the signal line
and ground plane of a microstrip line will be in contact with the central regions of the
two broad-side walls of the waveguide, which is desired as mentioned above. As the
two fin patterns are situated in the antipodal position to each other, it is often called
the antipodal fin-line transition. Also, it can be viewed as a double-ridge waveguide
transition in a sense, as we may regard the two metal patterns on the substrate as
variant types of ridge. CPW to rectangular waveguide transition can also be
implemented with fin-line structures. However, as the ground planes are placed on
both sides of the signal line on the same metal level for CPW, the formation of the
two fin patterns on the (laterally) opposite side is not straightforward. There have
been various structural solutions reported, a couple of which are mentioned here.
Figure 4.39c shows a structure where two fins on the opposite sides are formed, one
with the signal line and the other with one of the ground planes [75]. The other
ground plane is linked to the other ground with air bridges. In the case shown in
Fig. 4.39d, one fin is patterned with one of the ground planes, which is regular. The
other fin, however, is formed with the signal line and the other ground plane
combined together, the gap between which is terminated with a slot-line radial
stub that serves as a CPW-to-slot line transition [76]. The fin-line transition is
compatible with high-frequency bands as the fin structures are realized with planar
4.3 Waveguides 231
Toward Toward
waveguide microstrip line
(a) (b)
(c) (d)
Fig. 4.39 Fin-line transitions: (a) Waveguide-to-microstrip line, (b) Waveguide-to-microstrip line
fixed inside the waveguide, (c) Waveguide-to-CPW with airbridge links, (d) Waveguide-to-CPW
with slot line radial stub
232 4 THz Propagation and Related Topics
E-plane
probe
Toward
Microstrip line
Waveguide
metal patterning that easily scales with increasing frequency, as opposed to the ridge
waveguide structures that requires a precision machining process. For this reason,
fin-line transitions have been widely employed in the THz band for various appli-
cations [77–79].
While the two transitions described above rely on a contact between the signal line of
the planar waveguides and the rectangular waveguide wall, E-plane probe does not
involve such a direct contact between the two conductors [80, 81]. Instead, the signal
line is modified to serve as a radiating component, from which electromagnetic
waves are radiated and coupled into one of the modes supported by the rectangular
waveguide at the given frequency. In a reciprocal way, of course, the same structure
can be used to detect electromagnetic waves propagating inside the rectangular
waveguide. A typical structure of E-plane probe is shown in Fig. 4.40. The signal
line is inserted and suspended inside the rectangular waveguide, lying on the E-plane
of the waveguide (hence the name “E-plane probe”), typically with the line width
modulated for impedance transformation. Although rare, it is possible for the surface
of the signal line to face the propagation direction of the waveguide [82], which is
not an E-plane probe in a strict sense as the probe lies on the H-plane in this case. In
most cases, dielectric substrate is also stretched into the waveguide together with the
signal line to provide a mechanical support to the thin metallic “probe”. Because it is
typical for the signal line to protrude into the waveguide from the side, the
References 233
References
19. D. Yoon, J. Kim, J. Yun, M. Kaynak, B. Tillack, J.-S. Rieh, 300-GHz direct and heterodyne
active imagers based on 0.13-μm SiGe HBT technology. IEEE Trans. Terahertz Sci. Technol. 7
(5), 536–545 (2017)
20. A. Babakhani, X. Guan, A. Komijani, A. Natarajan, A. Hajimiri, A 77-GHz phased-array
transceiver with on-Chip antennas in silicon: Receiver and antennas. IEEE J. Solid State
Circuits 41(12), 2795–2806 (2006). https://doi.org/10.1109/JSSC.2006.884811
21. N.G. Alexopoulos, P.B. Katehi, D.B. Rutledge, Substrate optimization for integrated circuit
antennas. IEEE Trans. Microw. Theory Tech. 31(7), 550–557 (1983). https://doi.org/10.1109/
TMTT.1983.1131544
22. H.-J. Song, J.-Y. Kim, K. Ajito, N. Kukutsu, M. Yaita, 50-Gb/s direct conversion QPSK
modulator and demodulator MMICs for terahertz communications at 300 GHz. IEEE Trans.
Microw. Theory Tech. 62(3), 600–609 (2014)
23. J. Kim et al., Three-dimensional terahertz tomography with transistor-based signal source and
detector circuits operating near 300 GHz. IEEE Trans. Terahertz Sci. Technol. 8(5), 482–491
(2018). https://doi.org/10.1109/TTHZ.2018.2851542
24. M. Asada, S. Suzuki, N. Kishimoto, Resonant tunneling diodes for sub-terahertz and terahertz
oscillators. Jpn. J. Appl. Phys. 47(6), 4375–4384 (2008). https://doi.org/10.1143/jjap.47.4375
25. R. Han, E. Afshari, A CMOS high-power broadband 260-GHz radiator array for spectroscopy.
IEEE J. Solid State Circuits 48(12), 3090–3104 (2013). https://doi.org/10.1109/JSSC.2013.
2272864
26. Z. Hu, M. Kaynak, R. Han, High-power radiation at 1 THz in silicon: A fully scalable array
using a multi-functional radiating mesh structure. IEEE J. Solid State Circuits 53(5), 1313–1327
(2018). https://doi.org/10.1109/JSSC.2017.2786682
27. H.G. Booker, Slot aerials and their relation to complementary wire aerials (Babinet’s principle).
J. Inst. Electric. Eng. IIIA Radiolocation 93(4), 620–626 (1946)
28. K. Carver, J. Mink, Microstrip antenna technology. IEEE Trans. Antennas Propag. 29(1), 2–24
(1981)
29. E.O. Hammerstad, Equations for microstrip circuit design, in 1975 5th European Microwave
Conference, (IEEE, New York, 1975), pp. 268–272
30. D.R. Jackson, N.G. Alexopoulos, Simple approximate formulas for input resistance, bandwidth,
and efficiency of a resonant rectangular patch. IEEE Trans. Antennas Propag. 39(3), 407–410
(1991). https://doi.org/10.1109/8.76341
31. E. Seok et al., Progress and challenges towards terahertz CMOS integrated circuits. IEEE
J. Solid State Circuits 45(8), 1554–1564 (2010). https://doi.org/10.1109/JSSC.2010.2049793
32. E. Ojefors, J. Grzyb, Y. Zhao, B. Heinemann, B. Tillack, U.R. Pfeiffer, A 820GHz SiGe chipset
for terahertz active imaging applications, in IEEE International Solid-State Circuits Conference
(2011), pp. 224–226
33. D. Yoon, K. Song, J. Kim, M. Kaynak, B. Tillack, J.-S. Rieh, A D-band active imager in a SiGe
HBT technology. J. Infrared Millimeter Terahertz Waves 36(4), 335–349 (2015)
34. E. Seok et al., A 410GHz CMOS push-push oscillator with an on-chip patch antenna, in IEEE
International Solid-State Circuits Conference (ISSCC), (IEEE, New York, 2008), pp. 472–629
35. R. Han et al., A 280-GHz Schottky diode detector in 130-nm digital CMOS. IEEE J. Solid State
Circuits 46(11), 2602–2612 (2011). https://doi.org/10.1109/JSSC.2011.2165234
36. S. Chai, S. Lim, S. Hong, THz detector with an antenna coupled stacked CMOS plasma-wave
FET. IEEE Microw. Wirel. Compon. Lett. 24(12), 869–871 (2014). https://doi.org/10.1109/
LMWC.2014.2353211
37. C. Li, T. Chiu, 340-GHz low-cost and high-gain on-chip higher order mode dielectric resonator
antenna for THz applications. IEEE Trans. Terahertz Sci. Technol. 7(3), 284–294 (2017).
https://doi.org/10.1109/TTHZ.2017.2670234
38. J.M. Edwards, G.M. Rebeiz, High-efficiency elliptical slot antennas with quartz superstrates for
silicon RFICs. IEEE Trans. Antennas Propag. 60(11), 5010–5020 (2012). https://doi.org/10.
1109/TAP.2012.2207353
References 235
39. W. Choe, J. Jeong, Broadband THz CMOS on-chip antenna using stacked resonators, in 2017
IEEE International Symposium on Radio-Frequency Integration Technology (2017),
pp. 208–210. https://doi.org/10.1109/RFIT.2017.8048251
40. K.M. Lee, I.J. Lee, S. Jeon, M. Kim, J.S. You, 300 GHz InP rectangular cavity antenna, in 2015
IEEE International Symposium on Antennas and Propagation (2015), pp. 2105–2106. https://
doi.org/10.1109/APS.2015.7305442
41. V. Rumsey, Frequency independent antennas. IRE Int. Conven. Record 5, 114–118 (1957).
https://doi.org/10.1109/IRECON.1957.1150565
42. P.E. Mayes, Frequency-independent antennas and broad-band derivatives thereof. Proc. IEEE
80(1), 103–112 (1992). https://doi.org/10.1109/5.119570
43. J. Dyson, The equiangular spiral antenna. IRE Trans. Antennas Propag. 7(2), 181–187 (1959)
44. R. DuHamel, D. Isbell, Broadband logarithmically periodic antenna structures. IRE Int.
Conven. Record 5, 119–128 (1957). https://doi.org/10.1109/IRECON.1957.1150566
45. A.D. Semenov et al., Terahertz performance of integrated lens antennas with a hot-electron
bolometer. IEEE Trans. Microw. Theory Tech. 55(2), 239–247 (2007). https://doi.org/10.1109/
TMTT.2006.889153
46. S.-P. Han et al., Compact fiber-pigtailed InGaAs photoconductive antenna module for terahertz-
wave generation and detection. Opt. Express 20(16), 18432–18439 (2012)
47. S. Lepeshov et al., Boosting terahertz photoconductive antenna performance with optimised
plasmonic nanostructures. Sci. Rep. 8 (2018)
48. S. Verghese, K.A. McIntosh, E.R. Brown, Highly tunable fiber-coupled photomixers with
coherent terahertz output power. IEEE Trans. Microw. Theory Tech. 45(8), 1301–1309
(1997). https://doi.org/10.1109/22.618428
49. R. Mendis, C. Sydlo, J. Sigmund, M. Feiginov, P. Meissner, H.L. Hartnagel, Tunable CW-THz
system with a log-periodic photoconductive emitter. Solid State Electron. 48(10–11),
2041–2045 (2004)
50. S. Yang, M.R. Hashemi, C.W. Berry, M. Jarrahi, 7.5% optical-to-terahertz conversion effi-
ciency offered by photoconductive emitters with three-dimensional plasmonic contact elec-
trodes. IEEE Trans. Terahertz Sci. Technol. 4(5), 575–581 (2014). https://doi.org/10.1109/
TTHZ.2014.2342505
51. D.M. Pozar, Microwave Engineering (John Wiley & Sons, Hoboken, NJ, 2009)
52. R. Garg, I. Bahl, M. Bozzi, Microstrip Lines and Slotlines (Artech house, Boston, MA, 2013)
53. I.J. Bahl, D.K. Trivedi, A designer’s guide to microstrip line. Microwaves 16(5), 174–182
(1977)
54. R.A. Pucel, D.J. Masse, C.P. Hartwig, Losses in microstrip. IEEE Trans. Microw. Theory Tech.
16(6), 342–350 (1968). https://doi.org/10.1109/TMTT.1968.1126691
55. L. Maloratsky, Passive RF and Microwave Integrated Circuits (Elsevier, Amsterdam, 2003)
56. L.J. van der Pauw, The radiation of electromagnetic power by microstrip configurations. IEEE
Trans. Microw. Theory Tech. 25(9), 719–725 (1977). https://doi.org/10.1109/TMTT.1977.
1129201
57. C.P. Wen, Coplanar waveguide: A surface strip transmission line suitable for nonreciprocal
gyromagnetic device applications. IEEE Trans. Microw. Theory Tech. 17(12), 1087–1090
(1969). https://doi.org/10.1109/TMTT.1969.1127105
58. A. Komijani, A. Hajimiri, A wideband 77-GHz, 17.5-dBm fully integrated power amplifier in
silicon. IEEE J. Solid State Circuits 41(8), 1749–1756 (2006)
59. G. Ghione, C. Naldi, Analytical formulas for coplanar lines in hybrid and monolithic MICs.
Electron. Lett. 20(4), 179–181 (1984). https://doi.org/10.1049/el:19840120
60. S.S. Gevorgian, Basic characteristics of two layered substrate coplanar waveguides. Electron.
Lett. 30(15), 1236–1237 (1994). https://doi.org/10.1049/el:19940861
61. S.B. Cohn, Slot line on a dielectric substrate. IEEE Trans. Microw. Theory Tech. 17(10),
768–778 (1969). https://doi.org/10.1109/TMTT.1969.1127058
236 4 THz Propagation and Related Topics
62. E.A. Mariani, C.P. Heinzman, J.P. Agrios, S.B. Cohn, Slot line characteristics. IEEE Trans.
Microw. Theory Tech. 17(12), 1091–1096 (1969). https://doi.org/10.1109/TMTT.1969.
1127106
63. P. Toulios, R. Knox, Rectangular dielectric image lines for millimeter integrated circuits, in
Western Electronic Show and Convention (1970), pp. 25–28
64. W.V. McLevige, T. Itoh, R. Mittra, New waveguide structures for millimeter-wave and optical
integrated circuits. IEEE Trans. Microw. Theory Tech. 23(10), 788–794 (1975). https://doi.org/
10.1109/TMTT.1975.1128684
65. T. Itoh, R. Mittra, New waveguide structures for millimeter-wave integrated circuits, in IEEE-
MTT-S International Microwave Symposium (1975), pp. 277–280. https://doi.org/10.1109/
MWSYM.1975.1123359
66. T. Itoh, Inverted strip dielectric waveguide for millimeter-wave integrated circuits. IEEE Trans.
Microw. Theory Tech. 24(11), 821–827 (1976). https://doi.org/10.1109/TMTT.1976.1128967
67. E.A.J. Marcatili, Dielectric rectangular waveguide and directional coupler for integrated optics.
Bell Syst. Tech. J. 48(7), 2071–2102 (1969). https://doi.org/10.1002/j.1538-7305.1969.
tb01166.x
68. P. Bhartia, I.J. Bahl, Millimeter Wave Engineering and Applications (Wiley, New York, 1984)
69. D. Marcuse, Theory of Dielectric Optical Waveguides (Elsevier, Amsterdam, 2013)
70. S. Fukuda et al., A 12.5+12.5 Gb/s full-duplex plastic waveguide interconnect. IEEE J. Solid
State Circuits 46(12), 3113–3125 (2011). https://doi.org/10.1109/JSSC.2011.2168870
71. N.V. Thienen, W. Volkaerts, P. Reynaert, A multi-gigabit CPFSK polymer microwave fiber
communication link in 40 nm CMOS. IEEE J. Solid State Circuits 51(8), 1952–1958 (2016).
https://doi.org/10.1109/JSSC.2016.2580605
72. M.V. Schneider, B. Glance, W.F. Bodtmann, Microwave and millimeter wave hybrid integrated
circuits for radio systems. Bell Syst. Tech. J. 48(6), 1703–1726 (1969). https://doi.org/10.1002/
j.1538-7305.1969.tb01147.x
73. G.E. Ponchak, R.N. Simons, A new rectangular waveguide to coplanar waveguide transition, in
IEEE MTT-S International Microwave Symposium Digest (1990), pp. 491–492. https://doi.org/
10.1109/mwsym.1990.99626
74. L.J. Lavedan, Design of waveguide-to-microstrip transitions specially suited to millimetre-
wave applications. Electron. Lett. 13(20), 604–605 (1977). https://doi.org/10.1049/el:19770434
75. J.V. Bellantoni, R.C. Compton, H.M. Levy, A new W-band coplanar waveguide test fixture, in
IEEE MTT-S International Microwave Symposium Digest (1989), pp. 1203–1204. https://doi.
org/10.1109/mwsym.1989.38940
76. V.S. Mottonen, Wideband coplanar waveguide-to-rectangular waveguide transition using
fin-line taper. IEEE Microw. Wirel. Compon. Lett. 15(2), 119–121 (2005). https://doi.org/10.
1109/LMWC.2004.842855
77. C. Groppi, C.Y.D. d’Aubigny, A.W. Lichtenberger, C.M. Lyons, C.K. Walker, Broadband
finline ortho-mode transducer for the 750–1150 GHz band, in Proc. 16th Int. Symp. Space
Terahertz Technol. (2005), pp. 513–518
78. L.A. Samoska, An overview of solid-state integrated circuit amplifiers in the submillimeter-
wave and THz regime. IEEE Trans. Terahertz Sci. Technol. 1(1), 9–24 (2011). https://doi.org/
10.1109/TTHZ.2011.2159558
79. P. Kittara, P. Grimes, G. Yassin, S. Withington, K. Jacobs, S. Wulff, A 700-GHz SIS antipodal
finline mixer fed by a Pickett-Potter horn-reflector antenna. IEEE Trans. Microw. Theory Tech.
52(10), 2352–2360 (2004). https://doi.org/10.1109/TMTT.2004.835976
80. T.Q. Ho, Y.-C. Shih, Y.-C. Shih, Spectral-domain analysis of E-plane waveguide to microstrip
transitions. IEEE Trans. Microw. Theory Tech. 37(2), 388–392 (1989). https://doi.org/10.1109/
22.20065
81. Y.-C. Leong, S. Weinreb, Full band waveguide-to-microstrip probe transitions, in IEEE MTT-S
International Microwave Symposium Digest, vol. 4 (1999), pp. 1435–1438. https://doi.org/10.
1109/MWSYM.1999.780219
References 237
82. Y. Shih, T. Ton, L.Q. Bui, Waveguide-to-microstrip transitions for millimeter-wave applica-
tions, in IEEE MTT-S International Microwave Symposium Digest (1988), pp. 473–475. https://
doi.org/10.1109/MWSYM.1988.22077
83. L. Samoska, A. Peralta, H. Ming, M. Micovic, A. Schmitz, A 20 mW, 150 GHz InP HEMT
MMIC power amplifier module. IEEE Microw. Wirel. Compon. Lett. 14(2), 56–58 (2004).
https://doi.org/10.1109/LMWC.2003.822575
84. P.H. Siegel, R.P. Smith, M.C. Graidis, S.C. Martin, 2.5-THz GaAs monolithic membrane-diode
mixer. IEEE Trans. Microw. Theory Tech. 47(5), 596–604 (1999). https://doi.org/10.1109/22.
763161
85. K.M.K.H. Leong et al., A 340–380 GHz integrated CB-CPW-to-waveguide transition for sub
millimeter-wave MMIC packaging. IEEE Microw. Wirel. Compon. Lett. 19(6), 413–415
(2009). https://doi.org/10.1109/LMWC.2009.2020043
86. W. Choe, J. Jeong, A broadband THz on-chip transition using a dipole antenna with integrated
balun. Electronics 7(10), 236 (2018)
Chapter 5
THz Optical Methods
It was mentioned earlier in this book that there can be two approaches to the THz
research, one upward approach from the electronics and the other downward from
the optics. While the purpose of this book is to cover the topics relevant to the
electronic approach, it is true that the THz research has long been dominated by the
approach based on the optics, which has inevitably influenced THz electronics.
Therefore, it is desired for the intended readers of this book, who are expected to
have their backgrounds in the electronics, to be aware of the basic concepts com-
monly shared in the optics-driven THz territory. It will promote the interaction with
the researchers from the other side of the THz spectrum with a lowered “language
barrier” and also help to import the ideas on that side for a productive blending.
When those with electronics background attempt to learn the basics of optics-based
THz, however, they are often overwhelmed by the amount and details carried by the
references and books targeted at the optics majors. This chapter is prepared to serve
as a quick guide to optics-based THz for the electronics-based THz researchers.
Those who are seeking for a comprehensive coverage of optics-based THz are
referred to other references intended mostly for such purposes [1–3]. In this chapter,
two major topics will be discussed, which are, naturally, the THz wave generation
and detection. The discussion on generation is divided into two sections, one on the
pulse and the other on the continuous wave (CW) generation. The section on
detection is described mainly for the pulse detection, as the CW detection is
commonly performed by electrical means even in the optical environment.
transform. When the pulses are tailored to have a very short time period, in the order
of roughly below 1 picosecond (ps), their corresponding band coverage in the
frequency domain will be extended to reach the THz spectrum, enabling various
exciting applications. Such extremely short pulses became readily available with
optical methods, in large part owing to the advent of femtosecond lasers [4]. Among
the various approaches developed for THz pulse generation, the two most widely
adopted techniques, which are based on photoconductive (PC) material and optical
rectification with nonlinear electro-optic (EO) materials, will be introduced in this
section. The former technique involves free electrons while the latter is based on
bound electrons for THz pulse generation, although they are both driven by femto-
second lasers. Although less popular, THz pulses can be generated based on the
semiconductor surface field, which also rely on femtosecond lasers. All these
techniques will be briefly discussed in this section.
t
− 2 − 2
( )µ cos ( )µ
(a) (b)
Fig. 5.1 Profiles of an ultrashort optical pulse in the time domain: (a) Electric field (b) Averaged
intensity
carrier lifetime. The former is achieved with a femtosecond laser while the latter is
typically realized with low-temperature (LT)-GaAs.
A femtosecond laser is a pulse laser that generates ultrashort optical pulses whose
temporal duration is well below 1 ps, typically in the range of 10–100 fs
(1 fs ¼ 1015 s ¼ 103 ps). Femtosecond lasers are usually based on solid-state
gain media, the most popular one being Ti:sapphire (Ti-doped Al2O3). Ti:sapphire
lasers can be tuned over a wavelength of 650–1100 nm, while the most efficient
operation is obtained near 800 nm, for which a separate pump laser with ~520 nm
wavelength is used for optical pumping. Considering that a wavelength of 800 nm
corresponds to a time period of ~2.7 fs, we can see that a typical optical pulse from a
femtosecond laser contains several to a few tens of cycles for the alternating electric
(or magnetic) field. However, it will be the averaged intensity, which shows a
Gaussian profile in the time domain, that will affect the THz (electrical) pulse
generation. Figure 5.1 shows the profiles of electric field and the averaged intensity
of an ultrashort optical pulse in the time domain. Although a relatively recent
invention with its first appearance around early 1980s, femtosecond lasers have
been rapidly matured and are now readily available as commercial products, which
can be conveniently purchased to serve as a key component in ultrafast optics setups.
LT-GaAs commonly refers to a GaAs epitaxial layer grown at a low temperature.
In the epitaxial growth of semiconductor layers, typically with an MBE or MOCVD,
the crystal quality is critically affected by the growth temperature. A higher temper-
ature will result in a higher quality crystal as there is a higher chance for atoms to find
the right site on the crystal surface. As our purpose is to obtain a low-quality crystal
with lots of defects for this particular application, a semiconductor layer grown at a
low temperature is desired, the opposite direction from the conventional approaches.
GaAs is the most popular and readily available compound semiconductor with a
photoconductive characteristic, and thus widely adopted for this purpose. Typical
growth temperature of LT-GaAs with MBE is around 200 C [5–7], which can be
compared to the conventional high-quality GaAs growth temperature of around
500 C. As-grown LT GaAs will have too many defects to achieve a reasonable
242 5 THz Optical Methods
Beam spot
Photoconductive
layer
DC
bias
(a)
Si hyper-
Photoconductive Substrate hemispherical lens
layer
(b)
electron mobility that is required for THz generation with photoconductive material.
Hence, a post-growth annealing is routinely carried out, after which a typical
mobility of 200–300 cm2/Vs is achieved with a carrier lifetime of around
0.2–0.5 ps [5–7]. While RD-SOS (Radiation-Damaged Si on Sapphire) is also
employed for the application, in which a Si layer grown on Sapphire is intentionally
damaged by a high-energy ion implantation of O+ ions [8], the mobility by this
method is lower and the carrier lifetime is longer than the LT-GaAs case. This makes
LT-GaAs a more popular choice. The following discussion is made assuming an
LT-GaAs layer is employed as the photoconductive material needed for the
application.
With a femtosecond laser and an LT-GaAs layer available at hand, THz (electri-
cal) pulses can now be obtained. For the actual generation of THz pulses, a more
sophisticated structure than simply an LT-GaAs layer is needed, which is called the
photoconductive antenna. A schematic of a typical photoconductive antenna is
illustrated in Fig. 5.2. An LT-GaAs layer, which is inherently photoconductive, is
grown on a substrate, and a planar metallic electrode is patterned on top of the
5.1 THz Pulse Generation 243
∂I photo ðt Þ
E THz ðt Þ / : ð5:1Þ
∂t
1.0
0.5
0.0
0.0
0 5 10 15 0 1 2 3 4 5
Time (ps) Frequency (THz)
(a) (b)
Fig. 5.4 (a) THz pulse in the time domain. (b) Corresponding profile in the frequency domain.
(Reprinted with permission from [11] © The Optical Society)
electric field of the radiated THz pulse (ETHz) with the electric field of the incident
optical pulse (Eopt). The latter is based on a sinusoidal waveform with a period of a
few fs, or hundreds of THz in frequency, originated from a femtosecond laser. An
optical pulse would comprise several to a few tens of oscillation cycles of the optical
field as mentioned earlier (see Fig. 5.1), while the transient field of the generated
THz pulse covers a time span in the same order of (but longer than) the optical pulse
duration (see Fig. 5.3). The efficiency of THz pulse generation with the photocon-
ductive antenna is known to be higher than with the electro-optic method to be
described below, which can be further improved by employing an optimized antenna
structure [12].
is tripled, so does the electron displacement. However, when the applied field has a
very strong intensity, the linear relation does not hold any more and the response of
the electrons becomes nonlinear. Consequently, various nonlinear optical phenom-
ena begin to take place, including optical rectification.
Optical rectification is typically observed with electro-optic (EO) crystals, in
which the optical properties are affected by the presence of electric field. The most
popular electro-optic crystal is ZnTe (zinc telluride) with a zincblende crystal
structure [13, 14], which is taken as an example for the description below [1]. In
this EO crystal, the electrons that form bonds between Zn and Te atoms experience
an asymmetric potential profile as the electronegativity of Te is stronger than that of
Zn, as is depicted in Fig. 5.5a. The potential profile dictates the pattern of the electron
oscillation in response to the illumination with light. If the illumination is weak and
the magnitude of the oscillating electric field is moderate, the electrons will show a
nearly symmetric harmonic oscillator-like behavior despite the asymmetric potential
profile. However, when the illumination is strong, the displacement of the electrons
will exhibit an obvious skew toward the negative direction, as is depicted in
246 5 THz Optical Methods
Fig. 5.5b. A closer look reveals that the asymmetric displacement of the electrons
can be decomposed into two parts: a linear fundamental-mode oscillation (at the
frequency identical to that of the applied electric field) and a nonlinear second
harmonic oscillation that also includes a DC component. While the availability of
the second harmonic oscillation is one thing useful for some applications by itself,
the major consequence that interests us here is the existence of the DC component in
it. The DC component in the electron displacement gives rise to a DC or rectified
polarization. This process is called the optical rectification. Here, the polarization is a
quantity defined as the volume density of the electric dipole moments, which is
different from the polarization for electromagnetic waves that was discussed in Sect.
4.2.
Let us now consider a special case that is more relevant to our current interest, in
which the optical illumination on the EO crystal is based on an ultrashort optical
pulse from a femtosecond laser instead of continuous wave (CW). In a single optical
pulse, the electric field will oscillate with a magnitude confined by a Gaussian
envelop (see Fig. 5.6). If the intensity is strong enough, a shift in polarization will
occur due to optical rectification as expected with the electron behaviors described
above. However, differently from the CW case we assumed in the previous discus-
sion, the rectified polarization induced by the pulse is not a constant but has a
temporal variation. Its profile will also be Gaussian as illustrated in Fig. 5.6, since
the induced rectified polarization (second-order nonlinear by its nature) is known to
be proportional to the averaged incident field intensity [15]. So, what will be the
major consequence of this very fast time-domain variation in the rectified polariza-
tion? The variation in the rectified polarization means the variation in the average
electron displacement. Therefore, due to the electron movement, a pulse radiation
will occur, and it is short enough to be a THz pulse. The radiated field will show a
following relation with the time-dependent rectified polarization [2]:
2
∂ Pðt Þ
E THz ðt Þ / : ð5:2Þ
∂t 2
Pðt Þ / e2at
2
With a rectified polarization with an ideal Gaussian profile , the
generated THz pulse will take on a similar pulse shape but flanked with a shallow
5.1 THz Pulse Generation 247
negative swing on both sides as depicted in Fig. 5.6. Because the electro-optical
properties of EO materials including ZnTe depend on the orientation of the fields, an
optimal incident angle against the crystal axis needs to be maintained for the
incoming optical field to achieve the best efficiency. While the description above
is based on a time-domain picture, the THz pulse generation with optical rectification
can be equivalently explained in the frequency domain as a difference-frequency
mixing between two frequency components within the broad frequency spectrum of
the incident optical pulse.
At the surface of semiconductors, where the lattice periodicity stops and lots of
dangling bonds are formed, energy levels are not prohibited anymore in the middle
of the energy bandgap, resulting in almost continuous surface states distributed
across the bandgap. This situation gives rise to the Fermi energy pinning at the
surface, and this in turn leads to a band bending near the surface, resulting in
the surface depletion region with a built-in field normal to the surface. When the
semiconductor surface is illuminated with an ultrashort optical pulse from a femto-
second laser, lots of electrons and holes will be generated near the surface, and they
will be accelerated by the built-in field inside the surface depletion region, building
up a surge current normal to the surface. Note that the surge current is the drift
current by nature. This process is described in Fig. 5.7 for both n-type and p-type
semiconductors. As the current will last for a very short time with the ultrashort
optical pulse, in part helped by the rapid recombination via the abundant surface
states, the radiation will be a pulse in the THz regime, again dictated by the relation
given in Eq. (5.1). In a sense, this situation is similar to the case for THz pulse
generation with photoconductive antennas. The difference is that, for this surge
current case, the generated carriers are accelerated by the intrinsic surface field
instead of the external DC bias, and the rapid recombination is due to the surface
states rather than the bulk states due to defects. This technique can be applied to both
n-type and p-type semiconductors, although the time-domain profile of the generated
248 5 THz Optical Methods
Optical pulse
EV
Air holes
n-type
Surface depletion semiconductor
region
(a)
Drift Current
electrons
EC
EF
EV
Optical pulse
holes p-type
Air
semiconductor
Surface depletion
region
(b)
THz pulses will be slightly different (vertically flipped when shown along the time
axis), which is due to the opposite direction of the surge current for the two types of
semiconductor as shown in Fig. 5.7.
Let us maintain our focus on semiconductor surface, but with a selected group of
semiconductors, which have a narrow bandgap and a large mobility difference
between electrons and holes. InAs (indium arsenide) and InSb (indium antimonide)
belong to this category. For InAs, which has a bandgap around 0.35 eV, electron
mobility is ~40,000 cm2/Vs, but hole mobility is only ~500 cm2/Vs at room
temperature. For InSb with a bandgap of ~0.17 eV, electron and hole mobilities
are ~77,000 and ~850 cm2/Vs, respectively. When these semiconductors are illumi-
nated, the drift current due to the built-in field as discussed above, for which electron
and hole current components have the same direction, will exist but only with a small
5.2 THz CW Generation 249
Optical pulse
Air Semiconductor
amount. This is because the built-in potential in the depletion region is weak for the
narrow-bandgap materials as a result of the small band bending. Now recall another
current component in semiconductors, the diffusion current. It is driven by the
gradient of carrier concentration, either of electrons or holes. When electrons and
holes are generated near the surface by photo excitation caused by pulse illumina-
tion, both electrons and holes will diffuse toward the bulk due to the established
concentration gradient, building up diffusion currents for both carrier polarities.
Although the direction of the electron and hole fluxes should be the same due to
the same gradient vector, the direction of the two diffusion currents will be opposite
because of the different charge polarities, as shown in Fig. 5.8. In such a situation, if
there is no difference between the mobilities (thus the diffusion constants) of
electrons and holes, the two diffusion currents would cancel each other, assuming
an identical concentration gradient for the two carriers. However, when they are
different, there will exist a finite net diffusion current, which will be greater with the
larger mobility difference. Hence, for those semiconductors with highly imbalanced
mobilities for electrons and holes, a strong net diffusion current will develop upon
the illumination, which will dominate over the surface field-driven drift current if
band bending is small. A dipole will also be established normal to the surface due to
the separation of the distributions of electrons and holes. This is called the photo-
Dember effect. The fast surge and relaxation of the developed diffusion current will
end up generating a THz pulse when illuminated by an ultrashort optical pulse from a
femtosecond laser.
Optical methods can also be applied to generate continuous wave (CW) signal that
falls within the THz spectrum. In Chap. 2, we have overviewed various methods to
generate CW signals based on electronic approaches. In a sense, we can say that the
two different approaches, electronic and optical, are competing for the same techni-
cal solution. After all, however, we benefit from having multiple options for THz
CW generation that have their own pros and cons. Most optics-based THz CW
250 5 THz Optical Methods
When there exist two periodic optical fields with slightly different frequencies, an
optical beat is created with a frequency that corresponds to the difference of the two
original frequencies. This is very similar to the acoustic beat generated from two
acoustic waves with slightly different pitches, as is often utilized for tuning of
musical instruments. For practical purposes, the optical beats are usually obtained
with two aligned beams from a pair of CW lasers with slightly different wavelengths.
When such optical beats are incident on a photoconductive device, carriers will be
generated to form a photocurrent, in a similar way as described in Sect. 5.1.1. With
the optical beat, which can be tuned for THz frequency range by adjusting the
wavelengths of the pair of lasers, the photocurrent will also be modulated at the
same frequency. This will result in a CW THz signal radiated through a properly
designed radiating structure, typically a wideband antenna. This process is known as
photomixing or optical heterodyne down-conversion [16]. The photoconductive
device that generates THz CW signals in response to the optical beat is called the
photomixer. The most common type of photomixers is based on the photoconductive
material deposited on top of a semiconductor substrate, as was employed for THz
pulse generation discussed before, which will be the main subject of this section.
Additionally, photomixers based on p-i-n diode will also be described with focus on
UTC-PD as an example.
The process for CW THz wave generation with photoconductive material can be
described as follows [1]. For two interfering optical fields with frequencies of ω1 and
ω2, the total field applied to the photomixer can be expressed as:
5.2 THz CW Generation 251
where E1 and E2 are the amplitudes of the fields corresponding to optical frequencies
of ω1 and ω2, respectively. The associated intensity of the total field is given as:
1 pffiffiffiffiffiffiffiffi
I i ðt Þ ¼ cε0 jE i ðt Þj2 ¼ I 0 þ 2 I 1 I 2 ð cos ðω1 þ ω2 Þt þ cos ðω1 ω2 Þt Þ ð5:4Þ
2
I photo,AC
I photo ðωÞ ¼ I photo,DC þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi cos ðωt ϕÞ, ð5:5Þ
1 þ ω2 τ 2
2
1 I photo,AC RA
PTHz ðωÞ ¼ , ð5:6Þ
2 1 þ ðωτÞ2 1 þ ðωRA C A Þ2
where RA is the antenna radiation resistance and CA is the device capacitance. The
power roll-off with increasing frequency, as is expected from this expression,
includes the effect of the finite carrier lifetime as well as the RC time constant.
The former reflects the material property that is also considered in Eq. (5.5), while
the latter is rather an extrinsic property that stems from the device structure [16, 17].
The actual structure of the photomixers based on photoconductive material very
much resembles the photoconductive antenna described earlier for THz pulse gen-
eration, which is not surprising as the radiation mechanisms are virtually identical.
These photomixers also employ an LT GaAs layer grown on a semi-insulating GaAs
substrate, which is mounted on a hyper-hemispherical Si lens. When it comes to the
antenna structure, it is typical for the photomixers to employ broadband antennas
252 5 THz Optical Methods
such as log-spiral or bowtie antennas, rather than the simpler dipole-based structures
common for THz pulse generation. The gap at the center between the two conductor
patterns in the broadband antennas, where the incoming beams are focused, usually
takes on an interdigitated shape to enhance the efficiency of the conversion. This can
be seen in an example case shown in Fig. 5.9. In some cases, photomixers without an
antenna structure are employed. For them, the operation relies upon the direct THz
radiation from the generated carriers, which are accelerated by the incoming beams
with modulated intensity. As the dimension of the illuminated area is large for these
cases, comparable to or even larger than the wavelength of the radiated THz wave,
they are called large area emitters (LAEs) [17].
As for the lasers that generate the beams to pump the photomixers, there is an
obvious difference from the case for THz pulse generation. While femtosecond
lasers are used for pulse generation, it should be CW lasers that are employed for
CW THz signal generation. As a fractional change in the wavelength in the optical
band will result in a significant variation in the generated THz wave frequency,
tunable lasers are preferred to allow broadband THz wave generation. In most
practical cases, one of the two lasers is tuned, while the other is fixed at a reference
wavelength. The coupling of a pair of beams from the two lasers can be made
through a proper combination of a mirror and a beam splitter. As the requirements
for CW lasers are much relaxed compared to the case of femtosecond lasers,
commercially available diode lasers can be readily employed, whose wavelength
typically ranges over 800–850 nm [2]. Note that 1% change in the wavelength in this
region (~8 nm), which is readily available in typical cases, results in nearly 4 THz
variation in the down-converted frequency. Of course, it is not this entire frequency
range that is converted to the tuning bandwidth of the generated THz radiation, due
to the internal limiting factors of the photomixers such as the carrier lifetime and RC
time constant as discussed above with Eq. (5.6). Still, the tuning bandwidth of
photomixers is far wider than those of electrical sources, a major advantage of
photomixers.
The radiation power of the THz waves from the photomixers based on photo-
conductive material currently reaches up to ~10 μW for emission frequencies larger
than 100 GHz [18, 19]. This can be compared with the output power levels available
5.2 THz CW Generation 253
with electrical signal sources, which exceed ~10 mW and ~1 mW at 100 GHz and
300 GHz, respectively, with IMPATT or Gunn diodes. RTDs also exhibit output
power larger than ~0.1 mW at these frequencies (for the output power of the diode-
based sources, see Fig. 2.9). The output power from semiconductor integrated
circuits based on oscillators and power amplifiers now reaches beyond 10 mW
around 300 GHz (for the output power of the transistor-based sources, see
Fig. 2.33). Hence, it is fair to say that the photomixers fall behind the electrical
sources in terms of the generated output power at the moment, while they provide
much wider tunable bandwidths. The output power of the photomixers can be
improved with increased beam intensity from the lasers, but an excessive optical
power may end up damaging the LT GaAs structure with increased device temper-
ature. Substrates with a higher thermal conductivity may allow higher incident beam
power, for which Si substrate may be used in place of GaAs as the thermal
conductivity of Si is higher than that of GaAs (1.41 W/cmK for Si vs 0.46 W/cmK
for GaAs at T ¼ 300 K) [20]. LAE mentioned above can partially address this issue
by spreading out the beam, and thus the beam power, over a wide area. The
optimization of the antenna structure can enhance the output power as well. For
example, it was shown that a dual dipole structure slows down the roll-off by
resonating out the device capacitance with inductive tuning, as was achieved by
controlled dipole length [21]. A structure employing a Fabry–Pérot cavity for
photomixers was also reported, which resulted in an output power of 0.35 mW
around 300 GHz [22].
(a)
Diffusion block layer
electrons
p Optical signal
p-contact EC
holes
n
Absorption layer
n-contact layer
(b)
In typical p-i-n diodes, the velocity of the holes is much smaller than that of the
electrons. Hence, the slow hole transport will impose a negative impact on the photo-
response speed and thus the bandwidth of the p-i-n diode photomixer. Additionally,
the slow drift of the holes will effectively promote the accumulation of holes in the
drift region, resulting in substantial space charge density formed in the intrinsic
region. The space charge will cause a field modulation and, consequently, a band
bending, which will induce a local reduction of the field in the intrinsic region near
the n-type region boundary. The reduced electric field will further slow down the
carrier motion, eventually limiting the maximum current allowed in the diode. This
will serve as a main limiting factor of the output power available with conventional
p-i-n diode photomixers.
In view of this situation, if only the electrons are selected to contribute to the
operation as the active carrier, while the effect of the hole transport is suppressed,
one would expect improved photomixer bandwidth as well as higher saturation
current. The UTC-PD was invented to serve this purpose [23]. In an UTC-PD, the
absorption layer is p-type doped and separated from the intrinsic layer. Besides, it
has a narrower bandgap than the intrinsic region as shown in Fig. 5.10b [24]. There-
fore, when the absorption layer is illuminated with incoming optical beam, the
photo-generated holes are rather confined in the p-type region and quickly relaxed
5.2 THz CW Generation 255
as the majority carrier. On the other hand, the generated electrons enter the intrinsic
region and are collected by the n-type region after drifting through the depleted
intrinsic layer. Hence, the photocurrent is mostly composed of electrons, and its
speed is governed by the electron velocity, leading to an improved operation speed
and bandwidth. Also, a higher current level can be achieved compared to the
conventional p-i-n diode case, since it does not suffer from the adverse band bending
effect that limits the current saturation of the photomixer. The separation of the
absorption and drift regions also helps to optimize the structure to reduce the device
RC time constant that affects the bandwidth. It is noted that the band bending does
occur in UTC-PD as well (in the opposite direction), but its effect is not as
outstanding as the p-i-n diode case owing to the higher velocity of electrons,
which is helped by the electron velocity overshoot. In fact, the average traveling
distance for generated electrons to reach the n-type region is longer for UTC-PD due
to the separate absorption layer, but it is compensated for by other merits including
the higher electron velocity. The overall advantage of UTC-PDs leads to a wider
bandwidth and, probably more importantly, higher output power compared to the
conventional p-i-n diode photomixers.
A more detailed look into UTC-PD structure would help understand the opera-
tion. The absorption and intrinsic regions are usually composed of InGaAs and InP
layers, respectively. The p-type absorption layer can be modified to create a quasi-
electric field to help the electron acceleration, which can be achieved by bandgap
grading or doping modulation. On top of the p-type absorption layer, another layer
with a wider bandgap is routinely placed, typically an InGaAs layer with a higher Ga
composition, which would block the diffusion of electrons toward the anode (top
contact). Between the absorption and the intrinsic regions, an n+ InP layer can be
inserted to suppress the bandgap spike, facilitating a smooth launching of electrons
into the drift region. This feature is similar to the layer structure in typical III-V
HBTs, which are designed to help electron acceleration at the base-collector bound-
ary. The ballistic transport of electrons with velocity overshoot in the intrinsic region
in UTC-PDs as mentioned above also resembles the situation at the base-collector
depletion region in HBTs. The difference is that the electrons are photo-generated in
UTC-PDs, while they are injected from the n-type emitter in HBTs.
The output power of UTC-PD is higher than that of the conventional p-i-n diodes
by a factor of an order or two, as indicated by the results reporting ~20 mW near
100 GHz [25] and ~100 μW close to 300 GHz [26]. It is notable that the large output
power can be achieved simultaneously with a large bandwidth, decoupling the trade-
off between output power and bandwidth in the conventional p-i-n diodes. There are
some variants of UTC-PDs, including those integrated with a broadband or narrow-
band antenna for a more efficient THz wave radiation [27].
256 5 THz Optical Methods
The mixing between two beams with slightly different frequencies and the conse-
quent THz wave generation can be obtained with optical rectification based on
electro-optic materials as well. In Chap. 2, we observed that a mixing between two
signals can be obtained as they go through a nonlinear (electrical) device, and, as a
result, an output signal with a frequency that corresponds to the difference between
the frequencies of the two incoming signals can be generated. The principles of the
generation of THz CW signal by frequency difference mixing is conceptually
identical to such an electrical mixing with a nonlinear device. The only difference
for the frequency difference mixing is the fact that the operation frequency is much
higher, around the optical band, and the nonlinear “device” is in fact a plain optical
medium, rather than established structure such as diode or transistor, with a preferred
electro-optical property.
Consider two beams propagating together that are assumed to be linearly polar-
ized. For the two beams, the total optical field will be expressed by Eq. (5.3), as was
the case for photomixing case. When an electro-optic material with a second-order
nonlinearity is illuminated with the combined beam, a polarization will be
established due to optical rectification. The second-order nonlinear polarization
will be proportional to the square of the incident total optical field, as is given below:
Note that the expression is basically identical to the intensity of the incoming coupled
beams as was shown in Eq. (5.4). Hence, it can be equally stated that the achieved
polarization is proportional to the intensity of the incident optical field. Assuming the
sum frequency component is suppressed due to the low-pass filtering in the process of
optical rectification, as was the case for the photomixing, the time-dependence of the
polarization will be simply given by cos(ω1t ω2t), or cos(ωt) with ω ¼ ω1 ω2. If
we recall Eq. (5.2), which dictates that the generated THz field is proportional to the
second-order time-derivative of the polarization, we have ETHz(t) again proportional
to cos(ωt). Hence, the generation of a CW THz signal with a frequency equal to the
difference of the two input optical frequencies is achieved.
For difference frequency generation, various nonlinear materials have been
employed, such as GaSe, GaP, quartz, and LiNbO3. Among these, GaSe has been
found most effective owing to its strong second-order nonlinearity as well as the low
absorption rate at the THz frequency range [28]. As for the beam sources, conven-
tional CW lasers can serve the purpose, similar to the photomixing case, which
enables a compact setup for THz generation. The generation of CW THz signal from
difference frequency method also provides a very wide tuning range as is the case for
photomixing. However, the output power level is again a major issue for this
technique, smaller than those from photomixing.
5.2 THz CW Generation 257
Fig. 5.11 Three scenarios of the interaction between electrons and photons: (a) Absorption (b)
Spontaneous emission (c) Stimulated emission
258 5 THz Optical Methods
boundaries. This will further enhance the stimulated emission, leading to a rapid
growth in the photon density before reaching a saturation, which is responsible for
the high radiation power available with a laser. A key point here in the context of
terahertz lasers is the frequency of the radiation from the lasers. It is dictated by
ΔE ¼ E2 E1, with a relation f ¼ ΔE/h, where h is Planck constant. A major
challenge for THz lasers is to obtain ΔE that corresponds to the THz spectrum,
which is only a few meV, far smaller than what is usually available with typical laser
systems. However, it is still achievable with some selected types of lasers, which are
described below in this subsection.
It is possible to obtain CW THz radiation with a solid-state laser. For most solid-state
crystals, the energy bandgap between the conduction and valence band typically
ranges in the order of ~eV, which is far larger than the energy levels corresponding
to the THz spectrum. Therefore, the inter-band transition in solid-state materials
cannot be used for THz radiation. On the other hand, there are chances for the intra-
band transition between subbands internal to the valence or conduction bands to be
exploited for lasing, if the transition energy falls within a range of a few meV. Such a
transition can be achieved with p-type doped germanium (Ge) lasers [29, 30].
For Ge, the valence band is composed of the heavy hole (with a large effective
mass) band and the light hole (with a small effective mass) band, the former showing
a larger radius of curvature than the latter in the E k relation. There exists yet
another band, namely the spin-orbit split-off band, in the valence band, but it hardly
contributes to the conduction due to the negligibly small hole population, as it is
located significantly away (~290 meV at T ¼ 300 K) from the valence band edge.
Figure 5.12a shows the hole bands of Ge in the E k space on the bottom half. It
would be noteworthy that the maxima of these hole bands are all located at Γ point,
whereas the conduction band for electrons (shown on the upper half of the figure) has
its minimum at L point in the crystal Ge. Such an indirect bandgap structure,
however, does not preclude Ge for laser operation, since it is not the inter-band
transition that leads to its laser operation. For p-type Ge lasers, the intra-band
(or inter-subband) transition between Landau levels at cryogenic temperatures is
exploited for THz lasing, as described below.
We all know that a charged particle will be linearly accelerated in the presence of
an external electric field. In a slightly more involved situation where a magnetic field
is additionally applied in a direction orthogonal to the existing electric field (called
the cross field), the charged particle will become under the influence of Lorentz force
and experience an acceleration in the transverse direction, ending up with a circular
motion with a cyclotron orbit. Interestingly, the energy level associated with the
cyclotron orbit will be quantized to meet the quantum condition, and the generated
discrete energy levels are called Landau levels. This process will happen to charged
particles inside solid-state crystals as well. Therefore, Landau levels will be gener-
ated for hole bands in p-type Ge crystal when mutually orthogonal external electric
5.2 THz CW Generation 259
and magnetic fields are applied. Hence, for lasing, you will first need a biased p-type
Ge with an external magnet in a proper field orientation.
Next, you will need a cryogenic temperature. The readers should be aware that
there are two major scattering mechanisms inside semiconductors: phonon scattering
and ionized impurity scattering. As for the phonon scattering, there are two modes of
phonons, optical phonon and acoustic phonon, the former involving a larger energy
than the latter in scattering events. All added up, therefore, there are basically three
scattering mechanisms: optical phonon scattering, acoustic phonon scattering, and
ionized impurity scattering. For lowly doped p-type Ge placed in a cryogenic
environment, optical phonon scattering will show a much higher probability than
the acoustic phonon scattering and ionized impurity scattering. Interestingly, when a
certain combination of the magnitudes for the electric and magnetic fields is attained,
the heavy holes will enter a condition where optical phonon scattering will prevail
while the light holes will be dominated by acoustic phonon and ionized impurity
scatterings. As a result, for such an optimal range of electrical and magnetic fields,
the heavy hole states will show a short carrier lifetime due to the high scattering rate
of optical phonons, while the lifetime of the light hole states will be relatively long
with the low associated scattering rates with the given condition.
With these properties involving the holes in p-type Ge in mind, consider a hole
accelerated by electric field [31]. It will gradually climb the energy curve of the
heavy hole band as described in Fig. 5.12b, which is basically an electric pumping
260 5 THz Optical Methods
Fig. 5.13 (a) Multiple quantum wells in a superlattice, shown with the quantized energy states. (b)
Multiple quantum wells under bias, shown with electron transitions between energy states and
tunneling across energy barriers. (c) Actual structure and electron transitions inside a typical QCL
(quantum cascade laser). (From [33]. Reprinted with permission from AAAS)
pump level 3 to maintain a high population. The generated photons travel along the
lateral in-plane path (perpendicular to the layer stack direction), which is confined by
mirrors on both sides, typically realized with a cleaving of the semiconductor piece
containing the QCL structure.
For the operation of THz QCLs, cryogenic temperature is again needed. For THz
generation, the involved energy level ranges in the order of a few meV, which can be
easily exceeded by the thermal energy of electrons, roughly around kT. For example,
the transition energy corresponding to 1 THz is 4.1 meV, which equals to the thermal
energy of an electron at around 50 K. A large thermal energy would activate optical
phonon scattering, leading to a high probability for electrons to be relaxed by losing
energy to optical phonons instead of photons. This issue is not relevant only to THz
QCL but to THz lasers in general, imposing a great challenge to the realization of the
room-temperature THz lasers. This explains the absence of room-temperature oper-
ation of THz QCLs reported so far. At cryogenic temperatures, however, advances in
THz QCLs have continued [34]. A QCL with an output power beyond 1 W at 10 K
has been reported with a pulsed mode operation around 3.4 THz, which also
exhibited 420 mW at 77 K [35]. As for the efforts to increase the operation
temperature, a QCL operating near 200 K has been reported at 3.2 THz
[36]. There is a large number of other reports on THz QCLs, whose radiation
frequency ranges roughly from 1.5 to 4.5 THz.
While both of the THz lasers described above are solid-state lasers, THz lasing can
be achieved with gas lasers as well. Gas lasers employ gas-phase materials as the
gain media. While some gas lasers are based on monatomic gases, as in the case of
the popular HeNe (helium-neon) lasers, many of them rely on gas molecules to serve
as the gain media. For laser operation, we need quantized energy levels. For lasers
with gas molecules, it is the vibrational and rotational modes (or coupling of the two)
of molecules that are exploited for electron transitions required for stimulated photon
emission. While both modes typically involve a much smaller excitation energy
compared to the atomic energy level transition, it is the rotational modes that provide
smaller energy exchange for inter-mode transition, which is needed for long-
wavelength radiation. If the transition energy between the rotational modes is
small enough, THz radiation can be expected with gas lasers.
There indeed have been such gas lasers reported for radiation in the THz spectrum
[37, 38]. The major efforts have been made on CH3OH (methanol) lasers, the
118.9 μm (2.52 THz) line being the main target among other spectral lines available
with CH3OH [39]. CH3OH lasers are optically pumped with another laser, typically
with a high-power CO2 laser radiating at around 9.7 μm (31 THz). As illustrated in
Fig. 5.14, the pumping with a CO2 laser excites electrons to an upper vibrational
mode of the CH3OH gas, which is followed by transitions between rotational modes
bunched within the vibrational mode, leading to THz radiation. A CH3OH laser with
a CW output power exceeding 1 W at 118.9 μm has been reported with CO2 laser
5.3 THz Detection 263
CO2
(31 THz)
Upper vibrational mode
Transition between
Pumping
rotational modes
pumping [40]. Other gases known to be capable of THz lasing with a substantial
output power include NH3 (ammonia), CH2F2 (difluoromethane), CH3Cl
(chloromethane), and CH3I (methyl iodide) [1].
In Chap. 3, various techniques for THz radiation detection such as thermal detection
as well as detection with diodes or transistors were discussed. However, what they
detect is the time-averaged power of the THz radiation, mostly applicable to CW
radiation. In some cases, however, phase-sensitive detection is desired, especially
when the THz radiation is in the form of pulses. For example, the recovery of the
THz pulse shapes, typically after the transmission through or reflection from a target
object, is the core part of THz spectroscopy. Although it is extremely challenging to
precisely reconstruct the pulse profiles as short as those of THz pulses, it can be
achieved based on principles that are similar to those behind the THz pulse gener-
ation, with either photoconductive antenna or electro-optic crystals. The key element
employed for the phase-sensitive detection of ultra-short pulses is the sampling
(or gating) technique, which is detailed below.
THz pulse
electrical path
Emitter Detector
PC PC
Output
Antenna Antenna
THz pulse
recovered
Delay
optical path
Optical
pulse
adjustable
fs laser
Fig. 5.15 Typical setup for THz waveform detection with photoconductive antennas
variation of which results in THz pulse generation. The detection can be made with
the same antenna structure. For the detection, however, the antenna is not biased.
Instead, the electric field needed to generate the current is provided by the incoming
THz pulse incident on the antenna that is to be detected. With the incident THz pulse
only, there will be no current generation despite the field established across the
antenna gap due to the lack of carriers. However, when it is combined with an optical
pulse, current will be generated owing to the carrier generation by the photo
excitation. Since the current level will be higher with a stronger THz pulse intensity,
the measured current level will be a good indicator of the THz pulse strength at the
moment of the optical pulse arrival. Now, if such a current generation process is
repeated multiple times with progressively shifting arrival time of the optical pulse
with reference to the THz pulse, one can reconstruct the THz pulse shape by
connecting the current level points over arrival time. This is the basic principle of
sampling and the applied optical pulse for sampling is called the sampling or gating
pulse.
A typical setup to realize the sampling-based THz detection with PC antennas is
depicted in Fig. 5.15. It is basically composed of a pair of PC antennas, one for the
generation and the other for the detection of THz pulses, which are called the emitter
and detector, respectively. The two antennas are linked by an electrical path on one
side and an optical path on the other side. The generated THz pulse at the emitter
travels along the electrical path to reach and illuminate the detector, optionally
through a target sample. On the other hand, the optical pulse from the femtosecond
laser is split into two pulses: one directly arrives at the emitter for THz pulse
generation mentioned above, and the other reaches the detector for THz pulse
detection after traveling along the optical path. Placed in the middle of the optical
5.3 THz Detection 265
path is a time delay component, which is a critical element to precisely control the
timing of the optical pulse arrival at the detector. It is typically composed of a set of
mirrors, a part of which can be moved with high precision to accurately adjust the
total traveling path of the optical pulse (see the sketch in the bottom of Fig. 5.15).
The output of the detector antenna is terminated with an ammeter for current
detection, which is in contrast with the emitter antenna that is terminated with a
voltage source for bias supply. The detected current level is typically amplified to
boost the signal strength afterwards.
With the system described above, the generated THz pulse at the emitter can be
detected at the detector with the phase information retained as described below. The
sampling technique to recover the THz pulse profile, as mentioned earlier, requires a
series of identical measurements to trace the shape. In each round of the measure-
ment, a THz pulse is generated and arrives at the detector in synchronization with the
arrival of an optical pulse with a controlled delay. The delay is slightly different for
each round and sequentially increases in the course of the entire measurement cycle.
Hence, if the current level is recorded for each measurement round and the collected
current level points are plotted as a function of the delay time for the complete
process, it will show the profile of the original THz pulse. This sampling process is
illustrated in Fig. 5.16, which shows the sampling of the THz pulse shape with a
series of measurements assuming the time delay increases by a step of Δτ each round
(in the figure, only even-numbered sampling points are shown due to space limit,
although the sampling occurs for the every sampling point). If needed, the obtained
time-domain profile can be converted into the frequency-domain profile by Fourier
transformation. In practical cases, a target sample is often placed in the middle of the
electrical path, in which case the THz pulse is detected after the transmission through
the sample. By comparing the profiles of the THz pulses with and without the
sample, rich spectral information of the sample can be retrieved. This is in fact the
principle of the time-domain spectroscopy (TDS), which will be discussed in Sect.
6.1 in more detail.
While the sampling technique so far discussed has been successfully applied to
THz pulse measurements, there is a limitation as well. The extracted temporal
variation of the detected photocurrent well represents the THz pulse profile, but it
falls short of being truly identical to the original. There exists a finite difference
between the original and the recovered profiles, which arises from the fact that the
photocurrent is induced by the combined effect of the THz pulse (electric field) and
the optical pulse (carrier generation). This is better understood by observing the
expression of the photocurrent in the frequency domain as shown below [1]:
where σ is the conductivity, which is governed by the carrier concentration and the
mobility. If the conductivity is constant over the frequency, the photocurrent profile
would exactly match the field profile in the frequency domain, and consequently in
the time domain as well. If not, which is the case in reality, the true profile will be
filtered by the bandwidth of the conductivity, leading to a finite distortion. One major
266 5 THz Optical Methods
Fig. 5.16 Conceptual diagram for the recovery of a THz waveform with the sampling process. In
the figure, only even-numbered sampling points are shown due to the space limit, although the
sampling occurs for the entire sampling points in reality
factor that affects the frequency dependence of the conductivity is the bandwidth of
the carrier concentration, which is in turn influenced by the pulse width of the
incident optical beam. It has been shown that a short optical pulse significantly
improves the bandwidth of the carrier concentration [41]. This indicates that the
ultrashort pulse generation from the femtosecond laser is critical for both generation
and detection of THz pulses.
5.3 THz Detection 267
Fig. 5.17 Typical setup for THz waveform detection with EO sampling. Also shown on the bottom
are the polarization states at selected locations along the path (the locations are indicated by dotted
arrows). (Adapted by permission from Springer Nature: [1] © 2009)
to the pulse will trigger birefringence inside ZnTe, rendering the optical beam under
the influence of birefringence. Consequently, one polarization component of the
optical beam will experience a phase retardation against the other component,
resulting in elliptical polarization for the beam exiting the crystal. The larger the
field strength, the larger will be the phase retardation. Hence, if the extent of the
phase difference can be converted into measurable quantity, such as current, the field
strength of the THz pulse can be estimated, and eventually the pulse shape can be
restored with a sequential acquisition of the quantity [13].
A setup for the THz pulse detection with EO-sampling is illustrated in Fig. 5.17
[1]. A linearly polarized probe optical beam is incident on an EO crystal together
with a THz pulse with a 45 angle against the optic axis of the crystal. The crystal
will be momentarily birefringent due to the field induced by the THz pulse that
passes by, and thus the optical beam will emerge from the crystal elliptically
polarized. The beam will subsequently travel through a λ/4 plate and a Wollaston
prism. The λ/4 plate, which would convert linear polarization into circular polariza-
tion, will convert the elliptical polarization of the optical beam into nearly circular
polarization, with a slight deviation from the perfect. The extent of the deviation, or
References 269
distortion, from the perfect circular polarization will depend on the strength of the
field induced by the THz pulse. With the Wollaston prism, a beam of perfect circular
polarization (zero field inside ZnTe or the absence of THz pulse) will be split into
two beams with equal intensity. The two beams will have a phase difference of 90 ,
and the phase information will be lost in the end at the detection with a pair of
photodiodes, since the currents induced by the two balanced diodes will be canceled
out. For a beam of elliptical polarization (non-zero field inside ZnTe or the presence
of THz pulse), whose axis is aligned to the optic axis of the Wollaston prism, the two
split beams will have a different intensity (see Fig. 4.17e for the relative intensities of
the field in elliptical polarization). The difference increases with increasing deviation
from the perfect circular polarization and thus with increasing electric field inside
ZnTe. The two beams coming out of the Wollaston prism will be separately incident
on a pair of balanced photodiodes and the current difference between the two
photodiodes (i0 ¼ i1 i2, where i1 and i2 are the currents generated by the two
photodiodes) will be extracted at the output. Hence, a larger deviation from the
circular polarization, which corresponds to a larger THs pulse intensity, will result in
a large out current level. In this way, the output current will serve as the indicator of
the THz field at the moment of the optical beam arrival at the EO crystal. Repeated
measurements of the output current with progressively increasing time delay
imposed on the optical beam will lead to the recovery of the profile of the THz
pulse being detected.
References
1. Y.-S. Lee, Principles of Terahertz Science and Technology (Springer, Berlin, 2009)
2. K. Sakai (ed.), Terahertz Optoelectronics (Springer, Berlin, 2004)
3. E. Bründermann, H.-W. Hübers, M.F. Kimmitt, Terahertz Techniques (Springer, Berlin, 2012)
4. D.E. Spence, P.N. Kean, W. Sibbett, 60-fsec pulse generation from a self-mode-locked Ti:
sapphire laser. Opt. Lett. 16(1), 42–44 (1991). https://doi.org/10.1364/OL.16.000042
5. S. Gupta et al., Subpicosecond carrier lifetime in GaAs grown by molecular beam epitaxy at low
temperatures. Appl. Phys. Lett. 59(25), 3276–3278 (1991)
6. D.C. Look, D.C. Walters, G.D. Robinson, J.R. Sizelove, M.G. Mier, C.E. Stutz, Annealing
dynamics of molecular-beam epitaxial GaAs grown at 200 C. J. Appl. Phys. 74(1), 306–310
(1993). https://doi.org/10.1063/1.354108
7. K.A. McIntosh, K.B. Nichols, S. Verghese, E.R. Brown, Investigation of ultrashort photocarrier
relaxation times in low-temperature-grown GaAs. Appl. Phys. Lett. 70(3), 354–356 (1997).
https://doi.org/10.1063/1.118412
8. F.E. Doany, D. Grischkowsky, C.C. Chi, Carrier lifetime versus ion-implantation dose in silicon
on sapphire. Appl. Phys. Lett. 50(8), 460–462 (1987). https://doi.org/10.1063/1.98173
9. Z.-S. Piao, M. Tani, K. Sakai, Carrier dynamics and THz radiation in biased semiconductor
structures. Terahertz Spectrosc. Appl. 3617, 49–56 (1999)
10. L. Duvillaret, F. Garet, J.-F. Roux, J.-L. Coutaz, Analytical modeling and optimization of
terahertz time-domain spectroscopy experiments, using photoswitches as antennas. IEEE J. Sel.
Top. Quantum Electron. 7(4), 615–623 (2001)
270 5 THz Optical Methods
32. I. Hosako, N. Hiromoto, Research and developments on p-type germanium lasers in the
frequency range from 0. 1 terahertz to few terahertz. J. Commun. Res. Lab. 49(1), 81–94 (2002)
33. J. Faist, F. Capasso, D.L. Sivco, C. Sirtori, A.L. Hutchinson, A.Y. Cho, Quantum cascade laser.
Science 264(5158), 553–556 (1994). https://doi.org/10.1126/science.264.5158.553
34. B.S. Williams, Terahertz quantum-cascade lasers. Nat. Photonics 1(9), 517 (2007)
35. L. Li et al., Terahertz quantum cascade lasers with >1 W output powers. Electron. Lett. 50(4),
309–311 (2014). https://doi.org/10.1049/el.2013.4035
36. S. Fathololoumi et al., Terahertz quantum cascade lasers operating up to ~200 K with optimized
oscillator strength and improved injection tunneling. Opt. Express 20(4), 3866–3876 (2012).
https://doi.org/10.1364/OE.20.003866
37. G. Dodel, On the history of far-infrared (FIR) gas lasers: thirty-five years of research and
application. Infrared Phys. Technol. 40(3), 127–139 (1999). https://doi.org/10.1016/S1350-
4495(99)00005-5
38. N.G. Douglas, Millimetre and Submillimetre Wavelength Lasers: A Handbook of Cw Measure-
ments (Springer, Berlin, 2013)
39. G. Moruzzi, J.C. Silos Moraes, F. Strumia, Far infrared laser lines and assignments of CH3OH:
a review. Int. J. Infrared Millim. Waves 13(9), 1269–1312 (1992). https://doi.org/10.1007/
bf01009988
40. J. Farhoomand, H.M. Pickett, Stable 1.25 watts CW far infrared laser radiation at the 119 μm
methanol line. Int. J. Infrared Millim. Waves 8(5), 441–447 (1987)
41. S. Kono, M. Tani, K. Sakai, Ultrabroadband photoconductive detection: comparison with free-
space electro-optic sampling. Appl. Phys. Lett. 79(7), 898–900 (2001). https://doi.org/10.1063/
1.1394719
42. A. Yariv, P. Yeh, Optical Waves in Crystals (Wiley, New York, 1984)
Chapter 6
THz Applications
In the previous chapters, basic principles of the topics related to the THz frequency
band have been overviewed with a main focus on the electronics-based approaches.
While the efforts on exploiting the THz band have been so far rather skewed toward
the academic side, there are recent indications that the theories and knowledges
accumulated by THz research are now quite matured, so that they can be applied to
the real world with more practical applications. In this chapter, four most represen-
tative THz applications will be reviewed: spectroscopy, imaging, communication,
and radars. The first two topics are rather traditional application fields of the THz
band, and already a great amount of accomplishments have been made in the labs
and are moving toward the application-oriented front. The following two topics are
relatively new and emerging fields of THz application and expected to have a more
practical impact on our daily lives. For each application topic, basic principles will
be first provided, followed by notable example cases.
6.1.1 Overview
Spectroscopy refers to the analysis of the interaction between matters and electro-
magnetic waves presented as a function of frequency (or wavelength). The emission
or the absorption of electromagnetic waves by matters are characteristic conse-
quences of such interaction. Spectroscopy has long played a great role in various
scientific areas including physics, chemistry, astronomy, as well as in biology,
owing to the rich information it provides related to the composition and physical
structure of a wide range of materials in various scales. We know that, for example,
the atomic structure became understood to us through the analysis of the electro-
magnetic wave emission lines from transitions between atomic energy states. THz
spectroscopy concerns spectroscopy in the range of the THz spectrum. It has been
Fig. 6.1 Reflection from various surfaces: (a) Flat surface, (b) Grating with specular reflection, (c)
Grating with diffuse reflection
diffraction
grating
focusing
mirror
exit slit
Mirror2 (M2)
light
source
beam
splitter
detector
Fourier-transform spectrometer for practical use was enabled only when digital
computation techniques became readily available in the mid-twentieth century.
The core structure of Fourier-transform spectrometers is the Michelson interferom-
eter as illustrated in Fig. 6.3. It is composed of a beam splitter, a light source, a
detector, and a pair of mirrors, one of which is movable (M1). A beam from the light
source is split into two orthogonal branches by the beam splitter, which are com-
bined again after reflections by the pair of mirrors. The combined beam is eventually
detected by the detector, optionally passing through a target object. As the paths
traveled by the two beams are different, there will be interference between the two
beams upon arrival at the detector. As a result of the interference, the maxima will
occur when the following relation is satisfied:
Δl ¼ l1 l2 ¼ nλ ð6:2Þ
6.1 THz Spectroscopy 277
1 3 1
Amplitude
Amplitude
Intensity
Intensity
1 2
1
0 0 0
0
0 5 10 0 5 10 0 5 10 0 5 10
Displacement Frequency Displacement Frequency
(a) (b)
2 2
Amplitude
Amplitude
Intensity
Intensity
1
1 1
0 0 0 0
0 5 10 0 5 0 5 10 15 0 5 10
Displacement Frequency Displacement Frequency
(c) (d)
Fig. 6.4 Interferograms and corresponding frequency domain profiles for various cases (not to
scale): (a) Sine wave, (b) Sinc function, (c) Pulse, (d) Multiple pulses after multiple reflections.
(Reprinted by permission from Springer Nature: [1] © 2012)
where Δl is the path difference, l1 and l2 are the total paths of the beams reflected by
M1 and M2, respectively (each designated by a solid- and dotted-lines in Fig. 6.3),
n is an integer, and λ is the wavelength.
If the light source is monochromatic, maxima and minima will be alternately
observed at the detector as one keeps moving M1 toward the detector or away from
it. If the light source is broadband, which is mostly the case, maxima and minima
will be alternately observed for every wavelength component with M1 movement,
but with a different period, a longer period for a longer wavelength. This implies
that, for a given position of M1, a set of various wavelengths will be simultaneously
detected, and, as M1 position is varied, the combination of the detected wavelengths
will change accordingly. Now, if a target object is placed in the middle of the
combined beam path, we can record the absorption level by the object as a function
of the mirror position, which is called the interferogram. However, our ultimate goal
is to obtain the absorption given as a function of the wavelength. The required
conversion from the mirror-position domain (in cm) to the wavelength domain
(in cm1) (or, equivalently, the frequency domain in Hz) turns out to be the Fourier
transform, as the name of the spectrometer indicates. Figure 6.4 shows interfero-
grams and the corresponding frequency domain profiles for various cases [1].
There are various versions of Fourier-transform spectrometers, and the one for the
infrared spectrum is called FTIR (Fourier-Transform InfraRed). Figure 6.5 shows the
optical configuration of a typical FTIR setup. For the light source, a broadband IR
source such as a Globar is typically used. For the detector, cryogenic bolometers are
a popular choice, while a room-temperature Golay cell can also be used, partially
sacrificing the sensitivity in favor of the measurement convenience. After dozens of
278 6 THz Applications
Reference Mirror
Moving Mirror
Sample
Beam
Splitter
Detector
Light Source
FFT Filter
years since its first appearance, FTIR still serves as a useful tool for infrared and THz
spectroscopy.
The landscape of THz spectroscopy went through another major transformation
when THz time-domain spectroscopy (TDS) was invented in 1980s [4, 5]. Contrary
to the previous spectroscopic methods that were based on the frequency domain,
THz TDS is a time-domain technique making use of ultrashort time pulses. Owing to
a set of advantages over the Fourier-transform spectroscopy which will be discussed
later, THz TDS rapidly arose as a highly favored choice for a wide range of THz
spectroscopic applications [6]. Being a major topic, THz TDS will be treated as a
separate subject in Sect. 6.1.2. Yet another technique is the CW-based THz spec-
troscopy that is performed in the frequency domain, called CW THz frequency-
domain spectroscopy (FDS). The CW signals required for this technique may be
generated by optical methods but also with electronic components, which provides
an opportunity for compact and affordable solutions with a low power consumption.
This will also be described separately in Sect. 6.1.3.
sample
emitter detector
output
(a)
(b)
time pulse resembles the delta function, and the shorter the pulse duration, the wider
the corresponding bandwidth in the frequency domain. Hence, if we know the
response of a material to a short time pulse, we can extract the corresponding
frequency response by the Fourier transform. The obtained frequency response
will be over a wider bandwidth if the time pulse is shorter. This is the principal
idea behind THz TDS, and the availability of sub-picosecond THz pulses with the
invention of femto-second lasers naturally led to a very useful spectroscopy tech-
nique with a wide frequency coverage.
From a technical point of view, the generation and detection of ultrashort THz
pulses are the key elements of THz TDS. The related details can be found in Chap. 5,
while the main concept is recaptured here along with the actual setup diagram for
THz TDS. There are two approaches for the generation and detection of THz pulses,
one with photoconductive (PC) antennas and the other with electro-optic
(EO) crystals, both driven by a femtosecond laser. The TDS setup for the first
approach is depicted in Fig. 6.6a. A THz pulse is radiated from the emitter when
an optical pulse from a femtosecond laser is applied on a photoconductive antenna
with the help of the very short carrier lifetime of the photoconductive material in the
280 6 THz Applications
1.0
0.6
0.5
Time domain 0.4 Time domain
E-field [arb.units]
E-field [arb.units]
0.2
0
0
-0.5
-0.2
-0.4
-1.0
0 10 20 30 70 80 90
Time Delay [psec] Time Delay [psec]
1.0
0.6
Amplitude [arb.units]
Amplitude [arb.units]
0.8
Frequency Frequency
0.6
domain 0.4
domain
0.4
0.2
0.2
0 0
0 0.5 1.0 1.5 2.0 2.5 0 0.5 1.0 1.5 2.0
Frequency [THz] Frequency [THz]
(a) (b)
Fig. 6.7 THz waveforms in time domain and frequency domain: (a) Through free space, (b)
Through sapphire sample. (Reprinted with permission from [8] © The Optical Society)
out in IBM [8], the properties of various dielectric crystals and semiconductors have
been investigated for a frequency range of 0.2–2 THz. Based on 70-fs optical pulses
with a repetition rate of 100 MHz emitted from a colliding-pulse mode-locked dye
laser, THz pulses were generated with a GaAs-based photo-conductive antenna
illuminated by the optical pulses. THz pulse detection was made with a similar
antenna structure built on an ion-implanted silicon-on-sapphire substrate. The time-
domain waveform and its frequency profile obtained from the Fourier transform are
shown in Fig. 6.7a for a THz pulse that has propagated through a free space. In
comparison, Fig. 6.7b presents the same set of data obtained for a THz pulse that has
traveled through a 9.6-mm-thick sapphire sample, one of the materials examined in
the study. By comparing the two data set obtained with and without the sample,
various properties of the sample can be obtained. For example, the refractive index
of the sapphire sample can be extracted as shown in Fig. 6.8 [8]. As indicated by the
plot, a reasonable agreement with previously reported data points is obtained by THz
TDS. Additional information such as conductivity and carrier concentration of
semiconductors can also be provided by THz TDS with a similar setup [9, 10].
In another experiment reported in [11], various explosives and related compounds
were studied with THz spectroscopy. Based on a 800-nm Ti:Sapphire femtosecond
laser with a pulse duration of 80 fs and a repetition rate of 80 MHz, THz pulses were
generated using the surface effect (see Sect. 5.1.3) by focusing the optical pulses on a
282 6 THz Applications
p-type InAs crystal. With the generated THz pulses and electro-optic sampling
process based on ZnTe crystal for pulse detection, THz TDS was performed. A
frequency range of 0.1–2.8 THz was achieved after a fast Fourier transform with a
frequency resolution of 50 GHz. Seventeen compounds have been examined for
absorption properties with the setup in the study. Three of them are introduced in
Fig. 6.9 as an example to provide a basic idea [11]. It shows the absorption spectra of
trinitrotoluene (TNT), cyclotrimethylene trinitramine (RDX), and pentaerythritol
tetranitrate (PETN) along with their chemical structures. Multiple characteristic
absorption peaks were obtained for each compound as indicated by arrows, which
align well with the peak positions from the existing literature.
While THz TDS has been dominated by optical techniques, there have been
efforts to obtain THz TDS based on electronic approaches as well. A work in UCLA
demonstrated a 95–105 GHz spectroscopy system based on a CMOS transmitter and
receiver [12]. It detects the re-emission from gas molecules after excitation by the
pulse-modulated CW waves, instead of the absorption by the molecules, with a
mirror-ended cavity gas cell that spans only several centimeters. With the system, the
spectral line of NO2 was obtained. Although this approach involves the generation of
CW waves, the spectroscopic data were acquired in the time domain with the pulse-
modulated CW waves, indicating it is a TDS technique.
a (cm–1)
CH3 60
O2N NO2 40
20
0
0.5 1.0 1.5 2.0 2.5 3.0
NO2 Frequency (THz)
(RDX) 120
1.54 THz 47.0 cm–1
NO2 80 1.95 THz 77.7 cm–1
N 40 2.19 THz 59.0 cm–1
0
N N 0.5 1.0 1.5 2.0 2.5 3.0
O2N NO2 Frequency (THz)
(PETN)
150 2.84 THzb
NO2 100
O 50
O
O 2N 0
O NO2 0.5 1.0 1.5 2.0 2.5 3.0
O
O2N Frequency (THz)
Fig. 6.9 Absorption spectra of various explosive materials obtained with THz TDS. (Adapted with
permission from [11] © The Optical Society)
with a special technique such as cavity ringdown [16], while it can be as long as
hours if an extremely fine frequency resolution is desired.
There can be two approaches for CW FDS operating in the THz band, one based
on optics and the other on electronics, each of which will be described below.
Optics-based CW FDS usually employ a photomixing technique to attain the
required CW generation (see Sect. 5.2.1 for the details on photomixing). The beating
signal from two laser sources can provide a bandwidth exceeding 1 THz with a
wavelength tuning less than 1%, with typical near-infrared lasers employed for
spectroscopy. A frequency resolution in the order of MHz may be achieved,
although the actual resolution obtained in the practical setup may be much larger
than this [13]. It is noted that the typical frequency resolution of THz TDS is limited
around 1 GHz, which is dictated by the span of the delay stage (the resolution is
given by c/(2l), where l is the moving stage span and c is the speed of light [17]).
Recently emerging as an alternative is electronics-based FDS that employs an
electronics-based CW THz signal source, which is under rapid development as
described in Chap. 2. Compared to optics-based FDS, the available frequency
range with electronics-based FDS is still lower, limited to a few hundred GHz
range. However, it offers a much better frequency resolution, which can reach as
small as sub-kHz range [18]. Additionally, a low-cost compact spectroscopy system
with a tiny power consumption can be realized based on electronics-based FDS,
which is favored for various practical applications.
One application that attracts recent interests is the detection of the rotational states
of gas molecules. The interest was partially driven by its potential application to the
detection and analysis of the human exhaled breath. A breath of a human may
contain hundreds of volatile organic compounds (VOCs) [19], some of which can
serve as diagnostic biomarkers of various diseases [20]. For the spectroscopy to
detect the rotational states of gas molecules, which is called the rotational spectros-
copy, electronics-based CW FDS is very well suited. While the vibrational state
transition of gas molecules falls in the range of mid-infrared spectrum, the rotational
state transition corresponds to mm-wave and THz region due to its smaller energy
transfer involved, which can now be covered by electronics-based FDS. Besides, the
linewidth for rotational spectral lines is only in the order of 1 MHz, which requires a
frequency resolution well below this linewidth. This demanding target is also
satisfied by electronics-based FDS. The bandwidth that can be obtained with the
currently available electronic THz systems for spectroscopy, up to around 100 GHz
[18, 21], may limit the applications that require a wide frequency range. However,
most of chemical species of interest show repeated response pattern in the frequency
domain. Thus, most of the key spectroscopic information of those species is likely to
be included in the frequency window offered by the current electronics-based
systems [18].
There have been recent reports that demonstrated rotational spectroscopy based
on semiconductor-based THz sources and detectors [18, 22]. In a work in IHP [22],
transmitters and receivers based on SiGe BiCMOS technology operating at
238–252 GHz and 494–500 GHz were developed for gas spectroscopy of exhaled
breath. For the detection experiment, the gas sample was contained in a folded
6.1 THz Spectroscopy 285
absorption cell of a 1.9-m total length, through which the generated THz wave
traveled to experience the absorption. Figure 6.10 shows the experimental setup and
the absorption spectrum of CH3OH (methanol) measured with the 238–252 GHz
system [22]. The obtained spectrum reveals a nice match with the absorption lines
calculated based on the JPL database (Jet Propulsion Laboratory submillimeter,
millimeter, and microwave spectral line catalog) shown on the bottom of the plot.
Also, an MIT group developed a spectrometer operating in the range of
220–320 GHz based on a transceiver in CMOS technology [18]. The transceiver,
286 6 THz Applications
which serves as both transmitter and receiver, generates a signal with 10 comb
spectral lines over two frequency bands, reducing the scanning time with a factor of
20. With a 70-cm-long gas cell whose optical length depends on the propagation loss
in the cell, the absorption lines of CH3CN (acetonitrile) and OCS (carbonyl sulfide)
were obtained, which were consistent with the JPL database.
6.2.1 Overview
advantage of the shorter wavelength, which also enables a smaller component sizes,
including that of antennas.
There are various types of THz imaging, and it can be categorized from various
aspects. The types of THz imaging will be described below with relative strong and
weak points for each. Note that a large part of the categorization listed below can be
applied not only to THz but also to other imaging modalities as well. First, THz
imaging can be grouped into active imaging and passive imaging. In active imaging,
the target object is illuminated by THz waves generated from a signal source, and the
2D (sometimes 3D) spatial distribution of the response by the object is taken as an
image. It can be further sub-grouped into transmission imaging and reflection
imaging. In the former case, the image is obtained based on the THz waves
transmitted through the object, while the latter case makes use of the waves reflected
from the object. Active imaging benefits from a better control of the wave properties
such as frequency and intensity, as well as the availability of waveform modulation
if needed. However, it requires a dedicated signal source, preferably with a large
output power to attain a high SNR and a longer imaging range. Passive imaging is
based on the detection of the THz radiation from the target object, which obviates the
need for the signal source. The radiation can be either thermal radiation from the
object or the reflection of ambient THz waves by the object. While passive imaging
benefits from the simple system scheme without the source, the detector should
exhibit an extremely high sensitivity along with a wide bandwidth to catch the weak
signal from the object, a major challenge for THz passive imaging.
THz active imaging can also be grouped based on the type of the source signal:
pulse imaging and CW imaging. The basic properties of the pulse and CW signals
reviewed in the previous section for spectroscopy apply to imaging as well. Ultra-
short pulses inherently contain information over a broad frequency range. Hence,
imaging with such short pulses naturally provides images over a wide range fre-
quency, as is called spectral imaging. This implies that images can be reconstructed
at any desired frequency point included in the wide bandwidth offered by the pulses,
once the imaging is performed with the pulses. On the other hand, CW imaging can
be carried out with a simpler setup than pulse imaging owing to the absence of the
delay stage needed for the sampling process, possibly with both source and detector
realized with electronic components. Spectral imaging is in principle possible with
CW imaging as well, but the image capture needs to be repeated for every frequency
point desired. In both pulse imaging and CW imaging, images can be constructed
based on either attenuation or phase delay of the signal after passing through the
target object for the transmission-mode imaging.
Yet another categorization may divide THz imaging into 2D imaging and 3D
imaging. Conventional images are based on 2D imaging, which is basically a
process of mapping the target image seen from a single direction into a 2D space.
For some applications, however, 3D images of the target object are needed, in which
3D mapping of the object is carried out, as is the case for tomography. One well-
known modality of tomography is the computed tomography or CT, which is
ubiquitous in hospitals for medical inspection purposes. While CTs in medical fields
are based on X-ray to view the human body interior, THz CT can be realized with
288 6 THz Applications
THz waves based on the same imaging principles. There are other modalities for
THz tomography as well, based on both transmission and reflections imaging. They
will be detailed in Sect. 6.2.5.
The last categorization of THz imaging to be discussed here is single-pixel
imaging vs array imaging. A stand-alone detector, the simplest detector scheme,
can detect THz signal incident on a single point at a time. Hence, it can provide
information corresponding only to a single pixel of an image. For this reason, it is
called a single-pixel detector. To construct 2D (or higher-order) images with such
single-pixel detectors, scanning process is indispensable, either of the target object
or the detector. For some applications, real-time imaging is desired, which cannot be
supported by single-pixel detectors. To meet such a demand, array detectors are
necessary, with which a simultaneous acquisition of multiple-pixel data is possible.
1D detector arrays may significantly reduce the scanning time, but fall short of
providing real-time imaging with a reasonable video frame rate. From a practical
viewpoint, 2D arrays, preferably with sufficient pixel counts, are needed to realize
real-time imaging. As the pixel dimension increases with the wavelength, realizing
THz 2D arrays with a large pixel count is more challenging than those for visible
light, especially when size is considered as a constraint. Nevertheless, there have
been efforts to make large THz 2D detector arrays, which resulted in some successful
real-time imaging, as will be discussed in this section.
In the following subsections, various topics for THz imaging will be discussed in
further detail. The description will be first made for active imaging, including both
transmission and reflection modes, and passive imaging, followed by real-time
imaging and tomography. The focus will be made largely on CW imaging with
electronic sources and detectors.
As briefly mentioned above, active imaging provides a high SNR for THz imaging
by illuminating the target object with a THz beam of a desired frequency from a
signal source. With the apparent benefits, most of the reports on THz imaging are
currently based on active-mode imaging. Both transmission and reflection modes of
active imaging will be treated in this section.
Before moving on to the actual imaging setup and example results, it would be
helpful to review the definition of the resolution in imaging. The resolution is not a
term strictly reserved for technical communities as it is widely spread in our daily
lives. When you want sharp images on your phone or monitor/TV screens, you say
you need a higher resolution. In such cases, where the resolution concerns the
display performance, it is given as pixel counts, such as 3840 2160 pixels or
~8.3 mega pixel for UHD (ultra high definition, also called 4k display). There may
be various definitions of the resolution depending on specific applications. In optics,
the resolution is represented by different measures as described below.
One convenient and widely used parameter is the angular resolution, which
indicates the minimum angle required for two point objects to be resolved or
6.2 THz Imaging 289
Fig. 6.11 (a) A typical Airy function, shown together with the 2D and 3D representation of the
intensity. (b) The definition of the angular resolution θ, which corresponds to the case where the
center of Airy function 1 and the first minimum of the Airy function 2 coincides
λ
θ 1:22 ð6:3Þ
D
290 6 THz Applications
Focal
plane
Lens
Image Image
plane 2 plane 1
Object 2 Object 1
Fig. 6.12 The location of the focal plane and the image planes. As an object moves away from the
lens, its corresponding image plane approaches the focal plane. Note that, in geometric (ray) optics,
the incident light parallel to the lens axis passes through the focal point after the lens, and the light
passing through the center of the lens maintains a straight path. The image shows sharp patterns, or
“focused,” on the image plane where the multiple lights from a point source merge after different
paths
where λ is the wavelength and D is the lens diameter. This is called the Rayleigh
criterion (the factor of 1.22 is for a circular aperture, which becomes 1 for a single
slit). For instance, an image obtained with a THz wave of λ ¼ 1 mm (300 GHz) and a
lens with D ¼ 50 mm, will show a resolution of θ ¼ 0.024 rad or 1.4 . The resolution
will be improved with a small λ/D ratio, which can be easily understood since the
diffraction will be less pronounced for a small λ/D ratio. It is noted, though, that as λ/
D becomes smaller and the diffraction becomes less dominating, other factors, such
as aberration, begins to affect the resolution, resulting in an angular resolution worse
than Eq. (6.3).
If needed, the resolution may also be represented by the minimum distance
between the images of two point sources projected on the image plane for them to
be resolved. With the help of Fig. 6.11b, this spatial resolution can be simply
obtained by multiplying the angular resolution by the distance from the lens to the
image plane, t, as follows:
λt
d θt 1:22 ð6:4Þ
D
f-number of the lens. There are other definitions for spatial resolution, such as the
microscope resolution (d ¼ 0.61λ/NA, where NA is the numerical aperture). Based
on different assumptions, the various spatial resolution definitions may not show the
exactly same values, but they fall within an acceptable range.
60
50
Height [mm]
40
30
20
10
0
10 20 30 40 50 60 70 80
Length [mm]
(a) (b)
Fig. 6.14 (a) Image of various object hidden in a paper envelope obtained from electronics-based
CW THz transmission imaging. (b) Chip photo of the detector array used for the image. (© 2009
IEEE [24])
Fig. 6.15 (a) Images of a half-sided floppy disk with various input power level obtained with two
different types of detectors. (b) Chip photo of the 300-GHz SiGe HBT direct detector. (c) Chip
photo of the 300-GHz SiGe HBT heterodyne detector. (© 2017 IEEE [25])
detectors, direct and heterodyne, both designed for 300-GHz operation based on
SiGe HBT, were adopted and compared. The direct detector consists of an on-chip
dipole antenna and a differential-pair-based HBT detector circuit (Fig. 6.15b). The
heterodyne detector is composed of an on-chip dipole antenna integrated with a
mixer, an on-chip oscillator as LO, an IF amplifier, and an IF detector (Fig. 6.15c).
Both detectors were mounted on a Si lens for efficient backside radiation from the
chip. The imaging results show that clear images can be maintained down to much
lower input power levels for the heterodyne case, which is an obvious consequence
of the lower NEP of the heterodyne detector than that of the direct detector (3.9 pW/
Hz1/2 vs 21.2 pW/Hz1/2).
THz reflection imaging provides images of the reflectance on the object surface
based on the signal strength reflected back from the various regions over the object
294 6 THz Applications
after THz beam illumination. For the reflection imaging, the acquired information is
limited to the shallow layer close to the surface of the object. Also, it is highly
sensitive to the reflection angle on surface and affected by the surface scattering as
well. On the positive side, however, as opposed to the case of the transmission mode,
the reflection-mode imaging can be applied to objects with high THz attenuation
and/or large thickness. For example, biomedical samples, which contain high level
of water content, should be sliced thin for THz transmission imaging, allowing only
ex vivo analysis. With reflection imaging, on the other hand, in vivo imaging is
possible. Also, THz reflection imaging benefits from the fact that both THz source
and detector can be located at the same side of the object, which is an attractive
feature for many practical applications especially for the stand-off imaging. In fact,
radar systems require the source and detector to be located on the same side, and the
imaging radar belongs to the domain of reflection imaging. These observations
clearly indicate that there do exist THz applications that prefer the reflection
imaging, although the transmission imaging is still the dominant modality. As is
the case for the transmission mode, 2D imaging with object scan is the most widely
adopted scheme of the reflection mode, while real-time imaging and 3D tomography
imaging can also be realized with the reflection mode.
There have been quite a lot of reports on THz reflection imaging based on THz
pulses generated with optical means [26, 27], whereas CW-based results with
electronic sources and detectors have been relatively sparse. A reflection-mode
THz imaging based on an electronics-based CW source and detector has been
reported in [28]. For the THz source, an on-chip integrated 300-GHz oscillator
built with InP HBT was used, which exhibited a peak output power of 3.4 mW
[29]. The InP detector based on a differential pair exhibited an NEP lower than
35 pW/Hz1/2. The imaging setup is illustrated in Fig. 6.16a. The radiation and the
detection were made through a pair of horn antennas with a gain of 26 dBi via
on-chip probing of the source and detector circuits, which were placed on mobile
probe stations that enable beam alignment. The radiated beam was incident on the
sample fixed on a moving stage for raster scan and then consequently reflected and
collected by the detector. The beam path was controlled by lenses and off-axis
parabolic mirrors, with a mechanical chopper inserted for beam intensity modula-
tion. Images for various bio samples were obtained including the one shown in
Fig. 6.16b. It provides the surface images of a rat brain with a tumor bisected for the
experiment, obtained with both regular and THz reflection imaging. A clear image
with a high contrast for the tumor region was obtained with the THz imaging, which
is due to the slightly higher water content in tumor region. The measured reflectivity
was 45–46% for tumor and 40–41% for healthy tissue, the difference of which was
converted into the clear image contrast. The results indicate the possibility of
applying THz reflection imaging to medical applications to precisely identify the
cancerous regions.
6.2 THz Imaging 295
Fig. 6.16 (a) Setup for THz reflection imaging. (b) THz reflection image of a rat brain with tumor
cut by half. Also shown are the regular image and the reflectivity profile along a cutline. (© 2017
IEEE [28])
As briefly mentioned earlier, passive imaging relies upon the natural radiation from
the object. The conventional photography, which has become a core part of our daily
lives, is a good example of passive imaging (images with flashlight technically can
be considered as an active imaging, though). The imaging system is different from
that of active imaging. It does not require a signal source and consists of only a
detector. Also, because of this, the optics for signal path control is different, typically
with a simpler configuration. The performance requirement for the detectors is also
different from those used for active imaging. In addition to extremely high sensitiv-
ity to pick up the tiny signal, a large bandwidth is crucial to collect the incoming
signal spread over a wide spectrum range. Besides, a feature to suppress the long-
term fluctuation of the received signal level is desired. An overview of THz passive
imaging systems will be presented below, followed by examples obtained based on
electronic approaches.
296 6 THz Applications
T
NETD ¼ pSYS
ffiffiffiffiffi : ð6:5Þ
Bτ
The system noise temperature is the sum of the effective antenna noise temperature
(TA) and the radiometer noise temperature (TR), so that TSYS ¼ TA + TR. A rule of
thumb for passive imaging is that NETD needs to be below 1 K, while even much
smaller values may be required for enhanced image quality. A simple relation also
holds between NETD and NEP [32]:
NEP
NETD ¼ pffiffiffiffiffi : ð6:6Þ
kB 2τ
According to Eq. (6.5), the integration time directly affects NETD, along with the
system noise temperature and the bandwidth of the radiometer. A longer integration
time would reduce NETD, due to the noise averaged out during the integration. In
reality, however, the sensitivity improves with the integration time only to a certain
point (called Allan time), beyond which the sensitivity actually begins to degrade, a
typical pattern observed for Allan variance [33, 34]. This is due to the gain fluctu-
ation of the amplifier in the radiometer, which hinders the detection of ΔT as the gain
fluctuation can hardly be distinguished from object temperature fluctuation. Such a
6.2 THz Imaging 297
Fig. 6.17 Schematic diagram of a typical (a) Dicke radiometer, (b) Total power radiometer
gain instability is independent of the ratiometer noise, and Eq. (6.5) can be modified
to include the effect of the gain fluctuation as follows [31, 32]:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
1 ΔG
NETD ¼ T SYS þ , ð6:7Þ
Bτ G0
Hence, the effect of the gain fluctuation as well as the low frequency component of
the system noise will be suppressed. In fact, this function is similar to the chopping
of the signal source often employed for active imaging. Only difference is that, for
passive imaging, the modulation ought to be made at the detector side, requiring a
switch integrated in the receiver. While the modulation helps to suppress the low
frequency fluctuation, the received power is reduced by half as the detector is
connected to the antenna only for a half the cycle. In this case, the sensitivity is
degraded by a factor of two as follows:
2T SYS
NETDDicke ¼ pffiffiffiffiffi : ð6:8Þ
Bτ
This can be compared with the case for total power radiometers, in which the object
signal is received for the entire period of the integration time, rendering the “total
power” of the object to be integrated and NETD given as Eq. (6.5). However, this
apparent sensitivity degradation by a factor of 2 for Dicke radiometers applies only
to the ideal case of no gain fluctuation. At the presence of the gain fluctuation, NETD
is likely smaller for Dicke radiometers than the total power radiometers as the gain
fluctuation term in Eq. (6.7) will be negligible with the modulation, which is the
whole purpose of employing a Dicke switch.
There has been a recent surge of efforts to realize passive imager circuits based on
Si technologies at the frequency range around 100 GHz [32, 36–43]. There was also
a hybrid approach in which an InP-based LNA is co-packaged with a Si CMOS
receiver [44]. While most of them exhibited NETD of around 1 K or even below
0.5 K [38, 40–44], reports with real images taken have been rare so far [40, 44]. This
indicate the challenges in passive imaging at high-frequency bands. One of the
results is briefly introduced here [44]. The total power radiometer consists of two
chips: an InP chip with a low-noise pre-amplifier and a CMOS chip with a receiver,
which is driven by an on-chip LO composed of a frequency synthesizer followed by
a frequency doubler and a power amplifier. The photo of the CMOS chip is shown in
Fig. 6.18a. For a bandwidth of 90–100 GHz, the stand-alone CMOS chip showed a
noise temperature of 3000–25,000 K. However, when combined with the InP
pre-amplifier in a separate waveguide package, the entire system exhibited a noise
temperature of 300–500 K, reduced by a factor larger than 10. The NETD was 0.5 K
with a 50-ms integration time. The DC power consumed by the entire radiometer
system is 257 mW, while its weight is only 334 g, one of the main advantages of
chip-based radiometer system. With the radiometer system, a passive image of a
person was successfully taken with a scan, as is shown in Fig. 6.18b. While the
resolution is limited, it shows a possibility of passive imaging around 100 GHz
based on a chip-based radiometer.
6.2 THz Imaging 299
PGA DACs
PS
PS LNA
LO-PA
ADC
30
LO-X2
DACs Person 1 Head 0.984
20
Elevation (Deg)
0.981
0
CAL 0.980
ASIC
Synthesizer –10 0.979
–20 0.978
1.6mm 0.977
–30
–30 –20 –10 0 10 20 30
(a) (b)
Fig. 6.18 (a) Chip photo of the CMOS receiver for passive imaging. (b) Passive image of a person
obtained with the receiver. (© 2016 IEEE [44])
Scanning for imaging is a time-consuming process, and it may take several minutes
or even more than an hour for THz imaging with reasonable resolution depending on
the imaging size. The most effective way for scan time reduction is to employ an
array detector composed of multiple detector pixels, as was discussed in Sect. 3.4. A
simple math indicates that the scan time will be reduced by a factor corresponding to
the number of array pixels processed in parallel. That is, for an array detector with a
pixel count of n, the scan time will be 1/n of the single-pixel detector case. A popular
practical approach for scan time reduction is to employ a 1D array. With a sufficient
number of elements along the 1D array, only a scan over the direction normal to the
array axis will be required, which will significantly reduce the scan time. However,
for a full real-time imaging, 2D array will be eventually needed, which will be
described below.
The key requirement for real-time imaging is for the entire target scan to be
processed within a reasonable time span, or the inverse of the reasonable video frame
rate. The typical frame rate for practical imaging systems is a few tens of frame per
second (fps). For movies, the standard frame rate is 24 fps. For TVs, it ranges
between 24 and 60 fps depending on the standard. If we take 24 fps as a norm for the
frame rate, the allowed time for a full 2D scan of the object would be roughly around
40 ms. One may argue that real-time imaging is theoretically possible with a single-
pixel detector (or a detector with a small number of pixels) if the integration time
spent on each pixel is very small, or, equivalently, if the video bandwidth is very
high. However, from a practical viewpoint, a certain duration of integration time
300 6 THz Applications
should be spent to obtain a sufficient SNR for each pixel. Furthermore, extra time is
needed to mechanically move the moving stage for a raster scan, which would steal
time from the signal integration process at the detector. For example, the moving
stage adopted for the setup shown in Fig. 6.13 takes ~20 ms to move a 1-mm step at
the maximum rate, rendering the single-pixel option virtually impossible. Therefore,
the most desired and practical option for real-time imaging is to employ a 2D array
detector with a sufficient pixel counts, which can capture the full object image in a
“single shot” without mechanical scan (electronic scan over pixels internal to the
array is still needed). Figure 6.19 shows a typical imaging setup for real-time
imaging with a 2D array detector.
There are challenges to realize a large-sized THz 2D array detector, or THz
camera as is often called. First, the constraint on the array size would limit the
maximum number of pixels integrated in a single array. This limitation will become
more stringent as we move toward the lower frequency side of the THz band with a
longer wavelength. For on-chip detectors we are interested in, a single pixel dimen-
sion will be dictated by the antenna size rather than the circuit area, assuming
resonant antennas are employed. The adoption of electrically small antennas will
degrade the sensitivity (see Sect. 4.2.2). Besides, for diffraction-limited imaging
systems, the resolution will be roughly the dimension of resonant antennas, making
highly integrated small antenna options less effective due to redundancy. A drasti-
cally improved resolution can be obtained with near-field imaging [45], but we will
stay with the far-field case for the current discussion. Second, the THz signal power
delivered per pixel will be reduced as the incoming beam power will be spread over
the entire pixels, as was discussed in Sect. 3.4. In contrast, for a single-pixel imaging
described earlier, roughly the entire source power would be delivered to the single
pixel if we assume near-perfect focusing and negligible path loss through the
imaging system. If detectors with a similar sensitivity is used for arrays, this
would require a source with a much larger power than the single-pixel case. Third,
DC power consumption of the array detector will be larger. With direct detectors in
passive-mode operation, the power dissipation per pixel will be tiny, but the accu-
mulated total power may be considerable if the array size is large. A smart biasing
scheme to partially turn on detectors at a time in sequence and sharing as many
elements (such as amplifiers) as possible between pixels would be desired to reduce
the power dissipation. Additional challenges include longer data processing time for
entire pixels, extended signal loss and added noise along the path and the readout
circuit, nonuniformity over pixels, etc. (see Sect. 3.4.1 for related issues). The cost
could be an issue for 2D arrays if they are based on individually fabricated and
packaged detectors, leading to needs for imaging techniques with reduced number of
detectors, such as the interferometric imaging [46] or compressed sensing [47]. Sin-
gle-chip 2D arrays will not need techniques to reduce the number of detectors for a
6.2 THz Imaging 301
given array area, as partial removal of detectors within the area will not necessarily
reduce the cost.
Despite the challenges, there have been promising reports on real-time imaging
based on array detectors recently. The array detectors introduced in Sect. 3.4.2 are
well suited for real-time imaging. The Si CMOS single-chip 1k-pixel array detector
operating at 0.7–1.1 THz has demonstrated its successful function as a camera,
which is capable of acquiring still images at a single shot and, more importantly,
real-time images [48, 49]. Being composed of 32 32 detectors, each with an
80 80 μm2 area, the entire chip occupies an area of 2.9 2.9 mm2. With the THz
camera fabricated in a 65-nm CMOS technology, transmission-mode real-time
imaging was successfully performed at a frame rate of 25 fps. Figure 6.20 shows a
snapshot of a real-time image, which was taken on a metallic wrench as a test object.
Another 2D array described in Sect. 3.4.2, a Si CMOS 28 28 pixel multi-chip array
detector based on a modular approach with 16 sub-array chips each comprising
7 7 detectors, has also demonstrated its capability of real-time imaging at 300 GHz
[50]. Each sub-array chip has a dimension of 4 4 mm2, leading to a total array size
of 16 16 mm2. Figure 6.21 shows a snapshot of the real-time image taken with the
array, displaying the transmission-mode image of a metallic blade. The image is
based on total 961 pixels including the virtual pixels, which fall on the dead zone
along the peripheral area of each sub-array chip that is consumed by bonding pads
for inter-chip wiring. The image data on those virtual pixels were created by
interpolating the data from the adjacent pixels. The image shows no obvious
indication of the virtual pixels, nor the trace of the sub-arrays, demonstrating the
seamless integration of the multiple chips based on a modular approach.
The discussions so far have been made for 2D images, but it is obvious that 3D
images would provide richer information on the target object. There can be various
ways to obtain 3D images, while the most widely adopted modality is the tomogra-
phy. Tomographic imaging, or tomography, provides a set of cross-sectional slice
images of a target object, or tomograms. Based on the tomograms, the 3D internal
302 6 THz Applications
Fig. 6.21 Snapshot of a real-time image taken with a Si CMOS 300-GHz multi-chip 28 28 pixel
camera (total 31 31 ¼ 961 pixels including the virtual pixels). The boxed area of the metallic
blade on the right is imaged
or p(l), where l is the shortest distance from the origin to the projection line. If the
projection is repeated with multiple projection angles θ, p is eventually given as a
function of both l and θ, or p(l, θ). Hence, f(x, y) defined in the x y plane can be
transformed to p(l, θ) defined in the l θ plane. This process is called the Radon
transform and can be simply expressed as a line integral:
Z
pðl, θÞ ¼ f ðx, yÞdL, ð6:9Þ
Lðl,θÞ
where the integration is made along a straight line L connecting the source and the
detector. In practice, p(l, θ) is the raw data and we need to obtain f(x, y) from the raw
data. A widely adopted algorithm to reconstruct f(x, y) from p(l, θ) is the filtered
back-projection (FBP), which can be described as follows [52, 53]:
Z πZ 1
f ðx, yÞ ¼ Pðω, θÞjωj exp ð2πjωðx cos θ þ y sin θÞÞdω dθ, ð6:10Þ
0 1
where P(ω, θ) is the Fourier transform of p(l, θ) with respect to l, and ω is the spatial
frequency in the direction of l. A high-pass filtering is implicitly included in the
integration that helps to emphasize the high spatial frequency components of the
images, which is the basis for the name, “filtered” back-projection. The presentation
of p(l, θ) in the l θ plane is called the sinogram, and the reconstructed f(x, y)
presented in the x y plane is called the tomogram, which is basically the slice
image of the corresponding cross-section. Hence, the basic process of tomography is
to obtain a tomogram from a sinogram obtained at a fixed z point and repeat this step
over different z values along the z-axis. This full process will generate the 3D image
of the target object. While FBP is a popular choice for CT as mentioned above, the
304 6 THz Applications
Fig. 6.23 (a) Imaging setup for THz tomography. (b) The target object and its THz CT image
obtained with the given setup. (© 2011 IEEE [59])
mirrors for beam path control, as shown in Fig. 6.23a. The target object is metal and
plastic nuts enclosed in a polyethylene container. The obtained 3D CT image is
shown in Fig. 6.23b, which was acquired by raster-scanning the target object with
rotation. It clearly shows the internal structure of the target object, an indication of
the transparency of the container against THz waves. Another CT experiment by the
same group working at 500 GHz was reported in [60], which includes both source
and detector based on Si chips, the former in SiGe HBT and the latter in Si CMOS
technologies.
In another effort, 300-GHz THz CT experiments have been carried out based on
the source based on InP HBT and the detector based on SiGe HBT technologies
[61, 62]. For the detector, both direct detector and heterodyne detector [25] were
used and the CT images from the two types of detectors were compared. An imaging
setup modified from Fig. 6.13 to allow target rotation was used to image the target
object shown in Fig. 6.24a, which is composed of a metallic bolt and a washer
enclosed in an acrylonitrile butadiene styrene (ABS) box. Figure 6.24b shows a set
of 2D projection images (for various projection angles), from which sinograms and
tomograms were obtained (for various z-axis positions) with FBP algorithm. From
these images, the 3D image of the target object was reconstructed. The image
acquisition was carried out with both direct detector and heterodyne detector for
comparison. To investigate the effect of the incident CW power and the types of
detector, a series of 3D images were obtained by varying the source output power
level, which was repeated for both detector types. The results are shown in
Fig. 6.24c. With both detectors, there exists a lower limit of the source power
below which the images begin to degrade, while the limit is lower for the heterodyne
detector owing to its smaller NEP (4 pW/Hz1/2 vs. 21 pW/Hz1/2 of direct detector).
Instead, the heterodyne detector exhibited a sign of saturation at high input power
level, a result of the larger responsivity (320 kV/W vs. 6 kV/W of direct detector).
306 6 THz Applications
Fig. 6.24 (a) The target object composed of a metallic bolt and washer enclosed in a plastic box.
(b) 2D images taken from various rotation angles of the target. Also shown are sinograms and
tomograms obtained from the 2D images with FBP algorithm. (c) THz CT images with varied
source power, with both direct detector and heterodyne detector. (© 2018 IEEE [62])
6.3.1 Overview
from the fact that it is free of the bothersome wire connection across the link points.
One downside of the wireless systems, though, has been known to be its lower data
rate compared to the wired counterpart. However, we observe a rapid increase in
wireless data rate around our daily lives. In fact, the pace of data rate increase for the
wireless systems is so fast that some foresee that it will soon overtake that of the
wired systems and eventually end up surpassing it. For this optimistic prospect to
come true, the exploitation of the THz spectrum is an essential part to enable the
broadband needed for information mass transportation. In this section, a brief
overview of THz wireless communication systems is provided.
The opportunity of using the THz band for wireless communication is rather
straightforward. One effective approach to raise the data rate is to increase the
bandwidth of the system, as indicated by the famous Shannon’s theorem [63]:
S
C ¼ B log 2 1 þ , ð6:11Þ
N
where C is the channel capacity, or the maximum data rate allowed in the channel,
B is the bandwidth, and S/N the signal-to-noise ratio (SNR). The most effective way
to stretch the bandwidth is to increase the carrier frequency. For instance, if the same
fractional bandwidth is assumed, a 10 increase in carrier frequency would result in
a 10 increase in the bandwidth. This simple math has been the driving force behind
the incessant efforts for the carrier frequency increase in the past decades, and it has
now arrived at the THz regime [64–66].
Despite this obvious opportunity for THz communication, there are a great deal of
challenges lying ahead that need to be addressed for its realization. Those challenges
are quickly manifested with a simple link budget analysis as shown below. From the
Friis transmission equation (Eq. 4.57), the signal power at the receiver, Pin,RX, can be
expressed as follows in dB scale (including dBm and dBi):
4πR
Pin,RX ¼ Pout,TX þ GA,TX þ GA,RX 20 log , ð6:12Þ
λ
where Pout,TX is the signal power radiated by the transmitter, GA,TX and GA,RX are
the antenna gains at the transmitter and receiver, respectively. The last term is the
free-space path loss (FSPL) with distance R and wavelength λ. Another parameter
we need to consider for the analysis is the receiver sensitivity Psen,RX, or the
minimum required power received at the receiver for an SNR desired for the given
application. It is given as follows in dB scale at room temperature, which can be
readily derived from the definition of the noise figure (Eq. 3.10):
308 6 THz Applications
where B is the bandwidth given in Hz, NF is the noise figure, and SNRmin is the
minimum required SNR at the receiver output. Note that the first term corresponds
to the value of KT at room temperature expressed in dBm, which is obtained by
10 log(26 mJ 1.6 1019 / 1 mJ) ¼ 174 (dBm/Hz), where the relation
kT ¼ 26 meV at room temperature is used. Hence, the first two terms combined
indicate the total noise power at the receiver input for a given bandwidth, as the noise
is proportional to the bandwidth in linear scale. The sum of the first three terms are
often called the noise floor.
Now, for the link to be safely established, the received power should be larger
than the minimum required power at the receiver, or Pin,RX > Psen,RX. That is:
4πR
Pout,TX þ GA,TX þ GA,RX 20 log
λ
> 174 dBm=Hz þ 10 log B þ NF þ SNRmin : ð6:14Þ
forming, would be required to establish and retain the link. This situation will be
further aggravated by the weak diffraction of THz waves, which will prevent
non-line-of-sight connection. A complex relay mechanism may be needed to cope
with dynamic channel conditions. This long list of challenges is something we ought
to address to arrive at the reliable THz communication systems operating at a high
date rate as desired.
A brief comment will be worthwhile at this point regarding the effect of the
attenuation due to the absorption in the Earth’s atmosphere, which will potentially
affect the wireless links. Figure 6.26 presents the attenuation level in the Earth’s
atmosphere at the standard condition (see figure caption), which includes multiple
peaks that arise from water or oxygen molecule absorption [67]. It will be useful to
compare the loss due to FSPL and that of the atmospheric attenuation. For this
comparison, one may be tempted to combine Figs. 6.25 and 6.26 as a single chart.
However, the situation is not that straightforward. The atmospheric attenuation is
given by an exponential factor driven by the distance R, or exp.(αR), where α is the
attenuation constant given as a function of the frequency, or α( f ). This means that,
when the attenuation level is expressed in terms of dB scale, it is linearly propor-
tional to the distance (i.e., / αR). Hence, the attenuation can be represented by
attenuation (dB) per distance, as is the case of Fig. 6.26, where it is represented in
dB/km. For FSPL, however, this is not true. FSPL is given as (4πR/λ)2, or (4πRf/c)2,
and thus its dependence on the distance is not linear anymore when expressed in dB
scale (i.e., / log(4πRf/c)). Therefore, FSPL cannot be represented by dB per
distance, or by dB/km. This difference would not allow a direct comparison of the
atmospheric attenuation and FSPL in a single chart. More importantly, it implies that
the atmospheric attenuation increases more rapidly with distance than FSPL.
If you still desire a direct comparison between the atmospheric attenuation and
FSPL, one possible solution is to plot them together with a fixed distance. It is done
in Fig. 6.27, which presents the two losses plotted together at four selected distances
of interest. Also included in each plot are the cases where FSPL are compensated for
310 6 THz Applications
1000 1000
10 10
1 1
0.1 0.1
10 m 100 m
0.01 0.01
0.1 0.2 0.4 0.6 0.8 1 0.1 0.2 0.4 0.6 0.8 1
Frequency (THz) Frequency (THz)
1000 1000
Attenuation at 10 km (dB)
Attenuation at 1 km (dB)
1 1
0.1 0.1
1 km 10 km
0.01 0.01
0.1 0.2 0.4 0.6 0.8 1 0.1 0.2 0.4 0.6 0.8 1
Frequency (THz) Frequency (THz)
Fig. 6.27 Comparison of the atmospheric attenuation and FSPL, as well as FSPL compensated for
with various antenna gains, at fixed distances of 10 m, 100 m, 1 km, and 10 km
with various antenna gains (20, 40, and 60 dBi), assuming a pair of antennas of the
shown gain values are placed at TX and RX. At a distance of 10 and 100 m, it is clear
FSPL is dominating, almost totally eclipsing the atmospheric attenuation, even with
the cases with high-gain antennas. (For the entire frequency with the 10-m case and
below ~240 GHz with the 100-m case, negative loss, or net gain, will be obtained in
the free space with a 60-dBi antenna gain, which is not shown in the plot in the log
scale. In this situation, though, the loss comparison is not quite relevant because the
atmospheric attenuation is already too small.). Even with the case of 1-km distance,
FSPL still prevails over the atmospheric attenuation except at a few scattered
frequency intervals. The two losses become comparable beyond ~500 GHz only
with very high antenna gains. At 10 km, there is a cross-over point near 400 GHz,
beyond which the atmospheric attenuation is consistently larger than FSPL. How-
ever, with this level of distance, the total loss will be too large for any THz links to be
useful for practical applications. Hence, it will be fair to say that the loss is
dominated by FSPL for most practical THz applications. Exceptions will be very
long-distance wireless links based on pencil-like beams with a very high antenna
gain operating at limited frequency intervals with a very strong atmospheric
6.3 THz Communication 311
molecular resonance, which is highly unlikely and most likely avoided in realistic
cases. Note that the detailed atmospheric attenuation will vary with the weather
condition and altitude, as is partly shown with Fig. 1.2.
Despite the series of challenges mentioned above, people are marching toward real-
world THz communication systems. Optimistic-minded, they expect those technical
hurdles will be eventually resolved, as has been the case for many other engineering
frontiers. Various usage scenarios have been proposed for future THz communica-
tion systems, some of which are introduced here.
The most rational approach for THz communication systems would be to max-
imize the advantage of the THz band, the wide bandwidth, while circumventing the
weak point, the limited link budget. Naturally, the broadband short-range commu-
nication first appeared as the most immediate application scene of the THz commu-
nication. Much attention has been paid to the chip-to-chip communication
application [68]. A wireless link will get rid of the physical interconnection between
chips, avoiding related issues such as routing complexity and RC delay. THz link is
well suited for this purpose as it can support the increasingly heavy inter-chip data
traffic, while occupying a small chip area including on-chip antennas. The inter-chip
distance will be small, in the order of centimeter or less, which will lead to FSPL
around 40 dB for 300 GHz, for instance, as indicated by Fig. 6.25. This level of loss
can be managed even with low-gain on-chip antennas if RX and TX of reasonable
performances are employed. The idea can be extended toward more practical
applications, which includes wireless connectivity for high resolution displays. For
uncompressed data streaming, 4k (3840 2160 pixels) and 8k (7680 4320 pixels)
displays will require data rate as high as ~12 Gbps and ~48 Gbps, respectively. THz
link is a highly favored candidate to support this level of data rate between a set-top
box and a display. The link distance may be in the order of 10 cm, with which FSPL
will soar up to around 60 dB. However, unlike the chip-to-chip cases, high-gain
antennas such as horns can be employed here, which would be able to provide a loss
compensation of 40–50 dB as a pair. Another practical example of THz link is kiosks
for massive data download to personal mobile devices, as was recently demonstrated
by NTT [69]. For this vending machine concept for data download, the link budge
will be similar to the case for the display link described above, while alignment
condition will be less deterministic as the users are expected to place their mobile
terminal on the kiosk body by hands. Also, the antenna gain will be limited for the
mobile device side. This use case can be easily developed to links between mobile
devices and storage devices of any types in general including PCs or servers [64].
As opposed to such a short-range communication, there is rather a bold approach
as well, in which attempts are made to apply THz waves to a long-distance point-to-
point link application [66]. The main enabler for this long-distance THz link is the
high gain antenna. As briefly mentioned above, reflector antennas can provide a gain
sufficiently large to compensate for the high FSPL of the long distance. One favorite
312 6 THz Applications
tendency for THz applications is the increasing antenna gain with increasing fre-
quency (decreasing wavelength) when the antenna aperture size is fixed. For exam-
ple, with a dish antenna with a diameter of D, the gain formula defined in Eq. (4.54)
can be converted to the following form assuming that the maximum effective
aperture is equal to the dish area (Aem ¼ π(D/2)2):
2
4π 4π πD
G¼ A ¼ e A ¼ e : ð6:15Þ
λ2
e
λ2
r em r
λ
This relation is plotted as a function D for various frequencies in Fig. 6.28. A good
example exploiting this property was reported in [70], where a pair of Cassegrain
reflector antenna, each with 55-dBi gain, was employed to cope with FSPL of
139 dB near 240 GHz for an 850-m link. Although alignment may be an issue,
such a THz long-distance communication link will benefit from the wide bandwidth
of the THz band to enable high data rate. In this work, 64 Gbps was reported with the
given link. It can be mentioned that the 3-dB beam width is around 0.3 , or ~4.5 m at
the 850-m distance, for the 55-dBi antenna, which would require a dish diameter
around 20 cm near 240 GHz based on Eq. (6.15). In addition to FSPL, TX output
power and RX sensitivity are also issues for long-distance applications. It is certainly
true that the link budget will benefit from efforts to enhance the performance of the
THz TX and RX, which would probably lead to a few-dB enhancement in both PA
output power and LNA NF. However, a few dB of gain improvement will be
obtained by only a moderate expansion of the antenna dish size, which will have
the equal effect on the link budget. This implies that TX and RX performances may
not be the critical factors in such a long-distance THz link.
6.3 THz Communication 313
As opposed to wired links that have a fairly well defined channel conditions, wireless
links are based on the air channel, which is greatly affected by the surrounding
environment. The channel is composed of not only the line-of-sight path but also
reflection, refraction, as well as diffraction by various objects, stationary or dynamic.
Movement of transmitters and receivers also need to be taken into account for mobile
wireless applications. Hence, it is crucial for wireless links to have channels properly
modeled for the best prediction of the signal arriving at the receiver. The model will
help the reliable design of the communication systems with expected performance
requirements for antenna gains and other TX/RX parameters in realistic channel
environments. The channel modeling applies to both outdoor and indoor links, while
the latter is considered more critical as the former case will be mostly based on the
line-of-sight straight paths that are less affected by the environment.
There have been growing efforts to develop THz channel models for realistic
building interiors to pave the way for the actual deployment of THz communication
links [71–73]. The scattering and reflection properties of typical building interiors
were investigated with rough or multilayered surface in those reports. For THz
communication links, the utilized range of spectrum is extensive due to the wide
bandwidth available. Hence, in addition to the angle dependence, the frequency
dependence of the reflection properties will become also relevant for THz links
unlike the lower frequency cases, which may be caused by interference in the
multilayered surfaces [72]. Also, detailed behaviors of the reflection and the diffuse
scattering on the rough surfaces will be different for THz waves than the lower
frequency bands due to the small wavelengths [71]. A high molecular absorption in
the atmosphere at the THz band as mentioned earlier is another additional factor to
be considered for channel modeling. All these aspects suggest that channel modeling
dedicated to the THz links is critically needed.
6.3.1.4 Standardization
2.16 GHz
Ch Ch Ch Ch Ch Ch Ch Ch Ch Ch Ch Ch Ch Ch Ch Ch
4.32 GHz 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48
8.64 GHz Ch 49 Ch 50 Ch 51 Ch 52 Ch 53 Ch 54 Ch 55 Ch 56
12.96 GHz Ch 57 Ch 58 Ch 59 Ch 60 Ch 61
17.28 GHz Ch 62 Ch 63 Ch 64 Ch 65
25.92 GHz Ch 66 Ch 67
51.84 GHz Ch 68
69.12 GHz Ch 69
250 255 260 265 270 275 280 285 290 295 300 305 310 315 320
Frequency (GHz)
Fig. 6.29 Channelization of the 252–322 GHz frequency band by IEEE Standard 802.15.3d
range connectivity based on the band near 60 GHz for a data rate of around several
Gbps. A more advanced version recently developed for a higher data rate based on a
higher frequency band is IEEE 802.15.3d, which belongs to THz communication
and is described below.
IEEE 802.15.3d was driven by IEEE 802.15 Task Group 3d (TG3d) formed in
2014, which had evolved from the Interest Group THz (IG THz) (2008) and Study
Group 100G (SG 100G) (2013). IEEE 802.15.3d was approved in September 2017
and published in October 2018 [74]. The channel allocation by IEEE 802.15.3d is
shown in Fig. 6.29. Total 69 channels are defined in the frequency range between
252 and 322 GHz, based on eight different bandwidth types ranging from 2.16 GHz
to 69.12 GHz, all multiples of 2.16 GHz. For the case of the 2.16-GHz channels, for
instance, total 32 channels fit into the entire band provided. This can be compared
with IEEE 802.15.3c based on the 57–66 GHz band, which accommodates only four
channels with the same bandwidth of 2.16 GHz. IEEE 802.15.3d considers two
physical layers, single carrier (SC) and on-off-keying (OOK), with seven modulation
schemes, BPSK, QPSK, 8PSK, 8APSK, 16QAM, 64QAM, and OOK (see Sect.
6.3.2 for the modulation scheme details). The major target applications suggested by
IEEE 802.15.3d are kiosk downloading, intra-device communication, wireless
backhauling/fronthauling, and additional wireless links in data centers, which well
align with those described earlier in this section.
6.3 THz Communication 315
where bn is the input state of the nth bit and ωC is the carrier frequency. Another
modulation scheme, PAM (pulse amplitude modulation), is also an example of ASK,
which may involve more than two signals levels. With PAM, M ¼ 2k different levels
are allowed, where k is the number of bits modulated together. BPSK shares a
common property with ASK, in that they are both based on two signal states, or
binary levels. For BPSK, however, the two states are represented by two signals with
the same amplitude but the opposite phases, 0 and 180 :
QPSK and QAM are often called the quadrature modulation, as they require both
in-phase and quadrature signals. For the quadrature modulations, a stream of the
316 6 THz Applications
Fig. 6.30 Schematic diagram of the quadrature modulation showing the split of the baseband
binary signal. Shown in the inset table is the mapping between two consecutive baseband input bits
(each bit with Tb) and corresponding I- and Q-path bits (with 2Tb)
baseband binary signal bits is divided into two branches, in-phase (I) and quadrature
(Q). The way they are split is depicted in Fig. 6.30. Every pair of two consecutive
binary bits with a period of Tb are converted into two parallel binary bits with a
period of 2Tb on two separate signal paths, I and Q. The two parallel signal streams
with a relaxed rate of level transition (by a factor of two, from Tb to 2Tb) are
modulated by two orthogonal functions, sine function on one path and cosine
function on the other. The modulated signals on the two paths are combined at the
output and transmitted. Hence, for both QPSK and QAM, the modulated signal
output can be expressed as follows:
where α1 and α2 are the amplitude coefficients. It is this pair of coefficients that
distinguish between QPSK and QAM, as well as between different orders of QAM.
The allowed coefficient pairs for QPSK and various orders of QAM are:
QPSK : α1 ¼ 1 and α2 ¼ 1
16QAM : α1 ¼ 1, 3 and α2 ¼ 1, 3
64QAM : α1 ¼ 1, 3, 5, 7 and α2 ¼ 1, 3, 5, 7
256QAM : α1 ¼ 1, 3, 5, 7, 9, 11, 13, 15 and
α2 ¼ 1, 3, 5, 7, 9, 11, 13, 15:
ð6:19Þ
6.3 THz Communication 317
Fig. 6.31 (a) Constellation of various quadrature modulation schemes. (b) Example of waveform
development due to state transitions: state 1 ! 2 ! 3 ! 4, and corresponding waveforms along I-
and Q-paths in 64-QAM system
Hence, there can be four possible distinct states for QPSK, while 16QAM, 64QAM,
and 256QAM comprises 16, 64, and 256 distinct states, respectively. Of course,
there are QAMs with higher orders than shown in Eq. (6.19). Also, there can be
intermediate modulation orders for QAM other than those shown above, such as
32QAM or 128QAM, which omit some portion of the states. As indicated by
Eq. (6.19), a higher-order modulation scheme involves a larger number of amplitude
levels for a given time period, resulting in a higher information density. Equiva-
lently, from the spectrum point of view, this implies a higher spectrum efficiency.
A widely adopted graphical representation of modulations, especially useful for
quadrature modulations, is the constellation, which maps the location of each state of
the modulation in the two-dimensional I - Q space. The name comes from the fact
that the points representing the states resemble a collection of stars in the night sky
that constitute a constellation. The constellations of various quadrature modulation
schemes are shown in Fig. 6.31a. Each point on a constellation represents a signal
state as defined in Eq. (6.19). Some of signal waveforms in I and Q paths are
presented in Fig. 6.31b with 64QAM case with a few example points, which will
help understand the development of waveforms due to state transitions. Every state
transition in the time domain corresponds to a jump from one point to another on the
318 6 THz Applications
0.1
0.1 1 10 100 1000
Carrier Frequency (GHz)
The OOK modulation transfers the information by selectively transmitting the carrier
signal in the time domain depending on the state of the binary signal. The system that
supports this function can be separately reviewed with two parts: transmitter and
receiver.
On the transmitter side, the simplest approach to realize OOK modulation is to
involve a switch to turn on and off the continuous wave (CW) radio-frequency
(RF) carrier signal streaming out of the signal generator based on the base-band
(BB) signal. The most straightforward approach is to employ a dedicated electrical
6.3 THz Communication 319
switch inserted in the middle of the RF signal path, which is driven by the BB signal
(see Fig. 6.31a). For this purpose, the switch needs to show a low insertion loss and a
high isolation, as well as a fast switching speed to handle the data rate that may reach
as high as tens of Gbps. This method typically benefits from the high data rate
available with fast switching, but is affected by the inevitable loss from the switch.
Alternative approaches involve switching of one of the functional circuit compo-
nents that constitute the OOK transmitter core, such as oscillator or PA. The
switching of the oscillator that generates the carrier signal may result in a high
on-off isolation with a minimal loss, but it inherently suffers from the slow switching
speed arising from the finite oscillation start-up delay. The switching of the PA may
be a practical option as no extra switching loss is expected, but it typically shows a
limited on-off isolation. After all, the best option needs to be selected for a given
application scenario, considering loss, on-off isolation, switching speed, as well as
the chip area and power dissipation [76].
Another approach to attain the OOK modulation is to employ a mixer, with which
multiplication of the carrier signal and the BB signal is performed. Assuming the BB
signal is composed of a stream of binary bits of one and zero, the multiplication will
provide basically the same function as was obtained with the switching described
above. The multiplication in the time domain is equivalent to the convolution in the
frequency domain, leading to a frequency shift, or frequency conversion, of the BB
signal upward by the carrier frequency. From the physical implementation point of
view, this function can be realized with a heterodyne system including a mixer and
an LO (local oscillator), as depicted in Fig. 6.33b. For the modulation in the
transmitter, the RF carrier generated by the signal generator enters the mixer together
with a binary BB signal, and the modulated signal streams out at the mixer output.
There are issues surrounding the mixer when THz operation is assumed, which will
be discussed shortly when heterodyne detection is described. It is noted that
switching is in fact one practical way of achieving a mixing action, and thus the
both approaches described above are basically based on the same principle.
In either approach for OOK modulation, the modulated signal needs to be
amplified by a PA and transmitted into the air by an antenna. A major challenge
for the transmitter is the availability of PA operating at the THz band, as mentioned
in Sect. 2.5. Depending on the fabrication technology used for the design, amplifi-
cation may not be achieved at the targeted carrier frequency for the modulation. Or,
even when amplifiers are available with a desired gain, the saturation output power
and the bandwidth may fall short of the system needs. In this case, as an alternative
approach, a frequency multiplier (N ) can be placed at the final stage, driven by a
preceding PA that operates at a much lower frequency (by a factor of N ), as is
described in Fig. 6.33c. For this purpose, a frequency multiplier with a small
conversion loss and a high saturation power will be needed. The inclusion of a
frequency multiplier will also relax the frequency requirement of the signal gener-
ation by the same factor. This frequency-multiplier-last approach works fine for
OOK modulation, but will not be suitable for many of the quadrature modulation
schemes as will be discussed later. Another challenge is related to the radiation of the
modulated signal into the air, which will be handled by an antenna, on-chip or
320 6 THz Applications
off-chip. An on-chip antenna will be favored for compact implementation, but the
antenna gain will be too low for most THz applications. For such a situation, an extra
gain-enhancement scheme will be needed, for which a Si lens is often adopted. A
large bandwidth is also a desired aspect of antennas, especially for THz communi-
cation systems that benefit from wide channel bandwidths. This need will be more
pronounced for low-order modulation schemes such as OOK, as they would prefer a
large channel bandwidth to compensate for the low modulation order in achieving
the required data rate. Regarding the LO for signal generation, a large tuning range
would be preferred if the system supports multiple channels. For this purpose,
off-chip signal generation may be considered, which is followed by on-chip fre-
quency multipliers with buffers. For a compact system, however, on-chip LO will be
preferred with improved performance.
On the receiver side, a couple of topology options can be considered to perform
demodulation. As a first option, the binary signals can be recovered
(or demodulated) by the direct detection. In this case, the envelope of the waveform
incoming through an antenna will be detected, as described in Fig. 6.34a. The
envelope detector, which performs the detection, works in a similar way as the direct
detector used for imaging receiver described in Sect. 6.2: it converts the arriving
signal power into a voltage output. The difference is that a much higher temporal
resolution is required for the detectors in communication receivers than imaging
6.3 THz Communication 321
receivers. As the noise figure of the envelope detector is substantially high, it may
degrade the overall noise performance of the receiver chain if located at its front-end.
For this reason, a low noise amplifier (LNA) is desired at the front-end of the
detector, while it is not always available at the THz band for a similar challenge
faced by PAs at the transmitter. The envelope detector is often followed by a limiting
amplifier, which is a highly nonlinear amplifier with a high gain, so that the signal
peaks (at top and bottom) of the waveform are truncated, or “limited.” With the
limited waveform, usually obtained with intended signal saturation, a square-shaped
waveform, which is a replica of the original BB signal, can be recovered. The
recovered signal streams into the BB block, optionally through an on-chip clock
and data recovery (CDR) circuitry for improved data quality and clock extraction.
As an alternative approach for receivers, demodulation can be carried out by
multiplying the incoming signal with the LO signal with a mixer. Assuming the same
frequency for the LO and incoming signals, this process can be easily understood
based on a basic trigonometric identity:
1
yðt Þ ¼ AC ðt Þ cos ωC t cos ωC t ¼ AC ðt Þð1 þ cos 2ωC t Þ, ð6:20Þ
2
322 6 THz Applications
where Ac(t)cos ωct is the modulated incoming signal, cos ωct is the LO signal, and
y(t) is the demodulated output signal. It is clear from Eq. (6.20) that the BB signal,
AC(t), can be recovered with a linear factor if the high frequency component (2ωc) is
filtered out with a low pass filter. This is called the homodyne detection, as the carrier
frequency and the LO frequency are identical (homogeneous). This scheme is also
referred to as the zero-IF detection or direct down-conversion. A typical homodyne
receiver is shown in Fig. 6.34b. One caveat in Eq. (6.20), however, is that it was
implicitly assumed that the LO signal has the same phase as the incoming signal. In
general cases, however, there should be a finite phase difference between the
received signal and the LO signal, which leads to a more generic expression of cos
(ωct + ϕ) for LO signal where ϕ is the phase difference. This non-zero phase
difference will inevitably lead to a reduced recovery signal level, the worst case
occurring when ϕ ¼ π/2 that leads to a complete elimination of the signal at
DC. To mitigate this phase issue, a CDR or I-Q mixer may be needed.
For a slightly altered approach, the LO frequency may be set slightly off the
carrier frequency. In this case, the down-conversion does not directly result in the
BB signal, but in the BB signal “weakly” modulated by an intermediate frequency
(IF) that equals the difference between the carrier and LO frequencies. The process
will be governed by expression as follows:
where ωC1 and ωC2 are the carrier frequency and the LO frequency, respectively.
One can easily notice that Eq. (6.20) is a special case of Eq. (6.21) with ωC1 ¼ ωC2.
As was the case of homodyne detection, a low-pass filtering will leave only the
desired component, or the IF signal with ωIF ¼ |ωC1 – ωC2|. An additional mixing
will be required afterwards to arrive at the BB signal This scheme is called the
heterodyne detection (see Fig. 6.34c), as the carrier frequency and the LO frequency
are different (heterogeneous). Occasionally, however, in a broader sense, the het-
erodyne detection refers to both types of the detection described above. Note that it
was again assumed in Eq. (6.21) that there was no phase difference between the LO
and RF signals. A finite phase difference will cause a similar issue as mentioned
above, which requires the same solutions.
With both homodyne and heterodyne detection schemes based on frequency
down-conversion, a much higher sensitivity can be expected when compared with
the direct detection case. This is because a lower noise level is typically exhibited by
mixers than envelope detectors at the frequency range. The employment of an LNA
at the front-end will make the difference less pronounced, but, as mentioned earlier,
the availability of LNAs is not guaranteed for THz applications due to the demand-
ing operation frequency. Also affected by the high operation frequency in the THz
receiver is the LO that needs to drive the mixer at the carrier frequency. One
technique that relaxes the frequency requirement for the LO is to employ a
6.3 THz Communication 323
subharmonic mixer, instead of a fundamental mixer, which will reduce the required
LO frequency by a factor of the harmonic number adopted for the subharmonic
mixer. It is true that mixer performances such as the conversion gain will be
compromised with a subharmonic mixer [77], but it is still an attractive option
when the generation of the carrier frequency is not available at the receiver side.
Yet another way to reduce the LO frequency is to insert a frequency multiplier
between the LO and the mixer. This extra component will crowd the receiver circuit,
but mixer performance degradation may be avoided if the frequency multiplier
delivers a sufficiently large output power. The adoption of the frequency multiplier
also allows off-chip LOs. An off-chip LO, often a commercial signal generator,
would potentially provide a high output power along with a low phase noise.
However, on-chip LOs will be still favored if a compact chip solution is desired,
although an on-chip LO may require a signal-locking, on-chip or off-chip, to
improve the signal integrity that is crucial for communication applications. One
additional advantage to be mentioned for the heterodyne systems is that they can
possibly be operated as a coherent mode, in which case the phase information of the
signal can be extracted. Although OOK modulation does not necessarily require
phase information, a heterodyne-based OOK receiver can be easily extended for
quadrature modulation [78], for which the phase information is critical for informa-
tion recovery.
There have been several notable reports on THz communication systems based on
OOK (or ASK) modulation scheme [77–80]. In one of the earliest such reports [78],
heterodyne TX and RX chips operating at 220 GHz were implemented based on a
50-nm GaAs mHEMT technology (see Fig. 6.35). The TX chip is composed of a
mixer and an amplifier, together with a frequency doubler that up-converts the
sub-harmonic LO signal (110 GHz) injected from an external signal source. The
single-ended fundamental-mode mixer modulates the carrier signal with the IF
signal that carries the BB data. The modulated carrier signal at 220 GHz is amplified
before exiting the chip into a waveguide package, which leads to an off-chip horn
antenna used for radiation. The amplifier is in fact a low-noise amplifier that is also
used for the RX, which resulted in 1.4-dBm maximum TX output power with IF and
LO power of 5 dBm and 10 dBm, respectively. The RX is composed of the identical
set of circuits included in the TX, say, LNA, mixer, and frequency doubler. The
noise figure is estimated as low as 6.5–7.5 dB at 220 GHz, which reflects the
excellent low-noise performance of the HEMT technology. Based on the TX and
RX enclosed in waveguide packages, OOK link experiments were conducted, with a
pair of lenses placed along the air channel for beam collimation. With the link setup,
10-Gbps data transmission was successfully achieved across a 2-m distance with a
bit error rate (BER) of 3.2 1010 and 1.6 109 for pseudo random binary
sequence (PRBS) of 215 1 and 231 1, respectively. A similar experiment was
repeated at a shorter distance of 50 cm, resulting in a higher data rate of 25 Gbps. It is
noteworthy that the identical chips were also used for QAM modulation
experiments.
In another effort based on a 250-nm InP HBT technology, a 300-GHz heterodyne
chip set was developed that contains both TX and RX [77]. In a rare approach, the
324 6 THz Applications
Fig. 6.35 (a) Block diagram of a 220-GHz transmitter and receiver chip sets based on a 50-nm
GaAs mHEMT technology. (b) Photo of the fabricated chips. (© 2011 IEEE [78])
chip includes an on-chip local oscillator, which is shared for TX and RX to drive
up-conversion and down-conversion mixers, respectively (See Fig. 6.36). With the
oscillator operating at the carrier frequency of 300 GHz [29], the mixers can be
realized in a fundamental mode without the need for a frequency multiplier for
subharmonic mixing. No external signal generator is required, either. An RF ampli-
fier operating at 300 GHz, a variant of the one reported in [81], enables the
amplification of the carrier signal at both TX and RX. The transceiver circuit is all
based on the differential configuration, which requires the inclusion of baluns at the
RF and LO interfaces for connection with single-ended schemes. The TX exhibited a
conversion gain up to 25 dB with a 3-dB bandwidth of 18 GHz, with an output
power of 2.3 dBm after the output balun. At the RX, a conversion gain of 26 dB
was attained with a 3-dB bandwidth of 20 GHz. The noise figure, which was
measured with the Y-factor method based on a setup with liquid-nitrogen (LN),
showed a range of 12.0–16.3 dB. While not demonstrated in the paper, the chipset is
capable of OOK data transmission. There have been OOK transmitters and receivers
reported based on Si technologies as well operating beyond 200 GHz [82, 83].
6.3 THz Communication 325
the branch-line coupler described in Sect. 2.5.2, can be used [85], or simply a λ/4 line
section can be inserted in one of the two signal paths. Employing a quadrature VCO
(QVCO) is another option [86, 87], although not popular for THz applications yet
partly due to the complexity in locking for stable oscillation. As for the power
combiner that merges the signals from I and Q channels, the various power com-
biners introduced in Sect. 2.5.2 can be considered, including rat race couplers
modified for a large number of inputs [88–90] or smaller foot print [91]. The
requirements for the mixers will be similar to those in OOK transmitters discussed
earlier, and so will those for antennas. From the overall system point of view,
maintaining balanced characteristics between the I and Q channels is critical for
quadrature transmitters (and receivers), calling for extra efforts in circuit design
compared to the OOK case.
As mentioned above, when PA is not available at the carrier frequency, one
option is to perform modulation at a lower frequency, followed by a frequency
up-conversion. Probably the simplest way for this frequency up-conversion process
is to apply a frequency multiplier. However, this frequency-multiplier-last scheme
works fine for OOK transmitters (see Fig. 6.33c), but not necessarily for quadrature
transmitters. This scheme may corrupt the constellation of quadrature modulations
with misarranged signal points. This can be easily shown with a simple math.
Consider the modulated signal (before frequency up-conversion) in the form of
328 6 THz Applications
Eq. (6.18), which can be shown by a simpler expression without loss of generality as
follows:
where ω0 is used instead of ωC to distinguish from the final carrier frequency. The
condition required for this quadrature signal to maintain the same (relative) position
in the constellation after frequency up-conversion is that the I signal and Q signal
should scale by the same factor. In other words, the magnitude ratio of the I and Q
signals should remain the same after the frequency multiplication. Now, let us
consider the case where the frequency up-conversion is performed with a frequency
doubler based on nonlinearity. Then, the output signal will become:
side band (SSB) NF of 22.9 dB were attained in the absence of LNA. With the
developed TX and RX, a data rate of 80 Gbps was obtained with a 16QAM link at a
distance of 3 cm, while operations with QPSK and 32QAM were also reported.
In [92], a transmitter and a receiver operating at the 220–260 GHz frequency
range was developed based on a SiGe HBT technology, which features a PA
operating at the carrier frequency for both TX and RX, an indication of the superior
RF performance supported by the SiGe HBT technology. In TX, an externally
injected signal is converted into the 236 GHz carrier frequency with a 16 frequency
multiplier followed by a buffer amplifier, which drives the I-Q mixers to modulate
the BB signal after a split by a quadrature coupler. The modulated I and Q signals are
combined and radiated by an on-chip ring antenna after power-boosted by a power
amplifier. On the RX side, a reverse process is carried out for demodulation. The
received signal is first amplified by an amplifier (the same kind as in the transmitter)
then split into I and Q channels. The quadrature LO signal, generated by the same
procedure as in the TX, drives the I-Q mixers to demodulate the signals on the I and
Q channels. The TX showed a 3-dB RF bandwidth of 28 GHz and an output power
(Psat) of 8.3 dBm, while the RX exhibited a conversion gain of 24 dB, which were
greatly helped by the power amplifier employed as previously mentioned. A max-
imum data rate of 90 Gbps was achieved over a 1-m data link with 32QAM. A
similar structure was reported later on [84], in which the front-end amplifier was
6.4 THz Radars 331
removed from the RX to avoid the imbalance between the upper-side band (USB)
and lower-side band (LSB) that may arise from the asymmetric gain profile. Without
the RX amplifier, in fact, a higher data rate (100 Gbps) was achieved despite the
degraded receiver noise figure and conversion gain, which was helped by the
symmetric response. There have been other results that have successfully demon-
strated high-speed data links based on quadrature-modulation transmitters/receivers
beyond 200 GHz in Si CMOS [93–95], SiGe HBTs [96], GaAs mHEMTs [70, 78],
as well as in InP HBTs [91]. While the overview in this section has been focused on
the electronics-based approach, similar systems that rely on carrier signals generated
by optical means have also been reported exhibiting excellent data link performances
[97–99].
As a final note, a brief review on the units for the data rate can be made here. The
readers may have noticed in some reports that the data rate is quoted in terms of
“baud” instead of (or together with) bit per second (bps). Baud rate is basically the
rate of modulation (or the change rate of states), indicating the number of symbols
transferred per second, different from bps that represents the number of bits trans-
ferred per second. Hence, the relation between baud and bps is simple but tricky as it
depends on the modulation order. This is because the number of bits included in a
single symbol varies over the different modulation orders. Here is a simple conver-
sion rule between the two units: To obtain the data rate in bps from a given baud rate,
simply multiply a factor of N to the baud rate, where 2N is the modulation order of the
modulation scheme. For example, 1 Gbaud corresponds to 2 Gbps with QPSK, for
which the modulation order is 4 with N ¼ 2. More examples are presented in
Table 6.1 with various modulation schemes.
Radar (RAdio Detection And Ranging) is a system that determines the range
(distance) and velocity of a target object based on the detection of the reflected
radio waves from the object. The modern development of radars has allowed
332 6 THz Applications
imaging of the target objects by radar operation as well. The history of radars can be
traced back to the early twentieth century when they attempted to detect the presence
of a target object based on the reflection of radio waves, but the major development
of radar systems was made in the time period around World War II, which clearly
indicates the early needs driven by military purposes. This legacy continues till
today, as the main application area of radars is still heavily linked to military
applications. However, the commercial application of radars is growing fast. The
weather forecast of today is heavily dependent on radars, and the airport traffic
control radars are a critical part in modern airport operation. Major airplanes and
ships are equipped with radars, and speed guns based on radars are measuring car
speeds and ball speeds on the roads and ballparks. More importantly, the market for
automotive radars is growing rapidly, as indicated by recent mid-size cars rolling out
of the factories with radars installed, which used to happen only for limited luxury
cars several years ago. There are growing efforts to expand the application envelope
of radars into various commercial sectors including industrial, medical, and security-
related areas. Along with this trend, also increasing is the interest in radars operating
in extended frequencies, including THz radars.
There are many ways to categorize radars depending on various aspects of radar
operation. Most of all, they can be categorized into pulse radars and CW radars,
depending on the waveform of the radar signals. Pulse radars can be divided into
radars with short pulses and radars with pulse compression, while CW radars can be
grouped into unmodulated and modulated radars. They can be further sub-divided
based on the details of the waveform and modulation. We can define other types of
radars based on different features as well. There can be military and commercial
radars, as well as ground-based, ship-based, air-borne, and space-borne radars. They
can be grouped by operation frequency and the extent of the detection ranges, too.
There are radars with special features, such as imaging radars and synthetic aperture
radars (SARs). In this section, the discussion will be made in terms of pulse radars
and CW radars, which will be covered in the following two subsections.
Before embarking on the subsections for a separate treatment of pulse and CW
radars, it would be helpful to review the common basic principles of radars. A key
equation that governs the radar operation is the radar range equation. A brief
derivation to arrive at the equation will provide a useful snapshot over the radar
operation. Let us first consider a configuration depicted in Fig. 6.39. Electromagnetic
waves emitted from the antenna of a radar arrive at the target and are reflected back
to the antenna and then detected. The incident power density at the target can be
obtained as Eq. (4.55), which is repeated here for convenience:
Gt Pt
Starget ¼ , ð6:25Þ
4πR2
where Gt is the transmitter antenna gain at the radar, Pt is the power emitted from the
antenna, and R is the range or distance between the antenna and the target. The target
will have a finite dimension, and the intercepted power by the target will be
obviously given by the incident power density multiplied by the effective target
area, or:
Gt Pt
Ptarget ¼ σStarget ¼ σ , ð6:26Þ
4πR2
where σ is called the radar cross section (RCS), which can be roughly understood as
the effective surface area of the target, but is better defined as a measure of how
effectively the target reflects the waves back to the radar in that direction. The
incident power will be reflected and scattered by the target surface upon its arrival.
In fact, this scattering event can be considered as a re-radiation of the reflected power
by the target. According to this view, the illuminated target serves as an antenna,
which is assumed to be isotropic. Hence, the scattered power will propagate toward
all the directions, only a portion of which will travel back to the radar. The power
density arriving at the receiver antenna of the radar is given by:
Ptarget σG P
Sreceiver ¼ 2
¼ t t 2 : ð6:27Þ
4πR 4πR2
Hence, the power received by the antenna can be expressed in terms of the effective
aperture Ae such that:
σGt Pt λ2 σGr Gt
Preceiver ¼ Ae Sreceiver ¼ Ae 2
¼ Pt , ð6:28Þ
4πR2 ð4π Þ3 R4
where the relation for the effective antenna aperture of Ae ¼ Grλ2/4π is used
(Eq. 4.54). Here, Gr is the gain of the receiver antenna at the radar and λ is, of
course, the wavelength of the radar signal. Equation (6.28) is a key relation between
the transmitted and received powers at the radar after the round trip to the target. It
shows that the returned power is inversely proportional to R4, rapidly decreasing
with the increasing distance. This can be compared to the point-to-point FSPL that is
inversely proportional to R2.
Now, let us obtain the output SNR of the radar system. Note that the power
obtained in Eq. (6.28) is the input power that enters the radar system. From the
definition of the noise factor F as given in Eq. (3.10), the output SNR can be
expressed as:
334 6 THz Applications
1 1 Sin
SNRout ¼ SNRin ¼ : ð6:29Þ
F F N in
Here, Sin is the input signal power and simply given by Preceiver obtained in
Eq. (6.28). Nin is the input noise power and can be expressed as follows:
N in ¼ kT o B, ð6:30Þ
where k is Boltzmann constant, To is the reference temperature (290 K), and B is the
system bandwidth. Hence, SNR at the radar output (SNRout) is obtained by substitut-
ing Eqs. (6.28) and (6.30) into Eq. (6.29), leading to:
λ2 σGr Gt Pt
SNRout ¼ : ð6:31Þ
ð4π Þ3 kT 0 BFR4
If needed, the loss factor L can be included in denominator of Eq. (6.31), which is
omitted here. Now, the maximum detectable range is given as the distance that
corresponds to the minimum output SNR required for the radar systems [100], which
becomes,
!1=4
λ2 σGr Gt Pt
Rmax ¼ 3
: ð6:32Þ
ð4π Þ kT 0 BF SNRout, min
This relation is widely referred to as the radar range equation. A typical number for
SNRout,min in radar systems is 13 dB [101], while it can be as large as 20 dB [102].
Let us examine the maximum range given in Eq. (6.32) from the THz application
point of view. One parameter that is directly affected by the operation frequency is,
of course, the wavelength λ. The maximum detectible range decreases with decreas-
ing wavelength, or increasing frequency, which is not a favored trend for THz
application. Other parameters affected by the frequency is the transmission power
Pt at the transmitter and the noise factor F at the receiver. Both of them generally
degrade with increasing frequency, which makes the range of THz radars further
limited. In fact, for these reasons, the operation range of typical radars has been
mainly determined by the operation frequency. Starting from the long-range radars
(hundreds of km) based on L-band (1–2 GHz) and the medium-range radars (tens of
km) based on S-band (2–4 GHz), the operation range generally reduces with
increasing frequency. From this viewpoint, we can expect that the THz radars will
best fit for the very short-range applications, as was the case for THz communica-
tion. On a promising side, though, the range and image resolutions will be improved
with increasing frequency, in a similar way for THz imaging, which would be the
main motivation for THz radars. The issue of resolution in radars will be discussed
later in this section.
The topic of radar comprises not only the hardware but also lots of signal
processing issues. However, the discussion below will be mostly confined to the
hardware side, with a main focus on its application based on the THz band.
6.4 THz Radars 335
In pulse radars, a train of pulses are transmitted from the antenna and the reflected
signals (called “echo”) are received, from which the range and the velocity of the
target is determined. In typical cases for pulse radars, the same antenna is used for
both emission and detection, time-controlled by a duplexer, which is a main feature
of the monostatic radars. Let us first obtain the range of the target that can be
obtained with pulse radars. We can start with the simplest case, where a single pulse
(with a pulse width of τ) is transmitted and then received after reflection as is shown
in Fig. 6.40a. Then, the range is given by:
cΔt
R¼ , ð6:33Þ
2
where Δt is the time delay of the pulse and c is the speed of light. Hence, this is
basically a time-of-flight (ToF) technique as mentioned earlier in this chapter. Now
consider a more practical case with a train of multiple pulses transmitted and
received, as described by Fig. 6.40b. A couple of parameters can be defined for
this multiple-pulse situation. The pulse repetition interval (PRI), denoted as T in the
figure, indicates the time interval between the pulses. The pulse repetition frequency
(PRF), or fr, is simply the inverse of PRI, representing the frequency of the pulse
train. The readers, however, should be aware that a pulse is basically a switched CW,
which has its own frequency that needs to be distinguished from PRF. In the given
Received
t
2R
Δt =
c
(a)
t
Δt
Δt
Δt + T
Δt + 2T
(b)
336 6 THz Applications
configuration with a train of pulses, the range appears to be determined by Eq. (6.33)
again. However, this will be true only when the echo 1 is the response of pulse
1. There is a possibility that echo 2 is the response of pulse 1 for a longer range, in
which case R is given as c(Δt + T)/2. Likewise, there is a chance that echo 3 is the
response of pulse 1, leading to R ¼ c(Δt + 2 T )/2. Hence, there exists a range
ambiguity of cT/2, which can be avoided by allowing enough time between adjacent
pulses or increasing PRI.
Next, the range resolution is discussed. For this, consider the case given in
Fig. 6.41, where two targets are located at different locations, R1 and R2, along the
direction of signal transmission. If the separation between the two targets is not large
enough (see Fig. 6.41a), the two echo signals from target 1 and target 2 will partially
overlap and cannot be separated, depending on the pulse width. For the two targets to
be resolved (see Fig. 6.41b), the two echo signals need to be separated, which is
satisfied by the following condition:
2R2 2R1 2
¼ ðR2 R1 Þ > τ: ð6:34Þ
c c c
cτ c
ΔR ¼ ¼ , ð6:35Þ
2 2B
where B is the radar bandwidth which is defined as the inverse of the pulse width τ
(Note that B is not the RF bandwidth of the radar receiver included in the radar range
equation, Eq. (6.32)). Hence, to improve the resolution, τ needs to be small.
However, a small τ will lead to a reduced received power, degrading SNR of the
radar system. Therefore, there arises a trade-off between the resolution and SNR.
Fortunately, there is a technical solution that resolves this trade-off: modulating the
frequency or the phase of the pulses. Officially known as the pulse compression, this
technique will allow improved resolution together with a high SNR when combined
with the matched filtering. The details of the pulse compression and the matched
filtering can be found in standard radar textbooks (e.g., [100]) and will not be
discussed here further. Just one typical example of the pulse modulation will be
introduced with the help of Fig. 6.42. Called the linear frequency modulation (LFM),
the frequency of the CW wave within one pulse cycle is linearly increased. With this
frequency sweep from f1 to f2, the pulse width will be effectively reduced by a factor
of ( f2 f1)τ after matched filtering, improving the range resolution while
maintaining the receiver signal power. Apparently, a larger reduction factor can be
obtained with higher frequencies (because f2 f1 will be larger), which is an
attractive aspect for THz radars.
Another parameter to be acquired by pulse radars is the target velocity. The
principle behind the velocity acquisition is the Doppler effect, with which the readers
should be familiar: When the wave source is moving toward the observer, the
6.4 THz Radars 337
Fig. 6.41 Transmitted and received pulses of a pulse radar with two targets: (a) Small separation
between two targets, (b) Large separation between two targets
frequency of the wave increases, while the frequency decreases when the wave
source is moving away from the observer. It applies to the reflected waves as well, a
relevant situation for radars: When the wave reflector (target) is moving toward or
away from the wave source, the frequency increases or decreases, respectively.
Hence, the main equations for the Doppler effect apply to reflected waves as well.
From the radar perspective, when the target is moving toward the radar with a
velocity of v (in a rigorous sense, it is speed, not velocity, as it is considered as a
338 6 THz Applications
Pulse envelop
Pulse waveform
f f2
Frequency f1
t
Fig. 6.42 Pulse envelop, pulse waveform, and the frequency profile of the linear frequency
modulation (LFM)
scalar parameter in the range analysis; however, in the discussion throughout this
subsection, we will maintain the term “velocity” to follow the convention), the new
frequency f0 is related to the original frequency f with:
cþv
f0 ¼ f : target moving toward radar, ð6:36Þ
cv
where c is the speed of light. Likewise, when the target is moving away from the
radar, the relation becomes:
cv
f0 ¼ f : target moving away from radar: ð6:37Þ
cþv
Now let us express the relations in terms of the pulse width instead of the frequency.
If the frequency of the CW wave inside a pulse increases (decreases), the pulse width
will decrease (increase) by the same factor, when we assume that the number of
wave cycles inside a pulse remains unchanged, which is certainly true. In other
words, f0 /f ¼ τ/τ0 , where τ0 is the new pulse width. Hence, the following relations
hold for pulse radars:
cv
τ0 ¼ τ: target moving toward radar, ð6:38Þ
cþv
cþv
τ0 ¼ τ: target moving away from radar: ð6:39Þ
cv
Therefore, by detecting the ratio of τ0 /τ, the target velocity can be obtained. Here, it
was assumed that the target is moving along the radial direction. When the target is
6.4 THz Radars 339
moving in the direction off the radial direction by an angle θ, v in Eqs. (6.38) and
(6.39) needs to be replaced by v cos θ, or the radial component of v.
From the implementation point of view, the challenges for pulse radars comes
from the need for the fast and high-isolation switches, which are increasingly
difficult for higher frequencies, including the THz band. The heavy load on the
baseband circuit with the large bandwidth also leads to design challenges for the
baseband design. The limited transmitter output power and receiver noise figure in
high-frequency operation will reduce the maximum detectable range as mentioned
above, which is in fact a general issue that affects not only radars but also any high-
frequency transmitters and receivers.
Despite the challenges, there have been recent notable reports on pulse radars
operating beyond 100 GHz. In [102], a 160-GHz pulse radar implemented in 65-nm
Si CMOS technology was reported. The radar employs a 2 2 phased-array
transceiver architecture to increase the maximum detectable range with improved
directivity. The output signal was generated with a 40-GHz PLL, which was
subsequently followed by phase shifters, 4 frequency multiplier, 160-GHz injec-
tion locked oscillator, and a PA. The generated signal was merged with a pulse signal
inside the PA, leading to a pulse output. With a pulse width of 100 ps, a range
resolution of 1.5 cm was expected with a bandwidth of 20 GHz, while the pulse
repetition frequency of 100 MHz led to a range ambiguity of 1.5 m. With the
measurement, the transmitter exhibited an EIRP of 18.8 dBm for continuous wave
output, while the receiver showed a gain of 42.5 dB and noise figure of 22.5 dB with
a 7-GHz bandwidth, with a total DC power dissipation of 2.2 W.
Based on a similar but slightly different approach than pulse radars are the phase
radars that determine the target range by detecting the delay of the phase (instead of
pulse edges) in CW waves. This is the basis for the 144-GHz Si CMOS phase radar
reported in [103], which was used to acquire 3D images for raster-scanned targets
with a depth resolution of 0.76 cm. To deal with the range ambiguity, which is in the
order of the wavelength (much smaller than the case of regular pulse radars) and thus
a more serious issue with higher frequencies for phase radars, a successive-approx-
imation technique was employed in this work.
6.4.3 CW Radars
In CW radars, a sinusoidal CW signal is transmitted from the antenna and the signal
reflected from the target is received. As for the antenna configuration of CW radars,
the transmitter and receiver antennas are usually separated as they continuously
transmit and receive signals, differently from pulse radars. Hence, it is technically a
bistatic configuration as the two antennas are separate, but it is often considered
monostatic especially when the two antennas are integrated on a single chip. Or,
sometimes, it is considered quasi-monostatic since the distance between the two
antennas are in the order of a wavelength [104]. Still, there are cases with FMCW
radars employing a single antenna for both transmission and reception of the signal,
340 6 THz Applications
f0 f f
v¼c c D, ð6:40Þ
f0 þ f 2f
f b Δf
¼ : ð6:41Þ
Δt 2 1f
m
By substituting the relation Δt ¼ 2R/c into Eq. (6.41), the range can be expressed in
terms of the three frequency parameters and the speed of light c, so that:
cf b
R¼ : ð6:42Þ
4 f m Δf
Now let us consider the range resolution of FMCW radars. For this, assume that
two targets are located at ranges R1 and R2 (R2 > R1), which lead to beat frequencies
of fb1 and fb2, respectively, at the radar receiver. For R1 and R2 to be resolved in the
spatial domain, fb1 and fb2 need to be resolved in the frequency domain. If the two
echo signals from the two targets are arriving at the radar receiver simultaneously, it
takes a certain time interval to resolve the two signal profiles centered at fb1 and fb2 in
the frequency domain when they are obtained by the Fourier transform [107]. The
rule of thumb is that a frequency resolution of 1/T can be obtained by an observation
time interval of T. For FMCW radars, the maximum allowed observation time is the
maximum linear stretch of the frequency modulation profile in the time domain, or
the full ramp-up (or ramp-down) time, which corresponds to 1/2fm. This means that
the best frequency resolution available in the FMCW modulation profile given in
Fig. 6.43 is 2fm. Hence, for the two beating signals at fb1 and fb2 to be resolved, it
follows that:
f 1/f m
fb
Δf Transmitted
Received
Δt
t
(a)
f
fb,up fD
fb,down
Δt
t
(b)
Fig. 6.43 Transmitted and received signal of a FMCW radar: (a) Stationary target, (b) Target is
moving toward the radar
342 6 THz Applications
Δ f b ¼ f b2 f b1 > 2 f m : ð6:43Þ
With the relation between fb and R given in Eq. (6.42), Eq. (6.43) becomes:
4 f m Δf 4 f Δf
Δfb ¼ ðR2 R1 Þ ¼ m ΔR > 2 f m , ð6:44Þ
c c
c
R2 R1 ¼ ΔR > : ð6:45Þ
2Δf
c c
ΔR ¼ ¼ : ð6:46Þ
2Δf 2B
The readers will notice that this expression is identical to the range resolution
expression for pulse radars as shown in Eq. (6.35). In fact, this expression is valid for
any radar signals regardless the modulation type [108], although the exact definition
of B varies over modulation and radar types (for instance, B ¼ Δf for FMCW, B ¼ 1/
τ for pulse radars, as mentioned above). Equation (6.46) clearly indicates that a high
carrier frequency is desired for enhanced resolution, since a large Δf can be more
conveniently supported by a higher carrier frequency with a wide RF bandwidth
available. Note that higher carrier frequency will be favored for cross-range (normal
to the radial direction) resolution as well because it is limited by the diffraction,
which reduces with increasing frequency. This frequency dependence of the resolu-
tions that favor the high frequencies is the major opportunity for THz radars. While
Δf is typically in the order of hundreds of MHz for automobile radars operating near
77 GHz, it can be extended to be as large as tens of GHz for radars working in the
THz band as we will see shortly. In fact, for sub-cm range resolution, Δf larger than
15 GHz is needed according to Eq. (6.46), which can be reasonably accomplished
with operation frequencies above 100 GHz.
Now, let us consider a more general case, where the target is moving toward the
radar with a velocity v. In this case, the frequency profile of the received signal is
shifted up in the frequency axis due to the Doppler effect, in addition to the shift in
the time frame due to the time delay, which is depicted in Fig. 6.43b. In this case, the
beat frequency is split into two cases, fb,up and fb,down, which correspond to the beat
frequency for ramp-up and -down cycles, respectively. For the ramp-up cycle, the
beat frequency is reduced by the Doppler frequency fD, while the beat frequency is
increased by fD during the ramp-down cycle, with reference to the case with
stationary targets. One can easily show that the expression for the range in
Eq. (6.42) is modified as:
6.4 THz Radars 343
c f b,down þ f b,up c
R¼ ¼ f b,down þ f b,up , ð6:47Þ
4 f m Δf 2 8 f m Δf
which indicates that the averaged beat frequency, or ( fb,down + fb,up)/2, replaces fb of
the stationary target. What about the target velocity v? As mentioned above, the two
beat frequencies are shifted by fD in the opposite directions due to the Doppler effect,
so that:
c λ
v¼ f b,down f b,up ¼ f f b,up , ð6:49Þ
4f 4 b,down
where f and λ are the frequency and the wavelength of the transmitted signal. From a
rigorous point of view, f and λ are not constants and periodically change due to the
modulation, but the values at the center frequency can be taken in this case. It is
noted again that v obtained with Eq. (6.49) is the radial component of the velocity, or
the radial velocity. As mentioned earlier, frequency difference can be readily
obtained by a mixer, and thus fb,down and fb,up can be extracted by mixing the
transmitted and received signals during the ramp-up and -down cycles, respectively,
of the frequency chirping. Equation (6.49) is valid for targets moving away from the
radars as well, in which case v will assume negative values.
A typical implementation example of FMCW is shown in Fig. 6.44. The linearly
modulated signal, which can be achieved by a PLL that employs a direct digital
synthesizer (DDS) as a varied-frequency reference signal, is amplified by a PA and
transmitted by an antenna. There are other ways to generate modulated signals, such
as a fractional-N PLL in which the divide modulus of the frequency divider is swept
[109], a fixed-frequency PLL whose output is mixed with a varied-frequency VCO
[110], or a direct frequency multiplication of a DDS output [108]. On the receiver
channel shown in Fig. 6.44, the received signal from the antenna enters a mixer,
optionally after an LNA, and then mixed with a portion of the modulated signal that
is taken from the transmitter channel by a power divider or coupler. Optionally, a
quadrature topology based on I-Q mixers can be adopted for the mixing, which will
provide additional phase information of the echo signal. It will also help to reduce
noise at the receiver as the double-side band (DSB) noise will be effective instead of
the single-side band (SSB) noise, leading to an NF reduction up to 3 dB [111]. The
IF signal generated by the mixer is in fact the beat signal between the transmitted and
received signals and is used to obtain the range and velocity of the target based on the
relations provided above. The mixer output will also include the up-converted
signal, which will be filtered out with a low-pass filter (LPF) that is placed next to
the mixer. The beat signal amplified by an IF amplifier is accepted by the baseband
block for various signal processing steps to acquire the desired information and
present the data in the proper formats. Note that the structure of unmodulated CW
radars will be similar, except for the absence of the modulating components as well
as the less heavy BB block.
As mentioned earlier, FMCW radars benefit from the high operation frequency
for improved range resolution, for which THz radars are an attractive option. Yet, it
is still challenging to have a large voltage tuning at the THz band, which may
partially limit the efforts to attain a wide modulation bandwidth (Δf ). At the same
time, the components in the system, including the amplifiers, mixers, as well as
on-chip antennas, if employed, should have RF bandwidths large enough to cover
the modulation bandwidth. Another issue to be considered for FMCW radars,
especially when monolithically implemented on a single chip, is the signal leakage
from the transmitter to the receiver channels. As the leaked signal will have the same
frequency with the LO frequency for the receiver, it will cause a DC component in
the IF signal, which needs to be filtered out with a DC block capacitance. The
linearity in the frequency modulation is also important because any nonlinearity will
cause a spread in the demodulated frequency, leading to degradation in the range
resolution. One relief is that the nonlinearity can be compensated in the baseband
signal-processing steps.
Recently, a growing number of THz radars beyond 100 GHz are being reported,
majority of them being FMCW radars [105, 106, 108, 110, 112–114]. One recent
example is briefly introduced below, which reports a 220-GHz FMCW radar based
on a 55-nm SiGe BiCMOS technology [113]. The basic block diagram of the radar
and the chip photo are shown in Fig. 6.45a, b, respectively. For modulated ramp
signal generation at the desired frequency band, a 220-GHz push-push VCO with a
wide tuning range was employed, which was driven by a ramp generator that sweeps
the tuning voltage of the VCO with a rate fm ¼ 20 kHz. The generated signal is
transmitted by an on-chip slot antenna. For the receiver chain, the received signal
detected by an on-chip folded-dipole antenna is mixed with an LO signal coupled
from the VCO in the transmitter chain, then filtered and amplified. With a modula-
tion bandwidth of 62.4 GHz around the center frequency of 221 GHz, the radar
exhibited a range resolution of 2.7 mm, close to the theoretical value of 2.4 mm as
expected by Eq. (6.46). The lateral cross-range resolution was 2 mm. The transmitter
References 345
Fig. 6.45 A 220-GHz SiGe BiCMOS FMCW imaging radar: (a) Block diagram, (b) Photo of the
fabricated chip, (c) Example images acquired by the radar. (© 2018 IEEE [113])
exhibited an output power of 4.6 dBm and EIRP of 14 dBm with the help of a Si
lens. Total DC power dissipation was 87 mW. With the fabricated radar, 3D images
were successfully acquired based on the inverse synthetic aperture radar (ISAR)
technique, one example shown in Fig. 6.45c.
References
1. E. Bründermann, H.-W. Hübers, M.F. Kimmitt, Terahertz Techniques (Springer, Berlin, 2012)
2. R. Wood, The echelette grating for the infra-red. Philos. Mag. J. Sci. 20(118), 770–778 (1910)
3. P. Connes, Early history of Fourier transform spectroscopy. Infrared Phys. 24(2–3), 69–93
(1984)
4. D. Auston, K. Cheung, P. Smith, Picosecond photoconducting Hertzian dipoles. Appl. Phys.
Lett. 45(3), 284–286 (1984)
5. C. Fattinger, D. Grischkowsky, Point source terahertz optics. Appl. Phys. Lett. 53(16),
1480–1482 (1988)
6. J. Neu, C.A. Schmuttenmaer, Tutorial: an introduction to terahertz time domain spectroscopy
(THz-TDS). J. Appl. Phys. 124(23), 231101 (2018)
7. P. Han, M. Tani, M. Usami, S. Kono, R. Kersting, X.-C. Zhang, A direct comparison between
terahertz time-domain spectroscopy and far-infrared Fourier transform spectroscopy. J. Appl.
Phys. 89(4), 2357–2359 (2001)
346 6 THz Applications
28. J. Yun et al., Terahertz reflection-mode biological imaging based on InP HBT source and
detector. IEEE Trans. Terahertz Sci. Technol. 7(3), 274–283 (2017). https://doi.org/10.1109/
TTHZ.2017.2673549
29. J. Yun, D. Yoon, H. Kim, J.-S. Rieh, 300-GHz InP HBT oscillators based on common-base
cross-coupled topology. IEEE Trans. Microw. Theory Tech. 62(12), 3053–3064 (2014).
https://doi.org/10.1109/TMTT.2014.2364608
30. N. Skou, D. Le Vine, Microwave Radiometer Systems: Design and Analysis, 2nd edn. (Artech
House, 2006)
31. M. Tiuri, Radio astronomy receivers. IEEE Trans. Mil. Electron. 8(3), 264–272 (1964)
32. J.W. May, G.M. Rebeiz, Design and characterization of W-band SiGe RFICs for passive
millimeter-wave imaging. IEEE Trans. Microw. Theory Tech. 58(5), 1420–1430 (2010)
33. J.W. Kooi et al., Submillimeter InP MMIC low-noise amplifier gain stability characterization.
IEEE Trans. Terahertz Sci. Technol. 7(3), 335–346 (2017)
34. V. Ossenkopf, The stability of spectroscopic instruments: a unified Allan variance computa-
tion scheme. Astron. Astrophys. 479(3), 915–926 (2008)
35. R.H. Dicke, The measurement of thermal radiation at microwave frequencies. Rev. Sci. Instru.
17(7), 268–275 (1946)
36. A. Tomkins, P. Garcia, S.P. Voinigescu, A passive W-band imaging receiver in 65-nm bulk
CMOS. IEEE J. Solid-State Circuits 45(10), 1981–1991 (2010)
37. L. Zhou, C.-C. Wang, Z. Chen, P. Heydari, A W-band CMOS receiver chipset for millimeter-
wave radiometer systems. IEEE J. Solid-State Circuits 46(2), 378–391 (2011)
38. L. Gilreath, V. Jain, P. Heydari, Design and analysis of a W-band SiGe direct-detection-based
passive imaging receiver. IEEE J. Solid-State Circuits 46(10), 2240–2252 (2011)
39. Q.J. Gu, Z. Xu, A. Tang, M.-C.F. Chang, A D-band passive imager in 65 nm CMOS. IEEE
Microw. Wirel. Compon. Lett. 22(5), 263–265 (2012)
40. Z. Chen, C.-C. Wang, H.-C. Yao, P. Heydari, A BiCMOS W-band 2 2 focal-plane array with
on-chip antenna. IEEE J. Solid-State Circuits 47(10), 2355–2371 (2012)
41. E. Dacquay et al., D-band total power radiometer performance optimization in an SiGe HBT
technology. IEEE Trans. Microw. Theory Tech. 60(3), 813–826 (2012)
42. T. Kanar, G.M. Rebeiz, A low-power 136-GHz SiGe total power radiometer with NETD of
0.25 K. IEEE Trans. Microw. Theory Tech. 64(3), 906–914 (2016)
43. A. Tang, Y. Kim, Q.J. Gu, 25.4 A 0.43 K-noise-equivalent-ΔT 100GHz dicke-free radiometer
with 100% time efficiency in 65nm CMOS, in 2016 IEEE International Solid-State Circuits
Conference (ISSCC), (IEEE, 2016), pp. 430–431
44. A. Tang et al., A W-band 65nm CMOS/InP-hybrid radiometer & passive imager, in 2016
IEEE MTT-S International Microwave Symposium (IMS), (IEEE, 2016), pp. 1–3
45. P. Hillger et al., A 128-pixel 0.56 THz sensing array for real-time near-field imaging in 0.13
μm SiGe BiCMOS, in 2018 IEEE International Solid-State Circuits Conference-(ISSCC),
(IEEE, 2018), pp. 418–420
46. J.F. Federici et al., THz imaging and sensing for security applications—explosives, weapons
and drugs. Semicond. Sci. Technol. 20(7), S266 (2005)
47. D.L. Donoho, Compressed sensing. IEEE Trans. Inform. Theory 52(4), 1289–1306 (2006).
https://doi.org/10.1109/TIT.2006.871582
48. H. Sherry et al., A 1kpixel CMOS camera chip for 25fps real-time terahertz imaging applica-
tions, in 2012 IEEE International Solid-State Circuits Conference, (IEEE, 2012), pp. 252–254
49. R. Al Hadi et al., A 1 k-pixel video camera for 0.7–1.1 terahertz imaging applications in 65-nm
CMOS. IEEE J. Solid-State Circuits 47(12), 2999–3012 (2012)
50. K. Song, Development of THz Imaging Arrays Based on Si CMOS Technology, Ph.D.
Dissertation, Korea University, 2020
51. J.P. Guillet et al., Review of terahertz tomography techniques. J. Infrared Millim. Terahertz
Waves 35(4), 382–411 (2014)
52. B. Ferguson, S. Wang, D. Gray, D. Abbot, X.-C. Zhang, T-ray computed tomography. Opt.
Lett. 27(15), 1312–1314 (2002)
348 6 THz Applications
53. S. Wang, B. Ferguson, D. Abbott, X.-C. Zhang, T-ray imaging and tomography. J. Biol. Phys.
29(2–3), 247–256 (2003)
54. R. Gordon, A tutorial on ART (algebraic reconstruction techniques). IEEE Trans. Nucl. Sci. 21
(3), 78–93 (1974)
55. L.A. Shepp, Y. Vardi, Maximum likelihood reconstruction for emission tomography. IEEE
Trans. Med. Imaging 1(2), 113–122 (1982)
56. B. Recur et al., Investigation on reconstruction methods applied to 3D terahertz computed
tomography. Opt. Express 19(6), 5105–5117 (2011)
57. B. Ferguson, S. Wang, D. Gray, D. Abbott, X. Zhang, T-ray diffraction tomography, in The
Thirteenth International Conference on Ultrafast Phenomena, vol. 72, (Citeseer, 2002),
pp. 450–451
58. J. Takayanagi et al., High-resolution time-of-flight terahertz tomography using a femtosecond
fiber laser. Opt. Express 17(9), 7533–7539 (2009)
59. U. Pfeiffer, Sub-millimeter wave active imaging with silicon integrated circuits, in 2011
International Conference on Infrared, Millimeter, and Terahertz Waves, (IEEE, 2011),
pp. 1–4
60. P. Hillger, A. Schlüter, R. Jain, S. Malz, J. Grzyb, U. Pfeiffer, Low-cost 0.5 THz computed
tomography based on silicon components, in 2017 42nd International Conference on Infrared,
Millimeter, and Terahertz Waves (IRMMW-THz), (IEEE, 2017), pp. 1–2
61. D. Yoon et al., 3-D THz tomography with an InP HBT signal source and a SiGe HBT imaging
receiver operating near 300 GHz, in 40th International Conference on Infrared, Millimeter,
and Terahertz Waves (IRMMW-THz), (2015), pp. 1–2. https://doi.org/10.1109/IRMMW-THz.
2015.7327606
62. J. Kim et al., Three-dimensional terahertz tomography with transistor-based signal source and
detector circuits operating near 300 GHz. IEEE Trans. Terahertz Sci. Technol. 8(5), 482–491
(2018). https://doi.org/10.1109/TTHZ.2018.2851542
63. C.E. Shannon, Communication in the presence of noise. Proc. IRE 37(1), 10–21 (1949)
64. T. Kleine-Ostmann, T. Nagatsuma, A review on terahertz communications
research. J. Infrared Millim. Terahertz Waves 32(2), 143–171 (2011)
65. H.-J. Song, T. Nagatsuma, Present and future of terahertz communications. IEEE Trans.
Terahertz Sci. Technol. 1(1), 256–263 (2011)
66. S. Koenig et al., Wireless sub-THz communication system with high data rate. Nat. Photonics
7(12), 977 (2013)
67. ITU, RECOMMENDATION ITU-R P.676–9—Attenuation by atmospheric gases, 2012
68. J.-D. Park, S. Kang, S.V. Thyagarajan, E. Alon, A.M. Niknejad, A 260 GHz fully integrated
CMOS transceiver for wireless chip-to-chip communication, in 2012 Symposium on VLSI
Circuits (VLSIC), (IEEE, 2012), pp. 48–49
69. H.-J. Song et al., Demonstration of 20-Gbps wireless data transmission at 300 GHz for KIOSK
instant data downloading applications with InP MMICs, in 2016 IEEE MTT-S International
Microwave Symposium (IMS), (IEEE, 2016), pp. 1–4
70. I. Kallfass et al., 64 Gbit/s transmission over 850 m fixed wireless link at 240 GHz carrier
frequency. J. Infrared Millim. Terahertz Waves 36(2), 221–233 (2015)
71. R. Piesiewicz, C. Jansen, D. Mittleman, T. Kleine-Ostmann, M. Koch, T. Kurner, Scattering
analysis for the modeling of THz communication systems. IEEE Trans. Antennas Propag. 55
(11), 3002–3009 (2007)
72. C. Jansen, R. Piesiewicz, D. Mittleman, T. Kurner, M. Koch, The impact of reflections from
stratified building materials on the wave propagation in future indoor terahertz communication
systems. IEEE Trans. Antennas Propag. 56(5), 1413–1419 (2008)
73. C. Han, A.O. Bicen, I.F. Akyildiz, Multi-ray channel modeling and wideband characterization
for wireless communications in the terahertz band. IEEE Trans. Wirel. Commun. 14(5),
2402–2412 (2014)
References 349
74. IEEE, IEEE standard for high data rate wireless multi-media networks—Amendment 2:
100 Gb/s wireless switched point-to-point physical layer, in IEEE Std 802.15.3d-2017,
(2017), pp. 1–55. https://doi.org/10.1109/IEEESTD.2017.8066476
75. J.G. Proakis, M. Salehi, Digital Communications, 5th edn. (McGraw-Hill, New York, 2008)
76. C.W. Byeon, C.S. Park, A high-efficiency 60-GHz CMOS transmitter for short-range wireless
communications. IEEE Microw. Wirel. Compon. Lett. 27(8), 751–753 (2017)
77. S. Kim et al., 300 GHz integrated heterodyne receiver and transmitter with on-chip funda-
mental local oscillator and mixers. IEEE Trans. Terahertz Sci. Technol. 5(1), 92–101 (2015).
https://doi.org/10.1109/TTHZ.2014.2364454
78. I. Kallfass et al., All active MMIC-based wireless communication at 220 GHz. IEEE Trans.
Terahertz Sci. Technol. 1(2), 477–487 (2011)
79. M. Abbasi et al., Single-chip 220-GHz active heterodyne receiver and transmitter MMICs with
on-chip integrated antenna. IEEE Trans. Microw. Theory Tech. 59(2), 466–478 (2010)
80. H.-J. Song, J.-Y. Kim, K. Ajito, M. Yaita, N. Kukutsu, Fully integrated ASK receiver MMIC
for terahertz communications at 300 GHz. IEEE Trans. Terahertz Sci. Technol. 3(4), 445–452
(2013)
81. H.J. Park, J.-S. Rieh, M. Kim, J.B. Hacker, 300 GHz six-stage differential-mode amplifier, in
IEEE MTT-S International Microwave Symposium Digest, (2010), pp. 49–52
82. Z. Wang, P.-Y. Chiang, P. Nazari, C.-C. Wang, Z. Chen, P. Heydari, A CMOS 210-GHz
fundamental transceiver with OOK modulation. IEEE J. Solid-State Circuits 49(3), 564–580
(2014)
83. A. Tang, M.-C.F. Chang, A 294 GHz 0.47 mW caterpillar amplifier based transmitter in 65nm
CMOS for THz data-links, in 2013 IEEE 13th Topical Meeting on Silicon Monolithic
Integrated Circuits in RF Systems, (IEEE, 2013), pp. 27–29
84. P. Rodríguez-Vázquez, J. Grzyb, B. Heinemann, U.R. Pfeiffer, A 16-QAM 100-Gb/s 1-M
wireless link with an EVM of 17% at 230 GHz in an SiGe technology. IEEE Microw. Wirel.
Compon. Lett. 29(4), 297–299 (2019)
85. S. Lee et al., An 80Gb/s 300GHz-band single-chip CMOS transceiver, in 2019 IEEE Inter-
national Solid-State Circuits Conference—(ISSCC), (2019), pp. 170–172. https://doi.org/10.
1109/ISSCC.2019.8662314
86. J.-Y. Kim, H.-J. Song, K. Ajito, M. Yaita, N. Kukutsu, A 325 GHz quadrature voltage
controlled oscillator with superharmonic-coupling. IEEE Microw. Wirel. Compon. Lett. 23
(8), 430–432 (2013)
87. K. Song, J. Kim, H. Son, J. Yoo, M. Cho, J. Rieh, 300-GHz InP HBT quadrature VCO with
integrated mixer. IEEE Trans. Terahertz Sci. Technol. 10(3), 2020. https://doi.org/10.1109/
TTHZ.2020.2972539
88. K. Katayama et al., A 300GHz 40nm CMOS transmitter with 32-QAM 17.5 Gb/s/ch capability
over 6 channels, in 2016 IEEE International Solid-State Circuits Conference (ISSCC), (IEEE,
2016), pp. 342–343
89. K. Takano et al., A 105Gb/s 300GHz CMOS transmitter, in 2017 IEEE International Solid-
State Circuits Conference (ISSCC), (IEEE, 2017), pp. 308–309
90. S. Lee et al., An 80-Gb/s 300-GHz-band single-chip CMOS transceiver. IEEE J. Solid-State
Circuits 54(12), 3577–3588 (2019)
91. H.-J. Song, J.-Y. Kim, K. Ajito, N. Kukutsu, M. Yaita, 50-Gb/s direct conversion QPSK
modulator and demodulator MMICs for terahertz communications at 300 GHz. IEEE Trans.
Microw. Theory Tech. 62(3), 600–609 (2014)
92. P. Rodríguez-Vázquez, J. Grzyb, B. Heinemann, U.R. Pfeiffer, Performance evaluation of a
32-QAM 1-meter wireless link operating at 220–260 GHz with a data-rate of 90 Gbps, in 2018
Asia-Pacific Microwave Conference (APMC), (IEEE, 2018), pp. 723–725
93. Q. Zhong, Z. Chen, N. Sharma, S. Kshattry, W. Choi, K. Kenneth, 300-GHz CMOS QPSK
transmitter for 30-Gbps dielectric waveguide communication, in 2018 IEEE Custom Inte-
grated Circuits Conference (CICC), (IEEE, 2018), pp. 1–4
350 6 THz Applications
94. S. Kang, S.V. Thyagarajan, A.M. Niknejad, A 240 GHz fully integrated wideband QPSK
transmitter in 65 nm CMOS. IEEE J. Solid-State Circuits 50(10), 2256–2267 (2015)
95. S.V. Thyagarajan, S. Kang, A.M. Niknejad, A 240 GHz fully integrated wideband QPSK
receiver in 65 nm CMOS. IEEE J. Solid-State Circuits 50(10), 2268–2280 (2015)
96. D. Fritsche, P. Stärke, C. Carta, F. Ellinger, A low-power SiGe BiCMOS 190-GHz transceiver
chipset with demonstrated data rates up to 50 Gbit/s using on-chip antennas. IEEE Trans.
Microw. Theory Tech. 65(9), 3312–3323 (2017)
97. G. Ducournau et al., 32 Gbit/s QPSK transmission at 385 GHz using coherent fibre-optic
technologies and THz double heterodyne detection. Electron. Lett. 51(12), 915–917 (2015)
98. X. Yu et al., 60 Gbit/s 400 GHz wireless transmission, in 2015 International Conference on
Photonics in Switching, (2015), pp. 4–6. https://doi.org/10.1109/PS.2015.7328934
99. T. Nagatsuma, G. Ducournau, C.C. Renaud, Advances in terahertz communications acceler-
ated by photonics. Nat. Photonics 10(6), 371 (2016)
100. B.R. Mahafza, Radar Systems Analysis and Design Using MATLAB (Chapman and Hall/CRC,
New York, 2016)
101. M.C. Budge Jr., S.R. German, Basic Radar Analysis (Artech House, 2015)
102. B.P. Ginsburg, S.M. Ramaswamy, V. Rentala, E. Seok, S. Sankaran, B. Haroun, A 160 GHz
pulsed radar transceiver in 65 nm CMOS. IEEE J. Solid-State Circuits 49(4), 984–995 (2014).
https://doi.org/10.1109/JSSC.2014.2298033
103. A. Tang et al., A 144GHz 0.76 cm-resolution sub-carrier SAR phase radar for 3D imaging in
65nm CMOS, in 2012 IEEE International Solid-State Circuits Conference, (IEEE, 2012),
pp. 264–266
104. M. Jahn, A. Stelzer, A 120 GHz FMCW radar frontend demonstrator based on a SiGe chipset.
Int. J. Microw. Wirel. Technol. 4(3), 309–315 (2012)
105. S. Thomas, C. Bredendiek, N. Pohl, A SiGe-based 240-GHz FMCW radar system for high-
resolution measurements. IEEE Trans. Microw. Theory Tech. 67(11), 4599–4609 (2019)
106. J. Grzyb, K. Statnikov, N. Sarmah, B. Heinemann, U.R. Pfeiffer, A 210–270-GHz circularly
polarized FMCW radar with a single-lens-coupled SiGe HBT chip. IEEE Trans. Terahertz Sci.
Technol. 6(6), 771–783 (2016)
107. S. Rao, Introduction to mmWave sensing: FMCW radars, in Texas Instruments (TI) mmWave
Training Series, (2017)
108. T. Bryllert, V. Drakinskiy, K.B. Cooper, J. Stake, Integrated 200–240-GHz FMCW radar
transceiver module. IEEE Trans. Microw. Theory Tech. 61(10), 3808–3815 (2013)
109. J. Lee, Y.-A. Li, M.-H. Hung, S.-J. Huang, A fully-integrated 77-GHz FMCW radar trans-
ceiver in 65-nm CMOS technology. IEEE J. Solid-State Circuits 45(12), 2746–2756 (2010)
110. M. Hitzler et al., Ultracompact 160-GHz FMCW radar MMIC with fully integrated offset
synthesizer. IEEE Trans. Microw. Theory Tech. 65(5), 1682–1691 (2017)
111. K. Ramasubramanian, T. Instruments, Using a complex-baseband architecture in FMCW radar
systems. Texas Instruments, 2017
112. H.J. Ng, M. Kucharski, W. Ahmad, D. Kissinger, Multi-purpose fully differential 61-and
122-GHz radar transceivers for scalable MIMO sensor platforms. IEEE J. Solid-State Circuits
52(9), 2242–2255 (2017)
113. A. Mostajeran et al., A high-resolution 220-GHz ultra-wideband fully integrated ISAR
imaging system. IEEE Trans. Microw. Theory Tech. 67(1), 429–442 (2018)
114. A. Mostajeran, A. Cathelin, E. Afshari, A 170-GHz fully integrated single-chip FMCW
imaging radar with 3-D imaging capability. IEEE J. Solid-State Circuits 52(10), 2721–2734
(2017)
Index
A performance, 71, 72
ABCD matrix configuration, 173 RF model, 71
Absorption spectra, 283 single-ended on-chip antennas, 71
Accelerated electrons, 20, 30 SRF, 70, 71
Accumulation-mode MOS varactors, 53, 54 technology selection, 68
Active anti-parallel diode pair (APDP), 137 THz, 68, 70
Active mixers, 134, 135 topology, 68
Active-mode detectors, 123–126 tracking technique, 68
Air–dielectric boundary, 223 transistor circuits, 73
Aluminum oxide, 118 vacuum devices vs. solid-state devices,
AM radio signal, 193 73, 74
Amplification Amplifier-multiplier chain (AMC), 65
accurate device model, 71 Antenna-coupled bolometers, 99
bipolar transistors, 70 Antenna gain, 186
cascode topology, 68, 69 Antennas
circuit topologies, 71 bandwidth, 192
CMOS and SiGe HBT technology, 68 definition, 181
common-base (CB), 68 dipole (see Dipole antennas)
common-emitter (CE), 68 directivity, 185
common-gate (CG), 68 effective aperture, 187
common-mode noise rejection, 70 EIRP, 188, 189
common-source (CS), 68 electromagnetic field, 182
diode-based vs. transistor-based signal electromagnetic waves, 182
sources, 73 Friis transmission equation, 187, 188
efficiency, 68 patch (see Patch antennas)
EM simulation, 71 planar THz circuits, 181
gain vs. frequency, 71, 72 polarization, 189, 190
HBT, 68 radiation efficiency, 186
HEMT, 68 radiation pattern, 182–185
InP HEMT technology, 71, 72 THz applications, 181
lower-frequency operations, 73 types, 192
MAG/MSG, 69, 70 Aperture antennas, 195, 196
modification, 71 Array detectors
PA, 68 ALMA, 155
passive device model, 71 antenna-free dummy reference detector, 152
Frequency dividers, 57, 59, 60, 63 free space (see Free space, Gaussian beam)
Frequency-domain spectroscopy (FDS) function, 167
278, 282 Helmholtz equation, 165
Frequency doubler, 66 magnetic field, 165
Frequency enhancement techniques normalization, 168
frequency multipliers, 64–67 paraxial approximation, 165
harmonic signal, 61 paraxial wave equation, 165
n-push techniques, 61–63 phase shift, 167, 168
Frequency-independent antennas, 205, 208 plane waves, 165
Frequency multipliers propagation parameters, 164
AMC, 65 properties, 165, 171
baluns, 65 radius of curvature, 166
characteristics, 64 Rayleigh range/length, 167
circuit topology, 64 simplest optical principles, 163
diodes, 64 THz wave propagation, 164
even-harmonic multiplier, 65 transformation by lenses (see Ray transfer
harmonic signals, 64, 66 matrix (RTM))
ILFDs, 66 Gm-Boosting Technique, 137, 138
ILFM, 66, 67 Golay cells, 14, 104, 105
multiplication factor, 67 Group velocity of wave, 27
nonlinearity, 64, 66 Gunn diodes, 33–35
nonlinearity-based, 65 Gyro-BWOs, 29
odd-harmonic multiplier, 65 Gyro-devices, 29
output power, 67 Gyro-klystrons, 29
performance, 67 Gyro-monotron, 29
push-push oscillators, 65 Gyrotrons, 20, 27–30
quasi-optical method, 66 Gyro-TWTs, 29
SBDs, 64
solid-state, 64
transistor-based frequency multiplier H
topology, 65 Half-power beamwidth (HPBW or HP), 184
transistors, 64 Half-wave dipole antennas, 194, 197
traveling-wave, 66 Harmonic signals, 44, 61–66, 78, 87
wideband characteristics, 66 Helical antenna, 194
Frequency tuning, 51 Helmholtz equation, 164
Friis formula, 107, 108, 133, 144 Heterodyne arrays, 151
Friis transmission equation, 187, 188 Heterodyne detection
Frontend-of-the-line (FEOL) processes, 78 circuit components, 109
dB, 109
Friis formula, 107, 108
G IF, 106
GaAs, 112 noise factor, 108, 109
GaAs IMPATT diodes, 37 noise performances, 106, 107
GaAs mHEMT technology, 85 responsivity, 106, 107
Gating process, 15 transistor-based circuit, 106
Gaussian beam Heterodyne detectors, 14
confocal distance, 167 Heterodyne technique, 95
diffraction, 164 Heterojunction bipolar transistor (HBT), 68
electric field, 165, 168 High-electron mobility transistor (HEMT), 68
electromagnetic waves, 163 High-frequency semiconductor technologies, 41
expression, 167 Horn antenna, 196
formal derivation, 164 Hyperabrupt junction varactor, 52
356 Index