INVITED

PAPER

Finite-Element Method
Simulations of Guided Wave
Phenomena at Terahertz
Frequencies
Design of cylindrical-wire wave guides and other THz devices can be aided by finite
element models that characterize propagation and interaction characteristics.
By Jason A. Deibel, Member IEEE , Matthew Escarra, Student Member IEEE ,
Nicholas Berndsen, Student Member IEEE , Kanglin Wang, and
Daniel M. Mittleman, Member IEEE

ABSTRACT | As the science and engineering associated with and quality-control in manufacturing [10], [11]. Terahertz
terahertz time-domain spectroscopy and imaging evolves past time-domain spectroscopy (THz-TDS) systems use sub-
the use of conventional free-space optics, the continued picosecond pulses having a bandwidth that can span a
development of waveguides for terahertz pulses is increasingly large portion of the Bterahertz gap[ (100 GHz–30 THz).
relevant. The ability to model and simulate terahertz wave The generation of these broadband THz pulses is ac-
propagation aids in the development, visualization, and complished through the use of ultrafast lasers (pulses
understanding of novel terahertz devices and phenomena. widths G 100 fs) in combination with either photocon-
We discuss the use of the finite-element method, a powerful ductive antennas [12]–[14] or optical rectification in
computational tool for the modeling of guided wave phenom- nonlinear optical crystals [15], [16]. Likewise, broadband
ena and devices at terahertz frequencies. detection is accomplished using ultrafast photocon-
ductive antenna receivers [12], [14] or via electro-optic
KEYWORDS | Finite-element method (FEM); photonic crystals; sampling [17].
surface plasmon polaritons (SPPs); terahertz time-domain A standard THz time-domain spectroscopy and
spectroscopy (THz-TDS); waveguides imaging apparatus typically consists of an assembled
system of free-space optical components that include
lenses and mirrors for the guiding, focusing, and colli-
I. INTRODUCTION mating of both the near-infrared laser radiation (NIR) and
Terahertz time-domain spectroscopy and imaging are the terahertz radiation. The use of these bulky free-space
proven techniques for the characterization of semicon- optics hinders the development of THz technology due to
ductors [1], [2], biomedical imaging [3], [4], and trace- the complexities associated with alignment of the optical
gas detection [5]. There are potential applications for this components and the associated space requirements. As a
broadband technology in fields such as security [6]–[9] consequence, the need for effective guided wave
techniques has drawn considerable attention recently.
The development of waveguides for use with terahertz
pulses is complicated by the operational requirements of
Manuscript received March 16, 2007; revised April 17, 2007. This work was supported
low attenuation loss and low dispersion, and by the lack
in part by the National Science Foundation, in part by the R. A. Welch Foundation, of suitable materials with sufficient transparency at
and in part by the Intelligence Community Postdoctoral Fellowship Program.
The authors are with the Department of Electrical and Computer Engineering
terahertz frequencies [18]. Several research groups have
at Rice University, Houston, TX 77005 USA (e-mail: daniel@rice.edu). reported novel waveguiding structures including poly-
Digital Object Identifier: 10.1109/JPROC.2007.898817 meric photonic crystal fibers [19]–[21], parallel-plate

1624 Proceedings of the IEEE | Vol. 95, No. 8, August 2007 0018-9219/$25.00 Ó 2007 IEEE
 Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

metal guides [22]–[24], low index discontinuity wave- The completion of an FEM model of any type of
guides [25], and metallic slits [26]. We have recently electromagnetic phenomena generally involves several
described the use of bare metal wires, which can support a stages. The first is the definition of the model geometry.
surface-guided wave with low attenuation and relatively It can be as simple as defining the parallel plates of a
little dispersion [18], [27]–[29]. This new class of tera- capacitor or as complex as a coaxially fed patch antenna.
hertz waveguide is a promising step in the development of An essential component of this stage is the definition of
terahertz applications. In order for these waveguides to be the simulation domain. This domain bounds the spatial
implemented, it is vital that they be carefully characterized extent of the simulation whether it is in one, two, or
both theoretically and experimentally. three dimensions. Within the simulation domain, it may
The finite-element method (FEM) is a useful and be subdivided into smaller spatial extents known as
powerful computational technique for modeling how subdomains.
electromagnetic waves propagate and interact with their Following the geometry definition, physical properties
surroundings. With the FEM, one can determine propa- are assigned to the model. Material parameters such as
gation characteristics such as loss and dispersion and the dielectric permittivity, the magnetic permeability, the
electromagnetic field distributions as a function of spatial refractive index, or the conductivity are defined for each
location, frequency, or time. Advances in computer subdomain. For each plane, edge, and point in the model
technology and the availability of commercial FEM domain, boundary conditions are assigned. These bound-
software have lowered barriers to the use of the FEM for ary conditions can include defining a boundary plane as a
modeling of electromagnetic wave phenomena. As FEM continuous boundary, a perfect electrical conductor, or a
modeling has been applied to study electromagnetic transition impedance boundary condition. Sources are
propagation in both the microwave and optical portions defined at this point. For a dc electrostatics simulation, a
of the spectrum [30]–[32], it is a logical step to use this boundary can be given a voltage potential whereas for a
technique for the study of wave propagation at terahertz waveguide simulation, the boundary can be assigned an
frequencies. electric field source such as a port condition where a TE
In this paper, we discuss the applicability of FEM or TM mode is excited.
simulations to the study of terahertz guided wave pheno- The next stage in the FEM modeling process is the
mena. We present a brief overview of the FEM technique discretization of the simulation domain. This spatial region
and describe the critical aspects of modeling electromag- is subdivided into numerous discrete elements that are
netic wave propagation and interaction at terahertz fre- much smaller than the subdomains that make up the
quencies. Simulations of the excitation of terahertz simulation domain. The ensemble of all of these elements
surface waves on metal wire waveguides are presented. in the domain is often referred to as the mesh. The
Next, we show how FEM modeling is used to develop and geometric nature of the mesh elements is a function of the
test a novel photoconductive antenna with radial symme- dimension of the simulation i.e., triangular and rectangu-
try that greatly enhances the coupling efficiency of THz lar mesh elements for 2-D simulations and tetrahedral and
radiation to wire waveguides. Finally we discuss the emis- block elements in 3-D simulations. A system of nodes
sion of THz radiation at the end of the wire waveguide. arises from the creation of the mesh. Interpolation
functions are defined between the nodes that are chosen
to approximate the solution. This node system, in
II . FINITE-ELEMENT METHOD conjunction with the previously defined physical para-
MODELING OF ELECTROMAGNETIC meters and boundary conditions, forms a system of partial
WAVE PROPAGATION differential equations (PDEs).
The system of PDEs must be solved in order to
A. Basic Steps of Finite-Element Method Modeling determine the electromagnetic fields within the simulation
At its most basic level, the FEM provides an domain. Depending on the size and complexity of the
approximate solution to a system of linked partial simulation, the model solution can be arrived at using
differential equations. The FEM was first introduced by either direct solvers such as UMFPACK or SPOOLES or
Courant in 1943 and its development has matched the pace iterative solvers such as GMRES [37]–[39]. While most
of the growth of computer technology [33]. It was not until FEM simulations are completed in the frequency-domain,
the 1960s that the FEM was applied to the problem of time-domain modeling is also possible [40], [41]. The
electromagnetic wave propagation when it was used to computational time required to solve the system of PDEs
analyze waveguide modes [34]. As the focus of this paper is and arrive at a solution varies with the dimension of the
on the study of terahertz wave phenomena and devices simulation, the number of mesh elements, and the type of
using the FEM and not an explanation of the functionality electromagnetic phenomena being modeled. Once a
of the FEM, it is recommended that the reader consult Jin’s solution is found, parameters such as the electric field,
and Zimmerman’s work for a more detailed explanation of magnetic field, capacitance, etc., can be plotted and
the FEM [35], [36]. analyzed.

Vol. 95, No. 8, August 2007 | Proceedings of the IEEE 1625

Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

B. Electromagnetic Wave Propagation commercially available FEM software packages to choose

Modeling Issues from (HFSS, ANSYS, ANSOFT, COMSOL Multiphysics).
The modeling of electromagnetic wave propagation can Several of these offer graphical user interfaces (GUIs) that
be quite challenging. To properly simulate wave propaga- allow the user to visually construct the simulation domain
tion, the largest mesh element size must be no larger than and to perform postprocessing of the solution. Commer-
one-third of the radiation’s wavelength. Many models cial FEM software for electromagnetics simulations can
require an even higher mesh resolution constraining the vary from versions that only solve for one specific type of
largest mesh element to one-tenth of the radiation wave- problem such as electrostatics or wave propagation to
length [36]. For the latter case, to model a wave propa- ones that can solve a wide variety of electromagnetics
gating at a frequency of 100 GHz, the largest mesh element problems. Programs such as COMSOL Multiphysics
can be no larger than 300
m; for a wave at 1 THz, this allow the user to couple multiple application modes to-
critical element size shrinks to 30
m. However, a real- gether [45]. In this paper, all of the simulations were
world model can involve feature sizes not only that small performed using the COMSOL Multiphysics package.
but also as large as centimeters. Thus a small critical mesh-
element size results in a model with a huge number of
mesh elements. The number of mesh elements not only I II . MODELING THE EXCITATION
affects the computational requirements of the simulation, OF THz SURFACE WAVES ON
but can also determine whether or not a solution can be ME T AL WIRE WAVEGUIDE S
determined.
A. Metal Wire Waveguide
C. Computational Requirements Experimental Configuration
FEM simulations can be performed on any personal In our early work with the metal wire waveguides, the
computer or workstation that is capable of running radial mode of the waveguide was excited via a scattering
mathematical software packages such as Matlab. As with mechanism [18], [27]. In this method, horizontally
most computational techniques, the time required by an polarized terahertz pulses, generated using a GaAs
FEM solver to arrive at a solution (Bthe solution time[) photoconductive antenna, are focused onto the 0.9 mm
decreases with increasing processor speed. With some diameter stainless steel wire waveguide (Fig. 1). An
computational techniques such as the finite-difference identical stainless steel wire is placed perpendicular to
time-domain method, the use of multiple processors in a the waveguide near the focal spot of the beam. A small
parallel computing arrangement can greatly decrease the portion of the radiation scattered at this intersection point
solution time. FEM solvers are difficult to implement excites the radially polarized surface wave that propagates
across parallel computing geometries [42]. The primary along the surface of the metal wire. The radial polarization
computer component that governs FEM simulations is the of the surface wave is due to the circular cross section of
RAM memory. As an FEM model’s total number of mesh the metal wire, which causes the electric field of the
elements increases, the system of PDEs grows as well, thus weakly guided mode to point away from the surface. This
increasing the RAM memory required for the computer to radially polarized guided wave is known as a Sommerfeld
be able to solve the model. The advent of 64-bit computing wave [46], [47] and is equivalent to an azimuthally pola-
technology and multiprocessor systems that allow the rized surface plasmon polariton (SPP). The SPP arises
assignment of 8 GB of RAM per processor has significantly
advanced the capabilities of FEM modeling as more
complex models with larger mesh densities can be now
be solved. In addition, new solvers such as the Pardiso
[43], [44] solver are being developed that enable FEM
simulations to be completed using parallel processing.

D. Commercial FEM Software

Commercial FEM software did not exist for the
several decades following its first description in 1943
[33]. Even with the advent of the personal computer and
the computer workstation, FEM simulations could only
be accomplished by those who understood the mathe-
matical intricacies of the method and could develop the
necessary computer code to run the simulations. These
limitations hindered the application of the FEM for
electromagnetic device development and the understand- Fig. 1. Excitation of THz surface wave on a metal wire using a
ing of novel phenomena. There is now a vast array of scattering configuration.

1626 Proceedings of the IEEE | Vol. 95, No. 8, August 2007

 Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

from the collective oscillations of the conduction elec- the EM waves in the simulation are minimized. An
trons at the metal surface, in response to the driving absorbing boundary condition or perfectly matched layer
electromagnetic surface wave [48]. The terahertz pulse is would be more physically accurate, but is not used for
detected at the end of the wire with a fiber-coupled these simulations due the modeling complexities and
photoconductive receiver. The receiver used here was a computational constraints introduced through their either
conventional bow-tie antenna, sensitive only to the pola- implementation [36].
rization component oriented parallel to the bow-tie axis The 3-D simulation domain is discretized into approx-
[49]–[51]. As the THz radiation propagates along the imately 1.8 million tetrahedral mesh elements yielding a
surface of the wire waveguide, there is no terahertz field computational model consisting of 2.1 million degrees of
present at the central axis at the end of the wire. De- freedom. The large number of mesh elements is due to the
tection measurements are consequently performed at mesh density constraints that are a function of the
lateral offsets from the central axis. radiation wavelength, as noted above. The experiment is
simulated using a time-harmonic solver, so that only one
B. Finite-Element Method Model input frequency is considered at a time. The model
Experimental results from the coupling configuration problem is solved using an iterative generalized minimal
described in Fig. 1 indicate that less than 1% of the power residual (GMRES) iterative solver with symmetric succes-
of the incident terahertz beam is coupled to the wire sive overrelaxation (SSOR) matrix preconditioning [39],
waveguide by the scattering process described above. A [45]. For a waveguide 15 cm long, a workstation with dual
more accurate estimate of the coupling efficiency is re- 64-bit processors and 14 GB of RAM arrives at a solution in
quired. We have developed an FEM model of the dual- less than 20 h. The solution time could be lessened with
wire THz coupler that demonstrates the nature of the the use of a geometric multigrid preconditioner, which was
coupling and that also provides a method to more ac- unavailable at the time when these simulations were
curately quantify the coupling efficiency. The wire wave- performed. The multigrid technique employs a hierarchy
guide is modeled in three dimensions, with the outer of meshes each with different mesh densities [52].
surface of the metal wire defined as a perfect electrical Solution time is decreased as the software algorithms
conductor (PEC). To correctly model the ohmic loss of only employ the fine mesh when necessary and otherwise
the waveguide, it would be necessary to consider the use coarser meshes when appropriate.
surface impedance of the wire’s outer boundary and the
Drude conductivity of the metal [18]. However, at this C. FEM Simulation Results
point our focus is on modeling the coupling efficiency of Simulation results completed for a frequency of
the experiment and not to consider the waveguide’s 0.1 THz are presented in Fig. 2 [53]. The incident plane
inherent loss. wave is scattered at the gap between the wire coupler and
Fig. 1 shows the experimental configuration used to waveguide. It is evident that the majority of the plane
measure the waveguide properties. The coupler wire, wave propagates unimpeded and only a small amount of
having the same diameter and boundary conditions as the the incident radiation is coupled to the radial mode of the
waveguide wire, is placed in the vicinity of the wire waveguide. Fig. 3 is a plot of the x-component of the
waveguide, but in a direction perpendicular to it. The electric field at a point 300
m above the wire and along
distance between the closest outer surfaces of the two its z-axis. The large peak located at z ¼ 0 is part of the
wires is 500
m. The incident linearly polarized terahertz incident excitation wave. A sine wave with a frequency of
field is modeled as a plane wave whose k-vector is 0.1 THz is fit to the oscillating electric field along the
incident at a 45
angle from the axis of the wire wave- waveguide, demonstrating that the model is successfully
guide. This angle of incidence is chosen based on qua- simulating wave propagation. Both forward and backward
litative experimental results indicating that this geometry propagating modes are excited. Since the wires are
optimizes the input coupling. The propagation direction modeled as PECs, we expect little attenuation once the
of the plane wave is chosen so that its center is incident mode is excited, which is consistent with the results of the
at the center of the gap between the waveguide and calculation. Any noise or amplitude fluctuation present in
coupler wires. The size of the plane wave (2 cm di- the plot in Fig. 3 is due to an insufficient mesh density
ameter) mirrors the size of the loosely focused terahertz which is limited by the computational capabilities of the
beam used in the waveguide coupling experiment. The simulation computer. A plot of the x-component of
vector of the linear input polarization lies in the same the guided wave’s electric field at the end of the wire
plane (xz plane) occupied by the long axis of the wire (xy plane) is shown in Fig. 4(a). It is compared to
waveguide, again mirroring the experiment. The simula- experimental data in Fig. 4(b). The spatial profile of the
tion domain is bounded by a box of air confining the electric field at the end of the waveguide is measured by
waveguide and coupler wires and the input plane wave. A translating the THz receiver antenna in the plane normal
low-reflecting boundary condition is selected for the to the waveguide axis [27]. The two-lobed structure is what
outer walls of the domain so that any back reflections of one would expect in observing a radially symmetric

Vol. 95, No. 8, August 2007 | Proceedings of the IEEE 1627

Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

The simulation results demonstrate the propagation of

a surface wave, or SPP, supported by the wire waveguide.
The boundary condition chosen for the surface of the
wires is a perfect electrical conducting (PEC) condition,
meaning that the model treats the volume bounded by
the surface of the wire as if its conductivity were infinite.
However, the Sommerfeld model predicts that a PEC
wire cannot support a stable radial mode [46], [54]. Due
to limits on the size of the simulation domain imposed by
computational capabilities, we can only model propaga-
tion along less than 15 cm of wire. This distance may not
be long enough to adequately determine whether or not
the propagating mode is stable and is supported by the
waveguide. While the length of the wire (z) in our
simulation is limited, a more important condition is
imposed by the finite lateral (x and y) extent of the
Fig. 2. FEM simulation result of a 0.1 THz wave coupling to a wire
simulation domain which is also limited by computational
waveguide using a dual-wire coupling configuration (Ex shown here).
The inset shows a zoomed-in view of the electric field (xz plane)
capabilities. It is possible that the low-reflecting bound-
in the coupling region between the two wires. ary conditions chosen for the outer walls of the simu-
lation domain may perturb the simulation results. Future
simulations replacing the low-reflecting boundary con-
dition with a perfectly matched layer or absorbing
boundary condition would potentially yield insight into
polarization pattern with a polarization-sensitive detector.
this aspect of our simulation results. Furthermore, it is
Thus, this result demonstrates the radial nature of the
not possible from these simulation results to determine if
waveguide mode.
single-mode excitation is taking place. The surface waves
seen in the simulation can be described as a superposi-
tion of many modes, all but one of which exhibit ex-
tremely high radiation losses [46]. While these FEM
simulations do not demonstrate the excitation of stable
single-mode propagation along the waveguide, they do
effectively demonstrate the coupling of terahertz radia-
tion to the waveguide.

Fig. 4. (a) FEM simulation results a 0.1 THz continuous wave showing
the x-component of the electric field at the end of the metal wire
waveguide (0.45 mm radius) (xy plane shown). (b) Experimental
measurements showing the spatial profile of the electric field at the
end of metal wire waveguide (0.45 mm radius). This image is obtained
by translating the THz receiver antenna in the plane normal to the
waveguide axis. This data represents a spectrum-weighted average
as all of experimental data was obtained at the same time-delay.
Fig. 3. A plot of the x-component of the electric field along a line Red indicates positive values and blue indicates negative values of
300
m above the wire and parallel to the axis. The inset shows the electric field. The polarity reversal shown in both figures is
a 0.1 THz sine wave fit to the extracted simulation data. evidence of the radial nature of the waveguide mode.

1628 Proceedings of the IEEE | Vol. 95, No. 8, August 2007

 Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

It is now quite simple to quantify the coupling ductor substrates in a cleanroom environment, terahertz
efficiency at 0.1 THz. In our model geometry, the incident antenna fabrication can be time consuming and financially
excitation wave is defined at the base of a cylinder that can costly. Trial-and-error development and testing of novel
be seen in Fig. 2. The input power is determined by antenna designs is not cost-effective. In a situation such as
integrating the time-averaged power over the area defined this one, where novel antenna designs are being developed
by the circular base of the cylinder, where the input wave to function with novel waveguide designs, it is extremely
is excited. Likewise, the power coupled to the waveguide beneficial to model the new antenna designs and
mode is calculated by integrating the time-averaged power waveguide configuration prior to device fabrication and
over a circular area normal to the z-axis at the end of the testing. We show that the FEM is an important design tool
waveguide (15 cm from the point of excitation). These in the development of novel terahertz antenna and
calculations produce a simulated power coupling effi- waveguides.
ciency for the dual wire coupling configuration of 0.42%,
comparable to the estimate generated from experimental B. Analytical Model
data [18]. It should be noted that this number is probably Both analytical methods and FEM simulations show
the upper limit on any expected coupling efficiency. The that an idealized radial antenna can produce a radially
wires in this simulation were modeled as PECs, so no loss polarized terahertz beam [57]. The idealized radial
due to finite conductivity was considered. Also, while the antenna ignores the effects of the feed electrode and the
efficiency is expected to vary with wavelength, wire sepa- break in the outer electrode present in the actual design
ration, etc., there is no reason to expect it to increase (Fig. 5). A simple calculation illustrates the principle of a
greatly, since the polarization of the waveguide mode and Bradial array[ of Hertzian dipoles. Neglecting the influ-
the incident wave are so poorly mismatched. It is im- ence of the dielectric substrate, the radiated field can be
portant to note that it is virtually impossible to develop a thought of as a superposition of a large number of dipole
closed-form analytical approach that would produce the fields, emitted by a series of point dipoles distributed on a
coupling efficiency. The versatility and utility of the FEM circle, each pointing radially away from the origin. The
approach provides a convenient method of modeling field from each one of these point dipoles can be found by
laboratory results of this sort. simply shifting and rotating a classical far-field dipole
pattern [58]. The radiated power is computed using
superposition, as
IV. ENHANCED COUPLI NG OF
THz RADIATION TO METAL
1
~ Þ

WIRE WAVEGUIDES n  ð~
P  Re ~ EH
2 " !#
1 XN X
N
A. A Photoconductive THz Antenna With n
¼ Re ~ ~
Ei  ~i
H 
(1)
Radial Symmetry 2 i i
In order to address the low coupling efficiency
associated with the early terahertz metal wire waveguide
experiments, the poor spatial overlap between a linearly where ~Ei and H~i are the electric and magnetic fields for
polarized THz source and the radial mode of the wave- each dipole element. Fig. 6 shows the results of this com-
guide must be addressed. Prior to 2004, standard methods putation (solid curve) completed at the single frequency
of THz pulse generation, such as photoconductive an-
tennas or optical rectification, produce linearly polarized
light [55]. Nahata et al. addressed the coupling problem via
modification of the waveguide itself [56]. We have focused
our efforts on developing a novel photoconductive
terahertz antenna with radial symmetry that can ade-
quately addresses the coupling issue [47], [53], [57].
Traditional photoconductive antenna design employs a
linear dipole configuration or similar, resulting in linearly
polarized pulses. We instead have proposed an antenna
with a cylindrical symmetry to produce a Bradial array[ of
Hertzian dipoles (Fig. 5). A short optical pulse excites the
region between the two dc-biased electrodes creating an
annular current surge induced by the acceleration of the
photogenerated carriers in the illuminated region. As the
electrode patterns on photoconductive antennas are Fig. 5. (a) A schematic of the ideal radially symmetric antenna pattern.
typically photolithographically defined on the semicon- (b) A schematic of the actual radially symmetric antenna design.

Vol. 95, No. 8, August 2007 | Proceedings of the IEEE 1629

Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

C. FEM Simulation of an Idealized Radial Antenna

The analytical model described above is adequate for
describing the emission from an idealized radial antenna
in free-space. However, it can not easily be modified to
account for the effect of a dielectric substrate. To verify
the analytical model and account for the effects of a
high dielectric substrate, we model the ideal antenna
design and its generated electric field pattern using the
FEM. A current distribution is defined in the yz plane
(the plane of the antenna) in a disc-shaped domain
centered at the origin and bounded by an inner ðRinner Þ
and outer radius ðRouter Þ. The annular current pattern of
the antenna is modeled by assigning a time-varying
current to this disc

hx y i
~
J¼ e r x^ þ e r ^y "0 j!ej!t (2)
r r
Fig. 6. The calculated far-field radiation pattern in the xz plane,
neglecting substrate effects. The solid horizontal line indicates the
z axis, which is the rotational symmetry axis of the antenna. where  and  are constants chosen such that the
The open circles show the results of the FEM simulations. magnitude of J decays to zero at the outer edge of the disc.
This current distribution is surrounded by a spherical
surface in the far field of the antenna, on which the
emitted field is computed [Fig. 7(a)]. Low reflecting
of 1 THz. It is evident that no radiation is emitted in the boundary conditions are selected for the model domain
plane of the antenna or along the axis of cylindrical sym- boundaries in order to minimize reflections of the fields
metry perpendicular to the plane of the antenna. By rota- generated by the antenna. As with prior models, an ab-
ting this result around this symmetry axis, one obtains sorbing boundary condition or perfectly matched layer is
the full three-dimensional far-field radiation pattern, more physically appropriate, but is not chosen in order to
clearly a Bdonut[ mode, as expected for radially polarized minimize the computational load associated with the
emission. model. A mesh of approximately 60 000 elements is used

Fig. 7. (a) The model geometry of the FEM simulation of the idealized radial antenna. (b) The generated mesh from the same FEM simulation.

1630 Proceedings of the IEEE | Vol. 95, No. 8, August 2007

 Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

[Fig. 7(b)]. We first model the idealized radial antenna in antenna on a dielectric half-space, the lobed pattern is
free space, and then model the antenna situated on a largely preserved, except that most of the energy is ra-
dielectric half space. We have chosen a dielectric of diated into the substrate (greater than 98%) [Fig. 8(b)],
" ¼ 12:25, the approximate value for GaAs in the THz similar to the effect observed for conventional dipole sub-
range. Due to the low absorption in GaAs at terahertz strate antennas [59].
frequencies, the loss in the dielectric substrate is ignored
[2]. A stationary linear direct solver (UMFPACK) was D. FEM Simulation of the Actual Radial Antenna
employed to solve the model and analyze the electromag- A Multiphysics approach is used for the FEM
netic fields emitted by the antenna at a variety of simulations of the radially symmetric terahertz antenna
frequencies for both the free-space and dielectic half- [Fig. 5(b)] [45]. With this approach, two linked simula-
space models. tions are performed one after another. The first simu-
The results from the FEM simulations of the idealized lation is a dc electrostatics model in which the dc electric
radial antenna in free-space agree well with those from fields present in the radial antenna gap are determined.
the analytical model previously described (see open circles The second simulation is an electromagnetic wave
in Fig. 6). Good agreement is observed in both the angle propagation model. The solution from the first simula-
and width of the emitted lobe. Fig. 8(a) is a three- tion, which contains the static electric field in the gap
dimensional plot of the FEM simulation further demon- area, is used as the amplitude in the time-varying exci-
strating the radial polarization of the beam emitted by the tation field for the wave propagation simulation. Both
simulated ideal antenna in free-space. In the case of an simulations rely on the same model geometry and mesh,
but the physical parameters, boundary conditions, and
master equations are different for each.
The paired simulations are run in 3-D. The radial
antenna geometry is defined in a 2-D plane where the
center of the antenna consists of a circular electrode 5
m
in diameter, fed by an electrode 1
m wide that approaches
the center from the negative y-axis direction. The
separation between the center electrode and the 10
m
wide outer electrode is 100
m. There is a 7.5
m gap
on each side of the feed electrode at the bottom of the
antenna where the outer ring electrode approaches the
feed electrode. The boundaries of the electrodes are
defined as PECs. The 2-D geometry containing the radial
antenna structure is placed at the interface between a
section of air and a 0.5 mm thick dielectric substrate,
with a dielectric constant of 12.25, approximately equal
to that of GaAs. A substrate-matched hyperhemispherical
silicon dome (2 mm radius) is placed on the other side
of the GaAs substrate in order to couple the beam into
free space and collimate it. This model configuration is
chosen such that it closely mirrors a photoconductive
antenna transmitter assembly commonly used in tera-
hertz time-domain spectroscopy [60]. In order to ade-
quately characterize the influence of the feed electrode
on the radial antenna design, we also model an
idealized radial antenna (with no feed electrode and
no gap in the outer circular electrode) with a 100
m
radius on the same GaAs substrate with the same silicon
dome configuration.
In the first dc electrostatics simulation, the outer
electrode is grounded and a potential is assigned to the
center electrode. A charge density with a Gaussian dis-
Fig. 8. (a) FEM simulation of the power emitted by an Bideal[ tribution and a 1/e width of 40
m is defined at the
radial antenna in free space at 0.5 THz. The antenna is located at
center of the antenna. This is done to mimic the charge
the center of the sphere within the xy plane. (b) A polar plot of the
radiation pattern for the radial antenna on a dielectric half-space,
carriers generated in the GaAs by the optical pump pulse
extracted from FEM simulation results. The vast majority of the in a typical photoconductive generation scheme. The
power is radiated into the high dielectric substrate. electrostatic fields are computed using the FEM solvers

Vol. 95, No. 8, August 2007 | Proceedings of the IEEE 1631

Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

Fig. 9. FEM simulation results of the radial antenna in a typical THz configuration. (a), (b) Plots of the x and y component of the electric field
for the idealized radial antenna. (c), (d) Plots of the x and y component of the electric field for the actual radial antenna.

and then those fields are used as the time-varying input field being stronger in amplitude than the x-component
fields for the electromagnetic wave propagation model. and the strongest parts of the x-component of the
These models, both the ideal and the actual radial antenna generated field not being centered on the antenna. The
devices, typically consisted of approximately 90 000 mesh spatial overlap of the fields between these two sets of
elements and were run in less than 4 h on Pentium PC simulations (idealized antenna versus actual antenna)
with 2 GB of RAM. can be calculated. Approximately 60% of the power
Fig. 9 shows the simulated radiation fields at 0.1 THz generated by the actual antenna emerges in the form of a
in the xy plane immediately after the silicon dome, radial mode.
which is visible in addition to the GaAs substrate. The
FEM results show that the field from the idealized an- E. Enhanced Coupling Capability
tenna is perfectly radial. The generated field from the In order to fully evaluate the effectiveness of the radial
actual radial antenna structure [Fig. 9(c) and (d)] is not antenna design prior to actual device fabrication, it is
perfectly radially polarized. This is due to the lack of helpful to perform additional FEM simulations of the
symmetry in the actual antenna design. The break in the coupling capability of the radial antenna’s output to a
outer electrode and the presence of the feed electrode metal wire waveguide. The geometric models used in the
create asymmetry in the lower half of the antenna previous simulations are modified to include the addition
structure. This results in the y-component of the electric of a wire waveguide directly end-coupled to the center and

1632 Proceedings of the IEEE | Vol. 95, No. 8, August 2007

 Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

56%, an improvement of more than 2 orders of mag-

nitude over the 0.42% coupling efficiency obtained via
the scattering mechanism discussed earlier. Additional
simulations with the waveguide removed were also
performed and showed that, for both the case of the
idealized and actual antenna structures, only a very tiny
amount of power propagated to the end of the simulation
domain.
Following this simulation work, radially symmetric
photoconductive antennas were fabricated and tested
[53]. A radially polarized mode was excited and guided
over the length of a 27 cm wire waveguide. The magnitude
of the peak terahertz electric field detected at the end of
the waveguide was twenty times larger than the radiation
detected at the same position with the waveguide re-
moved. A quantitative comparison of the two coupling
Fig. 10. Geometry of the FEM simulation model of radial antenna
schemes presented here is difficult. In order to determine
coupling to a wire waveguide. the coupling efficiency with either the scattering config-
uration or the radial antenna one, a single time-domain
measurement at a single spatial location before and after

exterior of the silicon domes (Fig. 10). The wire wave-

guide’s dimensions, 0.9 mm diameter, are identical to the
waveguides used in our previously published results [18],
[27]. The wire waveguide is modeled as a PEC. For simu-
lations involving the idealized antenna, the wire wave-
guide is 2.75 cm long; whereas those with the actual radial
antenna design the waveguide is 1.75 cm long. This dif-
ference is a result of computer capability constraints,
which limit the maximum number of mesh elements in
any given model. Due to its more complex electrode
structure, the radial antenna requires more mesh elements
than the idealized antenna. Consequently, the wire
waveguide is shortened in order for the simulation to be
capable of realizing a successful solution.
Fig. 11 presents the results of FEM simulations
performed at 0.1 THz of the idealized and actual radial
antennas coupled to wire waveguides. The idealized
antenna is clearly capable of exciting the low-order radial
mode of the wire waveguide. Likewise, the actual an-
tenna design is also shown to be capable of exciting the
radial mode, although with somewhat lower efficiency. A
more quantitative observation can be obtained by per-
forming a coupling efficiency calculation similar to the
one performed in Section III. The power available to be
coupled into the waveguide is calculated by integrating
the time-averaged power over a boundary in the xy plane
immediately after the silicon dome. This is the same
plane where the electric field is plotted in Fig. 9(c) and
(d). The power coupled into the waveguide is determined
in a similar manner. The time-averaged power is integ- Fig. 11. FEM simulation model of radial antenna coupling to a wire
waveguide. (a) Plot of Ex at the end of a waveguide coupled to an
rated over a boundary at the end of the waveguide and
ideal radial antenna. (b) Plot of Ey at the end of a waveguide coupled
normal to it. The areas integrated over in both cases are to an ideal radial antenna. (c) Plot of Ex at the end of a waveguide
identical. For the actual radial antenna design, this cal- coupled to an actual radial antenna. (d) Plot of Ey at the end of
culation yields a coupling efficiency of approximately a waveguide coupled to an actual radial antenna.

Vol. 95, No. 8, August 2007 | Proceedings of the IEEE 1633

Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

the waveguide is not sufficient. To determine the ex-

perimental coupling efficiency, measurements of the total
electric field over a large spatial region must be performed
and are difficult and time-consuming. A more qualitative
comparison of measurements from both coupling config-
urations shows that the signal-noise-ratio of the THz field
measured at the end of the waveguide excited using the
radial antenna is significantly higher than that of the field
measured at the end of wire excited using the scattering
configuration [18], [53].
It is noteworthy to stress again the importance of
conducting the FEM simulations prior to antenna device Fig. 12 (a) Electrode design of the second-generation photoconductive
fabrication and characterization. Our simulations of a THz antenna with radial symmetry. (b) FEM simulation plot of the
radial antenna at an air-dielectric interface show that x-component of the electric field generated by the radial antenna.
over 98% of the generated power is radiated into the (c) FEM simulation plot of the y-component of the electric field
generated by the radial antenna. (d) FEM simulation plot of
dielectric [57]. The conventional substrate þ matched
the power generated by the radial antenna.
silicon dome configuration that is modeled in our simu-
lation geometry permits most of the power available to be
coupled to the waveguide. Others have reported exper-
imental results using a radial antenna similar in design to
ours, but in a configuration involving optically pumping electrode fed above and below by 2
m wide lines. The
the antenna through the dielectric substrate [28], [47]. distance between the center electrode and the ring
This significantly limits the amount of power available to electrodes (10
m wide) is 75
m. As with the earlier
be coupled to the waveguide because most of the emitted simulations, the second-generation radial antenna is
THz power is lost into the high dielectric substrate. placed on a 0.5 mm thick GaAs substrate with a silicon
While other methods of generating radially polarized THz substrate dome lens on the opposite side. Simulation
radiation [61] and exciting the primary mode of metal results are plotted in Fig. 12(b) and (c). When examining
wire waveguides [56] have been reported, the FEM both the generated x-component and y-component of the
simulations in conjunction with experimental character- electric field, the polarity reversal and symmetry asso-
ization show that a properly designed photoconductive ciated with radial polarization are evident. The improve-
terahertz antenna can produce a radially polarized beam ment in the quality of the radial polarization emitted by
that couples very efficiently to the primary mode of a the second-generation antenna is further established
metal wire waveguide. upon examination of a plot of its power emission
[Fig. 12(c)]. Fabrication of these second-generation de-
F. Second-Generation Radial Antenna vices is currently under way.
The knowledge and insight gained from the FEM
simulations of the terahertz radial antenna suggests
design enhancements that could be made to the antenna V. THz E MI SS ION FROM ME T AL
that would improve its ability to generate radially WIRE WAVEGUIDES
polarized light. From Fig. 9(c) and (d), it can be in-
ferred that the presence of the feed electrode in the A. Motivation
antenna structure can be blamed for the asymmetry in In addition to the study of the excitation of terahertz
the generated terahertz electric field. We propose a surface plasmon polaritons on metal wire waveguides, it
design for a second-generation radial antenna that ad- is also important to understand what happens when the
dresses this issue. The proposed design [Fig. 12(a)] relies THz radiation propagates to the end of the wire. In most
on the addition of a 2nd feed electrode, placed directly of the reported experiments involving terahertz metal
opposite to the original feed electrode. The new design wire waveguides, the THz pulses are detected by
then possesses two breaks in the outer electrode struc- allowing them to propagate to the end of the wire and
ture, but now has symmetry about both x- and y-axes. emit into free-space [18], [27], [28], [47], [53], [56],
This design also provides the ability to adjust the cur- [62], [63]. In order for experiments involving THz metal
vature of the outer ring electrodes in order to tune the wire waveguides to be properly interpreted, it is
emitted polarization. necessary to characterize the spatial properties of this
The second-generation design is simulated in a manner emitted THz radiation as it transitions from the near-
very similar to the original design using the same 2 stage field at the end of the cylindrical wire to the far-field.
Multiphysics FEM approach. The simulated antenna While experimental measurements are necessary to ac-
structure consisted of a 10
m diameter circular central complish this, it is also essential that a theoretical model

1634 Proceedings of the IEEE | Vol. 95, No. 8, August 2007

 Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

of this near-field to far-field transition be developed. To long placed inside of a cylindrical subdomain (3.75 cm
our knowledge, closed-form analytical expressions that in length, 6.5 mm in diameter) which bounds the
describe this near-field to far-field transition do not exist. volume of the simulation. The large simulation domain
As a result, numerical modeling can again play an is necessary because a long extent of waveguide is re-
important role. quired for the excited THz surface wave to exhibit
Experimental characterization of the SPP emission at stable propagation. Electric field excitation of the
the ends of metal wire waveguides has found that the Sommerfeld wave is accomplished by assigning a time-
measured spectra of these terahertz pulses are spatially varying electric field to one of the circular bases of the
dependent [64]. Fig. 13 is a schematic of one type of bounding cylinder. This excitation is based on an ana-
such characterization experiment. Measurements are lytical solution describing the propagation of a Sommer-
performed by varying the distance d between the feld wave [46]. The external boundaries, with the
detector and the end of the wire at a fixed transverse exception of where the Sommerfeld wave is excited,
distance w. As the distance d is increased, the spectral are defined as Bmatched[ boundaries which minimize
maximum of the measured THz pulse shifts to higher the possibility of back reflections of any propagating
frequencies. In addition, the peak amplitude of the spec- electromagnetic waves. The metal surface of the wire
trum decreases. waveguide is assigned a transition boundary condition
that is dependent on the metal’s surface impedance and
B. Frequency Dependent Diffraction finite conductivity. The simulation mesh consists of over
To better understand such experimental results, we 1.5 million mesh elements. The solution is obtained
again employ the FEM to simulate the propagation of a using a time-harmonic iterative solver that employs a
Sommerfeld wave [46] at THz frequencies along a GMRES method with SSOR matrix preconditioning that
metal wire waveguide and the subsequent emission of solves for only one radiation frequency at a time [39]. A
THz radiation into the far-field [64]. The model geom- workstation with dual 64-bit processors and 16 GB of
etry consists of a 0.9 mm diameter waveguide, 2.5 cm RAM typically arrives at a solution after 3 h of compu-
tational time. Computer limitations do limit the fre-
quencies that can be simulated due to the increasing
number of mesh elements required for simulations at
higher frequencies. This issue is further complicated by
the nearly 4 cm long simulation domain which also re-
sults in a large number of mesh elements.
The FEM simulation results provide an explanation of
the aforementioned experimental measurements. Fig. 14
shows 2-D images of the calculated spatial distribution of
the THz intensity at and around a 0.9 mm stainless steel
wire, for a set of frequencies from 50 to 150 GHz. As
expected, the images show that no radiation is present
beyond the end of the wire along its axis. This is due to
the fact that the waveguide mode is radially polarized,
and therefore must vanish on the axis of symmetry. More
importantly, the simulation results show that at the end
of the wire, the spatial shape of the emitted field is a
conical emission pattern propagating into free space.
Inspection of this conical pattern reveals the presence of
frequency-dependent diffraction at the end of the metal
wire. The opening angle becomes smaller and the width
of the conical pattern narrows with increasing frequency.
Thus, the spatial size of the emitted beam at the end of
the wire is broader at lower frequencies. This effect is a
result of the terahertz surface waves scattering into
propagating waves, which diffract off the edges of the
cylinder surface. The diffraction is stronger for the low
Fig. 13. Schematic diagram used to measure the THz electric field frequencies than for the high frequencies. Additional
present near the end of a metal wire waveguide using fiber-coupled
time-domain simulations are being conducted to deter-
THz emitters and detectors. In these measurements, w is the
transverse separation between the center of the waveguide
mine if there is an end-face reflection of the THz surface
and the position where the field is measured; d is the longitudinal wave at the end of the wire due to an impedance
distance. mismatch with free space. In this case, FEM simulations

Vol. 95, No. 8, August 2007 | Proceedings of the IEEE 1635

Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

Fig. 14. Simulated THz intensity distribution in a plane containing the axis of a 0.9 mm thick wire, for five different frequencies.
Red indicates a high intensity, blue indicates a low intensity. The color scale in these images has been saturated to enhance the
visibility of the fields propagating away from the end of the wire. The higher intensities near the wire are therefore beyond
the limits of this saturated color scale, and are depicted in white.

provided a visualization of the frequency-dependent emanating from a wire waveguide provides a promising
diffraction at the end of a metal wire waveguide that direction for the future use of wire waveguides in tera-
aids in the interpretation of experimental results. hertz microscopy. In fact, recent simulations discussed by
Maier et al. have shown that periodic corrugations on
C. THz Emission From Tapered Wire Waveguides tapered metal wire waveguides can further confine the
A logical extension of this work is to modify the end terahertz field [65].
geometry of the wire waveguide. Using an FEM model, Tapered waveguide geometries have been utilized for
we investigate the effect of tapering a metal wire wave- various applications and radiation frequencies for many
guide to a tip size smaller than original diameter of the years [66]–[68]. The vast majority of this work has dealt
wire. The models consist of THz surface plasmon polar- with geometries in which the mode propagates via total
itons being excited at the beginning of a 1.5 cm long internal reflection or is confined by a metallic structure.
section of parallel wire with a 0.9 mm diameter that The use of a tapered structure for the guided propagation
slowly tapers down to much smaller tip diameters. Fig. 15 of a surface wave or surface plasmon polariton is a more
shows FEM simulation results of various tip diameters at recent development [65], [69], [70]. The treatment of
differing frequencies. Diffraction of the emitted terahertz Sommerfeld waves developed by Goubau indicates that
wave still occurs at the end of the tapered wire wave- the lateral extent of the electric field decreases with
guide. It is further evident that the spatial extent of the decreasing wire diameter, but a nontrivial analytical
field at the end of the wire can be manipulated by varying expression defining this relation does not exist [46]. The
the diameter of a tapered wire waveguide. The spatial rate of decrease in the field extent seen in our simu-
extent of the field narrows with decreasing tip diameters, lations of adiabatically tapered waveguides must be char-
as the propagating mode is adiabatically compressed. The acterized and compared to this analytic expression.
ability to manipulate the spatial extent of the field Clearly, FEM simulations play an important role in the

1636 Proceedings of the IEEE | Vol. 95, No. 8, August 2007

 Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

Fig. 15. FEM simulation results showing the power (log scale) wave propagating to the end of a tapered wire waveguide for differing
tip diameters and frequencies. (a) 20
m tip diameter at a frequency of 100 GHz. (b) 20
m tip diameter at a frequency of 200 GHz.
(c) 500
m tip diameter at a frequency of 150 GHz. (d) 50
m tip diameter at a frequency of 150 GHz.

ongoing study of tapered surface plasmon polariton that significantly improves the free-space to guided-mode
waveguides. coupling efficiency. In addition, FEM methods can
describe the nature of the near-field to far-field tran-
sition at the end of wire waveguides, which helps to
VI. CONCLUDING REMARKS explain the strong spatial dependence of the measured
The use of cylindrical metal wires as terahertz waveguides pulse spectra.
provides a low attenuation and virtually dispersionless In passing, we note that we have also made extensive
method of guiding and manipulating pulsed terahertz use of the FEM to study the interaction of THz radiation
radiation. In order to exploit terahertz metal wire wave- with photonic crystals. Photonic crystals exhibit periodic-
guides, it is essential to characterize both the coupling of ity in their dielectric function in one or more dimensions
free-space THz radiation to the surface plasmon polar- [71], [72]. This periodicity can be tuned in order to
itons and also the radiation of this energy back into free- manipulate the flow of optical radiation. Two-dimensional
space at the end of the wire waveguide. We have shown photonic crystals, consisting of periodic air holes etched
that the FEM, used in conjunction with ongoing ex- in high resistivity silicon, have been fabricated for study
perimental measurements, is a powerful tool for the with time-domain spectroscopy [73]–[75]. FEM simula-
modeling of terahertz guided wave phenomena. The tions of the interaction of THz radiation with photonic
FEM has been used to simulate the coupling of linearly crystals have been employed to calculate the transmis-
polarized terahertz radiation to a metal wire waveguide sion spectrum and showed excellent agreement with
and to develop a novel photoconductive antenna source experimental results [73]. The FEM has also recently

Vol. 95, No. 8, August 2007 | Proceedings of the IEEE 1637

Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

been compared with experimental studies of superprism interaction of THz pulses with novel systems like photonic
effects in terahertz photonic crystals [75]. crystals can provide an increased understanding of the
The FEM can be employed as an effective computa- interaction, as well as useful feedback for the design of
tional tool to aid in the research and development optimized devices. h
associated with terahertz time-domain spectroscopy and
imaging. FEM modeling of novel terahertz devices can
save considerable research time and resources by tuning Acknowledgment
the design parameters prior to fabrication. FEM simula- The authors would like to Vincent Cocula and
tions of terahertz phenomena such as the propagation of Magnus Olsson of COMSOL AB for their valuable
terahertz radiation in or along novel waveguides or the scientific and technical advice.

REFERENCES crystal,[ Appl. Phys. Lett., vol. 61, no. 26, [31] B. M. A. Rahman, F. A. Fernandez, and
pp. 3080–3082, 1992. J. B. Davies, BReview of finite element
[1] D. M. Mittleman et al., BNon-contact methods for microwave and optical
semiconductor wafer characterization [16] D. H. Auston and K. P. Cheung, BCoherent
time-domain far-infrared spectroscopy,[ waveguides,[ Proc. IEEE, vol. 79, no. 10,
with the terahertz hall effect,[ Appl. Phys. pp. 1442–1448, Oct. 1991.
Lett., vol. 71, pp. 16–18, 1997. J. Opt. Soc. Amer. B: Opt. Phys., vol. 2, no. 4,
pp. 606–612, 1985. [32] I. Tsukerman, F. Cajko, and A. P. Sokolov,
[2] D. Grischkowsky et al., BFar-infrared BTraditional and new simulation techniques
time-domain spectroscopy with terahertz [17] Q. Wu and X.-C. Zhang, BFree-space
electro-optic sampling of terahertz for nanoscale optics and photonics,[ Proc.
beams of dielectrics and semiconductors,[ SPIE, Plasmonics: Metallic Nanostructures
J. Opt. Soc. Amer. B, vol. 7, no. 10, beams,[ Appl. Phys. Lett., vol. 67, no. 24,
pp. 3523–3525, 1995. and Their Optical Properties III, vol. 5927,
pp. 2006–2015, 1990. pp. 128–137, 2005.
[3] D. Crawley et al., BThree-dimensional [18] K. Wang and D. M. Mittleman, BGuided
propagation of terahertz pulses on [33] R. L. Courant, BVariational methods for
terahertz pulse imaging of dental tissue,[ the solution of problems of equilibrium
J. Biol. Opt., vol. 8, pp. 303–307, 2003. metal wires,[ J. Opt. Soc. Amer. B, vol. 22,
pp. 2001–2007, 2005. and vibration,[ Bulletin of the American
[4] R. M. Woodward et al., BTerahertz pulsed Mathematical Society, vol. 5, pp. 1–23,
imaging of skin cancer in the time and [19] M. Goto et al., BTeflon photonic crystal
1943.
frequency domain,[ J. Biol. Phys., vol. 29, fiber as terahertz waveguide,[ Jpn. J. Appl.
Phys., vol. 43, pp. L317–L319, 2004. [34] P. P. Silvester, BFinite-element solution of
pp. 257–261, 2003. homogeneous waveguide problems,[ Alta
[5] R. H. Jacobson, D. M. Mittleman, and [20] H. Han et al., BTerahertz pulse propagation in
Frequenza, vol. 38, pp. 313–317, 1969.
M. C. Nuss, BChemical recognition a plastic photonic crystal fiber,[ Appl. Phys.
Lett., vol. 80, pp. 2634–2636, 2002. [35] W. B. J. Zimmerman, Multiphysics Modeling
of gases and gas mixtures using terahertz With Finite Element Methods, ser. Series on
radiation,[ Opt. Lett., vol. 21, pp. 2011–2013, [21] G. d. l. Reyes et al., BLow-loss single-mode
Stability, Vibration, and Control of Systems,
1996. terahertz waveguiding using Cytop,[ Appl.
Series A. London: World Scientific, 2006,
[6] M. R. Leahy-Hoppa et al., BWideband Phys. Lett., vol. 89, p. 211 119, 2006.
vol. 18.
terahertz spectroscopy of explosives,[ [22] S. A. Maier and H. A. Atwater, BPlasmonics:
[36] J. Jin, The Finite Element Method in
Chem. Phys. Lett., vol. 434, no. 4–6, Localization and guiding of electromagnetic
Electromagnetics. New York: John Wiley &
pp. 227–230, 2007. energy in metal/dielectric structures,[
Sonc, Inc., 2002.
[7] Y. C. Shen et al., BDetection and identification J. Appl. Phys., vol. 98, p. 011 101, 2005.
[37] C. Ashcraft, SPOOLES.
of explosives using terahertz pulsed [23] R. Mendis and D. Grischkowsky,
spectroscopic imaging,[ Appl. Phys. Lett., BUndistorted guided-wave propagation [38] T. Davis, UMFPACK, Gainesville, FL, 2005.
vol. 86, no. 24, p. 241 116, 1986. of subpicosecond terahertz pulses,[ [39] R. Barrett et al., Templates for the Solution of
[8] K. Kawase, Y. Ogawa, and Y. Watanabe, Opt. Lett., vol. 26, no. 11, pp. 846–848, Linear Systems: Bulding Blocks for Iterattive
BNon-destructive terahertz imaging of 2001. Methods, ser. Miscellaneous Titles in Applied
illicit drugs using spectral fingerprints,[ [24] R. Mendis and D. Grischkowsky, Mathematics Series No. 43. Philadelphia:
Opt. Express, vol. 11, no. 20, pp. 2549–2554, BTHz interconnect with low-loss SIAM, 1994.
2003. and low-group velocity dispersion,[ [40] D. Jiao and J.-M. Jin, BA general approach
[9] F. Huang et al., BTerahertz study of IEEE Microw. Wireless Compon. Lett., for the stability analysis of the time-domain
1, 3, 5-trinitro-s-triazine by time-domain vol. 11, no. 11, pp. 444–446, Nov. 2001. finite-element method for electromagnetic
and Fourier transform infrared [25] M. Nagel, A. Marchewka, and H. Kurz, simulations,[ IEEE Trans. Antennas
spectroscopy,[ Appl. Phys. Lett., vol. 85, BLow-index discontinuity terahertz Propag., vol. 50, no. 11, pp. 1624–1632,
no. 23, pp. 5535–5537, 2004. waveguides,[ Opt. Express, vol. 14, no. 21, Nov. 2002.
[10] D. Zimdars et al., BLarge area terahertz pp. 9944–9954, 2006. [41] J.-F. Lee, R. Lee, and A. Cangellaris,
imaging and non-destructive evaluation [26] M. Wachter, M. Nagel, and H. Kurz, BTime-domain finite-element methods,[
applications,[ InsightVNon-Destr. Test. BMetallic slit waveguide for dispersion-free IEEE Trans. Antennas Propag., vol. 45, no. 3,
Cond. Monit., vol. 48, no. 9, pp. 537–539, low-loss terahertz signal transmission,[ pp. 430–441, Mar. 1997.
2006. Appl. Phys. Lett., vol. 90, p. 061 111, 2007. [42] B. Butrylo et al., BA survey of parallel
[11] S. Wang and X.-C. Zhang, BPulsed terahertz [27] K. Wang and D. M. Mittleman, BMetal wires solvers for the finite element method
tomography,[ J. Phys. D, vol. 37, pp. R1–R36, for terahertz waveguiding,[ Nature, vol. 432, in computational electromagnetics,[
2004. p. 376, 2004. COMPEL: The International J. for
Computation and Mathematics in Electrical
[12] D. M. Mittleman, Ed., Sensing With [28] M. Wachter, M. Nagel, and H. Kurz,
and Electronic Engineering, vol. 23, no. 2,
Terahertz Radiation. Heidelberg, BFrequency-dependent characterization
pp. 531–546, 2004.
Germany: Springer-Verlag, 2002. of THz Sommerfeld wave propagation on
single-wires,[ Opt. Express, vol. 13, no. 26, [43] O. Schenk and K. Gartner, BOn fast
[13] P. R. Smith, D. H. Auston, and M. C. Nuss, factorization pivoting methods for
BSubpicosecond photoconducting dipole p. 10 815, 2005.
symmetric indefinite systems,[ Elec. Trans.
antennas,[ IEEE J. Quantum Electron., [29] T.-I. Jeon and D. Grischkowsky, BDirect
Numer. Anal., vol. 23, pp. 158–179, 2006.
vol. 24, no. 2, pp. 255–260, Feb. 1988. optoelectronic generation and detection
of sub-ps-electrical pulses on sub-mm-coaxial [44] O. Schenk and K. Gartner, BSolving
[14] M. v. Exter and D. Grischkowsky, unsymmetric sparse systems of linear
BCharacterization of an optoelectronic transmission lines,[ Appl. Phys. Lett., vol. 85,
no. 25, pp. 6092–6094, 2004. equations with PARDISO,[ J. Future
terahertz beam systems,[ IEEE Trans. Generation Computer Systems, vol. 20, no. 3,
Microw. Theory Tech., vol. 38, no. 11, [30] R. Coccioli et al., BFinite-element methods
pp. 475–487, 2004.
pp. 1684–1690, Nov. 1990. in microwaves: A selected bibliography,[
IEEE Antennas Propag. Mag., vol. 38, no. 6, [45] COMSOL Multiphysics. Stockholm,
[15] X.-C. Zhang et al., BTerahertz optical Sweden: COMSOL AB, 2007.
rectification from a nonlinear organic pp. 34–38, Dec. 1996.

1638 Proceedings of the IEEE | Vol. 95, No. 8, August 2007

 Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

[46] G. Goubau, BSurface waves and their lens-coupled terahertz antennas,[ [66] K. Matsumaru, BReflection of a pyramidally
application to transmission lines,[ J. Appl. J. Opt. Soc. Amer. B, vol. 18, tapered rectangular waveguide,[ IRE Trans.
Phys., vol. 21, pp. 1119–1128, 1950. pp. 1524–1533, 2001. Microw. Theory Tech., vol. MTT-7, no. 2,
[47] T.-I. Jeon, J. Zhang, and D. Grischkowsky, [56] H. Cao and A. Nahata, BCoupling of terahertz pp. 192–196, Apr. 1959.
BTHz Sommerfeld wave propagation on pulses onto a single metal wire waveguide [67] F. Sporleder and H.-G. Unger, Waveguide
a single metal wire,[ Appl. Phys. Lett., using milled grooves,[ Opt. Express, vol. 13, Tapers, Transitions, & Couplers, ser. IEE
vol. 86, p. 161 904, 2005. no. 18, pp. 7028–7034, 2005. Electromagnetic Waves Series. London,
[48] Q. Cao and J. Jahns, BAzimuthally polarized [57] J. Deibel, M. Escarra, and D. M. Mittleman, U.K.: IEE, 1979.
surface plasmons as effective terahertz BPhotoconductive terahertz antenna with [68] N. N. Qaddoumi, M. Abou-Khousa, and
waveguides,[ Opt. Express, vol. 13, no. 2, radial symmetry,[ Electron. Lett., vol. 41, W. M. Saleh, BNear-field microwave
pp. 511–518, 2005. pp. 9–10, 2005. imaging utilizing tapered rectangular
[49] K. L. Shlager, G. S. Smith, and J. G. Maloney, [58] J. D. Jackson, Classical Electrodynamics, waveguides,[ IEEE Trans. Instrum.
BOptimization of bow-tie antennas for 3rd ed. New York: Wiley, 1999. Meas, vol. 55, no. 5, pp. 1752–1756,
pulse radiation,[ IEEE Trans. Antennas Oct. 2006.
[59] P. U. Jepsen and S. R. Keiding, BRadiation
Propag., vol. 42, no. 7, pp. 975–982, patterns from lens coupled terahertz [69] D. F. P. Pile and D. K. Gramotnev, BAdiabatic
Jul. 1994. antennas,[ Opt. Lett., vol. 20, no. 8, and nonadiabatic nanofocusing of plasmons
[50] D. B. Rutledge, D. P. Neikirk, and pp. 807–809, 1995. by tapered waveguides,[ Appl. Phys. Lett.,
D. P. Kasilingam, BIntegrated-circuit vol. 89, p. 041 111, 2006.
[60] C. Fattinger and D. Grischkowsky,
antennas,[ in Infrared and Millimeter Waves, BTerahertz beams,[ Appl. Phys. Lett., [70] M. I. Stockman, BNanofocusing of optical
K. J. Button, Ed. New York: Academic, vol. 54, pp. 490–492, 1989. energy in tapered plasmonic waveguides,[
1983, pp. 1–90. Phys. Rev. Lett., vol. 93, no. 13, p. 137 404,
[61] G. Chang et al., BGeneration of
[51] J. V. Rudd and D. M. Mittleman, 2004.
radially polarized terahertz pulses via
BThe influence of substrate lens design velocity-mismatched optical rectification,[ [71] S. John, BStrong localization of photons
in terahertz time-domain spectroscopy,[ Opt. Lett., vol. 32, no. 4, pp. 433–435, in certain disordered dielectric superlattices,[
J. Opt. Soc. Amer. B, vol. 19, pp. 319–329, 2007. Phys. Rev. Lett., vol. 58, pp. 2486–2489,
2002. 1987.
[62] M. Walther et al., BEmission and detection
[52] Y. Zhu and A. Cangellaris, Multigrid of terahertz pulses from a metal-tip [72] E. Yablonovitch, BInhibited spontaneous
Finite Element Methods for Electromagnetic antenna,[ J. Opt. Soc. Amer. B, vol. 22, emission in solid-state physics and
Field Modeling, ser. The IEEE Press Series pp. 2357–2365, 2005. electronics,[ Phys. Rev. Lett., vol. 58,
on Electromagnetic Wave Theory, pp. 2059–2062, 1987.
[63] M. Walther, M. R. Freeman, and
D. G. Dudley, Ed. Hoboken, NJ: [73] Z. Jian and D. M. Mittleman,
F. A. Hegmann, BMetal-wire terahertz
Wiley, Inc., 2006. BCharacterization of guided resonances
time-domain spectroscopy,[ Appl.
[53] J. A. Deibel et al., BEnhanced coupling Phys. Lett., vol. 87, p. 261 107, 2005. in photonic crystal slabs using terahertz
of terahertz radiation to cylindrical time-domain spectroscopy,[ J. Appl. Phys.,
[64] J. A. Deibel et al., BFrequency-dependent
wire waveguides,[ Opt. Express, vol. 14, vol. 100, p. 123 113, 2006.
radiation patterns emitted by THz
pp. 279–290, 2006. [74] Z. Jian and D. M. Mittleman, BOut-of-plane
plasmons on finite length cylindrical
[54] F. Yang, J. R. Sambles, and G. W. Bradberry, wires,[ Opt. Express, vol. 14, pp. 8772–8778, dispersion and homogenization in photonic
BLong-range surface modes supported by 2006. crystal slabs,[ Appl. Phys. Lett., vol. 87,
thin films,[ Phys. Rev. B, vol. 44, p. 191 113, 2005.
[65] S. A. Maier et al., BTerahertz surface
pp. 5855–5572, 1991. [75] T. Prasad et al., BThe superprism effect
plasmon-polariton propagation and
[55] J. V. Rudd, J. L. Johnson, and focusing on periodically corrugated in a metal-clad terahertz photonic crystal
D. M. Mittleman, BCross-polarized metal wires,[ Phys. Rev. Lett., vol. 97, slab,[ Opt. Lett., vol. 32, pp. 683–685,
angular emission patterns from p. 176 805, 2006. 2007.

ABOUT THE AUTHORS

Jason A. Deibel (Member, IEEE) was born in Matthew Escarra (Student Member, IEEE) was
Louisville, KY, in 1974. He received the B.A. degree born in New Orleans, LA, in 1984. He received the
in physics and mathematics from Transylvania B.S. degree in electrical engineering from Rice
University, Lexington, KY, in 1997 and the Ph.D. University, Houston, TX, in 2006. He is currently
degree in applied physics from the University of working toward the Ph.D. degree in the Depart-
Michigan, Ann Arbor in 2004. ment of Electrical Engineering, Princeton Univer-
From 1997 to 2004, he was a Graduate Student sity, Princeton, NJ.
Research Assistant in the Ultrafast Technology As an undergraduate at Rice, he was involved in
Group at the University of Michigan Center for the development of a new generation of photo-
Ultrafast Optical Science. Since 2004, he has been conductive terahertz antennas for use with metal
a Postdoctoral Research Associate in the Department of Electrical and wire waveguides. As a graduate student at Princeton, he is a member of
Computer Engineering at Rice University, Houston, TX, where he is a Professor Claire Gmachl’s Mid-Infrared Photonics research group, a part
member of Professor Daniel Mittleman’s research group. He was an of the Mid-Infrared Technologies for Health and the Environment
Intelligence Community Postdoctoral Fellow from 2004–2007. He spent (MIRTHE) NSF Engineering Research Center.
two months in 2006 at the University of Leeds, U.K., as a Royal Society
Visiting Fellow. He will join the Department of Physics as an Assistant Nicholas Berndsen (Student Member, IEEE) was
Professor at Wright State University, Dayton, OH, in Fall 2007. His born in St. Louis, MO, in 1985. He is currently
research interests include terahertz transmission and emission spec- working toward the B.S. degree in electrical
troscopy of novel inorganic and organic semiconductors, terahertz engineering at Rice University, Houston, TX, and
microscopy and plasmonics, terahertz waveguide development, and will graduate in 2008.
finite-element method simulations of terahertz radiation phenomena He is a member of Professor Daniel Mittleman’s
and devices. research group and is also a Co-President of the
Dr. Deibel is a member of the Optical Society of America, the American 2007–2008 Rice chapter of the IEEE.
Physical Society, and the IEEE Lasers and Electro-Optics Society.

Vol. 95, No. 8, August 2007 | Proceedings of the IEEE 1639

Deibel et al.: Finite-Element Method Simulations of Guided Wave Phenomena at Terahertz Frequencies

Kanglin Wang received the B.S. degree in Daniel M. Mittleman (Member, IEEE) was born in
applied physics from Tongji University, China, in Berkeley, CA, in 1966. He received the B.S. degree
1998, the M.S. degree in physics from Fudan in physics from the Massachusetts Institute of
University, China in 2001, and the Ph.D. degree in Technology, Cambridge, in 1988 and the M.S.
applied physics from Rice University, Houston, and Ph.D. degrees in physics from the University
TX, in 2006. of California, Berkeley, in 1990 and 1994,
From 2001 to 2006, he was a graduate respectively.
student in the Department of Electrical and Com- He spent two years as a Postdoctoral Member
puter Engineering at Rice University, and studied of the Technical Staff at AT&T Bell Laboratories
novel devices and systems for terahertz spec- (now Lucent Technologies), where he was in-
troscopy and imaging in professor Daniel Mittleman’s research group. volved in the development of terahertz BT-ray[ imaging. In 1996, he
He then joined Shell Oil Company, Houston, TX, and took a research joined the faculty of the Electrical and Computer Engineering Depart-
position in the Potential Field and Remote Sensing team. ment at Rice University, Houston, TX, where he is now an Associate
Professor. His research involves ultrafast optoelectronic generation of
terahertz radiation and its applications for spectroscopy and imaging. He
is also active in the study of new terahertz techniques, devices, and
systems.
Prof. Mittleman is a member of the Optical Society of America,
the American Physical Society, and the IEEE Lasers and Electro-
Optics Society. He is the Founder and Chair of the Houston chapter
of IEEE/LEOS.

1640 Proceedings of the IEEE | Vol. 95, No. 8, August 2007