Spie proceedings final_prof_eb_lpw

Advancements in Photomixing and Photoconductive Switching for

THz Spectroscopy and Imaging

E.R. Brown

Wright State University, Dayton OH 45324

Physical Domains, LLC, Glendale, CA, 91214, Dayton, OH 45324

ABSTRACT

This paper reviews the design methodology and some of the applications space of standard photomixers and

photoconductive switches. The methodology falls into three categories: (1) photoelectrostatics, (2) terahertz (THz)

electromagnetics, and (3) laser coupling and thermal management. The applications space of ultrafast photoconductive

devices, as for any device technology, is the best measure of their utility. At present photomixers are being used

worldwide in at least these two instruments: (1) broadly tunable sweep oscillators for THz diagnostics, and (2) broadly

tunable coherent transceivers for high-resolution THz spectroscopy. Photoconductive switches are being used in at least

these two systems applications: (1) time-domain spectrometers, and (2) illuminators for THz impulse radars. Each of

these applications will be addressed in turn, and some commercialization challenges facing ultrafast photoconductive

devices will be addressed.

Keywords: ultrafast photoconductors, photoconductive switches, photomixers, photoelectrostatics, terahertz, THz,

electromagnetics, spectroscopy, frequency-domain spectrometer, time-domain spectrometer, impulse radar.

I. INTRODUCTION

The THz portion of the electromagnetic spectrum occupies the spectral range from 300 GHz to 3 THz (or

beyond, depending on who is defining it) and has long been the realm of gas-phase molecular spectroscopy and

astrophysics and, to a lesser extent, earth sensing and materials science. This situation has changed dramatically in the

past decade with the heightened interest in concealed weapon and contraband detection for homeland security,

biological-agent detection, and biomedical imaging. Along with these world-event-related interests have come

heightened scientific interests in molecular chemistry, biochemistry, and biology.

A second factor in the recent advancement of the THz field is the maturation and commercialization of the

fields of high-speed electronics and optoelectronics, photonics, and materials science, many of which are now being

“pulled” by industrial applications in broadband wireless and fiber-optic communications. Two examples are the

engineering of nanostructures by molecular-beam epitaxy, and deep-submicron lithography to fabricate devices having

THz speeds. Another example is the commercial availability of high-performance sources, such as near-infrared single-

frequency semiconductor and solid-state lasers and optical amplifiers. This is particularly true of optical-fiber

components and amplifiers in the telecommunications band around 1550 nm.

A third factor is the advent and rapid development of ultrafast photoconductive devices. Arguably their impact

during the past two decades has been on par with Schottky diodes as a building block for new THz components and

systems. Photoconductive switches have become the workhorse in time-domain systems, and photomixers have been

widely implemented in high-resolution frequency-domain systems of various types. The primary photoconductive

material has been low-temperature-grown gallium arsenide (GaAs). More recently, this has been rivaled by erbium

arsenide-gallium arsenide (ErAs-GaAs): a nanocomposite consisting of ErAs nanoparticles embedded in a GaAs matrix.

ErAs-GaAs photomixers have produced very useful THz output power levels between 1.0 and 10.0 microwatts when

pumped by low-cost distributed feedback (DFB) lasers operating around 780 nm. ErAs-GaAs photoconductive switches

have produced average output power up to ~1 mW, and peak power exceeding 10 W when pumped by frequency-

doubled fiber mode-locked lasers.

Device performance is always important, but system applications is another matter. To be useful in systems,

devices must have unique capabilities, and be reliable and affordable. Without a doubt, the unique feature of ultrafast

photoconductive devices is bandwidth. Photomixers are continuously tunable over at least 1.0 THz, usually limited by

the drive lasers (when using DFBs or similar laser diodes). Photoconductive switches generally have a huge

instantaneous bandwidth of 0.5 THz or more depending on the pulse width of the mode-locked laser driver and the

impulse response of the photoconductive switch in its THz embedding circuit. Bandwidth is important in THz

spectroscopic instruments of all sorts since spectral signatures from interesting materials, such as explosives or toxic

gases, can be spread over a decade of frequency or more. It’s also important in impulse radar where instantaneous

bandwidth determines the pulsewidth in the time-domain, which, in-turn, defines the range resolution.

II. BACKGROUND ON THz PHOTOCONDUCTIVE DEVICES

As is now well understood, photomixing (short for photoconductive mixing) entails the driving of an ultrafast

photoconductive two-terminal structure with two single-frequency, frequency-offset lasers. The result is a highly-

tunable, continuous-wave (cw), coherent source of radiation contained in a single spatial mode, either in a transmission

line or free space. Fig. 1 shows a microphotograph of a typical THz photomixer used today. In contrast,

photoconductive (PC) switching entails the pumping of an ultrafast photoconductive two-terminal structure with a single

mode-locked laser. The result is a train of subpicosecond pulses whose power spectrum is a “comb” peaked in the sub-

THz region, but still produces useful power well beyond 1.0 THz. Most of the photomixer and PC-switch research over

the past decade has been carried out on devices made from low-temperature-grown (LTG) GaAs, or ErAs:GaAs.

Photomixers and PC switches have become a very useful and successful THz device technology during the past

decade. They are now being used worldwide and have been integrated into commercial systems in both the United

States [1, 2] and Europe [3,4]. The two devices complement each other to a large extent. The PC switch is well suited to

time-domain THz spectroscopy with modest resolution requirements, ~10 GHz, but very broad spectral coverage, up

to 3 THz or greater. The photomixer is well suited to high-resolution ( < 1 GHz) spectroscopy over a more modest

spectral range of ~2 THz. PC switches generally have greater spectral coverage than photomixers because of their

lower capacitance and lower RC time constant under laser operating conditions.

A big difference between photomixers and PC switches is average power. In devices fabricated from the same

material and coupled to the same planar antenna or transmission line, the photomixer is limited to just a few W below

1.0 THz. The corresponding optical-to-THz conversion efficiency is less than 10-4 [5]. The PC switch typically

produces ~100x higher average power than a photomixer, and a similar margin in optical-to-THz conversion efficiency

[6]. After analysis and large-signal equivalent-circuit modeling, the difference can be primarily attributed to impedance

matching. Both devices have very high “dark” differential resistance, photomixers between 107 and 108 Ohms, and PC

switches between 108 and 109 Ohms at their respective bias voltages. Under illumination, however, the PC switch's

instantaneous resistance will drop to 100 Ohms or even less because of the high peak power that mode-locked lasers

typically provide. In contrast, the photomixer will drop to a minimum of 10 k, depending on the laser drive power,

which is usually taken from single-frequency distributed feedback (DFB) semiconductor lasers. Attempts to reduce this

resistance further by increasing the laser power usually leads to device burnout. As such, photomixers generally present

a poor impedance match to their THz load circuits, which has a major impact on the THz delivered power and the

optical-to-electrical conversion efficiency.

Given these issues, great care must be exercised in the design and fabrication of THz photoconductive devices,

particularly photomixers. The first and foremost issue is the choice of ultrafast material. Unfortunately, 20 years after

the advent of LTG-GaAs and more than a decade after ErAs:GaAs, these materials are still rather exotic and difficult to

Active Region, 9x9 Micron Desired Polarization

Fig. 1. Top view of a typical GaAs photomixer showing the interdigitated-electrode active region at the driving gap of a square spiral antenna.

obtain. There are several reasons for this, not the least of which is the unusual growth materials or conditions required to

do the molecular beam epitaxy (MBE). Because MBE challenges are difficult to overcome, this paper will focus on

important issues that the THz engineer or scientist has more control over, which are: (1) photoelectrostatics, (2) THz

electromagnetics, and (3) laser coupling and thermal management.

II.A. Photoelectrostatics

As the name suggests, ultrafast photoconductivity is a balancing act between the internal photoelectric effect

and the collection of photogenerated carriers by drift and diffusion between two electrodes under bias. The internal

photoelectric effect produces more carriers as the thickness of the semiconductor increases, which in-turn reduces the

collection efficiency, increases the device capacitance, or both. This tradeoff is captured by the following expression for

the maximum difference-frequency power Pdiff generated from photomixers (below the frequencies where rolloff starts to

occur):

21

2

2

2

1PP

h

egRiP Ldiff

(1)

where i is the difference-frequency photocurrent amplitude, is the external quantum efficiency (i.e., the fraction of

incident photons that produce photoelectrons or photoholes in the active region), and g is the photoconductive gain.

From the Shockley-Ramo theorem of device electrostatics, the photoconductive gain is the mean distance an electron or

hole drifts in the dc bias field before recombination, divided by the physical distance between the electrodes.

(a)

Contacts

InterdigitalElectrodes

(a)(a)

Contacts

InterdigitalElectrodes

Decreasing Field Magnitude

Gap

Semi-Insulating InP Substrate

~1 m

+

(b)

UltrafastPhotoconductor

(a)InterdigitalElectrode

Decreasing Field Magnitude

Gap

Semi-Insulating InP Substrate

~1 m

+

(b)

UltrafastPhotoconductor

(a)

Decreasing Field Magnitude

Gap

Semi-Insulating InP Substrate

~1 m

+

(b)

UltrafastPhotoconductor

(a)InterdigitalElectrode

Decreasing Field Magnitude

Gap

Semi-Insulating InP Substrate

~1 m

+

(b)

UltrafastPhotoconductor

(a)InterdigitalElectrode

Decreasing Field Magnitude

Gap

Semi-Insulating InP Substrate

~1 m

+

(b)

UltrafastPhotoconductor

(a)

Decreasing Field Magnitude

Gap

Semi-Insulating InP Substrate

~1 m

+

(b)

UltrafastPhotoconductor

(a)InterdigitalElectrode

Fig. 2. (a) Top view of interdigital electrode structure commonly used in THz ultrafast photoconductive devices. (b)Cross-sectional view of active region along dashed line shown in (a). The electric lines of force are represented by thecurved loci with the largest magnitude of electric field occurring at the top air-semiconductor interface.

A better qualitative understanding can be had by inspecting the popular interdigital-electrode THz

photoconductor structure shown in Fig. 2(a). Its popularity rests on simplicity of fabrication, low capacitance, and very

short interconnects to balanced planar antennas and coplanar transmission lines of all sorts. The photoconductive

tradeoff becomes clearer in the cross-sectional view of Fig. 2(b) which shows the elliptical electric lines of force

between two adjacent electrodes. The highest magnitude of electric field occurs at the top of the structure at the

semiconductor-air interface, and drops monotonically with depth in the photoconductive epitaxial layer. This means that

photons absorbed near the surface will have the most contribution to the THz photocurrent in the structure, and that

photons absorbed deeper in the structure will have a progressively weaker effect. As the bias voltage is increased to

enhance the THz generation from the deeper-absorbed photocarriers, surface breakdown tends to occur from the

combination of dc leakage current and photocurrent under the high bias voltage and high laser drive power. Because of

the short electron-hole recombination time (<< 1 ps) inherent to ultrafast photoconductors, the gain averaged over the

active volume in Fig. 2(b) is typically ~0.1 in optimized structures, whether made from LTG-GaAs or ErAs:GaAs.

II.B. THz Electromagnetics

As the ultrafast PC devices have such high differential resistance, a simple way to increase the THz power is to

increase the driving-point resistance of the THz load circuit. At microwave frequencies this would be a relatively

straightforward application of a transformer circuit of some sort. But at THz frequencies, and with the decade or more

bandwidth these devices offer, transformation is not so easy. So, to date, the common strategy has been to embed PC

devices in the driving gap of planar antennas, or nearby, to minimize coupling losses caused by the parasitic impedances

that invariably affect THz integrated circuits. Some of the first useful PC-coupled antennas came from the class of

traveling wave antennas, such as the tapered dipole (that is, bow-tie) [7] and tapered slot (called Vivaldi if the taper is

9-microngaps

9-micronarms

BiasLead

ActiveRegion

9-microngaps

9-micronarms

BiasLead

ActiveRegion

-150

0

150

0.2 0.6 1.0 1.4 1.8 2.2 2.6 3.0

Frequency [THz]

Imp

edan

ce [

Oh

ms] Real

Imaginary

modifiedBooker’sresistance

240

0

(a) (b)

Fig. 3. (a) Square spiral antenna used for THz photomixers and PC switches. (b) Real and imaginary parts ofdriving-point impedance of square-spiral antenna on left.

exponential) [8]. Comparable bandwidth, but superior THz efficiency and beam patterns, were then demonstrated from

log periodic [7] or log-spiral [8] designs. If designed with special symmetry properties such as equiangularity or log-

periodicity, such antennas have a frequency independent radiation pattern and a nearly real and frequency-independent

radiation resistance RA=Re{ZA}. If the antenna is also self-complementary in form, then its impedance can approach the

modified form of Booker’s relation RA ≈0[2(eff)1/2], applicable to an air-dielectric interface, where 0 is the

characteristic impedance of free space and eff is the effective dielectric constant [9]. The high THz permittivity (r =

12.8) of GaAs yields a Booker resistance RA ≈more than 5x lower than 0 and undesirable for the standpoint of

THz generation efficiency. On the other hand, a large r makes most of the radiation propagate into the substrate side of

the interface, which then makes simple spherical-lens coupling quite effective on the backside of the substrate [11].

Remarkably, some 30 years after its first demonstration at THz frequencies, spherical-lens coupling is still the preferred

way of coupling radiation from ultrafast PC devices to free space.

Another drawback of the common self-complementary antennas is physical size, which generally must extend

over at least one free-space wavelength in diameter for good wideband performance. A simpler but less explored

antenna structure is the square spiral shown in Fig. 3(a). While lacking the equiangular-or log-periodic symmetry

properties, it is still self-complementary and thus offers the possibility of large bandwidth [10]. Fig. 3(a) shows a

baseline square-spiral design for THz frequencies, and Fig. 3(b) shows its radiation impedance computed using a

commercial Method-of-Moments code. The real part of the computed resistance above 200 GHz varies between about

100 at the valleys and 240 at the peaks. This is in good agreement with the frequency-dependent variations

observed experimentally. Somewhat surprising, but beneficial, is the large deviation of the resistance from the 72-

Booker’s value. As expected from the Kramers–Kronig relations, the imaginary part is always significant, starting out

mostly inductive between ~200 and 500 GHz, and becoming capacitive at higher frequencies. The real part stays above

the modified Booker formula until 1.9 THz, and then falls below it at all higher frequencies.

With its high average driving point resistance below 1.0 THz, the square spiral has produced the highest power

levels we have ever achieved from photomixers and photoconductive switches. This includes a photomixer cw power of

over 10 W around 100 GHz [5], and a PC switch average power of over 1 mW spread over the range from ~0.1 to 1.0

THz. The latter result is discussed in more detail later.

II.C. Laser Coupling and Thermal Management

As in all optoelectronic devices, external laser coupling is an important factor for photonic-to-THz conversion

efficiency and laser stability too since even back-reflection from a photomixer, for example, can create diode-laser

instabilities if not isolated to a very high degree. In addition all semiconductors are imperfect absorbers with absorption

coefficients typically in the range between 5,000 and 10,000 cm-1 (depending on the proximity of the drive wavelength

to the band-gap wavelength). Thus a significant amount of laser power is absorbed ~1 micron or deeper in the active

layer where according to Fig. 2(b) the electrostatic collection of photocarriers is much worse than at the top.

Good top-side laser coupling entails some simple optical procedures. The first applies to interdigital-electrode

PC devices such as that shown in Fig. 1 whereby the polarization of the incident laser beam is oriented perpendicular to

the electrodes to minimize reflection by grating effects. Of course there is still specular reflection from the electrode

metal that increases with the metal fill-fraction, but this can generally be kept at 10% or less. The second procedure is

just an antireflection coating as shown in the cross-sectional view of Fig. 4. At the laser wavelengths typically used for

GaAs (~780 nm) or In0.53Ga0.47As (~1550 nm), it relatively easy to deposit a /4-wave-thick film of silicon nitride,

silicon dioxide, or some ternary alloy that can reduce the air-semiconductor reflection to well below 10%. If properly

deposited, such films can also act as surface passivants and protective coatings for both GaAs and InGaAs.

Improving the laser coupling within the active layer is more difficult, but made feasible by the molecular-beam

epitaxy growth technique commonly used to grow ultrafast PC materials. As shown in Fig. 4, one can grow a dielectric

mirror between the active layer and the substrate that reflects laser radiation not absorbed on the first pass through the

active layer. This takes advantage of the availability of aluminum and its ternary alloys AlGaAs and InAlAs in the MBE

process, and the fact that the optical refractive index of the Al-bearing compounds is significantly lower than GaAs or

InGaAs. About 10 alternating layers of GaAs and Al0.9Ga0.1As, for example, creates a dielectric mirror having a

reflectivity of ~90%. It is also important to judiciously locate the mirror with respect to the top air-semiconductor

interface to create a constructive interference. By so doing, one can create a “resonant optical cavity (ROC)” in which

for a given incident laser power, far more photoelectron hole pairs are created per unit volume than in a single-pass

device [11]. This then allows one to make the active layer much thinner than 1 micron, which according to Fig. 2(b)

UltrafastLayer

Semi-insulating GaAs substrate

Silicon nitride film

AlxGa1-xAsHeat

Spreader

AlAs/AlGaAs

DielectricMirror

1.09 m

0.31 m

10repeatunits

Dielectric LensSilicon Lens

THz Output BeamTHz Output Beam

Goldelectrode

h

Fig. 4. Improved THz photoconductive-switch and photomixer device structure in which the ultrafast(subpicosecond-lifetime) photoconductive layer is separated from the GaAs substrate by an AlGaAs heat-spreadinglayer and an AlAs/GaAs dielectric-mirror stack.

allows more of the photoelectron generation to be near the surface where dc fields are stronger and electrostatic

collection is more efficient. Such an ROC structure was, in fact, used in the most powerful photomixer that we have

ever tested [5].

Like practically all solid-state THz sources, PC devices are ultimately limited in output power and performance

by thermally-related failure. The two primary sources of heat are the optical power absorption and the Joule heating

from photocurrent flowing in the bias field. A secondary source prevalent in the narrow-band-gap THz PC materials like

InGaAs, is Joule heating from dark current. The “junction temperature” TJ (at the top air-semiconductor interface at the

center of the active area) can then be estimated from elementary thermal analysis as TJ = T0 + PQ·RTH where T0 is the

ambient temperature, PQ is the total power dissipation by heat, and RTH is the device thermal resistance. Being a planar

device and assuming a round heating area of radius REQ, we can re-write this as TJ = T0 + PQ/[(2)1/2REQ], where is

the bulk thermal conductivity [12]. For a typical GaAs photomixer, for example, ≈ 0.45 W/cm-K and REQ ≈ 5 m, so

that RTH ≈ 964oC/W ! If we then estimate the maximum rise above ambient as 120oC (a rule-of-thumb for some GaAs

devices, corresponding to a maximum junction temperature of 150oC), the maximum total power dissipation of PQ = 124

mW. Indeed, this is close to what is observed experimentally in GaAs devices, where the combined laser power is

generally limited to 80 mW or less, and the photocurrent is typically about 1.0 mA at a maximum bias voltage of 30 V,

for a total PQ of 110 mW. To extend the lifetime of critical devices such as those packaged into sophisticated

instruments, the total laser power must be backed off about 2x below this.

MicroscopeObjective

SHGUnit

Mode-LockedEDFA

BiasSupply

Si Hyperhemisphere

PhotoconductiveSwitch

SMPMFiber

ThermopileHead

PowerMeter

EDFA specs:49 MHz PRFCenter = 1572 nmPave = 100 mWPulsewidth < 100 fs

SHG specs:49 MHz PRFCenter = 782 nmPave = 20 mWPulsewidth < 200 fs

(a)

0

0.5

1

1.5

2

2.5

3

3.5

0 50 100 150

Bias Voltage [V]

Cu

rre

nt

[mic

roa

mp

]A

ve P

ow

er [m

W]

0

0.5

1

1.5

2

2.5

3

3.5

0 50 100 150

Bias Voltage [V]

Cu

rre

nt

[mic

roa

mp

]A

ve P

ow

er [m

W]

(b)

Fig. 5. (a). Experimental set-up for testing high-power photoconductive switches. (b) Dark I-V curve and THzaverage power vs bias voltage.

III. HIGH-POWER PHOTOCONDUCTIVE SWITCH

The photoconductive (sometimes called “Auston”) switch is the oldest and simplest of the ultrafast

photoconductive devices, but not as well characterized as photomixers in terms of THz power. The reason is simple:

from their introduction in the early 1990s, photomixers were contrasted against indigenous devices such as Schottky-

diode multiplier chains, because of their potential application as local oscillators in THz superheterodyne receivers. To

qualify for this application, it was important that the photomixers minimally supply a power level adequate for driving

cryogenic superconductive mixers, for example, which means roughly 1 microwatt cw. Not having such conventional

applications allowed the PC switch to evolve successfully as the transmit and receive element in time-domains systems

without a good understanding of its absolute power capacity.

Given this situation, the author embarked on characterizing the average power of PC switches with the same

level of scrutiny normally applied to photomixers, and with similar metrological methods. To make the comparison as

objective as possible, several design factors were kept constant, including the ultrafast material (ErAs:GaAs), the

antenna design (square spiral), and the pump wavelength (780 nm). The PC switch was embedded in the three-turn, self-

complementary, square spiral antenna shown of Fig. 3(a). The active area is the 9 x 9 micron driving gap at the center of

the antenna. The experimental set-up used to characterize the PC switch is shown in Fig. 5(a). The switch was driven

by an erbium-doped fiber mode-locked-laser with a PPLN doubler to produce ~780-nm pulses [13]. Initially, a Golay

cell was used to measure the power, but was quickly driven to saturation. So it was replaced with a small thermopile

(sensitive to the mW-level) which started recording at ~0.1 mW. The results for THz average power Pave vs PC switch

bias voltage are plotted in Fig. 5(b) along with the dark current-voltage characteristics. As in typical PC switches and

photomixers, Pave rises monotonically with bias voltage and approaches a maximum value of 1.6 mW. Higher bias

voltages were not attempted because of the onset of impact ionization seen in Fig. 2. To the best of our knowledge, this

is the highest Pave ever reported for a THz PC switch and exceeds by almost ten times the initial report from a device

having similar design [14]. The discrepancy is attributed to saturation of small-signal free-space-coupled THz detectors

(Golay cell in Ref. [16]) typically used to measure power. Thermopiles are well-known for large dynamic range and the

ability to measure pulses having high peak power Ppeak. In the present case, the maximum Ppeak can be estimated from

the 150-V bias data using Ppeak ~ Pave/(frep · tp) where frep = 49 MHz is the laser repetition frequency and tp ~ 1 ps is the

approximate THz pulse width into free space. The result is Ppeak = 33 W - an impressive number for the THz region

where powerful sources, pulsed or cw, are lacking.

IV. SYSTEM APPLICATIONS

IV.A. Frequency Domain

Arguably the most successful application of photomixers to date is high-resolution THz spectroscopy based on

the fully coherent photomixing transceiver [15,16]. It consists of two THz photomixers, each driven by the same pair of

single-frequency, single-mode, temperature-tunable distributed feedback (DFB) lasers. One photomixer acts as the

transmitter, and the other as the receiver. The temperature variable provides a continuously tunable coherent tone from

below 100 GHz to 1.5 THz or higher with instantaneous linewidth of ~100 MHz or better [17]. A block diagram of one

configuration of the transceiver is shown in Fig. 6. The radiation from the transmit photomixer is coupled from the

antenna to free space through a high-resistivity silicon hyperhemispherical lens. The THz beam is then collimated using

an aspherical optic, usually an off-axis paraboloid. The reciprocal process occurs between free space and the receive

photomixer. The sample under test can be mounted either in the collimated beam half-way between the two photomixers

where the beam is collimated, or as shown in Fig. 6, close to the transmit photomixer where the beam is quite small (~3

mm diameter) and more intense.

Fixed DFBLaser

780 nm

TunableDFB Laser >780 nm

WavemeterTransimpedance

Lock-In Amp

+

Isolator

Isolator

FocusingLens

BeamCombiner

SiliconLens

THz Path

TransmitPhotomixer

ReceivePhotomixer

NanofluidicChip

Limiting Aperture.Chip Holder

Off-Axis Paraboloids

Nanofluidic Cell (Top View)

THz circular polarization

SiO2Channel

SiO2

Wall

Fixed DFBLaser

780 nm

TunableDFB Laser >780 nm

WavemeterTransimpedance

Lock-In AmpTransimpedance

Lock-In Amp

+

Isolator

Isolator

FocusingLens

BeamCombiner

SiliconLens

THz Path

TransmitPhotomixer

ReceivePhotomixer

NanofluidicChip

Limiting Aperture.Chip Holder

Off-Axis Paraboloids

Nanofluidic Cell (Top View)

THz circular polarization

SiO2Channel

SiO2

Wall

Fig. 6. Improved THz photoconductive-switch and photomixer device structure in which the ultrafast (subpicosecond-lifetime) photoconductive layer is composed of ErAs:In0.53Ga0.47As and grown on a semi-insulating indium phosphide(InP) substrate and separated from the top-side electrodes and THz circuit (antenna or transmission line) by ablocking layer of In0.52Al0.48As, which has a direct bandgap significantly larger than the ultrafast material.

Because the lasers driving receive and transmit photomixers are mutually coherent, the THz beam into the

receive photomixer is mixed down in frequency by homodyne conversion. A simple amplitude modulation on the

transmit photomixer then allows for dc offset and straightforward synchronous detection with all the benefits of

traditional homodyne transceivers. As in any coherent system, the output of the transceiver maintains phase information.

Fig. 7 shows the in-phase (I2) response (gray curve), the power response ([I2 + Q2] (black curve), and the noise floor

obtained as the power response with the THz beam blocked but all other settings kept the same. The ratio of the power

response to the noise floor is the signal-to-noise (SNR) ratio, which is ~80 dB at 200 GHz, 60 dB at 1.0 THz, and 40 dB

at 1.8 THz. These are excellent SNR values for a room-temperature system with such wide tuning bandwidth and high

resolution, and can be attributed largely to the sensitivity advantage of coherent processing over incoherent (or direct)

techniques [18]. Furthermore, the photomixing transceiver has no moving parts, runs at room temperature, and requires

no high voltages or large magnetic fields.

The power response associated with Figs. 7 also exemplifies the complicated baseline that typically occurs in

coherent THz spectroscopy. Visible at 556, 752 and at several frequencies above 1.1 THz are absorption lines from the

ambient water vapor in the ~1-foot path between the transmitter and receiver. However, away from these are other

undulations associated with variations in the intrinsic system transfer function. Fortunately, these undulations are not so

deep or plentiful as to preclude high-resolution spectroscopy.

Frequency [GHz]

Po

we

r [A

rbU

nit

s]

60 dB

80 dB

40 dB

Noise Floor

Phase Sensitive

Magnitude

Frequency [GHz]

Po

we

r [A

rbU

nit

s]

60 dB

80 dB

40 dB

Frequency [GHz]

Po

we

r [A

rbU

nit

s]

60 dB

80 dB

40 dB

Noise Floor

Phase Sensitive

Magnitude

Fig. 7. Transfer function of coherent photomixing transceiver along with background noise floor. The phase-sensitivecurve plotted in gray is the in-phase (I) output of the receive photomixer (Ref. [21]).

A unique aspect of this instrument already utilized but not widely appreciated is the combination of spatial,

coherence, temporal coherence, and wide tunability. The vertical orientation in Fig. 6 allows one to orient small

samples, such as nanofluidic chips, horizontally. This facilitates the initial wetting and subsequent filling of the

channels. It also allows for locating the chip immediately below the transmit photomixer-coupling lens (a silicon

hyperhemisphere) where the spot diameter is small (~3 mm diameter), as determined by the photomixer spiral antenna

and the thickness of the lens. Assuming an average power of 1.0 W and instantaneous linewidth of 20 MHz, the THz

beam at this point has a spatial intensity of ~1.4x10-5 W/cm2, and the power spectral intensity is 0.7x10-3 W/cm2-GHz.

We have found the latter quantity to be a good performance metric for THz sources in wideband spectroscopy.

THz transmission experiments were carried out with the coherent photomixing transceiver customized for high-

resolution measurement of weak absorption signatures, and a nanofluidic chip designed for biomolecular spectroscopy.

By capillary action, the RNA-bearing solution filled the silica nanofluidic channels, which were 800 nm wide by 1000

nm deep, on a pitch of ~1200 nm. The raw experimental results are plotted in Fig. 8(a) in the frequency range 800 to

1200 THz - a band having two strong water vapor lines at 1098 and 1164 GHz, and a relatively weak line around 990

GHz. The top curve is the “background” signal PB through the nanofluidic chip containing buffer solution only, the

middle curve is the “sample” signal PS with RNA suspended in the buffer, and the bottom curve is the noise floor PN

obtained by blocking the THz path with a metal plate. In a typical experiment, the sequence of THz spectra acquisition

consisted of first mounting the chip within an auto-aligning rail that enables precise and repeatable positioning of the

nanofluidic chip sample in and out of the beam path of the THz spectrometer, followed by the acquisition of background

spectra of the chip in the absence of any fluids in the channels. Following this “dry-chip” background measurement, a

second “wet-chip” background was collected by placing a ~100-L drop of buffer in the nanofluidic chip fluid

reservoirs, and measuring the background spectra of the buffer-filled channels. Finally, si-RNA drops (~100-µL) were

added to the reservoirs, allowed to disperse, and the THz spectrum was measured. The measurement was repeated on

the same sample six times, and good reproducibility was obtained. The three curves in Fig. 8(a) are used to compute the

normalized and noise-referenced transmission function T vs plotted in Fig. 8(b) based on T() = [PS() – PN()] /

[PB() – PN()].

The transmission shows three prominent resonant signatures centered at 916, 962, and 1034 GHz, labeled (1),

(2), and (3), respectively in Fig 8(b). There is also a broad and weaker signature (perhaps a multiplet) between 830 and

875 GHz, and a narrow but weaker one centered at 1075 GHz. The feature around 1100 GHz is questionable since it is

mixed with a very strong water vapor line, evident from the background transmission in Fig. 8(a). These features are in

good qualitative agreement with our previous experimental results obtained by similar methodology but using silica

nanochannels fabricated on high-resistivity silicon substrates rather than fused quartz [19]. The previous results yielded

prominent resonances centered at 1034, 950, 875, and 1084 GHz, all having comparable spectral widths as presented

here but with weaker resonant absorbance and a lower signal-to-noise ratio.

IV.B. Time Domain

One of the most promising applications of THz today is in the field of biomedical imaging, particularly

burns and other lesions of human skin tissue. Most of these applications rest on the acute sensitivity of THz radiation to

water concentration. Water has long been a bane of the THz radiation region in both the vapor and liquid states. Water

vapor greatly attenuates the propagation through the terrestrial atmosphere, particularly between ~0.5 and 3.0 THz where

a large set of strong molecular rotational transitions occur. Liquid water is even worse because its attenuation occurs

over a broad continuum with absorption coefficient well exceeding 100 cm-1 above 0.5 THz [20,21,22,23]. In both

cases, the attenuation is absorptive and associated with the high built-in dipole moment (1.85 Debye) and relatively high

mobility of the H2O molecule. From the Fresnel equations, we know that strong absorption can affect the reflection too

if the associated imaginary part of the refractive index is comparable to the real part. This is exactly what happens with

water in the THz region. Furthermore, human tissue is generally a composite of water and some biomolecular material

(e.g., protein or polysaccharide, such as collagen). The biomaterials are not as polar as water, so they have little impact

on the reflection. Hence the composite reflection is a strong function of the water concentration, which is the basis for

our sensing technique.

Background

si-RNASample

Noise Floor

Frequency [GHz]

101

100

10-1

10-2

10-3

10-4

500 600 700 800 900 1000 1100 1200

Tran

smitt

ed P

ower

[Arb

Units

]

Background

si-RNASample

Noise Floor

Frequency [GHz]

101

100

10-1

10-2

10-3

10-4

500 600 700 800 900 1000 1100 1200

Tran

smitt

ed P

ower

[Arb

Units

]

(1)(2) (3)

Frequency [GHz]

si-RNA signatures

Tran

smis

sion

700 800 900 1000 1100

100

10-1

10-2

(a)

(1)(2) (3)

Frequency [GHz]

si-RNA signatures

Tran

smis

sion

700 800 900 1000 1100

100

10-1

10-2

(a)

(a) (b) Fig. 8. (a) Raw experimental data for the nanofluidic chip with pre-wet (glycerol-EDTA buffer), the nanofluidic chipwith si-RNA solution filling the channels, and the spectrometer noise floor. (b) Normalized transmission spectracomputed from the three raw data spectra in (a). The prominent attenuation signatures are labeled (1), (2), and (3).

Traditional THz time-domain imaging systems would work for this application, and has been addressed widely

in other review articles; however, this is not what we have focused on with the high-power PC switches. Instead we

have focused on a simpler and more portable type of sensor inspired by traditional radar design, specifically impulse

radar. A strong motivation for our design is affordability. At the cost of traditional time-domain systems, THz

biomedical imaging would likely only be done in medical research clinics. With a simpler impulse radar design, it might

be possible to reach a much broader medical arena, such as the urgent-care or general practitioner.

Our sensor is the THz impulse radar design presented in Fig. 9(a). The transmitter is a high-efficiency

photoconductive (PC) switch driven by a low-cost, 780-nm fiber mode-locked laser (MLL) having a pulse width of 230

fs and pulse repetition frequency (PRF) = 20 MHz. The radiation from the PC switch consists of a train of pulses, each

having ~1 ps width. In the frequency domain, the power spectrum is broadly spread over a “comb” starting with =

PRF and every harmonic thereof, and extending beyond 1.0 THz. While being a poor spectral match to molecular lines,

it is a good match to the inherently broadband reflection of liquid water. In other words, a large fraction of the THz

radiation from the PC switch contributes to the instantaneous reflected power from the sample. The reflected beam is

collected and rectified by a WR-1.5 waveguide-mounted (cutoff frequency 400 GHz), zero-bias Schottky barrier diode

having fast (<< 1 ns) impulse response and wide video bandwidth (> 10 GHz). The received power spectrum is plotted

in Fig. 9(b), showing a bandpass behavior centered at ~500 GHz. The resulting spatial resolution is far better than can

be achieved from typical ( ~ 3 mm) mm-wave imaging systems [25]. The diode output is then gated at the PRF with a

delay-controlled reference pulse, and the baseband DC component is time-averaged to achieve a good signal-to-noise

ratio (SNR). On specular surfaces such as smooth skin, the SNR reaches levels of 30 dB or higher with ~ 16 ms

integration time.

The best sensing metric for our radar is a variant of the noise-equivalent temperature difference (NET) used

widely in radiometric imaging, but here tailored to water detection – the noise-equivalent change in water concentration

(a) (b)(a) (b) Fig. 9. (a) Block diagram of THz impulse radar configured in reflection mode. (b) Power spectra obtained from the PC switch alone (dashed black line), and the portion collected by a WR-1.5 zero-bias Schottky diode (solid red line).

(NEWC). Performing calibrated evaporation experiments, we have determined NEWC 0.054%. The best

demonstration to date of this acute sensitivity was made by 2D imaging of in-vitro, “physiological” porcine skin

considered by medical researchers as a good simulant for human skin. Fig. 10(a) shows the visible photograph and 10(b)

the THz image of a branded burn with no obscuration. Fig. 10(c) shows the same burn through five layers of cotton

gauze. Fig. 10(b) displays interesting features not seen in the visible image of the burn, such as the "halo" that may

demark the spatial extent of tissue damage. Fig. 10(c) also supports the consensus that THz radiation can detect and

image through fabric materials that are opaque to infrared and visible radiation. The THz impulse radar imager is

currently in rapid engineering evolution, and our near-term performance specifications are 2D image acquisition with ~1

mm spatial resolution in <1 min over 1 KPixel and 1 sq. inch.

V. ACKNOWLEDGEMENTS

This work was sponsored by the U.S. Army Research Office, U.S. Army TATRC, and the National Science

Foundation. Special thanks goes to Dr. Dwight Woolard for supporting THz research consistently for over a decade.

“Halo”Feature

(b) (c)

(a)

“Halo”Feature

(b) (c)

(a)

Fig. 10. (a) Visible photograph of branded burn made on in-vitro porcine skin. (b) THz image of same burn. (c) THzimage of same burn through five layers of gauze. In the THz images, the spatial resolution is 2 mm, and the image sizeand acquisition time were 1 KPixel and 5 min, respectively.

REFERENCES

1 Emcore Corp, Model PB7100, http://www.emcore.com/fiber_optics/specialty_defense/terahertz_products?pid=65 2 Picometrix LLC, www.picometrix.com 3 Toptica Photonics AG, http://www.toptica.com/page/terahertz_thz_cw_packages_spectroscopy_kit_photomixers.php 4 Menlo Systems GmbH, http://www.menlosystems.com/home/products.html?cat=10 5 J. E. Bjarnason, T. L. J. Chan, A. W. M. Lee, E. R. Brown, D. C. Driscoll, M. Hanson, A. C. Gossard, and R. E.

Muller, “ErAs:GaAs photomixer with two-decade tunability and 12 µW peak output power,” Appl. Phys. Lett. , vol. 85 (no 18), p. 3983 [2004].

6 Z.D. Taylor, R.S. Singh, E.R. Brown, J.E. Bjarnason, M.P. Hanson, and A.C. Gossard, “A Photoconductive Switch with 1.3% Conversion Efficiency for Low SNR Sensing,” IEEE Sensors Journal.

7 S.S. Gearhart, J. Hesler, W.L. Bishop, T.W. Crowe, and G.M. Rebeiz, A wide-band 760-GHz planar integrated Schottky receiver, IEEE Microwave Guided Wave Lett 3(1993), 205–207.

8 K.A.McIntosh, E.R.Brown, K.B.Nichols, O.B.McMahon, W.F.DiNatale, and T.M.Lyszczarz, Terahertz measurements of low-temperature-grown GaAs photomixers coupled to resonant-planar antennas, Appl Phys. Lett 69 (1996), 3632.

9 D.B.Rutledge, D.P.Neikirk, and D.P.Kasilingam, Integrated Circuit Antennas, Infrared and Millimeter Waves, vol.10, K.J.Button (Ed.), Academic, New York, 1983, p.1.

10 E.R. Brown, A.W.M. Lee, B.S. Navi, and J.E. Bjarnason, “Characterization of a Planar Self-Complementary Square-Spiral Antenna in the THz Region,” Microwave and Optical Tech Lett, Vol. 48, No. 3, pp. 524-529 (2006).

11 E. R. Brown, “A photoconductive model for superior low-temperature-grown GaAs THz photomixers,” Appl. Phys. Lett., vol.75 (6), pp.769-71 (1999).

12 “Terahertz Generation by Photomixing in Ultrafast Photoconductors,” E.R. Brown, in Terahertz Sensing Technology, Vol. 1: Electronic Devices and Advanced Systems Technology, ed. by D.L. Woolard , W. R Loerop, and M.S. Shur (World Scientific, Singapore, 2003)

13 Mercury 780-020-INS, PolarOnyx Inc., Sunnyvale, CA. 14 Z. D. Taylor, E. R. Brown, J. E. Bjarnason, M. P. Hanson, and A. C. Gossard, “Resonant-optical-cavity

photoconductive switch with 0.5% conversion efficiency and 1.0W peak power,” Optics Letters, Vol. 31, Issue 11, pp. 1729-1731 (2006).

15 S. Verghese, K. A. McIntosh, S. Calawa, W.F. Dinatale, E. K. Duerr, and K. A. Molvar, “Generation and detection of coherent terahertz waves using two photomixers,” Appl. Phys. Lett. 73, 3824 (1998).

16 J. R. Demers, R. T. Logan Jr., N. J. Bergeron, and E. R. Brown, “A coherent frequency-domain THz spectrometer with a signal-to-noise ratio 60 dB at 1 THz,” Proc. SPIE Defense and Security, Paper# 6949-8, 16-20 March 2008, Orlando, FL

17 E. R. Brown, J. Bjarnason, T. L. J. Chan, D. C. Driscoll, M. Hanson, and A. C. Gossard, “Room temperature, THz photomixing sweep oscillator and its application to spectroscopic transmission through organic materials,” Rev. Sci. Inst., 75 5333 (2004).

18 J.E. Bjarnason and E.R. Brown “Sensitivity measurement and analysis of an ErAs:GaAs coherent photomixing transceiver,” Appl. Phys. Lett. 87, 134105 (2005).

19 E. R. Brown, E. A. Mendoza, S. R. J. Brueck, IEEE Sensors J., IEEE Sensors J. Vol. 10, 755 (2010). 20 H. J. Liebe, G. A. Hufford, and T. Manabe, Int. J. Infr. Millim. Waves, 12, 659 (1991). 21 J. T. Kindt, and C. A. Schmuttenmaer, J. Phys. Chem. 100, 10373 (1996). 22 J. Xu, K. W. Plaxco, S. J. Allen,, J. E. Bjarnason, E. R. Brown, Appl. Phys. Lett, 90, 031908 (2007). 23 Z.D. Taylor, R.S. Singh, E.R. Brown, J.E. Bjarnason, M.P. Hanson, and A.C. Gossard, IEEE Sensors Journal , vol.9,

no.1, pp. 3-8, (2009).