ARTICLE IN PRESS

Progress in Quantum Electronics 28 (2004) 1–66

Review

Terahertz fields and applications

D. Dragomana,*, M. Dragomanb
a
Physics Department, University of Bucharest, P.O. Box MG-11, 76900 Bucharest, Romania
b
National Institute of Microtechnology, P.O. Box 38-160, 72225 Bucharest, Romania

Abstract

Terahertz signals were until recently an almost unexplored area of research due to the
difficulties in generation and detection of electromagnetic fields at these wavelengths. Neither
optical nor microwave techniques are directly applicable in the terahertz range since optical
wavelengths are too short and microwave wavelengths are too long compared to terahertz field
wavelengths. The development of ultrafast optical techniques, the manufacturing of semi-
insulating semiconductors with very short lifetimes and of band-engineered heterostructures,
as well as the micromachining techniques and nanotechnology have boosted the terahertz
fields as a new area of research in quantum electronics with many important applications. The
paper reviews the most recent results in THz fields and is focused on the physical principles of
terahertz generators and receivers, underlining the link between terahertz devices and modern
technologies such as micromachining and nanotechnology.
r 2003 Elsevier Ltd. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Terahertz field generation . . . . . . . . . . . . . . . . . . . . . . . 6
2.1. Broadband THz generation . . . . . . . . . . . . . . . . . . . 6
2.1.1. Broadband THz generation/detection using
photoconductive effect . . . . . . . . . . . . . . . . . . 6
2.1.2. Broadband THz generation from semiconductor
surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.3. Broadband THz generation using optical
rectification . . . . . . . . . . . . . . . . . . . . . . . . 22

*Correspondence address. Department of Physics, University of Bucharest, P.O. Box 1-480, 70700
Bucharest, Romania. Tel./fax: +40-21-647-3382.
E-mail address: [email protected] (D. Dragoman).

0079-6727/03/$ - see front matter r 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/S0079-6727(03)00058-2
 ARTICLE IN PRESS
2 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

2.1.4. Broadband THz generation using nonlinear

transmission lines . . . . . . . . . . . . . . . . . . . . . 23
2.2. Narrowband THz generation . . . . . . . . . . . . . . . . . . . 25
2.2.1. Narrowband THz generation based on photomixing . . 26
2.2.2. Narrowband THz generation using optical
parametric conversion . . . . . . . . . . . . . . . . . . 31
2.2.3. Narrowband THz generation using electronic devices . . 32
2.2.4. THz generation using semiconductor lasers,
masers, tasers . . . . . . . . . . . . . . . . . . . . . . . 33
2.3. THz generation/detection using nanodevices . . . . . . . . . . . 36
3. THz propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4. THz detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.1. Detection of ultrashort electrical pulses . . . . . . . . . . . . . 49
4.2. CW THz heterodyne detection . . . . . . . . . . . . . . . . . . 51
4.3. Direct THz detection using micro and nanodevices . . . . . . . 54
5. Terahertz main applications . . . . . . . . . . . . . . . . . . . . . . 57
6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

1. Introduction

‘‘Terahertz (THz) fields’’ is a generic term for waves with a spectrum between 0.1
and 10 THz: Although, strictly speaking, THz waves are those with frequencies
between 1 and 10 THz the spectral range covered by the extended notion of THz
fields includes some millimeter- and submillimeter-waves. Sometimes, especially in
connection with imaging techniques, the THz spectrum is defined as T-ray. To get a
better grasp of the frequency region we are referring to it is useful to mention that the
frequency of 1 THz corresponds to a wavelength of 300 mm or 0:3 mm and to a
wavenumber of 33 cm
1 : THz fields have wavelengths extending from 3 mm
(0:1 THz or 100 GHz) up to 30 mm (10 THz); this wavelength interval ranges
between the top edge of the millimeter wave spectrum to the bottom edge of the
optical spectrum corresponding to the boundary of the far-infrared (FIR) spectral
region.
The location of the THz field spectrum between the electronic and photonic
domains implies that optical or electronic, or even better a mixture of optical and
electronic means, can be employed for THz field generation, detection and
processing. For example, THz fields can be generated with the help of a down-
conversion optical process or a photoconductive process, THz fields propagate into
free space using an antenna or are guided through a microwave-type waveguide such
as the coplanar wave waveguide (CPW), and so on. However, there are also all-
optical or all-electronic means to produce or receive THz fields; lasers or electronic
oscillators or multipliers are such examples.
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 3

The multidisciplinary character of the research area dealing with THz fields,
implied by their spectral location between frequency ranges covered by well-
developed and still separately developing research areas, requires a deep knowledge
of optics and photonics, microwave engineering and semiconductor physics. The
multidisciplinary character of the research in THz fields resides not only in the
disciplines that constitute its foundation, but also in the impressive areas of
applications, which includes astrophysics, plasma physics, spectroscopy, medical
imaging, biology, and communications.
Despite the fact that THz technology is at the boundaries of microwave and
photonic technology (or because of it), it is quite underdeveloped compared with the
achievements in microwave or photonics. There are very few commercially available
instruments for the THz frequency region and very often they lack the precision
required for performing accurate measurements. An illustrative example of this
regrettable state of affairs is that the highest oscillation frequency obtained at room
temperature using an electronic device such as an InAs/AlSb resonant tunneling
diode, which is 0:712 THz [1], was measured with an estimated error of 50% (the
emitted power is 0:3 mW). There are also no miniaturized and low-cost THz sources.
Although the value of 0:712 THz mentioned above is the highest performance ever
obtained with a single electronic device, ballistic transistors with a cutoff frequency
of several THz were recently reported; one should be optimistic that a single
electronic device will be able soon to oscillate at THz frequencies and provide an
output power greater than some mW [2]. Another promising direction of research
towards powerful THz sources involves wide bandgap semiconductors, which are
able to increase significantly the output power and the cutoff frequency of negative
differential resistance devices due to an increased electrical strength. In this respect,
GaN Gunn diodes are expected to oscillate at 740 GHz and even up to 4 THz: Wide
bandgap semiconductors can be used also in multipliers designed to generate THz
frequencies with a significant output power (0.1–1 mW); simulations show that a
single GaN Gunn diode can deliver 10 W at about 200 GHz [3].
Heterostructure semiconductor lasers based on interband transitions between
conduction and valence bands—one of the most popular laser sources in the visible
and near infrared spectral range—cannot extend their operation into the FIR or
THz range. In these lasers, called ambipolar semiconductor lasers because both
electrons and holes are involved in transitions, the energy of the generated radiation
is of the order of the bandgap, i.e. on the order of 1 eV: On the other hand, lasing at
THz corresponds to a difference in the energy levels involved in transition of about
4 meV; much smaller than the bandgap in semiconductor materials. Therefore, new
physical principles must be used for lasing at frequencies in the FIR or THz ranges.
One possible solution is to employ transitions between discrete levels or subbands
situated either in the conduction or the valence band, separated by much smaller
energy gaps than the bandgap in semiconductor materials. These intraband
transitions involve only one carrier type, the lasers based on them being called
unipolar lasers. By abandoning the ambipolar laser in favor of the unipolar laser the
mechanism for population inversion must be also changed: the electrical injection
common for ambipolar lasers is replaced by tunneling. These quantum tunneling
 ARTICLE IN PRESS
4 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

lasers demonstrated good performances first in the FIR when cooled, and nowadays
at THz frequencies near room temperature. Based on these principles a quantum
cascade laser generated more than 2 mW output power at 4:4 THz at a temperature
of 60 K [4]. This is an impressive achievement, although the working temperature is
still low. The main effort of many research teams is a room temperature THz laser;
such a laser may become reality sooner than it is expected.
Although there are encouraging perspectives regarding an electronic or photonic
single device able to generate THz signals at room temperature, THz continous-wave
(CW) sources are a rarity and are not miniaturized, requiring very expensive
components. Free-electron lasers (only a small number are working in the world!),
IR-pumped gas lasers (commercially available), or electronic tubes are the only THz
sources that provide output powers greater than 1 mW [5]. Moreover, these CW
THz sources are bulky, quite expensive and only few companies are producing them.
Therefore, many THz sources used in the laboratory are not of the CW type, being
based on short electrical pulses with picosecond (ps) or sub-ps duration and obtained
from photoconductive, rectification or other optical mechanisms when a femto-
second (fs) laser excites a semiconductor material. High power THz pulses with a
radiated peak power exceeding several hundreds of W can be obtained from these
pulsed lasers at fs excitation, although the devices are very expensive [6]. So, in the
THz range of frequencies we are confronted with a huge disadvantage: the
inexistence of miniaturized devices working at room temperature able to generate
CW THz fields with a relatively high output power.
The situation is much better for THz receivers. In the latest years very sensitive,
low noise, room temperature integrated receivers based on Schottky diodes were
demonstrated. However, these receivers must be pumped with a local oscillator (LO),
which should have a power exceeding 0:1 mW; requirement that is presently satisfied
only by an IR pumped gas laser or by electronic multipliers. The weird situation
arises that, in the case of THz Schottky receivers, the mixer is miniaturized, but not
the LO. If we want better noise performances cooled THz receivers such as hot-
electron bolometer (HEB) mixers are used, which require a LO with a power smaller
with 3–4 orders of magnitudes compared to that pumped in a THz Schottky mixer.
In this case, a fully integrated THz receiver could be possible, since a tunable THz
LO with a power average of 1 mW is achievable using existing integrated electronic or
photonic technologies. The price paid is the working temperature of the receiver,
which is not greater than a few K since in these receivers the Schottky diodes are
replaced with superconducting junctions or bolometers based on nano-sized metal
films. So, the receiver must be introduced inside a bulky and well-controlled cooling
system. The type of receiver depends on the specific THz application. For example,
in astrophysics a cooled receiver is a must since the astronomical sources are very
weak and accompanied by noise. On the contrary, if we intend to use a THz receiver
for spectroscopy or imaging we need a compact and even portable system so an
uncooled receiver is desirable.
The THz technology needs also many passive devices such as lenses, waveguides,
beam-splitters and antennas. The most common material for their realization is
presently the high-resistivity Si since at THz frequencies this material is practically
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 5

transparent, the absorption coefficient being of the order of only 0:04 cm
1 ; and has
an almost constant index of refraction of 3.42 [7]. This fact is of the considerable
interest for THz devices since Si technology is the most advanced and developed
semiconductor technology in the world and thus advanced technologies developed
for Si, such as MEMS fabrication techniques, can be successfully used to satisfy the
quite tight tolerances required for THz passive devices. Especially MEMS processing
techniques applied to THz devices have a very low costs compared with any other
THz technologies, and ensure an increased flexibility and complexity. Moreover,
using the 3D michromachining of Si, passive as well as active devices can be
integrated in a single THz circuit [8,9]. However, the best solution would be to base
the THz technology on Si for passive devices and integrate them with active devices
based on III–V semiconductor compounds or wide bandgap semiconductors.
The review will analyze the main methods and technologies for THz generation,
detection and processing. On the other hand, the most stringent problem for THz
technology is the quest for a CW THz source with reduced or even miniaturized size
able to generate a power greater than 1 mW: Therefore, a large part of the paper will
be devoted to the description of the main methods for THz field generation based on
either photonic or electronic means, or on a combination of them. Then, the paper
will present the major results concerning the propagation of THz fields. The
detection of THz fields will be reviewed and the last part of the paper will be
dedicated to THz applications. A simple enumeration of the emission/detection
methods for THz fields would certainly not deserve too much attention so that we
will focus our analysis of THz emitters/detectors on the physical principles and
effects on which THz emitters and detectors are based. An unexpected richness of
physical principles and effects are revealed, starting with the basic quantum
interference on which these devices are based on.
It is obvious from the above presentation that the THz technology has not yet
reached maturity and a lot of work remains to be done to improve the performances
of the existing devices in the THz range. The interest in the THz frequency range is
fuelled by the fact that this range of frequencies is the place where unique physical
phenomena with characteristic features are produced. Some of these unique features
are listed below.

* The spectral energy distribution in observable galaxies shows that 50% of the
total luminosity and 98% of photons emitted since Big-Bang are located in the
THz frequency range [5].
* THz fields interact strongly with polar substances but penetrate those non-polar.
Thus, the absorption spectra of many polar molecules, for example H2 O; C, N2 ;
O2 ; O3 ; HCl, CO, SO2 ; CH3 CN; etc., have many and distinct spectral peaks in the
THz range. This unique signature of molecules in the THz range is of highest
importance in monitoring the surrounding medium, air pollution detection, or gas
sensing.
* Biological tissues or other biological constituents have distinct signatures in the
THz range. For example, DNA signature, DNA manipulation, gene diagnostics
were demonstrated experimentally using THz techniques [10].
 ARTICLE IN PRESS
6 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

* Very small/miniaturized antenna arrays and circuits can be used in the THz range
since the corresponding wavelengths, which impose the dimensions of antenna
and circuits are much smaller than those encountered in the microwave and
millimeter-wave spectral intervals. This advantage is of great importance in
medical imaging and imaging devices such as THz cameras. Moreover, despite
their reduced size, THz devices are able to send or receive a huge quantity of
information that can be encoded in the ultra wideband of THz signals.
* THz signals are the information carriers in the 1–10 Tb=s optical communications
systems, which are developed now and are expected to become a commercial
reality in the next decade. THz modulators able to modulate ultrafast laser diodes
will use quantum devices such as ballistic diodes or transistors with a cutoff
frequency well beyond 10 THz:
* The 1 ps switching performance of the THz transistor is studied now by leading
companies using the latest discoveries in nanotechnology [11]. For example, the
dimension of the gate of this transistor is 90 nm thick (about 5 atomic layers) and
a SRAM cell based on it is smaller than 1 mm2 : If the power consumption
problems occurring at such huge data speed will be solved, a computer will run at
a speed that is unimaginable today making possible, for example, real-time speech
recognition and translation.
It is now clear that THz technology age will come soon, since the THz features
mentioned above are indeed tremendous and imply astonishing applications in
various areas of science such as astronomy, biology, computers and communica-
tions.

2. Terahertz field generation

2.1. Broadband THz generation

The common feature of the next paragraphs is the quest for the production of an
ultrashort electrical pulse with durations within the interval 0.2–2 ps; which has a
spectrum inside the THz range. The generation of this ultrashort electrical pulse
could be accomplished by mixed optical and electronic means (see THz
photoconductive devices), using only optical means (see THz generation using
optical rectification) or using only electronic means (nonlinear transmission lines).

2.1.1. Broadband THz generation/detection using photoconductive effect

Rapid (ps) photoconductors have been used in the last two decades to generate
ultrashort electrical signals with duration of several hundreds of fs that have a
spectrum situated in the THz region. Nowadays, this is the most encountered
method to generate/detect THz fields.
When a fs laser with an intensity IðtÞ excites a biased semiconductor with photon
energies greater than its bandgap electrons and holes are produced at the
illumination point in the conduction and valence bands, respectively. The rapid
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 7

changes of the photocarriers’ density and their acceleration due to the applied dc bias
Vb produce an electromagnetic field radiating into free-space with the help of an
antenna. The production of ultrashort currents with a full-width half-maximum
(FWHM) of 1 ps or less strongly depends on the carrier lifetime in the
semiconductor. Although intrinsic semiconductors have a carrier lifetime exceeding
hundreds of ps, some processing techniques such as annealing, ion implantation, and
radiation exposure, allows the reduction of the carrier lifetimes to sub-ps duration.
This category of semiconductors is referred to as semi-insulating semiconductors.
An updated review of semi-insulating semiconductors and their applications in
optoelectronics can be found in Ref. [12]. Nowadays, the most exploited
semiconductor with a very short carrier lifetime is the low-temperature grown
GaAs (LT-GaAs) that has a photocarrier lifetime of tt ¼ ðte þ th Þ=2 ¼ 0:25 ps;
where te ¼ 0:1 ps and th ¼ 0:4 ps are the lifetimes of electrons and holes,
respectively. Throughout this paper the subscripts e and h refer respectively to
electron and hole. LT-GaAs has also a quite high mobility ð103 cm2 =V sÞ and a high
breakdown field ð105 V=cmÞ:
So, the rapid biased photoconductor excited by a fs optical pulse (pump beam)
plays the role of a transient current source, which feeds an antenna propagating into
space transients with a short time duration. To detect such transients a device similar
to the one that emits them is needed, but the photodetector is no longer biased. The
current Iout is detected at the photoconductor pads when excited by a fs optical pulse
replica of that used at the emission point (probe beam). This optical pulse with an
intensity Iðt þ tÞ is subjected to a variable delay line that delays it with t compared to
the excitation pulse IðtÞ: The generation/detection principle of THz fields based on
the photoconductive effect is represented in Fig. 1.
Initially, photoconductors were used to feed nearby antennas and produce
transients with a few ps duration [13]. Then, the antenna was integrated on the same
substrate with the photoconductor, the device that resulted being called photo-
conductive antenna or Auston switch [14]. The integration of antenna and
photoconductor on the same substrate, combined with the search of semi-insulating
semiconductor materials with very short carrier lifetimes such as radiation damaged
silicon on sapphire (SOS), InP bombarded with He ions and later LT-GaAs resulted

emission
pump
+Vb antenna
beam detection
semiconductor
antenna
fs laser I(t)
Iout
semiconductor
I(t+τ )
FWHM <
probe
1ps
beam
variable
delay line τ

Fig. 1. Schematic representation of generation/detection of THz fields using the photoconducting effect.
 ARTICLE IN PRESS
8 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

semiconductor substrate
Si lens
dipole antenna

Vb

fs pump beam
(backside illumination)

Fig. 2. Auston switch.

in the generation of electrical pulses with sub-ps duration, which extended the
spectral response of the photoconductive antennas in the range 0.1–3 THz: A review
on initial researches focused on the development of THz photoconductive sources
can be found in Ref. [15]. One version of the Auston switch is presented in Fig. 2 and
consists of a coplanar-stripline (CPS) terminated with a dipole antenna metallized on
the semiconductor substrate and having two arms with lengths of about 40 mm: A Si
spherical lens is mounted above the antenna to collimate the emitted THz radiation.
The free carrier lifetime in a photoconductive antenna on a LT-GaAs
semiconductor can be approximated as equal to the carrier trapping time because
the trapping time in mid-gap states that trap the photocarriers is much shorter than
the recombination time between electrons and holes [16].
In these conditions, the carrier density behavior in time is given by
dn=dt ¼
n=tt þ GðtÞ; ð1Þ
where n is the carrier density and GðtÞ ¼ n0 expðt=DtÞ2 is the generation rate of
carriers due to laser pulse excitation, with Dt the laser pulse width that can be chosen
in the interval 30–150 fs and n0 the generated carrier density at t ¼ 0: The carrier
lifetime tt can be engineered to take values in the interval 0.1–5 ps by modifying the
annealing temperature for LT-GaAs containing different excess arsenic concentra-
tions [12]. The generated carriers are accelerated by the electric field bias with a
velocity rate given by
dve;h =dt ¼
ve;h =trel þ ðqe;h EÞ=meff;e;h ; ð2Þ
where ve;h are the average velocity of the carriers, qe;h is the charge of the electron or
hole, trel is the momentum relaxation time (equal to 30 fs in LT-GaAs), and E is the
local electric field, which is less than the applied bias Eb due to the screening effect of
space charges. More precisely,
E ¼ Eb
P=3er ; ð3Þ
where er is the dielectric constant and P the polarization induced by the separation
of electrons and holes. The polarization depends on time according to the
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 9

expression
dP=dt ¼
P=trec þ J; ð4Þ

where trec is the recombination time between electrons and holes (trec ¼ 10 ps for
LT-GaAs) and J ¼ envh þ ð
eÞnve is the current density.
The far-field radiation is given by
ETHz p@J=@tpev@n=@t þ en@v=@t; ð5Þ

where v ¼ ve
vh : The transient electromagnetic field ETHz consists of two terms: the
first term describes the carrier density charge effect while the second term describes
the effect of charge acceleration due to the electric field bias. Detailed simulations
have been carried out in Ref. [16] regarding Eqs. (2.1.1.1)–(2.1.1.5). The main results
can be summarized as follows:
(i) ETHz is proportional to the inverse of effective mass of the carriers. Since in LT-
GaAs the effective mass of the hole is five times larger than that of the electron,
the effect of holes in the THz radiation is significantly reduced compared to that
of electrons but cannot be ignored due to the screening effect.
(ii) The first term in Eq. (2.1.1.5) [16] is much greater than the second so that
the THz radiation is produced mainly due to the ultrafast change of
the carrier density v@n=@t; while the effect of carrier acceleration has a smaller
effect.
(iii) The pulse width of ETHz becomes larger when the width of the excitation laser
pulse is increased.
(iv) ETHz is a dynamic bias when the photoconductive antenna works as a detector.
The detector acts like a filter, filtering only the shortest wavelength components
of the transient electric field. The radiated THz field is thus considerably
distorted.

So, THz Auston switch performances depend on: (i) the optical pulse duration,
(ii) the semiconductor substrates which must have a very short carrier lifetime and a
high mobility, and (iii) the antenna geometry.
This last dependence has received a great deal of attention; Auston switches were
made with a multitude of antenna geometries such as Hertzian dipoles, resonant
dipoles, dual dipoles, slot antennas, tapered endfire antennas, log-periodic antennas,
etc. In Fig. 3 we present the dependence of the electrical pulse duration of the
transient current measured between two consecutive peaks (either positive or
negative) on the length L of the dipole. A linear decrease of the electrical pulse
duration is observed with the reduction of the dipole length. Since the geometry of
the antenna is so important for THz radiation we have presented in Fig. 4 some of
the most commonly encountered THz planar antennas. They are divided into two
large categories depending on the direction of the radiation pattern: antennas (a)–(d)
are named broadside antennas and are radiating in a direction perpendicular to the
substrate (>) while antennas (e)–(h) are named endfire antennas, radiating in a
direction parallel to the substrate ðjjÞ:
 ARTICLE IN PRESS
10 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

antenna length L [µm]

200

100

1 2 3
pulse duration (ps)
Fig. 3. Antenna length dependence on pulse duration.

The surface impendence of the metal from which all THz antennas are made is
given by
Zs ¼ ½iom0 ð1 þ iotÞ=s0 1=2 ¼ ð1=2Þðm0 =ts0 Þ þ ioðtm0 =s0 Þ ¼ Rs þ iXs ; ð6Þ
where s0 is the dc conductivity, t is the collision or relaxation time of electrons in the
metal and o is the frequency. At microwave frequencies ot51 and the formula
above is transformed into the formula that describes the well-known skin effect, the
imaginary part of the surface impedance being negligible. On the contrary, in FIR
and IR the complex part (reactance) of the surface impedance can be no longer
neglected and plays a major role in establishing the performance of antennas since it
increases linearly with the frequency while the real part of Zs is remaining constant.
Rs does not exceed some ohms, whereas Xs varies between 10 and 60 O for the
wavelengths in the range 1–70 mm: The Xs increase affects the performances of
antenna in several ways [17]. First, there are no longer orthogonal modes and
homogeneous boundary conditions in any radiating structure and the method of
separation of variables is no longer valid in many cases. This means that the
simulation results for THz antenna must be treated with extreme care, since the
effect of the surface impedance cannot be precisely taken into account in most
of the cases. If we consider a simple transmission line model applicable for dipoles or
bowtie (biconical) lines, the effect of the surface impedance is to slow down the
propagation constant with a factor Db ¼ ðX =2ÞðC=LÞ1=2 ; where C and L are the
capacitance and inductance per unit lengths, respectively. This slowing down can
destroy the main beam of traveling-wave antennas of the types presented in Fig. 4.
Unfortunately, in the case of THz antennas not only the metal, but also the
thickness of the substrate t has dramatic consequences on the performances. In
microwave circuits t5l; relation that guarantees the cancellation of substrate modes,
which has as a result the reduction of the radiation and the dielectric losses. On the
contrary, at THz and IR frequencies this inequality is no longer valid and very often
the substrate thickness well exceeds the wavelength: tXl: As a consequence the
energy generated above a critical angle is trapped in the semiconductor substrate due
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 11

(a) (e)

(b) (f)

(c) (g)

metal
(d) (h)
semiconductor substrate
Fig. 4. THz antennas. (a)–(d) broadside antennas: (a) dipole, (b) single-folded slot, (c) double-folded slot,
(d) bowtie; (e)–(h) endfire antennas: (e) Vivaldi, (f) slot V antenna, (g, h) tapered slot.

to the occurrence of substrate modes. Depending on the antenna thickness it is

possible that a large part of the radiated power (more than 90% in some situations)
is trapped into the substrate. Therefore, small losses can only be obtained when THz
antennas and propagating structures working at THz are patterned on very thin
substrates. For example, the substrate thickness must be smaller than 0:04l for a slot
antenna and smaller than 0:01l0 in the case of dipole antennas [18]. This means that
we need a substrate thickness less than 3 mm: This amazing achievement was possible
only in the last years using the micromachining techniques borrowed from MEMS
technology and adapted for the semi-insulating semiconductors. Even using these
thin substrates, a dielectric layer over the antenna is still needed to collimate the THz
radiation. This dielectric has a semispherical shape playing the role of lens also
preventing the refraction of waves emerging from the antenna at the dielectric-air
interface. A similar constraint regarding the thickness of the substrate exists for
endfire antennas. In this case, we have 0:005o½ðer Þ1=2
1t=lo0:03 giving also
thicknesses of few microns at THz frequencies.
 ARTICLE IN PRESS
12 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

The physical characteristics mentioned above are common for all antennas from
Fig. 4. However, their performances are quite different. Some of them show a
resonant behavior around a certain frequency and are therefore termed resonant
antennas. This is the case of the dipole and of the single- and double-folded slot
antennas. The dipole antennas are not compatible with CPW lines; the CPW line is
the standard configuration of the transmission lines in millimeter waves and THz
integrated circuits. In this type of antennas the radiation is coupled into the substrate
beyond the resonance. On the contrary, the resonant single-folded and double-folded
slot antennas are compatible with CPW lines and have an extended ground plane
shielding the dielectric from the free-space. Therefore, they are very frequently used
in the THz region. Moreover, they show a radiation pattern with a main symmetric
lobe and with secondary lobes situated at a moderate level (
15 to
20 dB) [19]. In
contrast with the narrow band antennas mentioned above, the broadband bowtie
antenna can be easily matched in a large range of frequency. It has a radiation
pattern consisting of two main lobes, which is its main disadvantage. This
disadvantage can, however, be overcome by positioning a silicon lens off the
symmetry axis of the antenna. In this way the radiation pattern shape is transformed
from a bi-lobed to a single lobe at the expense of a 3 dB reduction of the radiated
power. Due to its large bandwidth, it was experimentally proven that the detected
THz signal produced by a bowtie is greater with 26 dB than that produced by slot
antennas [20]. The endfire antennas are also broadband and have medium gain
values that can be increased by increasing the length of the antenna. Some of them,
like the Vivaldi antenna, cannot be directly excited with CPW lines, but with
balanced two-wires transmission lines like CPS. These antennas need a balun to
match them to a CPW line at the expense of loosing 3 dB of power. Since the endfire
antennas are not located on the same axis with the lens, additional integration
problems will be encountered. A review regarding the properties of endfire antennas
can be found in Ref. [21]. A recent photoconductive antenna configuration able to
radiate efficiently THz fields is based on the singularities of the electric fields that
occur at the tips of sharp metallic electrodes [22]. Its configuration is presented in
Fig. 5. This antenna was able to emit an average THz power of 3 mW under a 20 mW
optical excitation using a 45 mm dipole length; this power is one order of magnitude
greater than that emitted by other photoconductive antennas. High average THz
powers are obtained using metallic dipole antennas supported on InAs substrates. In

laser spot
Fig. 5. The singular electric field THz photoconducting antenna.
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 13

pump laser
Vb g h
w

LT-GaAs

Fig. 6. CPS dipole antenna.

this case, an average power of 650 mW was obtained placing the antenna in a 1:7 T
magnetic field and illuminating it with 1:5 W ultrashort pulses [23].
A CPS dipole antenna is also able to emit high THz powers in an ultrawide
bandwidth of 40 THz: Its configuration is displayed in Fig. 6 [24]. Typical values
for the antenna configuration are g ¼ 5–10 mm; w ¼ 10–20 mm; h ¼ 30–50 mm;
L ¼ 1 mm: The peak value of the THz field is given by
peak
ETHz ¼ emtint ½ð1
RÞ=hf ðPin =gÞðVb =gÞ; ð7Þ
where tint is the interval between pump pulses, m is the mobility of the carriers, R is
the reflection coefficient of the substrate, hf is the photon energy of the pump laser of
peak
frequency f and Pin is the average power of the pump laser. A high value of ETHz is
obtained if the substrate has a high mobility and a high resistivity, requirements that
enable the application of high bias values Vb : These criteria are satisfied by LT-
peak
GaAs. The peak value ETHz does not increases indefinitely, but saturates at a certain
intensity, I0 ; or power, P0 : We have
peak
ETHz pPin =ðP0 þ Pin ÞpgðPin =AÞ=ðI0 þ Pin =AÞ; ð8Þ
where the illuminated area A is proportional to g2 : From the above relation it follows
peak
that to obtain a maximum ETHz it is necessary to adjust the gap value so that the
pump intensity Pin =A equals the saturation intensity.
As we have mentioned above, THz antennas are almost always accompanied by
lenses, which cover them. When the lenses are made from the same material as the
antenna substrate they are sometimes termed substrate lenses. There are two main
types of substrate lenses: hemispherical and hyperhemispherical. Initially, these
substrate lenses have been used as a semi-infinite antenna dielectric substrate, which
eliminates the substrate modes. Since the substrate lenses also increases the gain of
antennas, the lenses accompany THz antennas even when they are realized on very
thin substrates. The first type of lens that appeared—the hemispherical lens—was
used as a collimating lens, the THz antenna or THz antenna array being positioned
in its focus. Although, the rays emerging from this lens are collimated, there are also
rays that are trapped inside the substrate lens due to the total internal reflection,
which occurs at large angles. The net effect is the wave-front aberration that is
unavoidable for this type of lenses. The distance between the emitter and the lens
tip is
demitter2lens ¼ Rn=ðn
1Þ: ð9Þ
 ARTICLE IN PRESS
14 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

Fig. 7. Hyperhemispherical lens. The dashed lines represent the path of the rays in the absence of the lens.

In the case of a Si hemisphere lens this distance is 1:41R; where R is typically

3–4 mm:
The second important substrate lens configuration is the hyperhemispherical
substrate lens, which is displayed in Fig. 7. It was designed to reduce the
disadvantages of the hemispherical substrate lens: there are no internal reflections in
the case of hyperhemispherical lens. Since the rays are no longer totally reflected
there is no wave-front aberration. The hyperhemispherical lens is an aplanatic lens,
which reduces the beamwidth of the radiation pattern of antennas and increases the
antenna gain by n2 ; where n is the dielectric constant of the substrate and the lens.
The distance between the THz antenna and the hyperhemispherical lens is given by

demitter2lens ¼ Rðn þ 1Þ=n: ð10Þ

This distance is 1:3R for a Si hyperhemispherical lens.

Many THz antenna arrays used for imaging or for receiving purposes were
designed with substrate lenses. Bowtie antenna arrays, Yagi-Uda arrays, CPW
slot antennas, single- and double-folded slot antennas are among them. For a
review of THz lenses properties and applications see Refs. [18,25,26]. A comparison
between the two main types of substrate lenses was recently made by Ref. [27].
It was found R that the directivity of the collimating lens, defined as D ¼
2 maxjEðyÞj2 = jEðyÞj2 sin y dy; increases with frequency and is about 30 dB at
1 THz; which means that the majority of THz energy is propagating in a beam with a
width of a few degrees. In deep contrast, the directivity of the hyperhemispherical
lens is quite low and independent of frequency; it has a value of about 3 dB at 1 THz;
the THz energy propagating in this case in a beam with a width of about 30 :
However, the hyperhemispherical lens couples much better to a gaussian mode.
While the collimating lens has no effect on the bandwidth of the THz emitted signal,
this bandwidth is dramatically restrained in the case of the hyperhemispherical lens.
In the last case diffraction and interference fringes have been detected even in a
broadside direction.
Although an optimized photoconductive antenna incorporates many innovative
ideas regarding the substrate, antenna, or lenses, the THz generation based on the
photoconductive effect shows a poor efficiency, i.e. ultrashort pulses with a power of
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 15

d
(a)

optical
excitation
θ θ ETHz(θ )

(b)
Fig. 8. THz photoconducting array. (a) Schematic representation, and (b) emitted field.

tens of mW are transformed in THz signals with a power not exceeding a few mW: In
what follows we describe some methods to improve this low efficiency.
An array of photoconductive antennas will radiate in free-space more THz power
than a single antenna and, depending on the type of optical excitation, we can steer
the THz beam or control its spectral content (for a review see Ref. [28]). Such an
array, the configuration of which is presented in Fig. 8, consists of a sequence of
dipole antennas, each of them being independently biased. The THz far-field pattern
of the array in the direction y is given by
XN
ETHz ðyÞ ¼ const
½ðVn
Vn
1 Þ=dIn ðOÞ expð
inkd sin yÞ; ð11Þ
n¼1

where Vn is the bias applied on the nth electrode, the total number of electrodes
(dipoles) being N: In ðOÞ is the optical excitation, k is the THz free-space wavenumber
and d the spacing between two consecutive photoconductive antennas. If an optical
pulse train with period between ultrashort pulses inversely proportional to the
emitted THz frequency illuminates the photoconductive array, the direction of the
radiation pattern can be scanned about 45 by changing the dc voltage applied to
each antenna. If the period of the bias voltage is periodically varied, the array is
analogous to a grating that changes the beam direction to a prescribed value given by
the bias period. If the bias has a sinusoidal variation Vn ¼ ðE0 =kb Þ cosðnkb dÞ; where
kb ¼ 2p=Lb with Lb the bias spatial period, the radiation pattern is given by
ETHz ðyÞ ¼ const
cos y sin½Ndðkb 7k sin yÞ=2=sin½dðkb 7k sin yÞ=2: ð12Þ
Such an array with 32 antenna elements working at 0:5 THz occupies an area of
2
3 mm2 : Each dipole is 2 mm long, 25 mm wide and is separated from the next one
by d ¼ 100 mm: The entire array has a beamwidth with a FWHM of about 10 :
 ARTICLE IN PRESS
16 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

If the array is illuminated by a single optical pulse, the spectral content of the
radiated THz signal can be changed by changing the bias. From (10) it follows that
the radiation has a maximum at kb ¼ 7k sin y: Since k ¼ 2p=l; where l is the center
wavelength of the THz radiation, we have l ¼ 7Lb sin y; which shows that the THz
wavelength is changed when the bias period is varied. For an array with the same
dimensions as those given above it was experimentally demonstrated that the
frequency can be tuned in the interval 0.140–1:06 THz by changing the bias period
from 3 to 0:4 mm:
Recently, miniaturized photoconductive THz sources and probes (detectors) were
realized using metal–semiconductor–metal (MSM) interdigited structures on SOS
[29] or LT-GaAs thin substrates [30]. MSM is playing the role of a photoconductive
switch, but the transient current is produced due to the ultrafast carrier drift across
the gaps formed by the consecutive metal fingers. A part of the optical pulse is
directed to the MSM emitter, while the delayed part (probe) is directed to the MSM
detector. Thus, we can detect and characterize the radiation of the THz emitter
propagated in free-space or in a transmission line due to probe sampling. The
photoconductive MSM probe can be used (i) to measure the propagation
characteristics of transmission lines at THz frequency emitted by another
photoconductive MSM emitter, (ii) to characterize the field distribution of a THz
antenna or other THz devices. The THz miniaturized MSM emitter and detector are
represented in Fig. 9. The widths of the MSM fingers, as well as the spacing between
them, can have a few microns or can even have sub- micron dimensions.
More sophisticated photoconductive antennas can be designed for near-field THz
microscopy [31]. For example, a photoconductive antenna realized on a very thin
LT-GaAs substrate glued on a thicker sapphire substrate and terminated with a
GaAs taper is represented in Fig. 10. The sapphire substrate as well as the taper is
designed to reduce the reflections. The THz radiation propagates through a small
metallic aperture, a spatial resolution of 60 mm being achieved for a 50 mm aperture.
The latest trends regarding THz photoconductive devices are the integration of the
emitter and the receiver on the same chip. The entire device is then called a

sharp
tip
MSM
switch

(a) (b)

optical
illuminated area
Fig. 9. (a) THz MSM emitter and (b) THz MSM probe (detector).
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 17

photoconductor
(LT-GaAs)
sapphire
taper radiation
pattern

optical
pulse

biased dipole aperture

antenna

Fig. 10. THz emitter for near field microscopy.

parabolic mirror substrate lens

THz field

A Vb
dipole
antennas

LT-GaAs optical
substrate pulse
delay line
Fig. 11. A photoconductive THz transceiver.

photoconductive transceiver. Since such a single chip device is practically a

miniaturized FIR spectrometer, it plays the role of a lab-on-a chip for gas detection
or air monitoring. In Fig. 11 we have displayed such a recently reported transceiver
using dipole antennas [32]. The distance between the two antennas was 500 mm; each
antenna being 10 mm wide and having a gap of 5 mm between each arm of the
antenna connected to a feed line that is 100 mm long. The photoconductive LT-GaAs
substrate, which had a thickness of 2 mm; was grown on semi- insulating GaAs and
the substrate lens had a diameter of 26 mm:
The THz generation method using a CPW line excited by an ultrashort optical
pulse (pump) is displayed in Fig. 12. THz detection is realized with the help of a
delayed version of the pump (the probe), which shortens the line at a prescribed
location. During the photoexcitation produced by the pump the charges are
transferred from the CPW central conductor to its ground generating a current flow.
Charges of opposite signs accumulated on the two conductors produce a TEM
dipolar field. The output voltage is given by

vout ðtÞ ¼ Vb Z0 gðtÞ: ð13Þ

 ARTICLE IN PRESS
18 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

Vout
metal

Vb t

mA

optical
excitation beam

semi-insulating THz detection, probe

substrate or sampling beam

Fig. 12. THz generation using CPW lines.

where gðtÞ is the photoconductance at the pump location, Z0 is the characteristic

impedance of the CPW and Vb is the bias voltage. Typical values of these parameters
are Vb ¼ 1:5 V; and Z0 ¼ 100 O; the CPW length is about 20 mm: For a review of
THz generation using optically excited CPW see Ref. [33]. The output voltage shows
a capacitor-like behavior. The leading edge of the current has a risetime similar to
that of the optical pulse, while its trailing edge has a decay time proportional to the
carrier’s lifetime. An electrical pulse with a symmetrical shape can be obtained if two
laser pulses are focused simultaneously between the central and ground electrode of
the CPW. Electrical pulses with very short durations (0.4–0:5 ps) can be obtained
using this method. A biased CPS line consisting of two electrodes was also used to
generate ultrashort electrical pulses using a SOS substrate. By illuminating the
positively biased electrode with an ultrashort optical pulse, the transient current
produced due to the photoconductive effect can have only 380 fs duration [34].
However, the CPW line is more suitable for THz generation because it can be readily
integrated with other THz components. A recent review of THz production using
transmission lines can be found in Ref. [35].
High power ultrashort pulses with a spectrum within the THz bandwidth can be
generated using large aperture photoconductive antennas. The aperture, which is in
fact the gap between the two biased electrodes, as depicted in Fig. 13, allows an
illumination area with typical dimensions of a few mm, much greater than the
radiated center wavelength. The increased size of the aperture allows the utilization
of high optical energies and high values of the bias, high power of the radiated THz
field being thus expected. When the energy of the ultrashort optical pulse exceeds the
bandgap of the photoconductor, the THz field is generated due to accelerated
photocarriers, which produce a transient current at the surface of the biased
photoconductor. However, the THz power saturates at high excitation powers due to
charge screening effects. The photogenerated electrons and holes move in opposite
directions creating spatial regions of positive and negative charges that induce an
electrical field in a direction opposite to the applied electric field; for sufficiently large
densities of photocarriers the applied electric field can be totally screened [37].
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 19

Er,2(t) Er,1(t)
far-field
transmitter detector
r
Eb
mA
Hr,2(t) Hr,1(t)
JS(t) THz
field
Vb

Vb optical
excitation pulse
variable
(a) (b) optical delay

Fig. 13. (a) Distribution of electric and magnetic fields for a large aperture photoconductive antenna, and
(b) generation/detection of THz fields using large aperture photoconductive antennas.

Denoting by Er;1 ðtÞ and Hr;1 ðtÞ the electric and magnetic fields, respectively, inside
the photoconductor and by Er;2 ðtÞ and Hr;2 ðtÞ the electric and the magnetic fields near
the photoconductor surface (near-fields), respectively (see Fig. 13a), the boundary
conditions [38] imply that
Er;1 ðtÞ ¼ Er;2 ðtÞ ¼ Er ðtÞ; ð14aÞ

Hr;2 ðtÞ
Hr;1 ðtÞ ¼ Js ðtÞ; ð14bÞ

1=2
where Hr;1 ðtÞ ¼ Er;1 ðtÞe =Z0 and Hr;2 ðtÞ ¼ Er;2 ðtÞ=Z0 with Z0 ¼ 377 O the impedance
of the free space. With these boundary conditions the radiated electric field is
Er ðtÞ ¼ Z0 Js ðtÞ=ð1 þ e1=2 Þ: ð15Þ
Taking into account that the Ohm law can be written as Js ðtÞ ¼ sðtÞ½Eb þ Er ðtÞ;
where sðtÞ is the surface conductivity, we get
Js ðtÞ ¼ sðtÞEb =½1 þ sðtÞZ0 =ð1 þ e1=2 Þ: ð16Þ
1=2
When sðtÞZ0 =ð1 þ e ÞX1; i.e. at high optical fluence (time-integrated optical
intensity), the radiated current saturates and the far-field is given in these conditions
by
ErTHz ¼ A½dJs ðtÞ=dt=ð4pe0 c2 rÞ: ð17Þ
We can see that the near-field is proportional to the surface current, while the far
field is proportional to its derivative. Thus, the saturation effect is a near-field
phenomenon.
Experiments with large aperture photoconductive antennas have demonstrated
high power radiated fields. For example, an electric far field with a value of
0:9 kV=cm and a duration of 1:3 ps was obtained when a InP ð1 1 1Þ substrate was
used; when a GaAs ð1 1 1Þ substrate was used the strength of the electric far field
reached 1 kV=cm and had a duration of 1:8 ps: In both cases the gap had a value of
0:5 cm; Eb ¼ 8 kV=cm and the optical fluence was 1 mJ=cm2 [36]. High power and
 ARTICLE IN PRESS
20 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

narrow band THz radiation overcoming the saturation limitations described above
can be obtained if the excitation and the thickness of the photoconductive substrate
are optimized [38]. Ideally, a multi-pulse excitation should be employed, obtained by
splitting the original excitation pulse into N pulses with an interpulse spacing Ts : The
multi-pulse excitation is equivalent with single-pulse excitations at the frequencies
o ¼ 2pm=Ts ; m ¼ 1; 2; y; N; and produces a THz spectral narrowing when the
number of pulses is increased. When the emitter is completely saturated the THz
peak power is enhanced by a factor of N 2 : Ts must be 2–3 times larger than the
carrier lifetime. THz radiation saturation can also be overcome if the thickness of the
photoconductive substrate (LT-GaAs) is chosen such that the light is entirely
absorbed into it and not in the substrate on which LT-GaAs is grown. For example,
a thickness of at least 3 mm is necessary for the photoconducting substrate at an
optical excitation at 800 nm:
All types of photoconductive devices used for THz generation presented above are
strongly dependent on the properties of the photoconductive substrate and especially
on the carrier lifetime. On the contrary, the behavior of the photoconductive switch
presented in Fig. 14 does not depend on the properties of the photoconductive
substrate. This is possible if the thickness of the photoconductor substrate is smaller
than the penetration depth of the optical excitation and if the bottom of the
photoconductor substrate is metallized, playing the role of an optical mirror as well
as that of an electrical ground plane. For example, at 800 nm the penetration depth
in Si is 10 mm: If we choose this value as the thickness t of the substrate, the optical
excitation reaches the ground plane after td ¼ tn=c ¼ 110 fs; a time that is shorter
than the lifetime of the carriers. In this case the carriers have no time to recombine
[39]. The micromachining technology allows the fabrication of Si or GaAs substrates
with a thickness less than 2 mm [40] so that this photoconductor switch becomes
feasible. For a detailed simulation of this device, including a THz equivalent circuit
demonstrating the generation of electrical pulses with durations varying from
hundreds of fs up to a few ps see Ref. [39].

optical excitation
Vb

output signal

photoconductive substrate
t
ground line

ground plane
Fig. 14. Photoconductive switch that is independent of the lifetime of the carriers in the photoconductive
substrate.
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 21

bare
reflected semiconductor transmitted field
field (THz field)

θr
θt
θopt

optical pluse
excitation
Fig. 15. THz generation using bare semiconductors.

2.1.2. Broadband THz generation from semiconductor surfaces

THz generation from bare semiconductor surfaces is based on the fact that
the surface states of many semiconductors surfaces are entirely occupied. As a result
the Fermi level is pinned and the conduction and valence bands are bent near the
semiconductor surface/air interface creating a depletion region and a strong built-in
surface electric field Eb ; with a typical value of 105 V=cm; perpendicular to the
semiconductor/air interface. When an incident ultrashort optical pulse with photon
energy greater than the bandgap of the semiconductor hits the semiconductor
surface, the injected photocarriers at the semiconductor surface are depleted and
accelerated by the built-in field. As a result an ultrashort transient current is formed,
which radiates an electromagnetic beam with a spectrum in THz domain (see
Fig. 15). Moreover, this beam can be steered by changing the incidence angle of the
optical excitation. Many semiconductors such as InP, GaAs, GaSb, InSb, CdTe,
CdSe, Ge, and Ga1
x Alx As ðxo0:2Þ have been used to demonstrate the generation
of THz radiation [41]. Also Au/GaAs Schottky barriers can be employed to generate
THz radiation at their surface/air interface [42].
The THz field is given by

ETHz ðtÞ ¼ Zs Js ðtÞ sin yr =ðcos Wr þ n cos yt Þ

Z N
¼ eZs ½sin yr =ðcos Wr þ ns cos yt Þ nðx; tÞvðEb ðx; tÞÞ dx ð18Þ
0

where Zs is the characteristic impedance of the semiconductor, ns its index of

refraction, nðx; tÞ is the photocarrier density, and vðEb ðx; tÞÞ is their drift velocity.
The electric field that accelerates the photocarriers can be generated by the
piezoelectric effect in strained superlattices. In this respect, a misfit (111) oriented
GaSb/AlSb superlattice has been used to generate THz fields [43]. In comparison to
the previous method, the THz signals are generated without an applied bias and an
antenna.
In another optical method THz radiation originates from ballistic photocurrents
generated due to quantum interference of one and two photons in semi-insulating
GaAs and LT-GaAs. THz single-cycle with controllable phase is obtained at the
 ARTICLE IN PRESS
22 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

central frequency of 4 THz with a 3 THz spectral width [44]. In contrast with
previous methods, the unbiased photoconductor is excited with two copolarized
ultrashort optical pulses having the carrier frequencies o and 2o; with 2_o > Eg >
_o; that couple the same valence and conduction band states through two-photon
absorption processes. If the two beams are phase related, the interference between
transition amplitudes produces a phase-controllable photocurrent. So, the THz
center frequency and bandwidth can be tuned by the optical pulse width, while the
THz power is changed by tuning the phases of the two optical pulses. However, the
emitted power does not exceed 3:5 nW at room temperature.

2.1.3. Broadband THz generation using optical rectification

Optical rectification is a process inverse to the electro-optic effect, and consists in
the generation of an electrical waveform that is the envelope of an ultrafast optical
pulse, which excites an electro-optical material. Materials suitable for this method
include among others LiNbO3 ; LaTiO3 ; zinc-blende semiconductors (GaAs, ZnTe,
CdTe, InP) or organic crystals (DAST). The physical mechanism involved in optical
rectification is the production of a transient polarization PðtÞ; when a fs optical pulse
is focused on an electro-optical material. The THz radiation is proportional to the
second time derivative of the low-frequency part of PðtÞ analogous to the case of a
transient dipole. Detailed reviews of this method can be found in Refs. [45,46].
Very short electrical pulses are obtained with this method, much shorter than in
the case of the photoconductive method. For example, in Ref. [47] a bipolar pulse
with a FWHM of 180 fs is produced using optical rectification. The efficiency of this
method depends on the optical second-order nonlinear coefficients ðwð2Þ Þ of the
materials, and on the phase matching conditions. The power of the THz radiation
produced by this method is lower than that produced using the photoconductive
method, but its spectral content is much broader attaining 50 THz [48].
Zinc-blende semiconductors are among the most commonly used materials for
THz generation/detection by optical rectification. When the photon energy is greater
than the semiconductor bandgap (Ephoton > Eg ), as in the case of unbiased GaAs,
THz generation originates from two mechanisms: the carrier’s acceleration, as
described in the previous paragraph, and optical rectification. The first mechanism is
cancelled by illuminating the sample at normal incidence, so that only optical
rectification is contributing [49]. Also, optical rectification can be obtained by
illuminating the semiconductor with photons that have a smaller energy than the
semiconductor bandgap (Ephoton oEg ). The value of the generated signal depends
strongly on the orientation of the optical polarization and on the crystallographic
orientation of the sample. A THz system using optical rectification is presented in
Fig. 16. The detection technique is based on electro-optic sampling, which will be
described in Section 4.1. It is important to mention that the two ZnTe crystals at the
emission and detection are identical, and are separated by a few tens of centimeters,
which is the free space distance of THz fields propagation.
The key problem in optical rectification is the phase matching, which maximizes
the interaction between the optical and the THz pulse in the nonlinear material and
thus enhances the efficiency of THz generation. The optical rectification process, in
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 23

THz path
optical path
optical ZnTe
delay emitter

ZnTe
detector
fs optical
pulse

to detectors
Fig. 16. THz generation using optical rectification.

which a THz pulse with frequency oTHz collinear with the optical pulse is produced,
can be viewed as mixing of different spectral components oopt and oopt þ oTHz of the
optical pulse [50]. The phase matching condition for the wavenumbers at different
frequencies is
Dk ¼ kðoopt þ oTHz Þ
kðoopt Þ
kðoTHz Þ ¼ 0: ð19Þ
Neglecting optical dispersion, the coherence length is
lc ¼ p=Dk ¼ pc=ðoTHz j nopt
nTHz jÞ; ð20Þ
where nopt and nTHz are the refractive indices at the optical and THz frequencies,
respectively. Long coherence lengths can be obtained using either birefringence or
angle tuning in noncollinear configurations. The dispersion of the optical refractive
index is also a way to increase the coherence length in a large THz bandwidth [50]. In
this case, the phase matching is obtained when the THz pulse propagates with the
group velocity of the optical envelope. In this case, the coherence length is given by
pc
lc ¼ : ð21Þ
oTHz j nopt
lopt ðdnopt =dlÞjlopt
nTHz j
In the case of ZnTe at lopt ¼ 800 nm the coherence length is large in the bandwidth
0.5–2 THz: A simple and efficient phase matching technique demonstrated
theoretically and experimentally in the GHz–THz range is based on a rectangular
waveguide partially filled with a nonlinear crystal. Controlling the filling degree of
the rectangular waveguide very efficient collinear phase matching and thus more
efficient THz generation can be obtained [51].

2.1.4. Broadband THz generation using nonlinear transmission lines

Nonlinear transmission lines (NLTL) are distributed devices, which consist of a
high impedance transmission line, usually a CPW line, periodically loaded with
nonlinear elements, usually Schottky varactor diodes. The concept of a NLTL device
is presented in Fig. 17, where the nonlinear elements are diodes with a nonlinear
capacitance CðV Þ; with V the voltage applied on them. The entire device can be
monolithically integrated using the GaAs technology, which is necessary to build
Schottky diodes with a cutoff frequency beyond 1 THz: Shock waves or solitons can
occur in NLTL devices due to the balance between the nonlinearity and inherent
 ARTICLE IN PRESS
24 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

Zl Vout
Z0 Zl Zl
sub-ps
duration
Vin C(V) C(V) C(V) Rs

time
Fig. 17. A NLTL device.

dispersion. Both waves are compressed versions of the input excitation, which is a
high power microwave sinusoidal wave. We consider that the CPW line impedance
Zl is formed by a series inductance Ll and a shunt capacitance Cl and that the
Schottky diode can be modeled with a resistance ðRd Þ in series with a variable
capacitor ðCd Þ: The NLTL behavior can be characterized with the help of two
frequencies:
fd ¼ ½2pRd Cd ðV Þ
1 ; ð22Þ

1
fb ¼ ; ð23Þ
pLl ½Cl þ Cd ðV Þ1=2
where fd is the cutoff frequency of the diode and fb is the cutoff frequency of the
NLTL circuit. Shock waves with a shape similar to that shown in Fig. 17 are
generated if fd ¼ fb and soliton waves are formed if fd bfb : The soliton wave is
produced due to the balance between dispersion and nonlinearity of the propagating
wave in the NLTL. The solitons are described by a Korteweg de Vries (KdV)-like
nonlinear equation and have a sech2 shape.
In the case of shock wave generation, the delay between two consecutive cells is
t ¼ d=vCPW ; where d is distance between two consecutive diodes and vCPW ¼
1:13
108 m=s (for GaAs) is the propagation velocity in the CPW, Ll ¼ tZl and
Cl ¼ t=Zl : The shock wave effect can be described as follows: the negative part of
the input sinusoidal voltage propagates along the NLTL, its fall time decreasing as a
function of distance. After propagating through n-cells, the fall time is
tn ¼ tin
ntf½1 þ Cd ð0Þ=Cl 1=2
½1 þ Cd ð
VMAX Þ=Cl 1=2 g: ð24Þ
When the fall time decreases, the dispersion that broadens the fall time is balanced
by the nonlinearity, which compresses the fall time due the voltage-dependent
propagation velocity. A stable fall time of the input voltage (the shock wave) is
attained when the fall time compression/cell is equal to the fall time broadening/cell.
After this, the shock wave propagates unchanged in shape along the NLTL.
A step-like shock wave of 3:5 V amplitude and 480 fs 10–90% fall time was
obtained using delta-doped Schottky varactors; its spectral content was found to
exceed 3 THz [52]. Moreover, an all-electronic THz spectroscopic system based on
NLTL was implemented for the amplitude and phase measurements of various
samples in the range 0.2–1 THz [53]. Extensive reviews about nonlinear waves and
NLTL devices for millimeter and submillimeter waves can be found in Refs. [54,55].
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 25

Recently it was analytically and numerically shown that solitons with a duration
of 2:7 ps can propagate in a NLTL consisting of n ¼ 60 quantum barrier varactors,
named also heterostructure barrier varactors (HBV) [56]. The dependence of HBV
capacitance with the applied voltage can be expressed as

CðV Þ ¼ Cð0Þ þ Cmax sechðDc V Þ; ð25Þ

where Cð0Þ is the unbiased value of the HBV capacitance, Cmax is the maximum
value of the HBV capacitance and Dc is the width of the CðV Þ curve.

2.2. Narrowband THz generation

This section is dedicated to the generation of CW THz fields. In contrast with the
THz sources presented in Section 2.1, the THz sources presented below are
characterized by a spectrum consisting of a single spike centered within the THz
range with a very narrow bandwidth. A very large tunability of this CW component
inside the THz bandwidth and a power as large as possible are highly desirable. As in
the last section, these aims will be fulfilled using methods that combine optical and
electronic means (THz photomixing), only optical means (THz parametric
generation), or solid state devices such as resonant tunneling diodes, multipliers
and solid state lasers. The THz CW signals are of considerable importance for high-
resolution THz spectroscopy, THz sensors and ultrabroadband communications.
Traditionally, gas lasers generate CW THz signals in the frequency domain
0.9–3 THz with output powers in the range of 1–30 mW: A gas laser consists of a
carbon dioxide laser that pumps a cavity filled with a gas such as CH4 ; N2 ; etc.,
which dictates the lasing frequency. The gas sources show no tunability and are very
large, with dimensions exceeding 2:5 m: However, a ‘‘miniaturized’’ version of a gas
laser was recently reported, which delivers 30 mW at 2:5 THz; its dimensions are
75
30
10 cm and weight 20 kg [57].
Free-electron lasers generate either CW or pulsed high power THz radiation, but
they are very costly and have very large dimensions, functioning in large rooms
containing many additional facilities. Therefore, only a few are operating in the
world. However, backward wave oscillators (BWO) are based on the same principles
as an electron laser and are able to deliver a few mW in the range of 0.6–1:3 THz: In
contrast with THz gas lasers, BWO are frequency tunable (for example, between 0.78
and 0:97 THz or 1 and 1:25 THz) with a high sweeping rate. BWO requires a water-
cooling system and high bias voltages of 1–6 kV at 25–45 mA: The weight of a BWO
(without the cooling system and the power supplies) is more than 15 kg:
However, THz gas lasers and BWOs are commercially available, being the only
CW THz sources that can be bought from the market. They are both bulky and need
a lot of accessories such as high power supplies, water-cooling systems, etc.
It is now understandable why so many efforts were dedicated in the last decade in
the quest of a miniaturized CW THz source working at room temperature and
delivering a few mW output power. As we will see, this quest continues, because the
performances mentioned above were only partially achieved up to now.
 ARTICLE IN PRESS
26 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

2.2.1. Narrowband THz generation based on photomixing

Heterodyne mixing (photomixing) of two individual optical CW lasers (one of
them being tunable) in a photoconductor produces a photocurrent with a frequency
equal to the difference between the frequencies of the two lasers. When this
difference frequency is within the THz range of frequencies the photocurrent is
propagated along the transmission line or is radiated in free-space with the help of an
antenna. There are two types of photomixers: discrete-element photomixers and
distributed photomixers. Discrete-element photomixers are photoconductors, such
as micrometric photoconductive gaps or MSM interdigited structures, with a large
bias field applied between their electrodes, illuminated by the two lasers sources and
placed at the driving point of an antenna or an antenna array. Discrete-element
photomixers act like a current source with a very wide bandwidth and drive the
antenna at THz frequencies. The distributed photomixers are based on similar
principles, except that the optical field produced by the lasers propagates along the
photomixer structure and is not localized in a single point, as in the case of discrete-
element photomixers. Recent comprehensive reviews about photomixers are found
in Refs. [58,59].
Different configurations of discrete-element photomixers, which will be
analyzed in what follows, are represented in Fig. 18. Since there are a lot of
photoconductor and antenna geometries, many combinations between them can be
imagined; only a few of them are represented in Fig. 18. Basically, the photomixer
can be modeled as a photoconductor with a photoconductance GðtÞ variable in time.
This photoconductance is in parallel with the capacitance C; which depends on the
photoconductor geometry (gap, interdigited, etc.). The equivalent circuit model of
the photomixer is represented in Fig. 19, where ZA ¼ RA þ iXA is the antenna
radiation impedance.
We consider that the incident optical power, which illuminates the photomixer, is
given by
Pi ðtÞ ¼ P1 þ P2 þ ðP1 P2 Þ1=2 fcos½2pð f2
f1 Þt þ cos½2pð f2 þ f1 Þtg; ð26Þ
where P1 and P2 are the optical powers, and f1 and f2 the frequencies generated by
the first and the second laser, respectively. The photon energies of the two lasers hf1
and hf2 must be greater than the bandgap energy of the LT-GaAs photoconductor
ð1:4 eVÞ: The frequency which modulates the photoconductance is j f1
f2 j ¼ fTHz
since the term in Eq. (26) containing the sum of frequencies varies on a much shorter
time scale than the lifetime of the photoconductor, t: The time variation of the
photoconductance is given by [60]
GðtÞ ¼ G0 f1 þ b sinðoTHz tÞ½1 þ ðoTHz tÞ2 g
1=2 ; ð27Þ
where G0 and b are dependent on the input optical power P0 ¼ P1 þ P2 and the
geometry of the electrodes that form the photoconductor. The voltage dropping on
the photoconductance, v; is described with the help of the equivalent circuit and is
given by the equation
C dv=dt ¼ ðVb
vÞ=Z
GðtÞv: ð28Þ
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 27

+ -
+
λ1 λ2

-
(a) (b)

λ1 λ2

λ1 λ2
+ -

(d)
λ1 λ2
(c)
metal
LT-GaAs
Fig. 18. Discrete-element photomixer geometries: (a) MSM photoconductor with dipole antenna, (b) gap
photoconductor with dipole antenna, (c) MSM photoconductor with CPW double slot antenna, and
(d) MSM photoconductor with bowtie antenna.

Z A = R A + iX A

Vb G (t)
C

Fig. 19. Equivalent circuit of a discrete-element photomixer.

 ARTICLE IN PRESS
28 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

Solving this equation with the assumptions that v has a sinusoidal variation, that
P1 ¼ P2 ¼ P0 =2 and that the antenna impedance is resistive, i.e. ZA ¼ RA ; we obtain
the power radiated at THz frequencies as
ðG0 Vb Þ2 RA
PTHz ¼ : ð29Þ
2½1 þ ðoTHz tÞ2 ½1 þ ðoTHz RA CÞ2 
The formula above provides guidance for increasing the output power of THz
photomixers as much as possible. According to it, in the limit of ultrahigh
frequencies, where otb1 and oRA Cb1; the photomixer that acts as a current
source, i.e. for which G0 RA 51; must have a low capacitance C and a very low value
of the carrier lifetime t: To this end photoconductor electrodes with a small area
must be used. For example, in the case of MSM interdigited electrodes an area of
8
8 mm with a 1:8 mm gap between two consecutive fingers was used, the fingers
having a width of only 0:2 mm: A typical value for C; valid for any type of electrode
geometry, is 0.5–1:5 fF; while G0 ¼ 2
10
5 mho: Eq. (29) suggests also that an
antenna with a high RA will produce high THz output powers. Therefore, there are
many photomixers based on dipoles, dual dipoles or dual-slot antennas, as well as
bowtie antennas, which show radiation resistances of about 300 O when properly
biased and working near resonance [61–64].
The THz output power can be written also as
RA Z2 l1 l2 ðe=hcÞ2 P1 P2
PTHz ¼ ; ð30Þ
2½1 þ ðoTHz tÞ2 ½1 þ ð1 þ oTHz RA CÞ2 
where Z is the external quantum efficiency. This expression apparently indicates that
a higher THz power could be obtained if the power of the optical sources is
increased. However, it was observed that beyond a certain optical power (tens of
mW) the photomixer is destroyed due to thermal failures [65]. Despite all the efforts
the THz power obtained using photomixers is still very low: about 1 mW at 1 THz
and 0:2 mW at 2 THz:
The reason is that the external quantum efficiency is very low, which means that
the optical heterodyne process has a very poor efficiency. The external efficiency can
be written as Z ¼ gA where g is the number of electrons induced in the antenna per
absorbed photon (photoconductive gain) and A is the fraction of incident power
absorbed in the photoconductive substrate. Extensive numerical calculations have
shown that the optical intensity inside the photomixer is smaller than one-half of the
incident intensity and the majority of the photocarriers are generated deep in the
photoconductor, where g is low [66]. In this situation, the external efficiency is about
0.008. An increase of Z can be achieved if the thickness of the LT-GaAs layer is
smaller than 1=a; where a is the absorption coefficient at the optical field wavelength,
and is backed by a mirror. An optimal photomixer structure consists then from a
0.35–0:5 mm thick LT-GaAs layer, followed by a buried 2:5 mm thick AlAs layer,
which enhances thermal dissipation since the thermal conductivity of AlAs is two
times larger than that of LT-GaAs, and ends with a DBR AlAs=Al0:05 Ga0:95 As
mirror having 2–3 periods. All these substrates form an optical cavity between the
top of the mirror and the top of the LT-GaAs. Even employing this optimal
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 29

structure, in which Z is increased 3 times and thus the output power is enhanced by a
factor of 9, the THz power does not exceed a few mW:
Lowering the carrier lifetime could be another way to increase the output power of
a photomixer. In this respect, self-assembled ErAs islands in GaAs were used to
build a heterostructure consisting of alternative layers of GaAs and ErAs islands
[67]. Although the carrier lifetime was reduced to 0:1 fs; i.e. 2.5 times lower than in
the LT-GaAs, the output power was still only 0:1 mW at 1 THz before the thermal
failure of the device.
New concepts of photomixers are developing using, for example, asymmetric p–i–
n–p–i–n heterostructures consisting of consecutive d-doped n and p GaAs layers and
incorporating ultrathin LT-GaAs or ErAs layers between the n and p layers [68]. In
this device the carriers move ballistically due to the very high electric field (of about
20 kV=cm) applied on the structure, which produces a sharp peak of the drift
velocities ð108 cm=sÞ in a short time interval of a few hundreds fs after which the drift
velocity takes its static value of about 107 cm=s: If the transit time through the device
has a similar duration the ballistic transport is assured and thus the thorny problem
of the carrier lifetime is eliminated. To reduce the transit time the length of this
device should be of only 200 nm but, assuming a cross-section of 5
5 mm2 ; the
device is expected to deliver 0:1 mW at 1 THz for an optical power of 10 mW:
A vertically integrated photomixer was realized recently [69]. A thin layer of LT-
GaAs with a thickness of 1:8 mm is sandwiched between two metal plates. These
metal plates are connected via a semitransparent pad and a buried contact to the two
metallic arms of an antenna, which have a spiral shape and are located at top and
bottom surface of the photomixer, respectively. This MSM-like photomixer requires
for its implementation only 1 mm scale standard photolithography techniques, while
the interdigited MSM photomixers described above, with fingers with submicron
dimensions, can be realized using only electron-beam lithography. The MSM-like
photomixer has demonstrated an output power of 0:5 mW at 1 THz and a
responsivity of 0:04 A=W at a bias of 8V.
Discrete-element photomixers have small active areas not exceeding 10
10 mm2 ;
carriers with low lifetimes and electrodes with small capacitances and sub-micron
electrode gaps, which provide high photocurrents. These characteristics lower the
output power. In contrast, traveling-wave photomixers use much larger active areas,
of the order of 103 mm2 (see Fig. 20a and b). When the second laser beam with a
frequency f2 is superimposed on the same spot as the first laser beam, which has a
frequency f1 > f2 ; an interference fringe pattern oscillating at f ¼ f1
f2 is produced.
The spatial distribution of the photocarriers has the same shape as the interference
pattern and the output power reaches a maximum value when the velocity of the
optical interference fringes ðvopt Þ equals the group velocity ðvTHz Þ of the THz signal
(the photocurrent) [70]. In the case of the photomixer displayed in Fig. 20
vopt ¼ cð f1
f2 Þ=ð f1 sin y1
f2 sin y2 Þ; ð31Þ
and
vTHz ¼ c=½ð1 þ er Þ=21=2 : ð32Þ
 ARTICLE IN PRESS
30 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

discrete-element photomixer
illumination

travelling-wave
photomixer
illumination
LT-GaAs

(a)

f1 f2

θ1 θ2

(b) LT-GaAs 1.5 µm thick

Fig. 20. Traveling-wave photomixer: (a) the basic concept, and (b) the position of the two lasers.

It is obvious that the equality between the two speeds can be easily accomplished by
tuning the angle of incidence of one laser. This type of photomixer produces powers
larger than 10 mW:
The above photomixer was realized on a thick substrate, the LT-GaAs being
grown on a semi-insulating GaAs substrate, and thus a lot of power was lost in the
dielectric substrate due to unwanted radiation. An optimized version of this
traveling-wave photomixer, put forward recently in Ref. [71], employs a CPW line
(instead of a CPS) and is terminated with a double slot antenna (in the place of the
dipole), both supported on a very thin micromachined LT-GaAs membrane of
1:5 mm and separated by an air substrate from a metallic area acting as reflector and
ground plane.
A last but very important problem of photomixers is the synthesis of the difference
of the optical frequencies. This is done with two CW semiconductor lasers, which are
phase-locked, one of the lasers being tunable. Practically, most of the references in
this paragraph indicate the experimental laser configuration used to synthesize the
desired frequencies that excite the photomixer. For example, a very reliable and
relatively easy way to implement the scheme able to synthesize precisely the required
difference of frequencies is presented in Ref. [70]. An optical difference frequency
synthesizer up to 3:17 THz; very stable and with a very good signal-to-noise ratio,
based on optical combs, is described in [72]. A review focused on various solutions
for optical difference frequency generation can be found in Ref. [73].
The synthesis of optical difference frequencies is generally implemented with quite
complicated setups. A simpler solution is the utilization of the light produced by a
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 31

commercially available multimode laser diode modulated in intensity by the beat

frequency between cavity modes [74]. The best solution is a single optical device–a
coupled-cavity vertical emitting semiconductor laser (VCSEL)—in which the optical
frequencies f1 and f2 are separately generated in each of the coupled cavities and the
optical beating is directly produced in the two-mode operation of the VCSEL. An
additional advantage is that, in this case, the optical difference frequency is
independent of any thermal drift [75].

2.2.2. Narrowband THz generation using optical parametric conversion

Continous tunable CW THz frequencies can be obtained by parametric light
scattering from the stimulated polariton scattering in nonlinear crystals. Optical
nonlinear crystals such as LiNbO3 or MgO doped LiNbO3 produce stimulated
polariton scattering when they are strongly pumped with a ns pulsed laser in the
near-infrared region ðl ¼ 1:064 mmÞ that has a repetition rate of several Hz and a
pulse energy in the range 20–50 mJ=pulse: The pump wave with frequency oP
generates an idler wave with a different frequency, oI ; when it excites a cavity formed
by the nonlinear crystal positioned between two mirrors. On its turn, the idler wave
is beating with the pump wave and thus generates a THz wave according to the law
oP ¼ oI þ oTHz (see Fig. 21). The THz wave is outcoupled from the cavity by a Si
prism. This parametric process is possible because polaritons behave like photons in
their spectral low-frequency region, including the THz region. In stimulated
processes, the momentum is conserved; this requirement imposes the phase-matching
condition kP ¼ kI þ kTHZ ; which indicates that the THz frequency can be tuned
changing the angle of incidence, yin ; of the pump.
The envelope of the THz signal generated in the way indicated above is a pulse
with a duration of 3–4 ns; so that the THz signal oscillating at a ps scale can be
viewed as a CW source with a large range of tunability (0.7–3 THz) and high peak-
powers (100 mW). The conversion efficiency is with 3–4 orders of magnitudes greater
than that obtained using the photomixing method. However, since the pump is a
bulky Q-switch Nd:YAG laser and since the distance between the mirrors of the
optical cavity is 15 cm (only the nonlinear crystal is 6:5 cm long) this THz source is
not miniaturized, but fits on a tabletop. Extensive reviews of this method of THz
generation can be found in Refs. [76,77].

E THz
Si prism

EI

θin
EP
pump LiNbO 3 mirror
mirror
Fig. 21. Optical parametric generation of THz radiation.
 ARTICLE IN PRESS
32 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

2.2.3. Narrowband THz generation using electronic devices

Presently, there is no single electronic device able to oscillate in the bandwidth
1–3 THz: Only resonant tunneling diodes were able to oscillate around 700 GHz;
other microwave and millimeterwave active devices, such as Gunn diodes or Impatt
diodes, being not able to exceed oscillating frequencies beyond 400–500 GHz: An
updated review of these devices can be found in Ref. [78]. InP Gunn oscillators are
able to generate 30 mW at 193 GHz; 3 mW at 300 GHz and more than 1 mW at
315 GHz; while a GaAs tunnel injection transit time diode (TUNNETT) produces
10 mW at 202 GHz [79]. Very recently it was theoretically shown that a unipolar
TUNNETT heterostructure can oscillate in the THz range when the electrons are
injected through tunneling of a square barrier into a very short (50–100 nm) transit
space where they are ballistically transported to an anode. The anode is made from a
material that allows no reflection and no backscattering of incoming ballistic
electrons [80].
Therefore, multiplication circuits are used to generate THz frequencies. A
multiplier consists of a nonlinear electronic device, such as a Schottky varactor diode
or a HBV diode, placed between an input and an output-matching network. HBV
diodes made on a gold substrate are very appealing for multiplication purposes [81].
The gold substrate offers mechanical stability and is in the same time a heat sink for
the device. The literature dedicated to millimeterwave multipliers is huge and, due to
this reason, these circuits will not be described in detail here. In this respect, the
reader is advised to read two extensive reviews: Refs. [82,83].
The output frequency of a multiplication circuit can be designed, using specific
rules, to be a multiple of its input frequency (pump): fout ¼ mfin : Unfortunately, the
output power is much lower than that of the pump, which is a serious drawback for
the THz frequency range. A pump with a power of 200–300 mW at 100 GHz can be
produced by HEMT amplifiers, but a multiplier with a high-order of multiplication
from 100 GHz up to 1–3 THz is not feasible due to the very high losses. Much lower
losses are achievable only in multipliers with a low-order of multiplication, i.e.
doublers (
2) and triplers (
3), so that a THz multiplier could consist of a sequence
of doublers and triplers of the frequency up to the desired THz frequency.
The main problem encountered in any THz electronic circuit are the dielectric
losses in the semiconductor substrate, which supports the diodes and the metallic
circuitry necessary for matching or propagating THz waves. The solution is the
thinning of the semiconductor substrate up to a thickness of 1–3 mm using MEMS
technologies [84]. Only in this way the output power of THz multipliers can be
substantially increased. In this respect, planar THz multipliers were realized using
two techniques displayed schematically in Figs. 22 and 23. The entire passive
network of the planar multipliers was made either on a thin membrane of GaAs
(3 mm thickness) supported on a waveguide block or suspended in air by removing
entirely the substrate under the metallic circuitry (‘‘substrateless’’ technique).
Impressive results were achieved using these MEMS techniques utilized in the
micromachining of semiconductor substrates. For example, 400 GHz and 800 GHz
doublers with a few mW output power and efficiencies of 15–20% were realized
using the substrateless technique. The first planar Schottky multiplier working
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 33

waveguide block
Schottky
varactor diodes

THz

input

metallic matching
GaAs membrane network
Fig. 22. Part of a THz multiplier illustrating the thin membrane technique used for passive circuitry.

GaAs frame
Schottky
varactor diodes
air THz

input

metallic matching network

Fig. 23. Part of a THz multiplier illustrating the ‘‘substrateless’’ technique for passive circuitry (suspended
in the air).

beyond 1 THz was realized using the membrane technique. At 1:2 THz this device
produces 80 mW output power at room temperature, 200 mW at 120 K; and 250 mW
at 50 K: A tripler at 2:7 GHz with a 1 mW output power was fabricated using the
same technique [85,86]. THz multipliers will play a major role in future THz
technology since they are compact and show a high-degree of integrability.
Although, the planar THz multipliers using MEMS techniques are 2–3 years old,
they already surpass the performances of some optical techniques involved in THz
generation.

2.2.4. THz generation using semiconductor lasers, masers, tasers

The subject of this section is a hot topic in the area of THz fields, being boosted by
the need of a single miniaturized device working at these frequencies and able to
provide a few milliwatt of THz power at room temperature. Although there are an
impressive number of proposals for THz semiconductor lasers based on very
complicated calculations and simulations, there are very few experimental results
regarding their performances. THz lasers can be found in literature under various
names such as masers or tasers, all of these devices having in common the inversion
of population between two or more discrete energy levels and the generation of THz
stimulated emission.
 ARTICLE IN PRESS
34 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

The available experimental results show that, despite many efforts paid in the last
two decades to develop a THz semiconductor laser, all the existing THz
semiconductor lasers work well only at low temperatures. However, there are some
important results obtained in the last years, which show that a THz semiconductor
laser working efficiently at room temperature could be realized in the next years.
The first THz semiconductor laser was a single crystal of p-doped Ge placed
between two mirrors. The dimensions of the rectangular parallelepiped Ge crystal
are about 5
7
50 mm3 [87]. The THz field is produced due to inversion of the
hole population between the LH and HH bands induced by perpendicular electric
and magnetic fields, the amplitudes of which are in a ratio of about jE=Hj ¼
1:5 kV=cmT: The electric and magnetic fields accelerate the heavy holes above the
optical phonon energies; part of them are scattered in the LH band where light-holes
are accumulated on closed paths just below optical photon energy. The inversion of
the population produces THz radiation that can be tuned in the range 1–4 THz by
tuning the magnetic field. The p-Ge laser has a low efficiency and is cryogenically
cooled at 4–5 K:
The tremendous technological realizations in semiconductor heterostructures,
which allow the engineering of semiconductor bands and thus the engineering of the
heterostructure properties, have boosted a vigorous quest for the search of THz
lasers. The result of this quest is the quantum cascade laser, proposed in 1971 [88] as
a FIR radiation source. In a quantum cascade laser the light produced by one carrier
transition between two levels is amplified due to photon-assisted tunneling of a single
type of carriers in a sequence of coupled quantum wells (superlattice) that has a
staircase-like band energy. The number of amplification stages dictates the output
power. The practical implementation of this laser was achieved in 1994 (after 23
years!) at Bell Laboratories. This laser is very different compared to usual
semiconductor lasers. It is a unipolar laser where the carriers can be either electrons
or holes. Only in this way it is possible to use the transitions within the same band,
which can be either the conduction or the valence band. The discreteness of energy
levels, named subbands, inside the same band is a result of the spatial confinement of
carriers inside the heterostructure. The band energy of a cascade laser is presented in
Fig. 24.
The FIR radiation frequency is determined by the energy difference of subbands
between which radiative/lasing transitions occur. The energies of subbands are, in
turn, governed by the thickness of the semiconductor layer that plays the role of well

photons
hf
EF
hf

Fig. 24. Quantum cascade tunneling laser.

 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 35

for the carriers involved in the lasing transition. The radiative transitions can take
place between the excited state and the ground state of the same quantum well or
between discrete levels in two adjacent quantum wells, case in which the transition is
called oblique. In both cases, the applied field must align the lower and upper energy
levels of subsequent transitions such that the carriers can tunnel between adjacent
wells. Quantum tunneling is the fastest way to transfer carriers from one quantum
well to another with a very low scattering rate. In the structure in Fig. 24 the
quantum wells in which radiative transitions occur are separated by wells with a
single discrete energy level in resonance with the lower and upper energy levels in the
adjacent quantum wells; the quantum tunneling at this energy level is resonant.
Resonant tunneling is characterized by a transmission close to unity and by the
fastest tunneling time.
The inversion of population (in particular, electrons) between the subbands
involved in the radiative transition takes place inside every quantum well of the
quantum cascade structure, the population of the upper level increasing with respect
to that of the lower level due to the fast depletion of the lower level population
caused by tunneling of electrons to the upper level of the next quantum well. In this
way, like in a cascade, each electron generates (ideally) a number of photons equal to
the total number of quantum wells (or subsequent transitions).
Very recent reviews on the state of the art of quantum cascade lasers and their
applications can be found in Refs. [89,90]. An in-depth analysis of various types of
quantum cascade lasers and their modeling can be found in Ref. [91].
InGaAs/InAlAs/InP and GaAs/AlAs heterostructures were used to generate FIR
at room temperature in the wavelength range 17–90 mm: The first quantum cascade
laser working in the THz range was reported very recently [4]. The realization of a
quantum cascade laser at THz frequencies encounters a series of difficulties and
limitations due to the very large values of the wavelength. Among them are very
large free-carrier absorption losses and the necessity of growing a very thick
heterostructure. The 4:4 THz quantum laser mentioned above had 104 periods, each
period containing 7 coupled quantum wells, each quantum well having two AlGaAs
barriers (with a thickness of 1–4 nm) and one GaAs well (10–20 nm thick), resulting
in a total number of 728 quantum wells (!). It is not at all easy to manufacture such a
heterostructure. This laser delivers about 2 mW power at 4:4 THz and is operating at
50 K: The output power decreases dramatically with increasing temperature and
becomes nearly zero at room temperature. However, this THz laser is considered a
huge step forward towards THz miniaturized sources. A low threshold THz
quantum cascade laser was reported in Ref. [92]. It contains a three-quantum-well
chirped superlattice active region located inside a waveguide, the total hetero-
structure being 2:7 mm long. The threshold current of this THz laser is only
210 A=cm2 ; it operates at 66 mm and delivers 4 mW at 12 K: All these THz cascade
lasers are based on n-type carriers (electrons) and the photon emission is parallel to
the heterostructure plane (edge-emission).
The operating temperature of quantum cascade lasers can be increased by
replacing the InP or GaAs based heterostructure with the one based on Si. A THz
quantum cascade laser based on the Si/SiGe material system could attain room
 ARTICLE IN PRESS
36 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

temperature operation due to the absence of strong polar optical phonon scattering.
Such a THz laser is also unipolar and its frequency is determined by subband energy
spacing, but the carriers are p-type (holes) and the photons are emitted normal to the
heterostructure plane. Thus, a THz Si/SiGe laser is a THz VCSEL laser, which is
easily integrable with THz waveguides and can be easily arranged in 2D arrays.
There are some proposals of THz Si/SiGe lasers based on transitions between LH
and HH subbands using the inverted masses concept [93], or phonon pumping [94].
All lasers mentioned above are electrically pumped (the carriers are injected from
carrier reservoirs—leads—by applying a bias) but there are THz lasers, which are
optically pumped, the lasing being based on intersubband emissions in four-level
GaAs/AlGas asymmetric quantum wells. The optical pump source is a CO2 laser.
Although this type of THz laser is able to work at room temperature its main
drawback is the large dimensions of the optical pump source [95]. There are also
some THz masers proposals. A THz spin flip maser is based on the population
inversion between Zeeman-split levels, which flip their associated spin through
tunneling [96]. Another THz maser proposal is based on optical phonon transit time
resonance in bulk GaN [97]. Calculations show that these masers are able to work
efficiently at 30 K:

2.3. THz generation/detection using nanodevices

The involvement of nanotechnologies in the realization of THz generation/

detection devices is more than 10 years old. It started with the observation of the
emission of THz electromagnetic radiation from an asymmetric coupled quantum
well structure, which consists from a wide well (WW) and a narrow well (NW)
separated by a thin potential barrier [98]. Since the coupled quantum wells are not
identical, the interband transition frequencies have different values in the two wells
denoted by o1 for the WW and by o2 for the NW. The THz radiation is due to the
coherent oscillation of electrons between the two quantum wells. These oscillations
in the coupled quantum structure occur when the lower electronic levels in the WW
and the NW are aligned by applying a dc voltage; when alignment of the lower
electronic levels is achieved, i.e. at resonance, the electrons become delocalized. At
the bias value for which the resonance condition is attained for electrons, the LH and
HH lower energy levels in the coupled asymmetric wells remain generally misaligned,
so that the holes remain localized in the respective wells. In these conditions new
bonding and antibonding eigenstates are created at resonance, with corresponding
energies Eþ and E
; respectively. These HH–NW and HH–WW energy states show a
hyperbolic dependence on the applied field, intersect each other and have a minimum
splitting at resonance. All these physical properties allow the preparation of an
electronic wavefunction in the WW through the excitation of the coupled quantum
structure with a fs optical pulse that has a frequency equal to o1 and a spectral
content Do1 larger than the difference of antibonding and bonding eigenstates’
energies and smaller than the difference of interband frequencies in the two wells, i.e.
ðE

Eþ Þ=_oDo1 oðo2
o1 Þ: The electron wavepacket is in these conditions a
superposition of the two eigenstates and tunnels the two coupled quantum wells at
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 37

the frequency
oTHz ¼ ðE
þ Eþ Þ=_: ð33Þ
The oscillating wavepacket produces on its turn a time-varying polarization, which
can be associated with the electromagnetic radiation of a Hertzian dipole, i.e.
ETHz ðtÞ!@2 P=@t2 : In an asymmetric GaAs/AlGaAs/GaAs coupled quantum well
structure, oscillations at 1:5 THz were observed at 10 K by exciting the structure
with an optical pulse of 160 fs duration and a photon energy of 1:53 eV: The
generated THz frequency cannot be tuned by changing the bias, since such a change
will bring the aligned energy levels off resonance.
THz electromagnetic radiation can also be produced by charge oscillations in a
single quantum wells. In GaAs single quantum wells structures, for example,
separated by AlGaAs barriers, the charge oscillations originate in the quantum beats
resulting from the LH and HH excitons. In contrast with the case of coupled
quantum wells, in the case of single quantum wells the resulting THz frequency can
be tuned in the range 1–3 THz by changing the applied bias, which changes the
energy levels of excitons [99]. To observe THz due to charge oscillations the photon
energy must exceed both LH and HH levels of the GaAs quantum well.
These initial successes in the development of THz devices based on quantum
nanostructures were possible due to the amazing technological developments in the
area of band-engineered semiconductors. Other THz devices and effects based on
nanostructures can be imagined as well, one of the most beautiful being
the generation of THz Bloch oscillations. Bloch oscillations are the result of the
dynamics of electron wavepackets that propagate in periodic potentials in the
presence of an applied electric field. The electrons undergo a succession of carrier
accelerations due to the applied electric field, until the electron momentum satisfies
the Bragg reflection condition from the periodic component of the potential and is
reflected, followed by electron decelerations due to the electric field until it is
reflected by the linear component of the potential, completing the cycle. The
electrons oscillate at the Bloch oscillation frequency oB and are localized in periodic
orbits being Bragg reflected on one side and reflected by the uniform electric field on
the other. A quantum treatment of the above electron wavefunction dynamics shows
that Bloch oscillations are the result of the quantum beat of Wannier–Stark states.
Sixty-five years have passed between the theoretical discovery of Bloch oscillations
and the experimental proof of their existence. The reason is that in bulk solids
scattering phenomena destroy the coherence of Bloch oscillations. For the first time,
Bloch oscillations with a spectral width within the THz range were observed in a
superlattice consisting of 35 periods of a thick GaAs well ð9:7 nmÞ located between
two Al0:3 Ga0:7 As barriers ð1:7 nmÞ [100]. The entire superlattice structure is
presented in Fig. 25. At a low reverse bias, strong coupling between quantum wells
occurs and the electron and hole wavefunctions are delocalized producing energy
minibands. The lowest electron and hole miniband widths are 19 and 2 meV;
respectively. Increasing the voltage, the minibands split into a series of discrete levels
that form a Wannier–Stark ladder. The THz Bloch oscillations were generated by
exciting optically the superlattice with a 100 fs pulse at the wavelength of 802 nm;
 ARTICLE IN PRESS
38 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

THz field
ultrashort
buffer (Bloch oscillations)
optical pulse
Al0.3Ga0.7As

35 periods
GaAs

buffer

n-GaAs

Fig. 25. THz Bloch oscillator.

which corresponds to the 0hh exciton transition. Bloch oscillations were observed at
15 K in the linearly tunable range of 0.5–4 THz; according to the Bloch frequency
formula
oB ¼ eEd=_; ð34Þ
where E is the applied dc electrical field, and d is the period of the superlattice. A
review about optically induced Bloch oscillations tunable in the THz range can be
found in Ref. [101]. Very recently electrically induced Bloch oscillations were
theoretically predicted at THz frequencies through hot electron injection into the
upper part of a superlattice miniband [102]. This injection mechanism shifts the
region of negative conductivity towards higher frequencies, attaining 2–4 THz: In
this way, the dc differential conductivity is positive and therefore suppresses the
domain instability, while the high-frequency differential conductivity is negative
above the Bloch frequency oB :
Despite the inherent beauty of the physical effects involved in THz generation
based on charge oscillations, the devices described above show the same drawbacks
as many other THz generating devices; more precisely, the THz emitted power is low
even when the devices work at low temperatures. A possible solution to improve
dramatically the performances of THz devices is represented by transistors with
nanometer dimensions, named also nanotransistors or THz transistors.
The high electron mobility transistor (HEMT) represents a first category of such
transistors. The cutoff frequency of the HEMT, given by fT E1=2ptt ; where tt is the
electron transit time, does not exceed 0:6 THz: This cutoff frequency could be
enhanced towards 10 THz if the HEMT would work in a special operation mode
when electron plasma oscillations are propagating in the HEMT channel. In such a
channel electrons propagate ballistically and, when the sample length and the free
path of electrons resulted from collisions with impurities and/or phonons are both
greater than the electron free path resulting from collisions with other electrons, the
entire 2D electron gas existing in the HEMT channel behaves like a frictionless fluid
analogous with shallow water, which can be described by hydrodynamic equations
[103,104]. In these conditions plasma waves are produced with a dispersion relation
o ¼ vP k; where the plasma wave velocity vP ¼ ðeVG =meff Þ1=2 is of the order of
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 39

108 cm=s and VG is the gate-to-channel voltage swing. The entire HEMT acts like a
resonator for plasma waves at the resonance frequency vP =L; with L the channel
length; the quality factor of the resonator is given by Q ¼ vP t=L; where t is the
momentum relaxation time. When we have an ac short circuit at the HEMT source
end the eigenmodes of HEMT plasma oscillations are odd harmonics of the
fundamental plasma oscillation oosc ¼ pvp =2L: Many THz devices working at room
temperature such as resonant mixers, detectors or oscillators can be implemented
using the principles described above. For example, for gate lengths within the
interval 60–100 nm a HEMT THz oscillator is able to generate 1 mW within the THz
range. A HEMT detector working in the plasma regime can exceed with 2–3 orders
of magnitudes the responsivities of Schottky detectors, which are about 103 V=W:
THz resonant mixers and multipliers can also be implemented.
Very recently, it was shown that self-oscillations of plasma could take place in
HEMT-like heterostructures with tunneling or thermionic injection of electrons from
the channel into the gate layer [105]. The plasma instability is possible due to the
combination of plasma resonances with transit-time resonances of injected carriers in
the gate layers. In this case the plasma dispersion equation is different from that
given above, but plasma oscillations between source and drain of a AlGaAs/GaAs
HEMT-like structure are still within the THz range when the gate length is of the
order of 500 nm at room temperature.
Moreover, simple devices such as T-branch and rectifier diodes can be
implemented and realized using the same concept of ballistic channels in AlInAs/
InGaAs HEMT devices, where the ballistic transport is present at room temperature
due to the extremely long electron free path, which is of about 100 nm at 300 K: In
this way, multiplexors/demultiplexors are able to process THz signals [106].
A THz transistor able to switch 1012 times per second is now under development
by Intel. It is a Si-based FET transistor that incorporates some new features
compared to a classical FET transistor. For example, it has a new gate dielectric
placed under the gate, a layer of oxide buried in the silicon substrate and a raised
source and drain. The gate length is between 60 nm and 90 nm; i.e. it has a thickness
of 3–5 atomic layers. For more details see the paper of Teixeira and other papers
written by the Intel team at www.intel.com/research/silicon.
A room temperature THz transistor was recently realized [2]. This is a ballistic and
tunneling GaAs static transistor with a 10 nm scale channel. The measured electron
transit time was 2
10
14 s; i.e. 20 fs! The cutoff frequency is thus about 9 THz:
THz generation at very low powers was observed when two CW lasers emitting at
different frequencies, such that the difference between them is located within the THz
range, were focalized in the air-gap of a single tunneling microscope (STM) that
consists of a sharp nanotip positioned a few nanometers above a metal ground. This
is in fact a nanophotomixer (see Fig. 26), emitting THz radiation due to the nonlinear
characteristics of the STM-like diode, which is analogous to a metal–insulator-metal
(MIM) diode [107]. Such sharp nanotips can be realized using micromachining
techniques applied to Si or GaAs; then, the nanotips are metallized. In connection
with THz emission by a STM, it was theoretically demonstrated that resonant
photon-stimulated field emission is able to generate signals up to 100 THz [108].
 ARTICLE IN PRESS
40 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

lasers THz
f2
I0
f1

metal
Fig. 26. THz nanophotomixer.

The latest trend in THz nanodevices is their implementation with the help of
carbon nanotubes (CNT). Depending on the wrapping of a graphene sheet CNTs
can be either metals or semiconductors with a bandgap that can be engineered by
choosing a certain diameter of the tube, by applying a stress on the nanotube or by
doping it. CNTs with a diameter varying between 10 and 50 nm could be the
building blocks of many future THz devices. Schottky diodes and negative
differential conductivity diodes with unprecedented performances compared to
similar devices based on bare semiconductors or heterostructures were already
simulated [109–111].
CNTs were used as field emitters generating current densities of 100–1000 A=cm2
in a nanoklystron device designed to produce 3 mW at 1:2 THz [112]. An array of
CNTs is a very efficient electron source for low-power and high-current densities.
The entire reflex nanoklystron is made using Si micromachining techniques and is
composed from two parts, which are sealed in vacuum. The entire vacuum microtube
is very small; its dimensions do not exceed 100 mm: We have presented in Fig. 27 the
THz devices based on CNTs described above.
A single-walled CNT (SWCNT)-based RTD able to overcome the maximum
oscillation frequency attainable with RTD based on semiconductor heterostructures
is represented in Fig. 28a. The SWCNT is a functional device allowing the creation
of rectangular quantum wells beneath dc biased metallic gates whereas rectangular
barriers are produced between them. The widths of the barriers given by the distance
between the gate electrodes is taken as 10 A ( and the well width determined by the
(
length of the inner gate contact is 20 A: The device described above is simulated with
the method described in Ref. [113]. The dependence of the transmission coefficient
and the transit time of the RTD on E are displayed in Fig. 28b for a (17,0) SWCNT
with an effective mass of 0:216m0 ; where m0 is the free electron mass, and for
V1 ¼ V2 ¼ V3 ¼ 0:8 eV; considering the reference for the electron energy at the
bottom of the well. The transmission coefficient takes significant values only around
the two resonant energies of the quantum well: 0.193 and 0:72 eV: The transit times,
however, for these two resonant energies are very different, 1:5
10
13 and 1:35

10
14 s; respectively. To have a very short transit time through the device, the
contribution of the electrons tunneling through the lowest resonant state must be
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 41

semiconductor CNT
V/2
metal 1
semiconductor
CNT
dielectric
metallic
cylinder metal
CNT
metal 2

(a ) -V/2 (b)

100 µm
reflector reflector
bias
beam THz
bias buncing grids output
resonant vacuum
cavity

CNT array grid bias

(c )

Fig. 27. THz CNT devices: (a) Schottky diode, (b) negative differential resistance diode, and
(c) nanoklystron with a CNT array cathode.

suppressed. This cannot be done by controlling the position of the Fermi level in the
emitter, for example, because enabling the electrons to reach the highest resonant
level automatically implies that the electron states around the lowest resonant levels
are occupied, and thus tunneling on the lowest resonant level automatically takes
place. The only possibility to avoid this is to engineer the value of V1 such that the
electron states around the lowest resonant level in the well cannot be reached by
electrons in the emitter region. The simplest way to do this is to raise the bottom of
the conduction band in the emitter region above the lowest resonant level in the well.
Considering again the energy reference at the bottom of the quantum well, the
bottom of the conduction band in the emitter region can be raised, for example, with
0:35 eV above zero. The transmission and transit time dependences on E are in this
case given in Fig. 28c for the situation when the bias along the SWCNT axis is absent
(solid lines) or takes the value VDC ¼ 0:5 V (dashed lines); it was assumed that VDC
drops mainly across the RTD. From Fig. 28c it can be seen that the position of the
highest resonant state in the quantum well shifts to lower energies when the SWCNT
is biased, and the transit time at the position of the resonant state shifts also to lower
values, i.e. it becomes even faster than 1:35
10
14 s; attaining the value of 10
14 s:
A transit time of 10 fs; as that attained in this biased device, corresponds to a cutoff
oscillation frequency of about 16 THz; two times higher than the cutoff frequency of
 ARTICLE IN PRESS
42 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

Fig. 28. (a) Schematic representation of a SWCNT-based RTD and its energy diagram. (b) Transmission
coefficient and transit time energy dependence of a SWCNT-based RTD with V1 ¼ V2 ¼ V3 ¼ 0:8 eV and
no applied bias. (c) Same as in (b) for V1 ¼ 0:45 eV and V2 ¼ V3 ¼ 0:8 eV (solid line: no bias, dashed line:
VDC ¼ 0:5 eV). (d) Room temperature I
VDC characteristics for the structure in (c) with Fermi energy
levels at 0:6 eV (dotted line), 0:65 eV (solid line), 0:7 eV (dashed line) and 0:75 eV (dashed-dotted line)
measured from the bottom of the quantum well.

the THz ballistic transistor, which has a transit time of 20 fs [2]. The I
VDC
Rdependence of the device was computed using the Landauer formula IðV Þ ¼
TðEÞ½ fL ðEÞ
fR ðEÞ dE; where TðEÞ is the transmission coefficient and fL ðEÞ and
fR ðEÞ are the Fermi distributions of the left and right contact electrodes. The
I
VDC dependence is displayed in Fig. 28d for four values of the Fermi level:
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 43

0:6 eV (dotted line), 0:65 eV (solid line), 0:7 eV (dashed line) and 0:75 eV (dashed-
dotted line), at room temperature. These values of the Fermi energy level are
calculated from the bottom of the conduction band in the well region; the value of
0:35 eV must be subtracted to obtain the values with respect to the bottom of the
conduction band in the emitter region. A very pronounced region of negative
differential resistance can be observed in all cases, assuring that oscillations with
THz frequencies are produced along the SWCNT axis. An efficient THz oscillator
should have a small negative differential conductance, i.e. a small jDI=DV j ¼
jðIp
Iv Þ=ðVp
Vv Þj; where Ip ; Iv ; Vp ; Vv are the peak and valley values of the
intensity and voltage, respectively, a high DIDV ¼ ðIp
Iv ÞðVp
Vv Þ; i.e. a high
output power, and a high Ip =Iv for a high signal-to-noise ratio. These conditions are
contradictory. Therefore, the Fermi level can be chosen at 0:65 eV (0:3 eV above the
bottom of the conduction band in the emitter region), for which Ip =Iv ¼ 2:23:
The easiest way to couple the generated THz frequency to the propagation media
is through quasi-optically means. Therefore, the metallic electrodes through which
the voltage VDC is applied throughout the structure have the shape of bowtie
antennas, the bias being applied on the two arms of the antenna. These antennas
radiate the THz power into free-space if VDC is located in the negative differential
resistance region of the I
VDC characteristics. The bias value of 0:5 V used in the
simulations of Fig. 28c lies within the negative differential region of the SWCNT-
based RTD. The condition of oscillation of the SWCNT-based RTD is RN þ RC þ
RA o0; where RN is the negative resistance of the RTD, RC is the contact resistance
and RA is the radiation resistance of the bowtie antenna. This condition is satisfied
for RC ¼ 6 kO and RA ¼ 80 O; and for a value RN ¼
99:2 kO determined
from Fig. 28d for the Fermi level value of 0:65 eV: Thus, the device can indeed oscil-
late at a cutoff (intrinsic) frequency of oscillations of fRC ¼ ½G=ðRC þ RA Þ

G 2 1=2 =ð2pCRTD Þ; where CRTD is the capacitance of the SWCNT-based RTD. This
capacitance is CRTD ¼ 100 aF=mm; which for a 4 nm long structure gives 4 aF. In
these conditions fRC ¼ 16:4 THz; a value that is very close to the oscillation
frequency estimated from the transit time computation. The output power of the
SWCNT-based RTD oscillator, equal to DIDV ¼ ðIp
Iv ÞðVp
Vv Þ; is found to be
2:5 mW using the data from Fig. 28d. In order to increase the THz power an array of
equally spaced SWCNT-based RTD devices can be implemented, where the distance
between two antennas loaded with SWCNT-based RTD is c=2fRC : Then, the
radiated power in the directions for which the interference between the radiations
emitted by the antennas is constructive increases proportional to N 2 ; where N is the
number of the radiating elements. So, an array containing tens of bowtie antennas
loaded with SWCNT-based RTD can produce an output power of a few milliwatts
in the frequency range of 1–16 THz:

3. THz propagation

There are two distinct ways to propagate THz fields: (i) quasi-optical techniques
and (ii) guided-wave techniques. The most encountered is the quasi-optical
 ARTICLE IN PRESS
44 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

propagation, which consists in the transmission of THz fields in free-space between

the emission and reception planes. This is possible when the electromagnetic beam is
collimated and is not seriously distorted by diffraction. The quasi-optical method
was for many years the only solution for THz field propagation since guided-wave
methods based on metallic or dielectric waveguides have major drawbacks in the
submillimeter frequency range. The waveguides have very high losses, which increase
with ðfrequencyÞ2 and require extremely tight mechanical tolerances. In a metallic
rectangular waveguide the losses at 1 THz are about 0:4 dB=cm: The large majority
of dielectrics, with the notable exception of Si, have large absorption coefficients in
the THz range. Only recently the micromachining techniques were able to realize
quite thin substrates on which THz propagation experiences losses comparable to
those obtained in quasi-optical techniques.
Therefore, quasi-optical techniques are very spread in the THz range and a THz
setup has many similarities with an optical setup. THz quasi-optical techniques are
best described by geometrical optics and paraxial approximation. Since the
amplitude distribution transverse to the direction of propagation of many
transmitting/receiving THz antennas shows a Gaussian shape, the Gaussian optics
based on the ABCD matrix calculation is applicable to THz quasi-optical
techniques. A review of quasi-optical techniques can be found in Ref. [114], while
the Gaussian optics is very well explained in Ref. [115]. A 2
2 ABCD matrix is
assigned to each quasi-optical component (lens, filter, plate, etc.) and to each free-
space distance separating them. Then, the product of all ABCD matrices provides
the transfer function of the entire quasi-optical system, thus allowing a complete
description of the propagated beam properties. There are even analytical techniques
able to extend this method in the case of non-paraxial THz beams, i.e. beams with a
divergence generally exceeding 190 ; using beam-mode transfer matrices [116].
Despite the similarities with optics, there are some specific quasi-optical THz
devices used for beam shaping and processing that can differ considerably from their
optical counterparts. For example, wire grids are used as polarizers able to rotate
significantly the polarization plane of an incoming THz beam without any
distortion. Wave dielectric plates of various thicknesses are used as phase shifters.
Some quasi-optical components are displayed in Fig. 29 together with their
transmission functions. THz filters are represented in Figs. 29a–c, while in Fig. 29d a
Fabry–Perot cavity is displayed, which can also play the role of a filter. All these
quasi-optical components are based on 2D periodic arrays of metallic wires, metallic
plates or slots; using a modern terminology all are photonic bandgap devices (PBG).
Simple formulas describe their basic parameters. For example, in the case of the
resonant wire grid filter depicted in Fig. 29c, the length of the cross-like slot is
L ¼ l0 =2:1; while the spatial period between two slots is P ¼ lD =ð1 þ sin yi Þ; with lD
the longest wavelength at which diffraction can occur and yi the incoming THz beam
angle. Its transmission characteristic has a Lorentzian shape given by
Tð f Þ ¼ a=½ð f
f0 Þ2 þ a2 ; ð35Þ
where f0 Dpc=L is the resonant frequency and a is the half-power width of the
resonance profile [117].
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 45

1
Transmission

f [T Hz]
(a)

1
Transmission

(b) f [T Hz]

Transmission

f [T H z]
(c)

Transmission

(d) f [THz]

metal
Fig. 29. Quasi-optical THz devices: (a) wire grid filter, (b) capacitive-plate filter, (c) resonant wire grid
band-pass filter, and (d) Fabry–Perot cavity.

Diplexers are multi-port quasi-optical devices, which combine two or more

signals, e.g. the incoming THz signal and a local oscillator. Martin–Puplett
polarization rotating interferometers are used for such purposes [57].
The THz waveguided propagation has boosted in the last years due to
micromachining techniques able to process semiconductor substrates with micro-
metric thicknesses. This is the only way to reduce drastically the large losses
encountered in any dielectric substrate at THz frequencies.
The micromachining consists in removing a part of a bulk substrate or of a thin
film using various etching techniques. An ideal candidate for micromachining is
silicon, which has very good mechanical properties and a low absorption at THz
frequencies. Since Si can be micromachined using usual etching technologies
 ARTICLE IN PRESS
46 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

borrowed from the semiconductor technology, Si micromachining is the most

frequent procedure utilized to obtain quasi-optical THz devices such as lenses,
antennas, or filters. The micromachining of GaAs or InP is also possible and
desirable for THz applications since III–V semiconductor compounds are employed
in THz devices such as photoconductor emitters/receivers and THz semiconductor
lasers. CPW on micromachined semiconductor substrates should be able to drive to
an antenna the emitted power from a THz semiconductor laser. An array of a few
THz semiconductor lasers connected between them by micromachined CPW lines
terminated with slot antennas will be able to emit an output power of tens of mW.
Micromachined techniques are intensively used in THz electronic devices as, for
example, the substrateless technique described in Section 2.2.3. III–V semiconduc-
tors or heterostructures based on them show piezoelectric, piezoresistance and
thermoelectric properties due to the polar character of the bonding between different
atoms in the unit cell. These properties could be further used for THz MEMS
devices.
The various etching shapes obtained by applying different techniques are
represented in Fig. 30. The etching techniques are divided into two large categories:
wet or chemical (presented in Figs. 30a and b) and dry or plasma etching (Fig. 30c).
The aim is to obtain controllable, high-precision shapes with low roughness after the
etching process. In the case of Si, GaAs or InP anisotropic etching and plasma
etching are used. Perfect vertical walls can only be obtained using plasma etching
methods. More sophisticated methods, based on laser etching or syncrotron
radiation (called LIGA) can be employed to realize perfect walls or various
semiconductors shapes.
There are two large types of micromachining techniques. In the first technique,
called bulk micromachining, the back of a wafer is etched in order to obtain a
prescribed mechanical structure on its top. A simple but very important example is
shown in Fig. 31: the fabrication of a micromechanical membrane. Nowadays, very
large membranes with surfaces of 2
2 cm2 and a thickness of 1 mm can be obtained
by using chemical or plasma-etching methods. In the case of bulk micromachining

isotropic
etching

(a) removed by
etching
anisotropic etching –
dependent on orientation

(b)

anisotropic etching –
independent of orientation

(c)
Fig. 30. Basic micromachined shapes using different etching techniques.
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 47

membrane

Si

anisotropic removed from

chemical etching the bulk
Fig. 31. Illustration of bulk micromachining.

Table 1
Etch stop techniques for chemical etching procedures

Name of the etch stop technique Description Accuracy

Time stop The etch is stopped after a prescribed time Low

pþ A thin pþ layer remains after removing Si High

in combination with EDP etchants

Electrochemical The thickness is obtained through epitaxy. Very high

(photovoltaic, galvanic, etc.) An etch stop is realized through a p-layer
grown on a n-type substrate

the precision and the desired shape of mechanical structures realized at the micro
or nano-scale is determined by the etching-stop techniques. Some etching-stop
techniques are summarized in Table 1.
The second micromachining technique is surface micromachining. In this
technique, MEMS or NEMS are realized by successive thin film depositions on a
common substrate, followed by the selective etching of one of these thin deposited
layers, called the sacrificial layer. The realization of a cantilever in this technique is
schematically represented in Fig. 32. In the Si technology the sacrificial layer can be
of polysilicon or different types of resists, while the mechanical layer (located over
the sacrificial layer) can be of silicon nitride, silicon dioxide, gold or aluminum. The
pair mechanical-sacrificial layers is selected in such a way that after a specific etching
the mechanical layer remains intact while the sacrificial layer vanishes completely.
The applications of micromachining techniques for THz domain started in 1980
when a tapered-rod antenna was realized for a wavelength of 120 mm by using
anisotropic etching of the silicon rod supported on a silicon dioxide membrane on
which a bismuth bolometer was patterned [118]. Many THz devices were
 ARTICLE IN PRESS
48 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

sacrificial layer cantilever layer

removed by
selective etching

substrate

Fig. 32. Illustration of the surface micromachining technique.

movable metal CPW

plate

dielectric
coating

Fig. 33. MEMS sliding planar backshort.

subsequently realized: a slot endfire Vivaldi-like antenna at 802 GHz on a 1:75 mm

silicon nitride membrane, a 16
16 THz horn array forming a CCD-like imaging
array at 802 GHz [18], waveguides and mixers [8,9] multipliers and receivers [84].
There are also two other reviews of THz micromachined propagating and radiating
devices [119,120] such as waveguides, lenses, filters, CPWs and other transmission
lines, as well as antenna arrays, which are very useful in the understanding of why
micromachining techniques are so attractive for THz technologies.
Moreover, even movable microfabricated translators able to tune THz transmis-
sion lines were realized using surface micromachining techniques [120]. An example
of such a movable translator is a sliding planar backshort, which consists of a
rectangular metal plate that translates linearly along a CPW transmission line,
varying in this way its electrical length (see Fig. 33). This MEMS device was used to
tune a 620 GHz detector circuit.
A new trend in propagation of THz fields is the guiding of THz pulses. We have to
point out that all THz devices described above are suitable only for CW THz fields.
Recently, however, THz pulse propagation in single-mode waveguide sapphire fibers
with a diameter of 325 mm was experimentally tested [121]. Similar experiments were
performed in 240 mm diameter cylindrical stainless steel waveguides and in parallel-
plate copper waveguides [122,123]. In the case of sapphire fibers, THz pulses were
reshaped due to the dispersion and absorption encountered in the dielectric
waveguide. Very low-loss, but very dispersive THz propagation was observed in the
case of the THz metallic cylindrical waveguide. For the first time, dispersionless, and
low-loss propagation of an incoming 0:3 ps THz pulse was observed in the
bandwidth 0.1–4 THz for a length of 25 mm via parallel-plate waveguide.
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 49

4. THz detection

THz detection is a quite difficult task since the power of emitted THz signals is
weak. Moreover, due to the low photon energies of the THz band (1–10 meV)
ambient thermal noise prevails over the THz signal and thus cooling of detectors is
very often required. Other detection configurations that increase the signal-to-noise
ratio, such as the heterodyne detection, are frequently used either at room
temperature or cooled.

4.1. Detection of ultrashort electrical pulses

Ultrashort electrical pulses with a spectral content within the THz frequency
range, produced using various physical principles (see Section 2.1), are detected using
mainly two methods. The first method uses a gated photoconductive antenna and is
called photoconductive sampling or, briefly, PC sampling. The second method is
based on the detection of the polarization change of an optical probe beam produced
by the THz field when both fields are applied on an electro-optic crystal; this method
is called free-space-electro-optic sampling FS-EOS. The configurations of both
detection schemes are presented in Fig. 34.
In the case of PC sampling the charge generated at the antenna terminals is [124]
Z
qðtÞ ¼ vðtÞgðt
tÞ dt; ð36Þ

where vðtÞ is the voltage across the photoconductive gap, given by

Z
vðtÞ ¼ HðoÞEðoÞ expðiotÞ do: ð37Þ

Here EðoÞ is the Fourier transform of the incident electric field pulse ETHz ðtÞ and
HðoÞ is the transfer function of the antenna, i.e. the ratio between the voltage
induced at the antenna terminals and the incident electric field, both represented in

PC x
incoming THz antenna incoming electro-optic
probe (gate) z
signal crystal
I(t-τ ) THz signal
controlled by controlled by y prism
the pump I(t) the pump I(t)

mA
λ /4

polarizer

probe (gate)
to lock-in
(a) (b) I(t-τ)
amplifier
Fig. 34. Ultrashort electrical pulse detection. (a) PC-sampling, (b) FS-EOS.
 ARTICLE IN PRESS
50 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

the frequency domain. The conductance is given by

Z
gðtÞ ¼ Iðt0 Þf1
exp½1
expðt
t0 Þ=trel g exp½ðt
t0 Þ=t dt0 : ð38Þ

From Eq. (38) it follows that the PC sampling output signal is dependent on the
incident field ETHz ðtÞ; but also depends on the momentum relaxation time trel and the
carrier lifetime t of the PC substrate. It was experimentally found that when THz
detection is performed with a short dipole without a substrate lens [124], HðioÞ ¼ 1;
and so vðtÞ becomes directly proportional to the incident THz signal ETHz ðtÞ: When
the THz detector consists of a short dipole with a substrate lens, HðioÞ ¼ io:
The FS-EOS uses the linear electro-optic effect in an EO crystal excited by an
optical probe field and the THz field. Both fields propagate in the same direction but
have different polarizations. For example, if z is the propagation direction, the
optical probe is polarized at 45 in the ðx; yÞ plane perpendicular on z due to
birefringence of the EO crystal, while the THz field is perpendicular on the y axis.
Since the electro-optic effect is practically instantaneous at the THz scale, the output
of a FS-EOS detector is directly proportional to ETHz ðtÞ [125]. Due to the presence of
the THz field a phase retardation Dj of the optical field is produced over the distance
dz; which is strongly dependent on the electro-optic crystal type and orientation. FS-
EOS uses different types of electro-optic crystals: (i) uniaxial crystals like LaTiO3 or
LiNbO3 ; or (ii) isotropic crystals like (1 1 0) ZnTe with a zinc-blende structure. ZnTe
is a material for which a high signal-to-noise ratio was obtained. For ZnTe the phase
retardation is given by

DjðtÞ ¼ ðo=cÞn30 r41 ETHz ðtÞ dz ¼ constZnTe

ETHz ðtÞ dz; ð39Þ

where o is the optical frequency of the probe and r41 is the electro-optic coefficient.
From the above relation it follows that the THz field obtained after propagating over
a length L in a ZnTe crystal, material with small absorption and a refractive index
difference of Dn ¼ nTHz
nopt ¼ 0:22; is

ETHz ðtÞ ¼ DjðtÞ=ðL

constZnTe Þ: ð40Þ

Thus, measuring the phase change we are able to determine the time variation of the
THz signal ETHz ðtÞ:
Both detection methods, which are coherent methods, were compared using the
same laser power modulated by an acousto-optic modulator. At low-frequency
modulation the PC sampling method shows a better signal-to-noise ratio and
sensitivity, for an identical received THz average power. Increasing the modulation
frequency over 1 MHz; the performances of the FS-EOS method become
comparable, i.e. the signal-to-noise ratio of FS-EOS becomes greater than 104
[126]. However, FS-EOS is able to detect signals in a huge bandwidth 100 GHz–
37 THz [127], while in the PC sampling method the bandwidth is limited to 3–4 THz
due to antenna parameters.
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 51

4.2. CW THz heterodyne detection

The most sensitive receivers at microwave, millimeterwave and THz frequencies

are based on the heterodyne principle, which consists in mixing of two signals, the
incoming THz CW signal and the local oscillator (LO) signal. The LO has a fixed
output power that should be much greater than the power of the incoming signal.
The LO frequency is also different from that of the received signal. The heterodyne
process is realized by a nonlinear device, named mixer, which has an output signal
with a frequency proportional to the difference between the frequencies of the
incoming THz signal and the LO; the frequency of the output signal is called
intermediate frequency (IF). Heterodyne receivers can be described by a series of
parameters, but the most encountered one in the THz range is the receiver noise
temperature TR ¼ Tmixer þ LTIF : Here the indices indicate the noise contribution of
the mixer and the IF first amplifier stage, respectively, and L is the mixer conversion
loss.
Room-temperature heterodyne receivers are based on Schottky diodes, character-
ized by strong I–V and C–V nonlinear characteristics. In Schottky diodes operating
at moderate bias values the I–V and C–V characteristics are well-known:
I ¼ Isat exp½ðV =nVT Þ
1 and C ¼ C0 =ð1
VT =VD
V =VD Þ1=2 ; where n is the
ideality factor, and VD and VT are the flat-band voltage and the thermal voltage,
respectively.
However, the above expressions for the I–V and C–V characteristics are not valid
at THz frequencies, because the operating point is near the flat-band voltage
(V
VD o3VT ¼ 3kB T=e ¼ 80 mV at room temperature). The corresponding
expressions of these characteristics at THz frequencies are, respectively [128]:
I ¼ Isat expðVD =nTÞ=2 sinh½ðVD
V Þ=nVT ; ð41Þ

C0 f1
exp½
ðVD
V Þ=VT g
C ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi: ð42Þ
1
V =VD
ðVT =VD Þf1
exp½
ðVD
V Þ=VT g
At THz frequencies the planar Schottky diodes, which are based on GaAs, InP or
semiconductor heterostructures, have submicronic dimensions, the anode diameter
varying between 0.2 and 0:5 mm: Recently, a 2:5 THz monolithic mixer consisting of
a GaAs Schottky diode placed between an input and an output low-pass filter was
demonstrated, the filters decoupling the high frequency signals from the dc bias. The
IF signal was generated using the membrane technique described in the previous
section. The GaAs membrane was 3 mm thick and the anode area of the diode was
0:2
1 mm2 [129]. The LO, which was a gas laser at 2:522 THz with a power
of 3–5 mW; was injected in the mixer together with the 2:5 THz signal through a
Martin–Purplett diplexer. The insertion loss was
17 dB giving a receiver noise
temperature of 9000 K:
Much lower noise temperatures are obtained using cooled heterodyne receivers
based on superconducting devices. These heterodyne receivers are based on SIS
(superconductor–insulator–superconductor) tunnel junctions that have a pro-
nounced nonlinear I–V characteristic due to tunneling of quasiparticles in the
 ARTICLE IN PRESS
52 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

receiver noise
Schottky, room
temperature [K]
temperature
(DSB)
SIS (4 K)
104
NbN HEB (4 K)

103
10x quantum limit
102 (hf/2kB)

102 103 104

f(GHz)

Fig. 35. Receiver noise temperature for heterodyne receivers in the THz range.

superconducting gap, and operate up to a cutoff frequency fc D150 Tc ; where Tc is

the critical temperature at which the transition towards the superconducting state
takes place. Typical junctions are based on superconductors such as Nb ðTc E9 KÞ or
high-temperature superconductors (HTS) such as YBCO ðTc ¼ 100 KÞ: Even for Nb
the cutoff frequency is greater then 1:3 THz: The main drawback of the SIS
heterodyne receiver, namely its low operating temperature, is overcome by the fact
that the required LO power is one order of magnitude lower than that for THz
receivers based on Schottky diodes. Even lower receiver noise temperatures are
obtained using hot electron bolometer (HEB) mixers. The receiver noise for the three
types of mixers is represented in Fig. 35 [130].
HEB mixers are based on the heating of a superconducting microbridge with a
THz radiation such that the corresponding electron temperature is higher than the
phonon temperature of the device. The resistance of the bolometer is dependent on
the electron temperature, the detected voltage being proportional with the THz
power when the bolometer is dc biased. The short relaxation time of heated electrons
(about 10 ps) allows a very fast variation of the microbridge resistance at IF up to
several GHz. The high-speed operation makes HEB mixers the most sensitive THz
heterodyne receivers, their noise temperature almost approaching the quantum limit
(see Fig. 35). The nonlinear I–V curve necessary for mixing originates in this case in
the electron heating of the microbridge, which experiences the superconducting
transition. In Fig. 36 we have represented the three I–V characteristics of THz
heterodyne receivers.
There are two main types of HEB mixers [131]. The first type, named ‘‘phonon-
cooled’’ or ‘‘lattice-cooled’’, is based on the strong electron–phonon interaction and
the fast phonon escape time. In this type of HEB the relaxation time is strongly
dependent on the film thickness. The maximum IF frequency is proportional to
1=te2ph ; where te2ph is the electron–phonon relaxation time. If we desire an IF
bandwidth of 10 GHz; the thickness of the superconducting microbridge must be less
than 6 nm. Microbriges from Nb, NbN, Al, or YBCO have thus submicronic or
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 53

I/Isat I I

LO on

LO off
V/VT V V
(a) (b) (c)
Fig. 36. I–V characteristics of nonlinear elements on which THz heterodyne receiver are based:
(a) Forward biased Schottky diode, (b) SIS, and (c) HEB.

IF

Fig. 37. HEB mixer with CPW twin slot antennas.

nanometric thicknesses; for example, in the case of NbN film the thickness is 3–4 nm
if the first type of HEB is implemented. The HEB can be positioned at the center of a
twin-slot CPW antenna [132] (Fig. 37).
The second type of HEB mixers, named ‘‘diffusion cooled’’, is based on the
cooling mechanism produced by the out-diffusion of hot electrons to a heat sink.
This is realized by connecting the ends of the superconducting microbridges to
normal metal pads.
Both types of the HEB mixers can be described by the same equations [133]:

K d2 Te =dx2 þ ðCe =te2ph ÞðTe
T0 Þ ¼ j 2 rn þ PTHz ; ð42aÞ

K d2 Te =dx2 þ ðCe =te2ph ÞðTe
T0 Þ ¼ PTHz ; ð42bÞ

where Te is the electron effective temperature, Ce is the electron specific heat, K is the
thermal conductivity, j the bias current density, rn the mixer resistivity in normal
state and PTHz the absorbed LO power/unit volume. The first equation, Eq. (42a), is
valid inside the hot spot, while Eq. (42b) describes the electron temperature outside
this spot.
 ARTICLE IN PRESS
54 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

4.3. Direct THz detection using micro and nanodevices

Micro- and nanotechnologies have boosted many areas of applied sciences and, in
particular, the THz detection devices. In Fig. 38 we have presented a bolometer
working up to 3 THz; realized using MEMS techniques [134]. The bolometer is a
square sheet of bismuth with a length greater than 1:5l evaporated on a thin
dielectric membrane, SiO2 (thickness 700 nm)/Si3 N4 (thickness 350 nm)/SiO2
(thickness 450 nm). A 20 element array was implemented to detect with the required
precision the beam profile of a FIR laser.
A bolometer is able to measure only the power of the incoming radiation. Using
the conceptions of MEMS techniques it was shown very recently [135] that an array
of micromachined nanosized Fabry–Perot-like cantilevers can be used to sense
simultaneously the power and the frequency of the THz field. This device is based on
the electromagnetic actuation of micromachined cantilevers in the THz range. The
device, which is a combination of a cantilever and a Fabry–Perot resonator, is
schematically represented in Fig. 39. The THz field bends the entire microstructure,
producing an additional tunneling current, which flows between the tip and the

metallic
contact bismuth film
membrane

Si

Fig. 38. THz bolometer based on MEMS technology.

THz field y

x
h
t

tunneling
L tip

tunneling
substrate contact
Fig. 39. Fabry–Perot-like cantilever for THz power and frequency sensing.
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 55

contact below it. Due to the exponential dependence of the tunneling current on the
cantilever deflection even small deflections of about 1 A ( produce large changes in the
current (greater than an order of magnitude).
A THz field that excites uniformly the microstructure presented in Fig. 39 will
deflect it with the amount d ¼ RPL3 =ð6cEIÞ; where R is the reflectivity of the entire
microstructure, P the incident power, I ð¼ Wt3 =6 þ Wth2 Þ the moment of inertia,
and L the length of the arm of the Fabry–Perot like cantilever. To have a large
deflection it is necessary to have a large L; a small thickness t and a small width W :
The reflectivity of the microstructure is given by
4R0 sin2 ð2pfh=cÞ
R¼ ; ð43Þ
ð1
R0 Þ2 þ 4R0 sin2 ð2pfh=cÞ
where R0 is the reflectivity of one of its arms, i.e. of a slab with a thickness t: A large
R0 can be attained for metallic arms, whereas a large R implies additionally a high h:
If the incident THz field is normal to the microstructure, the reflectivity (and hence
the deflection) is independent on polarization; otherwise polarization effects of the
THz field should be considered. The incoming THz power is determined by
measuring the variation of the tunneling current that is produced by the deflection d:
The incident power P is not the power emitted by the THz source, Ps ; but is the
fraction of this power incident on the Fabry–Perot-like cantilever, i.e. P ¼ Ps Ac =Af ;
where Ac ¼ LW is the cantilever area and Ac the focalization area. A power of
Ps ¼ 15 mW; for example, can be focalized on an area of 3
0:3 mm2 ; so that on a
cantilever with L ¼ 3 mm; W ¼ 0:1 mm; t ¼ 0:01 mm; h ¼ 8 mm and R0 ¼ 0:9; the
incident power is P ¼ 5 mW: This power produces a deflection of about d ¼ 140 A ( at
a frequency of 1:5 THz for a structure made of gold with E ¼ 80 GPa: Recently, it
was demonstrated that large-scale metallic nanowires can be grown in the pores of
silica gel. Although the length of the device in the example above seems at first to be
too long, we must take into consideration that the device is not a simple cantilever,
but a microstructure with a much higher inertia. Therefore, under the action of its
own weight, the deflection of the microstructure is x ¼ qL4 g=8EI; where q is the mass
per unit length and g ¼ 9:8 m=s2 : x is approximately 45 nm for this device compared
to 350 mm for a simple cantilever with the same dimensions as an arm of the Fabry–
Perot-like cantilever structure. With such an arrow a simple cantilever will obviously
break under the action of its own weight.
Since the deflection is proportional to the incident power and the measured
tunneling current depends exponentially on the deflection, we have I ¼ I0 expðKPÞ;
with I0 determined by the applied bias and by the distance between the Fabry–Perot-
like cantilever and the metallic tip. A device sensibility independent on the biasing
conditions can be defined as @ ln I=@P ¼ K ¼ ð2m0 fÞ1=2 RL3 =ð3cEI_Þ; sensibility
which is independent on the incident power, and which has the value 28:8 mW
1 for
the Fabry–Perot-like cantilever in the example above.
To measure the wavelength an array of such Fabry–Perot-like cantilevers is
needed, with h varying linearly along the array. For an array of N cantilevers along
the z direction and separated by dW ; the distance between the arms of the jth Fabry–
Perot-like cantilever is hj ¼ h0 þ A½ð j
1ÞðW þ dW Þ þ W =2; j ¼ 1; 2; y; N with A;
 ARTICLE IN PRESS
56 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

h0 constant parameters. Since the total reflection coefficient is periodic with h; taking
maximum values at hj ¼ ð2m þ 1Þc=4f and minimum values at hj ¼ mc=2f ; the
frequency of the THz source can be determined monitoring the position in the array
for which the reflection coefficient is minimum and/or maximum (in simple metallic
cantilevers the reflectivity does not depend on its thickness.). The resolution in the
bandwidth 1.2–58 THz is Df =f D10
3 in the example considered here.
Semiconductor heterostructures play also an important role in THz detection. For
example, a HEB bolometer detector and mixer was proposed using the nonlinearities
provided by a heated 2D electron gas medium [136]. This is a HEB of the second
type, where the electrons are cooled by diffusion into the contact. For a device that is
0:8 mm long, the time constant is 1 ps and the responsivity is 3000 V=W: At the
operating temperature of 77 K; with a 100 GHz as IF and a 1 mW LO, the receiver
noise temperature is 1000–2000 K at 1 THz: HEMT transistors are also used to
detect THz radiation for a constant drain bias when plasma waves are launched in
the short transistor channel. A good sensitivity was obtained when a HEMT
structure, biased so that the drain–source current has a constant value IDS ¼ 0:1 mA;
was illuminated with a 2:5 THz CW beam [137].
MSM interdigited photodetectors can be used to detect THz radiation when the
widths of the metallic fingers and of the space between them have submicronic
dimensions (25–900 nm). The Schottky photodetector is realized on a 200 nm thick
GaAs/AlGaAs substrate, the later being used for carrier absorption. The MSM
device is laterally illuminated by the THz beam (a waveguide-like excitation of the
absorbing layer) to obtain a maximum quantum efficiency [138].
A latest trend in THz detection is the single-photon detection. This was already
accomplished using a nanodevice—a single-electron transistor (SET)—at an
operating temperature of 50 mK: The SET is made from two parallel quantum
dots (see Fig. 40) [139]. One of the dots is coupled to a dipole antenna, which is
illuminated by the THz radiation. Inside this dot an electron–hole plasma is formed,
which produces a photomultiplication effect (108 –1012 electrons/photons) that, on its
turn, shifts the conductance of the other dot due to electron tunneling. The
nanodetector has a sensitivity 0:1 photons=0:1 mm2 : The noise equivalent power
NEP is 10
17 W ðHzÞ
1=2 ; three orders of magnitude better than any bolometer.

THz bowtie conductance

antenna of dot 1 THz shift

G1 reservoir
VSD
dot 1

dot 2

VG1

Fig. 40. THz SET detector.

 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 57

5. Terahertz main applications

There are an increasing number of papers dealing with THz applications. Their
number exceeds by far the number of papers dealing with the thorny problems of
THz devices for emission or detection. Since the applications of THz are so well
represented in the literature, we mention here only briefly the main areas where THz
fields play a major role, underlining the physical problems encountered in each
application.
The main two applications in which THz fields are involved are THz spectroscopy
and THz imaging. These applications have contributed to a better knowledge of
condensed matter, material properties and biology.
The most common THz spectroscopy method is based on time domain techniques,
which employ either photoconductive or electro-optical methods for emission (see
Sections 2.1.1 and 2.1.3) and detection (see Section 4.1). The setup for THz time
domain spectroscopy (TDS) in a transmission configuration is realized by placing a
sample between the THz emitter and receiver, the THz radiation passing through it.
There are some important advantages of THz TDS:

(i) Both photoconductive and electro-optical detection methods are coherent time-
gated detection methods that detect the electrical field and not its intensity.
(ii) The system works at room temperature because the noise background, which is
stronger than the THz signal, is cancelled by averaging through time sampling
and lock-in detection. This is possible because the polarization of the
background noise is randomly distributed, while the THz signal has a
predetermined polarization. In this way, high SNR are obtained (see Section
4.1), which exceed by orders of magnitudes the performances obtained with
bolometers.
(iii) The amplitude and phase of the detected THz signal, which results from the
interaction of the emitted THz signal with the sample, are available. Using
either analytical or numerical techniques the real and imaginary part of the
dielectric permittivity are then available in a large frequency range.

Taking into account these important advantages it was believed that TDS will
surmount in many ways the conventional spectroscopic method used in the THz
range of frequencies, i.e. the Fourier-transform infrared (FTIR) spectroscopy. The
FTIR spectrometer has as a source of radiation an incoherent arc lamp and is based
on the two-beam interferometry principle. In contrast to TDS, the output signal of
the FTIR spectrometer is proportional to the intensity of the FIR field. However,
recent results have demonstrated [140] that the SNR of TDS is better than that of
FTIR for frequencies in the range 10 GHz–4 THz: In this range the field amplitude
SNR of THz TDS is 104 while that of FTIR is about 300. Beyond this frequency
range (the FTIR method is able to work from FIR up to the visible spectral range),
the SNR of FTIR is better. The SNR of TDS is drastically decreased when its
bandwidth is increased up to 40 THz [141]. Both methods show a similar spectral
resolution E0:1 cm
1 :
 ARTICLE IN PRESS
58 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

Thus, TDS is an extremely powerful method of spectroscopy, working at room

temperature with an impressive sensitivity. TDS is performed in two steps. First, a
reference signal Er ðtÞ is detected in the absence of the sample. Then, a signal in the
presence of the sample, Es ðtÞ; is detected. The transmission function of the TDS is
given by
TðoÞ ¼ F ðEs ðtÞÞ=F ðEr ðtÞÞ ¼ Es ðoÞ=Er ðoÞ ¼ jTðoÞj exp½ijðoÞ; ð44Þ
where F ð:::Þ signifies the integral Fourier transform. The index of refraction of the
sample N ¼ n þ ik is related to the transmission function through the equation:
XP
TðoÞ ¼ 4N exp½ioðN
1Þ d=c f½ðN
1Þ=ðN þ 1Þ expðioN d=cÞg2i ; ð45Þ
i¼0

where d is the thickness of the sample. Expression (45) is the transfer function
resulting from the passage of the THz field through the sample, the THz field being
reflected P times inside the sample. These repeated reflections inside the sample,
which can be modeled as a Fabry–Perot-like effect, are seen in the time domain as
distinct pulses accompanying the main time domain response. Thus, P is easily
determined at least in the case of thick samples (see Fig. 41).
When P ¼ 0; Eq. (45) allows a simply determination of the real and imaginary
parts of the refractive index of the material as
jTðoÞj ¼ 2ðn2 þ k2 Þ1=2 expð
okd=cÞ=½ðn þ 1Þ2 þ k2 ; ð46aÞ

jðoÞ ¼ oðn
1Þd=c þ arctanf
kðn2 þ k2
1Þ=½nðn þ 1Þ2 þ k2 ðn þ 1Þg: ð46bÞ

Solving the system formed by the last two equations we can extract the real and
imaginary part of the index of refraction from the known amplitude and phase of the
transmission function at a certain frequency. The procedure is repeated for any
frequency with the bandwidth of the TDS system. Thus only two time-domain
measurements are necessary to determine TðoÞ; from which the frequency behavior
of the complex index of refraction of the material is determined. Both analytical
methods [142] and numerical algorithms [143] are used to determine simultaneously

input output

sample
Fig. 41. THz TDS spectroscopy.
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 59

the index of refraction and the thickness of the sample, a procedure useful mainly for
thin samples, where it is difficult to estimate the sample thickness.
In this manner, an impressive number of materials were measured in the THz
domain, among them being semiconductors, dielectrics, ferroelectrics, high-
temperature superconductors, gases and liquids. The study of each material type
has conducted to an additional knowledge of other parameters. For example, in the
case of semiconductors, carrier concentration and carrier mobility can also be
determined. In the case of gases, each component of a gas mixture can be identified
and in this way air composition monitoring becomes possible. A recent review
about the applications of TDS in material characterization can be found in Ref.
[144]. TDS is also an invaluable tool in the investigation of fundamental
physical properties of semiconductors and heterostructures. In this respect,
TDS was used extensively in the study of coherent phenomena, such as
quantum interference of hole states, quantum interference of Landau levels,
Bloch oscillations, THz emission by coherent photons, etc. The review of
Roskos [145] is a valuable tool for those interested to find additional details,
including TDS applications in atomic physics. Among the latest achievements
of TDS in this area we mention the observation of dressing of bare particles
inside a semiconductor plasma excited by an ultrafast optical pump (of about 10 fs)
and probed by a THz pulse. The THz pulse probes the polarization of the
electron–hole plasma formed by the fs optical pulse after a fixed time delay, then the
THz output signal is recorded in time using a FS-EOS techniques. In this way,
collective behaviors, such as Coulomb screening and plasmon scattering, were
observed to occur at time delays of about 10
14 s; inversely equal to plasma
oscillations [146,147].
The TDS was built due to a lack of continuous and tunable THz sources. Since
there are now some THz sources (see Section 2.2), a new THz spectroscopy
technique called CW THz spectroscopy (CWS) is being developed; for a review see
Ref. [58]. Extremely high resolutions (linewidths less than 1 MHz) and high SNR
ratio (better than 100:1) in narrow bandwidths are the main features of CWS, which
surpass TDS. It was demonstrated that the CW power available in 1 MHz
bandwidth is 104 times greater than that of a TDS system, and this happens when a
photomixer delivers only 1 mW output power at 1 THz: Tremendous results could be
obtained if the photomixer power could be increased at 1 mW: The CW THz field
was detected with either photoconductive or FS-EOS sampling techniques using as a
gating signal the delayed optical signal excitation (see Eq. (26)) that contains the
frequency difference of the two lasers and thus extends TDS detection techniques to
CW. In the case of CW photoconductive sampling detection two identical
photomixers were used, which form a transceiver configuration. The photomixer
that plays the role of the detector was gated by a delayed version of the optical signal
excitation. Both photomixers were excited by laser diodes, and the optical signal was
carried out through optical fibers up to the photomixer surfaces [148]. It is important
to note that THz CW detectors are also able to determine the amplitude and the
phase of the transmitted THz signal. The amplitude is measured as the average
output of the photomixer detector when the delay is changed and the phase is
 ARTICLE IN PRESS
60 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

determined measuring the change in paths between two consecutive zeros of the
output fringes. A heterodyne receiver will offer even better performances.
Imaging of composite materials, such as biological tissues, boxes containing food,
leaves, integrated circuits, bank notes, teeth, floppy disks, paintings, etc., reveals
hidden details inside these materials due to the fact that the THz radiation penetrates
non-metallic and non-polarizing materials such as semiconductors, plastics and
textiles. The imaging techniques were a natural development of THz spectroscopy
techniques. In the case of TDS, the object to be scanned is placed in the focal plane
of the THz beam and is translated across the x–y plane for image acquisition, the
THz intensity being recorded at each point. The image is then constructed pixel by
pixel, and requires much time for completion. 2D THz intensity distributions can,
however, be directly recorded using EO technique. In this case, a THz beam (pump)
and a readout optical signal (probe), which probes the electric field distribution
within the EO crystal, are applied to a ZnTe plate. The 2D THz field distribution is
converted into a 2D optical intensity and is recorded by a CCD camera, after passing
the optical readout through an analyzer [149].
CW imaging systems are expected to have better spatial resolutions and image
qualities than the imaging systems based on ultrashort pulses. There are many THz
review papers dedicated only to THz imaging. Two of them are in particular very
appealing: Refs. [150,151].
There are other configurations that can be employed for THz spectroscopy or
imaging, such as reflection configuration, differential TDS, or chirped probe THz
pulse. Their performances are described in the references indicated above and in
Refs. [152,153].
There are an increasing number of papers dealing with the imaging of biological
tissues and genetics. It is worth mentioning what we can expect from THz
imaging in comparison with X-ray imaging or X-ray tomography. X-rays can
penetrate inside the body obtaining images of interior organs. However,
no clear images are obtained for low-index materials. Thus, THz imaging is an
additional investigation tool besides X-ray. However, THz radiation cannot
penetrate deep inside the body due to the water content in any cell, and only skin,
hair, teeth or dried biological samples can be investigated; this is a serious drawback
for THz imaging. The spatial resolution of 0:3 mm is much worse compared to X-ray
and it is another major drawback. However, the THz field is much less scattered, due
to the much longer wavelength, allowing the visualization of objects hidden in
granular materials (powders) [24]. THz beams are not harmful for living beings,
while X-rays, which are ionizing radiations, become dangerous beyond a certain
dose. Recently an entire special issue of Physics in Medicine and Biology (vol. 47,
number 21, November 2002) was dedicated to biological imaging using THz
techniques.
In genetics the first steps involving THz radiation have been already made. Label-
free analysis of DNA, i.e. hybridization detection, was recently performed with a
very high sensitivity, allowing the detection of a single base mutation of DNA
molecules. The DNA sample is placed on a planar microstrip bandpass filter acting
as a THz resonator and the genetic diagnostic is realized by monitoring the
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 61

transmission of the filter when DNA is denatured or hybridized. The changing of the
resonance frequency of the filter allows a sensitivity up to a femtomol level [10].
Astronomy is a special area of THz applications. However, we will not review
these applications here. Interested readers can consult Ref. [5] for a comprehensive
review of this type of THz applications.

6. Conclusions

A simple inspection of the references shows that their large majority is not older
than two years. This demonstrates how emerging the THz technology really is. The
present THz technology is strongly dominated by ultrafast optical techniques from
which THz emission or detection results as a down-conversion process, but these
techniques are in serious competition with electronic techniques. These later
techniques benefit from the most modern technologies developed in electronics,
such as MEMS or nanotechnologies, having as an ultimate consequence a serious
decrease of the costs of a THz system, a significant increase of the sensitivity and a
large reproducibility. This is the only chance for THz technology to spread in
industrial applications such as gas detection and air monitoring miniaturized
systems, portable spectrometers, or medical tools. One of the largest markets in the
world is the communication market. The development of optical communications at
Terabits speeds will force the parallel development of specific THz modulators and
detectors for optical signals. THz transistors switching 1012 times in a second are an
achievable goal in a couple of years. THz communications are not developed at its
full potential due to the high attenuation encountered in the THz range, but short-
distance communications systems are still achievable.
The micro and nanotechnologies found numerous applications in the THz range.
Devices with unprecedented performances can be built based on them. A must for
the THz technology is a miniaturized THz tunable source and a THz amplifier. The
development of THz transistors and THz multipliers will make this task realizable in
a couple of years. However, only a robust presence in the communication and
medical markets will turn THz into a mature technology. Some small companies in
the THz area have already appeared in the last two years, most of them as spin-off
companies of university departments. However, only the development of THz
technology inside important companies will boost the THz technology from the
academic level to industrial applications.

References

.
[1] E.R. Brown, J.R. Soderstr .
om, C.D. Parker, L.J. Mahoney, K.M. Molvar, T.C. McGill, Applied
Physics Letters 58 (1991) 2291–2293.
[2] J. Nishizawa, P. P"otka, T. Kurabayashi, IEEE Transactions on Electron Devices 49 (2002)
1102–1111.
[3] E. Alekseev, D. Pavlidis, IEEE Microwave Technique and Techniques Symposium, Boston, USA,
2000, pp. 1905–1909.
 ARTICLE IN PRESS
62 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

.
[4] R. Kohler, A. Tredicucci, F. Beltram, H.E. Beere, E.H. Linfield, A. Gilles Davies, D.A. Ritchie,
R.C. Iotti, F. Rossi, Nature 417 (2002) 156–159.
[5] P.H. Siegel, IEEE Transactions on Microwave Theory and Techniques 50 (2002) 910–928.
[6] B.I. Greene, J.F. Federici, D.R. Dykaar, R.R. Jones, P.H. Buksbaum, Applied Physics Letters 59
(1991) 893.
[7] D. Grinschkowsky, S. Keiding, M. van Exter, C. Fattinger, Journal of Optical Society of America
B7 (1990) 2006–2015.
[8] J.W. Digby, C. McIntosch, G. Parkhurst, B.M. Towlson, S. Hidjiloucas, J.W. Bowen, J.M.
Chamberlain, R.D. Pollard, R.E. Milles, P. Steenson, L.S. Karatzas, N.J. Cronin, S.R. Davies,
IEEE Transactions on Microwave Theory and Techniques 48 (2000) 1293–1302.
[9] C. Mann, 30th European Microwave Conference, Paris, France, 2000, pp. 1–4.
[10] M. Nagel, P. Haring Bolivar, M. Brucherseifer, H. Kurz, Applied Physics Letters 80 (2002)
154–156.
[11] R. Chau, B. Doyle, J. Kavalieros, D. Barlage, A. Murthy, M. Doczy, R. Arghavani, S. Datta,
International Conference on Solid State Devices and Materials, Nagoya, Japan, 2002, pp. 68–69.
[12] D.D. Nolte, Journal of Applied Physics 85 (1999) 6259–6289.
[13] G. Mourou, C.V. Stancampiano, A. Onetti, A. Orszag, Applied Physics Letters 39 (1981) 295–296.
[14] D.H. Auston, K.P. Cheung, P.R. Smith, Applied Physics Letters 45 (1984) 284–286.
[15] P.R. Smith, D.H. Auston, M.C. Nuss, IEEE Journal of Quantum Electronics 24 (1988) 255.
[16] Z. Piao, M. Tani, K. Sakai, Japanese Journal of Applied Physics 39 (2000) 96–100.
[17] D.B. Rutledge, S.E. Scwartz, A.T. Adams, Infrared Physics 18 (1978) 713–729.
[18] G.M. Rebeiz, Proceedings of IEEE 80 (1992) 1748–1770.
[19] T.M. Weller, L.P. Katechi, G.M. Rebeiz, IEEE Transactions on Antennas and Propagation 43
(1995) 1423–1428.
[20] D.W. van der Weide, Journal of Optical Society of America B11 (1994) 2553–2560.
[21] K.S. Yngvesson, D.H. Scaubert, T.L. Korzeniowski, E.L. Kolberg, T. Thungren, J.F. Johansson,
IEEE Transactions on Antennas and Propagation 33 (1985) 1392–1400.
[22] Y. Cai, I. Brener, J. Lopata, L. Pfeiffer, J. Federici, Applied Physics Letters 71 (1997) 2076–2078.
[23] N. Sarukura, H. Ohtake, S. Izumida, Z. Liu, Journal of Applied Physics Letters 84 (1998) 654–656.
[24] M. Tani, M. Herrmann, K. Sakai, Measurement Science and Technology 13 (2002) 1739–1745.
[25] D.F. Filipovic, S.S. Gearhart, G.M. Rebeiz, IEEE Transactions on Microwave Theory and
Techniques 41 (1993) 1738–1749.
[26] D.F. Filipovic, G.P. Gauthier, S. Raman, G.M. Rebeiz, IEEE Transactions on Antennas and
Propagation 45 (1997) 760–766.
[27] J. van Rudd, D.M. Mittlemann, Journal of Optical Society of America B19 (2002) 319–329.
[28] N.M. Froberg, B.B. Hu, X.C. Zhang, D.H. Auston, IEEE Transactions on Quantum Electronics 28
(1992) 2291–2301.
[29] T. Pheiffer, H.M. Heiliger, E.S. von Kamienski, H.G. Roskos, H. Kurtz, Journal on Optical Society
of America B11 (1994) 2547–2552.
[30] R.K. Lai, J.R. Hwang T.B. Noriss, J.F. Whiteker, Applied Physics Letters 72 (1998) 3100–3102.
[31] O. Mitrofanov, I. Brener, M.C. Wanke, R.R. Ruel, J.D. Wynn, A.J. Bruce, J. Federici, Applied
Physics Letters 77 (2000) 591–593.
[32] M. Tani, M. Watanabe, K. Sakai, Electronics Letters 38 (2002) 5–6.
[33] D. Grischkowsky, M.B. Ketchen, C.C. Chi, I. N. Dulling III, N.J. Halas, J.M. Halbout, P.G. May,
IEEE Journal of Quantum Electronics 24 (1988) 221–225.
[34] N. Katzenellenbogen, D. Grischkowsky, Applied Physics Letters 58 (1991) 222–224.
[35] D. Grischkowsky, IEEE Journal of Selected Topics in Quantum Electronics 6 (2000) 1122–1135.
[36] P.K. Benicewicz, J.P. Roberts, A.J. Taylor, Journal of Optical Society America B11 (1994)
2533–2546.
[37] J.T. Darrow, X.C. Zhang, D.H. Auston, IEEE Journal of Quantum Electronics 28 (1992)
1607–1616.
[38] S.G. Park, A.M. Weiner, M.R. Melloch, C.W. Seiders, A.J. Taylor, IEEE Journal of Quantum
Electronics 35 (1999) 1257–1268.
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 63

[39] J.F. Holzman, F.E. Vermeulen, A.Y. Elezzabi, IEEE Journal of Quantum Electronics 36 (2000)
130–136.
[40] D. Dragoman, M. Dragoman, Progress in Quantum Electronics 25 (2001) 229–290.
[41] X.C. Zhang, B.B. Hu, J.T. Darrow, D.H. Auston, Applied Physics Letters 56 (1990) 1011–1013.
[42] X.C. Zhang, J.T. Darrow, B.B. Hu, D.H. Auston, M.T. Schmidt, P. Tham, E.S. Yang, Applied
Physics Letters 56 (1990) 2228–2230.
[43] X.C. Zhang, B.B. Hu, S.H. Xin, D.H. Auston, Applied Physics Letters 57 (1990) 753–755.
# J.M. Fraser, M. de Camp, P.H. Bucksbaum, H.M. van Driel, Applied Physics Letters 75
[44] D. Cote,
(1999) 3959–3961.
[45] D.H. Auston, M.C. Nuss, IEEE Journal of Quantum Electronics 24 (1988) 184–197.
[46] Q. Wu, X.C. Zhang, IEEE Journal of Selected Topics in Quantum Electronics 2 (1996)
693–700.
[47] A. Nahata, Optics Letters 26 (2001) 385–387.
[48] A. Bonvalet, M. Joffre, J.L. Martin, A. Migus, Applied Physics Letters 67 (1995) 2907–2909.
[49] A. Rice, Y. Jin, X.F. Ma, X.C. Zhang, D. Bliss, J. Larkin, M. Alexander, Applied Physics Letters 64
(1994) 1324–1326.
[50] A. Nahata, A.S. Weling, T.F. Heinz, Applied Physics Letters 69 (1996) 2321–2323.
[51] A.S. Nikogossian, E.M. Laziev, A.A. Hakhoumian, N.G. Pogosyan, The Digest of the 25th
International Conference on Infrared and Millimeter Waves, Beijing, China, 2000, pp. 65–66.
[52] D. van der Weide, Applied Physics Letters 65 (1994) 881–883.
[53] J.S. Bostak, D.W. van der Weide, D.M. Bloom, B.A. Auld, Journal of Optical Society of America
B11 (1994) 2561–2565.
[54] M.J.W. Rodwell, S.T. Allen, R.T. Yu, M.G. Case, U. Bhattacharya, M. Reddy, E. Carman,
M. Kamegawa, Y. Konishi, J. Pusl, R. Pullela, Proceedings of IEEE 82 (1994) 1037–1059.
[55] M. Dragoman, D. Dragoman, An overview of nonlinear microwave and millimeter wave generation
in magnetic, acoustic and electromagnetic distributed nonlinear physical systems, in: R. Marcelli, S.
Nikitov (Eds.), Nonlinear Microwave Signal Processing: Towards a New range of Devices, NATO
ASI Series, Vol. 20, 1996, pp. 13–43.
[56] X. Orlos, F. Martin, Journal of Applied Physics 90 (2001) 2595–2600.
[57] M.C. Gaidis, H.M. Pickett, C.D. Smith, S.C. Martin, P.R. Smith, P.H. Siegel, IEEE Transactions
on Microwave Theory and Techniques 48 (2000) 733–739.
[58] K.J. Siebert, H. Quast, H.G. Roskos, Perspectives of continous-wave optoelectronic THz
imaging, in: R.E. Miles (Ed.), Terahertz Sources and Systems, Kluwer Academic, Dordrecht,
2001, pp. 127–143.
[59] S.M. Duffy, S. Verghese, K.A. McIntosh, Photomixers for Continous-Wave Terahertz Radiation,
in: D. Mittleman (Ed.), Sensing with Terahertz Radiation, Springer, Berlin, 2002.
[60] E.R. Brown, F.W. Smith, K.A. McIntosch, Journal of Applied Physics 73 (1993) 1480–1484.
[61] K.A. McIntosh, E.R. Brown, K.B. Nichols, O.B. McMahon, W.F. DiNatale, Applied Physics
Letters 69 (1996) 3632–3634.
[62] S. Matsura, M. Tani, K. Sakai, Applied Physics Letters 70 (1997) 559–561.
[63] S. Verghese, K.A. McIntosh, E.R. Brown, IEEE Transactions on Microwave Theory and
Techniques 45 (1997) 1301–1309.
[64] S.M. Duffy, S. Verghese, K.A. McIntosh, A. Jackson, A.C. Gossard, S. Matsuura, IEEE
Transactions on Microwave Theory and Techniques 49 (2001) 1032–1038.
[65] S. Verghese, K.A. McIntosh, E.R. Brown, Applied Physics Letters 71 (1997b) 2743–2745.
[66] E.R. Brown, Applied Physics Letters 75 (1999) 769–771.
[67] C. Kadow, A.W. Jackson, A.C. Gossard, S. Matsuura, G.A. Blake, Applied Physics Letters 76
(2000) 3510–3512.
[68] M. Eckardt, A. Schwanh.auXer, F. Renner, L. Robledo, A. Friedrich, P. Pohl, P. Kiesel, S. Maltzer,
.
G.H. Dohler, D. Driscol, M. Hanson, A.C. Gossard, Third Symposium on Non-Stoichiometric
III–V Compounds, Erlangen, Germany, 2001.
[69] E. Peytavit, S. Arscott, D. Lippens, G. Mouret, S. Matton, P. Masselin, R. Bocquet, J.F. Lamplin,
L. Desplanque, F. Mollot, Applied Physics Letters 81 (2002) 1174.
 ARTICLE IN PRESS
64 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

[70] S. Matsura, P. Chen, G.A. Blake, J.C. Pearson, H.M. Pickett, IEEE Transactions on Microwave
and Techniques 48 (2000) 380–387.
[71] D. Pasqualini, A. Neto, R.A. Wyss, Microwave and Optical Letters 33 (2002) 430–435.
[72] K. Kourogi, B. Widiyatmoko, M. Ohtsu, IEEE Photonics Technology Letters 8 (1995) 560–562.
[73] K. Siebert, F. Siebe, M. Thomson, J.Z. Baghbidi, R. Leonhardt, H. Roskos, Proceedings of SPIE
3828 (1999) 234–243.
[74] O. Morikawa, M. Tonouchi, M. Tani, K. Sakai, M. Hangyo, Japan Journal of Applied Physics 38
(1999) 1388–1389.
[75] L. Chusseau, G. Almuneau, L.A. Coldren, A. Huntington, D. Gasquet, Proceedings of IEE J 149
(2002) 88–92.
[76] J.-I. Shikata, K. Kawase, K.-I. Karino, T. Taniuchi, H. Ito, IEEE Transactions on Microwave
Theory and Techniques 48 (2000) 653–661.
[77] K. Kawase, J.I. Shikata, H. Ito, Journal of Applied Physics D 34 (2001) R1–R14.
[78] H. Eisele, A. Rydberg, G.I. Haddad, IEEE Transactions on Microwaves and Techniques 48 (2000)
626–630.
[79] H. Eisele, Proceedings of the 10th IEEE International Conference on Terahertz Electronics,
Cambridge, UK, 2002, pp. 13–18.
[80] Z.S. Gribnikov, N.Z. Vagidov, V.V. Mitin, G.I. Haddad, Journal of Applied Physics 93 (2003)
5435–5446.
.
[81] M. Saglam, B. Schumann, V. Mullerwiebus, A. Megej, U. Auer, M. Rodrigues-Giron!es,
R. Jurdaschke, F.J. Tegude, H.L. Hartnagel, Electronics Letters 38 (2002) 657–658.
[82] S.A. Mass, Nonlinear Microwave Circuits, Artech House, London, 1988.
[83] E.I. Kolberg, J. Stake, L. Dillner, Philosophical Transactions of the Royal Society A354 (1996)
2383–2398.
[84] H. Hartanagel, 32nd European Microwave Conference, 2002, Milano, Italy, pp. 249–252.
[85] A. Maestrini, J. Bruston, D. Pukala, S. Martin, I. Mehdi, IEEE International Symposium Digest,
Phoenix, USA, 2001, pp. 20–25.
[86] F. Maiwald, S. Martin, J. Bruston, A. Maestrini, T. Crawford, P.H. Siegel, IEEE International
Symposium Digest, Phoenix, USA, 2001, pp. 1637–1640.
[87] J.N. Hovenier, M.C. Diaz, T.O. Klaassen, W.T. Wenckebach, A.V. Muravjov, S.V. Pavlov,
V.N. Shastin, IEEE Transactions on Microwave Theory and Techniques 48 (2000) 670–676.
[88] R. Kazarionov, R.A. Suris, Soviet Physics Semiconductors 5 (1971) 707–709.
[89] F. Capasso, R. Paiella, R. Martini, R. Colombelli, C. Gmachl, T.L. Myers, M.S. Taubman,
R.M. Williams, C.G. Bethea, K. Unterrainer, H.Y. Hwang, D.L. Sivco, A.Y. Cho, A.M. Sergent,
H.C. Liu, E.A. Whittaker, IEEE Journal of Quantum Electronics 38 (2002) 511–532.
[90] J. Faist, D. Hofstetter, M. Beck, T. Aellen, M. Rochat, S. Blaser, IEEE Journal of Quantum
Electronics 38 (2002) 533–546.
[91] D. Dragoman, M. Dragoman, Advanced Optoelectronic Devices, Springer, Berlin, 1999.
[92] M. Rocahat, L. Ajili, H. Willenberg, J. Faist, H. Beere, G. Davies, E. Linfield, D. Ritchie, Applied
Physics Letters 81 (2002) 1381–1383.
[93] L. Fridman, G. Sun, R.A. Soref, Applied Physics Letters 78 (2001) 401–403.
[94] G. Sun, R.A. Soref, J.B. Kurghin, IEEE Journal of Selected Topics in Quantum Electronics 7 (2001)
376–380.
[95] P. Kinsler, P. Harrison, R.W. Kelsall, Journal of Applied Physics 85 (1999) 23–28.
[96] J.B. Kurghin, K.L. Wang, Supperlattices and Microstructures 22 (1997) 551–557.
[97] S. Starikov, P. Shiktorov, V. Gruzinskis, L. Reggiani, L. Varani, J.C. Vassiere, J.H. Zhao,
Transactions on IEEE Electron Devices (2001) 438–443.
[98] H.G. Roskos, M.C. Nuss, J. Shah, K. Leo, D.A.B. Miller, A.M. Fox, S. Schmitt-Rink, K. Kohler, .
Physical Review Letters 68 (1992) 2216–2219.
[99] M.S.C. Luo, S.L. Chuang, P.C.M. Planken, I. Brener, H.G. Roskos, M.C. Nuss, IEEE Journal of
Quantum Electronics 30 (1994) 1478–1488.
[100] C. Waschke, H.G. Roskos, R. Schwedler, K. Leo, H. Kurtz, K. Kohler,. Physical Review Letter 70
(1993) 3222–3319.
 ARTICLE IN PRESS
D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66 65

[101] H.G. Roskos, R. Martini, G. Klose, F. Wolter, C. Schwarz, H. Kurz, Tunable coherent THz
radiation pulses from optically excited Bloch oscillations, in: J.M. Chamberlain, R.E. Miles (Eds.),
New Directions in Terahertz Technology, NATO ASI Science Series E, Vol. 334, Kluwer,
Dordrecht, 1997, pp. 369–375.
[102] D.A. Ryndyk, N.V. Demaria, J. Keller, E. Schomberg, Physical Review B 67 (2003) 0033305.
[103] M.I. Dyakonov, M.S. Shur, IEEE Transactions on Electron Devices 43 (1996) 380–387.
[104] M.I. Dyakonov, M.S. Shur, IEEE Transactions on Electron Devices 43 (1996) 1640–1645.
[105] V. Ryzhii, M. Shur, Semiconductor Science and Technology 17 (2002) 1168–1171.
[106] J. Mateos, B.G. Vasallo, D. Pardo, T. Gonzales, J.S. Galloo, Y. Roelens, A. Cappy,
Nanotechnology 14 (2003) 117–122.
.
[107] S. Grafstrom, Journal of Applied Physics 91 (2002) 1717–1753.
[108] A. Mayer, J.P. Vigneron, Physical Review B 62 (2000) 16138–16145.
[109] A.A. Odintsov, Physical Review Letters 85 (2000) 150–153.
[110] A.S. Maksimenko, G.Ya. Slepyan, Physical Review Letters 85 (2000) 362–365.
[111] F. L!eonard, J. Tersoff, Physical Review Letters 85 (2000) 4767–4770.
[112] H.M. Manohara, P.H. Sigel, C. Marrese, B. Chang, J. Xu, Far IR, submillimeter and millimeter
detector technology Workshop, Monterey, USA, 2002.
[113] D. Dragoman, M. Dragoman, IEEE Journal of Quantum Electronics 32 (1996) 1150–1154.
[114] P.F. Goldsmith, Proceeding of IEEE 80 (1992) 1729–1747.
[115] J.A. Arnaud, Beam and Fiber Optics, Academic Press, New York, 1976.
[116] D.H. Martin, J.W. Bowen, IEEE Transactions on Microwave Theory and Techniques 41 (1993)
1676–1690.
[117] M.E. MacDonanld, A. Alexandrian, R.A. York, Z. Popovic, E.N. Grossmann, IEEE Transactions
on Microwave Theory and Techniques 48 (2000) 712–718.
[118] D. Rutledge, IEEE Microwave Symposium, 1996, San Francisco, 1889–1892.
[119] L.P. Katehi, Proceedings of IEEE 80 (1992) 1771–1787.
[120] V.M. Lubecke, K. Mizuno, G.M. Rebeiz, IEEE Transactions on Microwave Theory and Techniques
44 (1998) 1821–1831.
[121] S.P. Jamison, R.W. McGowan, D. Grischkowsky, Applied Physics Letters 76 (2000) 1987–1989.
[122] R.W. McGowan, G. Gallot, D. Grischkowsky, Optics Letters 24 (2000) 1431–1433.
[123] R. Mendis, D. Grischkowsky, Optics Letters 26 (2001) 846–848.
[124] S.G. Park, M.R. Melloch, A.M. Weiner, Applied Physics Letters 73 (1998) 3184–3186.
[125] C. Winnewisser, P.U. Jepsen, M. Schall, V. Schyja, H. Helm, Applied Physics Letters 70 (1997)
3059–3071.
[126] Y. Cai, I. Brener, J. Lopata, J. Wynn, L. Pfeiffer, J.B. Stark, Q. Wu, X.C. Zhang, J.F. Federici,
Applied Physics Letters 73 (1998) 444–446.
[127] P.Y. Han, X.C. Zhang, Applied Physics Letters 73 (1998) 3049–3051.
. V. Krozer, H.L. Hartnagel, IEEE Transactions on Microwave Theory and
[128] A. Jelenski, A. Grus,
Techniques 41 (1993) 549–557.
[129] P.H. Siegel, R.P. Smith, M.C. Gaidis, S.C. Martin, IEEE Transactions on Microwave Theory and
Techniques 47 (1999) 596–604.
[130] E. Gerecht, C.F. Musante, Y. Zhuang, K.S. Yngvesson, T. Goyette, J.C. Dickinson, J. Waldman,
P.A. Yagoubov, G.N. Gol’tsman, B.M. Voronov, E.M. Gershenzon, IEEE Transactions on
Microwave Theory and Techniques 47 (1999) 2519–2527.
[131] J. Kawamura, R. Blundell, C.Y.E. Tong, D. Cosmo Popa, T.R. Hunter, S.C. Paine, F. Patt,
G.N. Gol’tsman, S. Cherednichenko, B.M. Voronov, E.M. Gershenzon, IEEE Transactions on
Microwave Theory and Techniques 48 (2000) 683–689.
[132] P. Focardi, A. Neto, W.R. McGrath, IEEE Transactions on Microwave Theory and Techniques 50
(2002) 2374–2383.
[133] A.D. Semenov, G.N. Gol’tsman, R. Sobolewski, Laboratory Laser Energetics Review 87 (2001)
134–152.
[134] C.C. Ling, J.C. Landry, H. Dave!e, G. Chin, G.M. Rebeiz, IEEE Transactions on Microwave
Theory and Techniques 42 (1994) 758–760.
 ARTICLE IN PRESS
66 D. Dragoman, M. Dragoman / Progress in Quantum Electronics 28 (2004) 1–66

[135] D. Dragoman, M. Dragoman, Applied Physics Letters 79 (2001) 581–583.

[136] K.S. Yngvegsson, Applied Physics Letters 76 (2000) 777–779.
[137] . M. Shur, Applied Physics Letters 78 (2001) 2587–2588.
J.Q. Lu,
[138] M. Ryzhii, M.W. Willander, I. Khmyrova, V. Ryzhii, Journal of Applied Physics 84 (1998)
6419–6425.
[139] O. Astafiev, S. Komiyama, T. Kutsuwa, Applied Physics Letters (2001) 1199–1200.
[140] P.Y. Han, M. Tani, M. Usami, S. Kono, R. Kersing, X.C. Zhang, Journal of Applied Physics 89
(2001) 2357–2359.
[141] Q. Wu, X.C. Zhang, Applied Physics Letters (1997) 1285–1287.
[142] L. Duvillaret, F. Garet, J.L. Coutaz, IEEE Journal of Selected Topics in Quantum Electronics 2
(1996) 739–746.
[143] T.D. Dorney, R.G. Baraniuk, D.M. Mittleman, Journal of Optical Society America A18 (2001)
1562–1571.
[144] B. Ferguson, X.C. Zhang, Nature Materials 1 (2002) 26–33.
[145] H.G. Roskos, Physica Scripta T86 (2000) 51–54.
[146] H. Haug, Nature 414 (2001) 261.
[147] R. Huber, F. Tauser, A. Brodscheim, M. Bichler, G. Abstreiter, L.A. Leittenstorfer, Nature 414
(2001) 286–289.
[148] S. Verghese, K.A. McIntosh, S. Calawa, W.F. Dinatale, E.K. Duerr, K.A. Molvar, Applied Physics
Letters 73 (1998) 3824–3826.
[149] Z. Jiang, X.C. Zhang, IEEE Transactions on Microwave Theory and Techniques 47 (1999)
2644–2650.
[150] D.M. Mittleman, M. Gupta, R. Neelamani, R.G. Baraniuk, J.V. Rudd, M. Koch, Applied Physics B
68 (1999) 1085–1094.
[151] S. Mickan, D. Abbott, J. Munch, X.C. Zhang, T. van Dorn, Microelectronics Journal 31 (2000)
503–514.
[152] S. Mickan, K. S. Lee, T.M. Lu, J. Munch, D. Abbott, X.C. Zhang, Microelectronics Journal 33
(2002) 1033–1042.
[153] B. Ferguson, S. Wang, D. Abbott, D. Gray, D. Abbot, X.C. Zhang, Microelectronics Journal 33
(2002) 1043–1051.