IEEE JOURNAL ON MULTISCALE AND MULTIPHYSICS COMPUTATIONAL TECHNIQUES, VOL.

3, 2018 29

Fast Modeling of Terahertz Plasma-Wave Devices

Using Unconditionally Stable FDTD Methods
Shubhendu Bhardwaj , Member, IEEE, Fernando L. Teixeira , Fellow, IEEE, and John L. Volakis , Fellow, IEEE

Abstract—Modeling of terahertz plasmonic device is a multi- than that in free-space. This would require a mesh-size in the
scale and multiphysics problem that requires fine mesh in the order of ≈ λ/5000. As per Courant-Friedrich-Levy (CFL) con-
electron transport regions, inevitably leading to long simulation dition, FDTD time-step for this simulation would be ≈ 10−17 s.
times. In this paper, we tackle this problem by employing un-
conditionally stable FDTD methods for these simulations. Specif- Such small step sizes could result in long simulation times.
ically, we present alternating direction implicit (ADI) FDTD and In this paper, the time-efficiency of the traditional FDTD-
iterative-ADI-FDTD based efficient multiphysics model that inte- based model [9] is improved using unconditionally stable FDTD
grate hydrodynamic equations and Maxwell’s equations for mod- algorithms. Specifically, we consider two unconditionally sta-
eling terahertz plasmonic devices. Using the proposed methods, ble models in this paper. First, we present alternating direc-
we demonstrate 50% time-reduction with a nominal 3% errors in
the solutions. Faster times may also be achieved at higher cost to tion implicit (ADI) FDTD method [14] coupled with hydrody-
solution-accuracy, suggesting an ability to adjust time-costs and namic equations for terahertz plasmonic applications. Second,
accuracy, useful for different terahertz modeling scenarios. an iterative-ADI based FDTD method [15], [16] coupled with
hydrodynamic equations, is presented for efficient yet accurate
I. INTRODUCTION modeling. In the paper, these methods are referred to as ADI-
HERE has been strong interest in terahertz plasma-wave FDTD-HD and it-ADI-FDTD-HD, respectively.
T devices, such as mixers, detectors and oscillators. In gen-
eral, these devices rely on plasma-wave oscillations in the two-
We note that the ADI-FDTD based hybrid algorithms have
been used for a transistor-modeling at microwave frequencies
dimensional (2-D) electron gas (2DEG) channel and have shown in [17]. But their application for terahertz plasmonic applica-
interesting physical properties [1]–[3]. We note that the origin tions is not considered yet. We find that ADI-FDTD-HD meth-
of such 2DEG channels could be varied, i.e., they could be ods only allow modest time-reductions due to errors associated
based on semiconductor-heterojunctions such as GaN/AlGaN with larger time-steps (as discussed in Section III-A). As is
or GaAs/AlGaAs [1] or could be based on 2-D material, such well known, these errors are due to truncation of second order
as graphene layers and bilayers [2]–[5]. In any case, modeling (Δt2 ) terms as per ADI algorithms. On the other hand, time-
of this class of devices requires multiphysics tools that can in- reductions can be significant if iterative corrections are applied
tegrate the electromagnetic propagation effects as well as the at large time-steps. This leads to it-ADI-FDTD-HD method
electron-transport effects in the devices [6]–[10]. discussed in Section III-B. We demonstrate that with it-ADI-
In the past, such multiphysics solutions have been developed FDTD-HD method, a simulation-time reduction of up to 50%,
using FDTD algorithm and have often been applied for mi- with nominal error of 3% is achieved for the considered exam-
crowave applications[11]–[13]. But for terahertz applications, ple device. Furthermore, it is shown that time-step value and
the problem becomes more challenging due to dominant plasma- iteration-count can become parameters to adjust various levels
wave oscillations in the 2DEG region. This ultimately requires of time-efficiency and accuracy.
much finer mesh in the 2DEG regions making this a multi- In the following sections, we will discuss the theory and im-
scale problem (also explained in [9]). For example, in modern plementation of ADI-FDTD-HD and it-ADI-FDTD-HD mod-
plasmonic devices, the 2DEG’s electron density is found to be els, respectively. The developed algorithms will be applied
≈ 1011 − 1013 cm−2 . At terahertz frequencies the correspond- to a typical HEMT geometry for performance evaluation and
ing plasmonic wavelengths would be 10 to 1000 times smaller comparison with traditional full-wave-hydrodynamic solver [9].
This will be followed by discussions and concluding remarks.
Manuscript received November 12, 2017; revised March 26, 2018; accepted
April 1, 2018. Date of publication April 16, 2018; date of current version
May 7, 2018. This work was supported by the Office of Naval Research Devices II. FULL-WAVE HYDRODYNAMIC HYBRID SOLVER USING
and Architectures for Terahertz Electronics Multi-University Research Initiative ADI METHODS
project, under Dr. Paul Maki under Grant N00014-11-1-0721. (Corresponding
author: Shubhendu Bhardwaj.) The presented frameworks use hydrodynamic equations for
S. Bhardwaj and J. L. Volakis are with the Department of Electrical and
Computer Engineering, Florida International University, Miami, FL 33174 USA the channel modeling and Maxwell’s equation for the modeling
(e-mail:,[email protected]; [email protected]). of the dielectric/metallic media. The solution steps are shown
F. L. Teixeira is with the Department of Electrical and Computer Engineer- in flowcharts in Fig. 1. First, the ADI based FDTD method is
ing, The Ohio State University, Columbus, OH 43210 USA (e-mail:, teixeira.5
@osu.edu). used for solving electrodynamic equations and electric/magnetic
Digital Object Identifier 10.1109/JMMCT.2018.2825427 field terms are updated in the domain. Then, by using updated

2379-8793 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

Authorized licensed use limited to: SASTRA. Downloaded on July 12,2024 at 04:04:41 UTC from IEEE Xplore. Restrictions apply.
30 IEEE JOURNAL ON MULTISCALE AND MULTIPHYSICS COMPUTATIONAL TECHNIQUES, VOL. 3, 2018

Fig. 1. Flowcharts for (a) ADI-FDTD-HD and (b) it-ADI-FDTD-HD methods. The corresponding electrodynamic and hydrodynamic equations are also listed.

values of electric field (Ex ), hydrodynamic equations are solved

for the same time iteration. Thus, new current values are ob-
tained. Newly obtained current values are used as source terms
in electrodynamic solution for the next time iteration. Thus, by
in-tandem updating of these variables, selfconsistent solution
is achieved (this is similar to prior works [6], [11]–[13]). We
note that typically field-solution is required in the device cross
section. Thus, associated a 2-D FDTD model and 1-D hydro-
dynamic model will be pursued in this paper. The specific solu-
tion steps and relevant equations are discussed in the following
Fig. 2. GaN/AlGaN High Electron Mobility Transistor (HEMT) model used
sections. for the performance evaluation of the algorithms. Left: Schematic showing
device-dimensions and excitation-method. Right: Simulation domain and posi-
tion of PML boundaries.
A. ADI-FDTD-HD Method
The algorithm used for the ADI-FDTD-HD method is shown
in Fig. 1(a). We follow the formulation from [15] to write
with fields and current terms given by matrices u and v (note
Maxwell’s electrodynamic equations for a TEz case while also
that here the vector sign represents a matrix, not a space-vector)
introducing current term as well as anisotropic conductivity
as
term
⎡ ⎤ ⎡ Je x ⎤
∂u Ex − 
= [A]u + [B]u + v n . (1) ⎢ ⎥ ⎢ Je y ⎥
∂t ⎢ Ey ⎥ ⎢−  ⎥
Here, operators [A] and [B] are u = ⎢ ⎥
⎢ H ⎥and v = ⎢
⎢ ⎥
⎥ (4)
⎡ σy ⎤ ⎣ zx ⎦ ⎣ 0 ⎦
1 ∂ 1 ∂
− 2 0  ∂y  ∂y Hz y 0
⎢ ⎥
⎢ 0
⎢ − σ2x 0 0 ⎥ ⎥
[A] = ⎢ ∗ ⎥and (2) Here, Jex and Jey are electric current density components in x
⎢ 0 0 − σ2μx 0 ⎥ and y directions, respectively. Symbols σx/y denote the elec-
⎣ ⎦
σ∗
1 ∂
0 0 − 2μy trical conductivity for wave traveling in x and y directions.
μ ∂y ∗
Likewise σx/y is the magnetic conductivity in the respective
⎡ σ ⎤
− 2y 0 0 0 directions. The anisotropic conductivities allow for the mod-
⎢ ⎥ eling of perfectly matched layers (PMLs) as shown in Fig. 2
⎢ 0
⎢ − σ2x − 1 ∂∂x − 1 ∂∂x ⎥
⎥ (refer [18], [19]). Furthermore, split field formulation is used
[B] = ⎢ ∗ ⎥ (3)
⎢ 0 − μ1 ∂
− σ2μx 0 ⎥ for Hz (= Hz x + Hz y ) to facilitate these PML layers [18]. Out-
⎣ ∂x ⎦
σ∗ side PML regions, the media chosen to be isotropic with σx =
0 0 0 − 2μy
σy and σx∗ = σy∗ .

Authorized licensed use limited to: SASTRA. Downloaded on July 12,2024 at 04:04:41 UTC from IEEE Xplore. Restrictions apply.
BHARDWAJ et al.: FAST MODELING OF TERAHERTZ PLASMA-WAVE DEVICES USING UNCONDITIONALLY STABLE FDTD METHODS 31

For time discretization, we apply Crank Nicholson scheme at B. Iterative-ADI-FDTD-HD Method

time-step n + 1/2 to obtain
As noted in previous sections, to reach the tridiagonal system
 
Δt Δt Δt Δt of equations, we ignored the second order term
I− [A] − [B] un +1 = I + [A] + [B] un
2 2 2 2 Δt2
[A][B] un +1 − un (10)
+ Δtv n . (5) 4

Here, Δt is the time-step used for difference equations. Above in (6). In iterative-ADI-FDTD method, we reconstruct this term
mentioned equation can be rewritten in factorized form as by using an iterative process, applied within each time step
  [15]. The model algorithm is shown in Fig. 1(b). Specifically,
Δt Δt the splitting (8) and (9) are rewritten to read
I− [A] I− [B] un +1
2 2  
  Δt tm p Δt
I− [A] uk +1 = I + [B] un
Δt Δt 2 2
= I+ [A] I+ [B] un
2 2
Δt2 Δt n
+ [A][B] unk +1 − un + v
Δt2 8 2
+ [A][B] un +1 − un + Δtv n . (6) (11)
4
In the above-mentioned, in addition to the factorized terms, a and
second-order Δt term is also present. Under ADI method, this  
Δt Δt
term is ignored providing I− [B] unk +1
+1
= I+ [A] utm p
  2 2
Δt Δt
I− [A] I− [B] un +1 Δt2 Δt n
2 2 + [A][B] unk +1 − un + v .
  8 2
Δt Δt (12)
= I+ [A] I+ [B] un + Δtv n (7)
2 2 In the above-mentioned, k denotes the iteration count for field
which can be solved in exactly two steps, i.e., correction within each time-step. Parameter k could vary as
  0 ≤ k ≤ it, where it is the total number of iterations specified by
Δt tm p Δt Δt n the user. Thus, the term utm p
k +1 for k + 1st iteration is calculated it
I− [A] u = I+ [B] un + v and (8)
2 2 2 times, using the field values from the previous iteration (unk +1 −
  un ). For the first iteration, the initial guess is chosen to be
Δt n +1 Δt Δt n (un0 +1 = un ). Note that for it = 0, it-ADI-FDTD-HD method
I− [B] u = I+ [A] utm p + v . (9) simply becomes ADI-FDTD-HD method, since no corrective
2 2 2
iterations are applied.
Here utm p is an intermediate solution, denoting field values at a Of course, the simulation time increases linearly with the to-
sub-time-step. Equations (8) and (9) are tridiagonal systems of tal number of iterations it, but the error reduces and ADI-FDTD
linear equations that can be solved using Gauss-Seidel method. solution converges to explicit-FDTD solution. Thus, the method
Thus, by introducing Δt2 approximation, matrix inversion is provides a way to fine-tune the accuracy by adjusting it. On the
avoided. This ADI-FDTD method is then integrated with so- other hand, CN is used for increasing the time-step leading to
lution of hydrodynamic equations (discussed in Section III-C) faster simulations. Overall, the parameters it and CN can be
together constitutes the ADI-FDTD-HD model. used to reach the fidelity versus time-cost compromise for the
Note that, the accuracy of the model depends on the field modeling of these devices. We will consider these issues while
derivatives and time-step sizes as per the ignoredterm in (6). That evaluating the performance of the two models in Section III.
is, in the regions of strong spatial field variations and/or for large Before doing so, we will consider the equations used for hydro-
time-steps, the solution could become inaccurate. Unfortunately, dynamic solver in the following section.
strong field variations occur in 2DEG regions of the device,
which are essentially the regions of our interest. Therefore, the
C. Hydrodynamic Modeling of the 2DEG Channel
usefulness of this method for plasmonic applications needs to be
justified. Following that, an examination of the solution fidelity For channel modeling, Boltzmann transport equations (BTE)
with increasing time-step size will be considered in Section IV. provide a complete description of the electron-transport phe-
As expected, we will conclude that for large time-steps, the ac- nomenon in a 2DEG channel. Since BTE solution is computer
curacy of ADI-FDTD-HD solution is compromised. Thus, the memory and time expensive, we consider its first two moments
fidelity issue needs to be addressed. To that end, we introduce only, i.e., particle conservation (or continuity equation) and mo-
it-ADI-FDTD-HD method, as discussed in the following mentum conservation equations [13]. We note that higher order
sections. moments of BTE can be integrated into the solutions as well. For

Authorized licensed use limited to: SASTRA. Downloaded on July 12,2024 at 04:04:41 UTC from IEEE Xplore. Restrictions apply.
32 IEEE JOURNAL ON MULTISCALE AND MULTIPHYSICS COMPUTATIONAL TECHNIQUES, VOL. 3, 2018

2-D confined electron gas, we write the hydrodynamic equations TABLE I

CPU TIMES FOR VARIOUS CN VALUE CASES FOR ADI-FDTD-HD METHOD
∂nsh ∂j
+ = 0, and (13)
∂t ∂x
∂j ∂j ∂v qnEx j KT ∂nsh
+v +j =− − − . (14)
∂t ∂x ∂x me τ me ∂x
Here nsh is the sheet carrier density, j(= nsh v) is the sheet cur-
rent and v is the electron velocity within the 2DEG channel. Ex
is the x-directed electric field along the channel (see Fig. 2 left),
τ refers to the momentum relaxation time and me is the effective
electron mass. q = 1.6 × 10−19 C is the charge-magnitude of
and if v|ni < 0, we define
a single electron. T is the electron temperature in the channel,
and K is the Boltzmann constant.    
Equation (13) is the particle conservation equation, signifying Δt n Δt n n
Ψ|ni = n n
j| v|i+1 − v|i + n
v| j|i+1 − j|i (19)
the equality between rate change of electrons and the gradient of Δx i Δx i
the sheet current at any given point in space. Equation (14) rep-
In above-mentioned equations, i refers to the grid-point index
resents conservation of momentum under an externally applied
along x̂-direction and n represents the time-iteration index. To
force (given by the term qnsh Ex /me ). The term j/τ represents
implement the conversation equations, vector quantities, e.g., v,
loss of momentum due to scattering within the channel. In this
j, are defined at the grid points (i, i + 1) and scalars are defined
paper, we use τ as phenomenological parameter, obtained either
at the cell midpoints (i − 1/2, i + 1/2 etc).
from experimental results or from separate particle simulations.
2) Time-Step for Hydrodynamic Solver: Upon integration
The current term Jex in (4) is related to j via
with ADI methods, the time-step of the upwind solver also
−q(j − jo ) increases, as the two solvers must run in tandem. With regards
Jex = (15) to that, solution of hydrodynamic equations was found to be
t2DEG
robust for increased time-steps. We noticed instability in the up-
where t2DEG is the thickness of the 2DEG layer. Term j − jo
wind equations only for very large time-steps (CN > 500). For
represents the ac current responsible for radiation from the chan- such cases, a smaller time-step, defined by ΔtHD = Δt/R, with
nel, with jo being the initial dc current in the channel that does R > 1, may be specified to bring the upwind solver in stable
not contribute to electromagnetic radiation. regime. Thus, the upwind equation solver would run R-times
1) Time-Space Discretization of HD Equations: For the nu- for each Δt time-step of the FDTD solver. This is also shown
merical solution, the partial derivatives in (13), (14) are con- in flowcharts of Fig. 1. Nevertheless, this time scaling was not
verted into their corresponding time and space difference equa- necessary for this paper since our simulations did not use CN >
tions. It is known that HD equations are nonlinear and are highly 500 cases.
coupled advection equations [20]. Their numerical solutions
could be unstable when center difference approximation is used. III. MODEL IMPLEMENTATION AND
This is especially true if we consider cases with strong dc bias PERFORMANCE EVALUATION
across the channel. Therefore, upwind difference scheme [20],
[21] was applied to maintain the stability of the algorithm. The For performance benchmarking, we modeled a gallium nitride
obtained difference equations for (13), (14) are based high electron mobility transistor (HEMT) device under
the influence of an incident terahertz plane-wave with excitation
Δt n
nsh |ni+1/2
+1
= nsh |ni+1/2 − [j| − j|ni−1 ] (16) frequency of 5 THz. A small gate discontinuity (similar to [8],
Δx i+1 [9]) is used for coupling of the incident terahertz waves to the
and 2DEG channel. The simulation set-up and the device schemat-
qnsh |ni Ech |ni j|n ics are shown in Fig. 2. As per GaN material system, channel
j|ni +1 = j|ni − Δt − Δt i
me τn parameters were chosen to be nsh = 5×1012 cm−2 , r = 9.5,
me = 0.2mo , and τ = 1.14 ps (considering low-temperature
nsh |ni+1/2 − nsh |ni−1/2 operation). The thickness of the 2DEG channel (same as the
KT
− Δt − Ψ|ni (17) smallest cell size in vertical direction) was chosen to be Δy = 1
me Δx nm. The cell size along x-axis was chosen to be Δx = 4 nm in
where, for upwind scheme we define the convective term Ψ the channel area. For outside the channel regions, nonuniform
based on the direction of flow of electrons at that point. At any meshing scheme was used, resulting in 1460 and 740 cells,
given cell index i and at the time iteration n, if v|ni > 0, then we respectively, in the horizontal and vertical directions. As per
define smallest Δy value, the maximum allowable time-step ΔtFDTD
    of 3.2×10−18 s was obtained in the traditional FDTD-HD solver.
n Δt n n n Δt n n n The simulations were allowed to run for 3 ps, in order for tran-
Ψ|i = j| v|i − v|i−1 + v| j|i − j|i−1 (18)
Δx i Δx i sients to die out and the simulations to reach a steady state.

Authorized licensed use limited to: SASTRA. Downloaded on July 12,2024 at 04:04:41 UTC from IEEE Xplore. Restrictions apply.
BHARDWAJ et al.: FAST MODELING OF TERAHERTZ PLASMA-WAVE DEVICES USING UNCONDITIONALLY STABLE FDTD METHODS 33


Fig. 3. (a) E-field profiles (calculated as E = E x2 + E y2 obtained at t = 3 ps for the various cases of ADI-FDTD-HD method compared with the reference
solution. Here, the reference case corresponds to explicit FDTD-HD method. (b) Current profile in the channel at t = 3 ps using the reference FDTD-HD method
and ADI-FDTD-HD methods.

MATLAB programming language was used for all simulations

in this paper.

A. Performance of the ADI-FDTD-HD Method

We conducted the simulations using the ADI-FDTD-HD for
the cases CN = 100, 200, and 300, as tabulated in Table I.
The resulting CPU times for a 3-ps-long simulation are also
listed in the table. As expected, the CPU times are reduced with
increasing CN. However, solution accuracy also degrades. This
is noted in Fig. 3(a) by showing the 2-D E-field plots of the Fig. 4. Current profile in the channel at t = 3 ps using the reference FDTD-HD
device. As shown, the CN = 100 case shows good agreement method and ADI-FDTD-HD methods.
with the reference data. But for CN = 200 and 300 cases, field
profiles start to differ from the reference case, especially so in accurate by visual inspection, there are errors in the calculated
the region under the gate. This is consistent with the theory, as current values. Spurious oscillations are also present near the
the ignored second order terms are dominant in the regions of gate region at x = 1.2 μ m for CN = 200 case.
large spatial derivatives. For a closer look, we also examine the We conclude that care should be taken while using ADI-
1-D current data in the channel for varying CN values (refer FDTD-HD method for the plasmonic applications. Although
Fig. 4). As shown, the calculated current spuriously exhibits the model shows appreciable time-reductions in the simula-
decreased wavelength as CN is increased to 200 and 300. Even tions (Table I), the resulting fields are only approximately cor-
for the case of CN = 100, where the 2-D-fields seemed to be rect. In any case, smaller CN should be used to maintain the

Authorized licensed use limited to: SASTRA. Downloaded on July 12,2024 at 04:04:41 UTC from IEEE Xplore. Restrictions apply.
34 IEEE JOURNAL ON MULTISCALE AND MULTIPHYSICS COMPUTATIONAL TECHNIQUES, VOL. 3, 2018

TABLE II
CPU TIMES FOR VARIOUS CN-it COMBINATIONS FOR
IT-ADI-FDTD-HD METHOD

oscillations under the gate region are corrected (compare CN

= 200 cases in Fig. 3(a) and (b)). For detailed comparison, the
obtained channel currents are also plotted in Fig. 5. These again
show excellent agreement between the reference data for the
cases CN, it = 100, 2, 200, 4, and 300, 7. Clearly, the
use of iterative corrections has allowed high fidelity. Further-
more, the CPU times for these cases are tabulated in Table II.
Remarkably, the time-costs are farsmaller (almost by a factor of
half) than in explicit FDTD-HD solver.
As noted, it-ADI-FDTD-HD provides several choices for run-
ning these simulations, as CN and it values can be varied. There-
fore, it is desirable to further understand a time-cost versus error
compromise in each of these simulations. We will examine this
in Section III-C.

C. Fidelity Versus Time-Cost Compromise

For this examination, we first define an error term for char-
acterizing the solution fidelity. The error % is defined as the
normalized percent deviation of the current from reference cur-
Fig. 5. Channel current profile using iterative-ADI-FDTD-HD method, com- rent (from traditional FDTD-HD solver). Thus,
pared to the reference data. Three plots refer to values of CN = 100, 200 and
L c
300, respectively. [Jdata (x) − Jref (x)]
Error% = x=0L c × 100 (20)
x=0 Jref (x)
integrity. Although, higher CN models may be useful for fast where, Jdata (x) is the current calculated along the channel for
but qualitative results. Following these arguments we will next
examine the iterative ADI-FDTD-HD algorithm for improving the specific case under examination. Likewise, Jref (x) is the
the accuracy for higher CN values. current for the reference case. Lc is the length of a channel.
The error term and time-cost for various CN-it combinations
are plotted in the Fig. 6. As shown, the error-term reduces with
B. Performance of it-ADI-FDTD-HD Method
increasing it. However, for larger CN, this rate of reduction is
To study the performance of it-ADI-FDTD-HD method, we smaller. That is, for CN = 300, the improvement in accuracy
use the same simulation cases as before with added variation in with increasing it is not as good as that for CN = 100. Previ-
the number of iterations. A few representative cases are chosen ously we discussed three cases with CN, it = 100, 2, 200,
to illustrate the effect of iterations: (1) CN, it = 100, 2 (2) 4, and 300, 7 that showed the best error-performances for
CN, it = 200, 4, and (3) CN, it = 300, 7. Note that these corresponding CN values. Using the quantitative error data, it is
choices are based on the smallest iteration count that provided further clear that 100, 2 case outperforms the other two cases.
good agreement with the reference data for specified CN. As Therefore, using lower CN and lower it values outperforms
before, the data are compared with the traditional FDTD-HD using higher CN and higher it values, for same time-cost cases.
solver for benchmarking. We notice an exponential decrease in the error with increasing
The simulation results showing 2-D fields are shown in it for each CN, but this improvement saturates faster for larger
Fig. 3(b) and time-costs are listed in Table II. As shown, the CN. E.g., for CN = 200 and it = 4, 5 the smallest error of around
obtained 2-D fields are in good agreement with the reference 8% persists even with increasing it. Possible reason is higher
data. The plots demonstrate that indeed the field-corrections dispersion experienced by larger CN cases [22], [23]. Note that
are achieved in the 2DEG regions since we observe a well- iterations only affect the splitting-error terms. The larger CN
defined plasmonic field profile. Specifically, the spurious field cases are ultimately limited by dispersion errors due to large

Authorized licensed use limited to: SASTRA. Downloaded on July 12,2024 at 04:04:41 UTC from IEEE Xplore. Restrictions apply.
BHARDWAJ et al.: FAST MODELING OF TERAHERTZ PLASMA-WAVE DEVICES USING UNCONDITIONALLY STABLE FDTD METHODS 35

REFERENCES
[1] M. Dyakonov and M. Shur, “Detection, mixing, and frequency multipli-
cation of terahertz radiation by two-dimensional electronic fluid,” IEEE
Trans. Electron. Devices, vol. 43, no. 3, pp. 380–387, Mar. 1996.
[2] A. Tredicucci and M. S. Vitiello, “Device concepts for graphene-based
terahertz photonics,” IEEE J. Select. Topics Quantum Electron., vol. 20,
no. 1, pp. 130–138, Jan./Feb. 2014.
[3] T. Otsuji, V. Popov, and V. Ryzhii, “Active graphene plasmonics for
terahertz device applications,” J. Phys. D, Appl. Phys., vol. 47, no. 9,
2014, Art. no. 094006.
[4] V. Mitin, “Graphene-based terahertz devices: Concepts and characteris-
tics,” Future Trends in Microelectronics: From Nanophotonics to Sensors
to Energy. Hoboken, NJ, USA: Wiley, pp. 293–306, 2010.
[5] F. Rana, “Graphene terahertz plasmon oscillators,” IEEE Trans. Nanotech-
nol., vol. 7, no. 1, pp. 91–99, Jan. 2008.
[6] S. Bhardwaj, N. K. Nahar, S. Rajan, and J. L. Volakis, “Numerical analy-
sis of terahertz emissions from an ungated hemt using full-wave hydrody-
namic model,” IEEE Trans. Electron. Devices, vol. 63, no. 3, pp. 990–996,
Mar. 2016.
Fig. 6. Variation of error and time-costs with increasing iteration-count for [7] S. Bhardwaj, B. Sensale-Rodriguez, H. G. Xing, S. Rajan, and J. L.
CN = 100, 200 and 300. The error-terms are marked in red and time-variations Volakis, “Resonant tunneling assisted propagation and amplification of
are plotted in blue. plasmons in high electron mobility transistors,” J. Appl. Phys., vol. 119,
no. 1, 2016, Art. no. 013102.
[8] S. Bhardwaj, S. Rajan, and J. L. Volakis, “Analysis of plasma-modes of a
Δt. Examination of dispersion errors for terahertz plasmonic gated bilayer system in high electron mobility transistors,” J. Appl. Phys.,
vol. 119, no. 19, 2016, Art. no. 193102.
applications is separate area of discussion that may be pursued [9] M. A. Khorrami, S. El-Ghazaly, H. Naseem, and S.-Q. Yu, “Global mod-
further but is beyond the scope of current work. eling of active terahertz plasmonic devices,” IEEE Trans. Terahertz Sci.
Even with better error performance for low-CN-low-it case, Technol., vol. 4, no. 1, pp. 101–109, Jan. 2014.
[10] T. Otsuji, M. Hanabe, T. Nishimura, and E. Sano, “A grating-bicoupled
higher CN simulations could also find applications. If time-cost plasma-wave photomixer with resonant-cavity enhanced structure,” Opt.
is a priority over accuracy, especially when only a rough estimate Express, vol. 14, no. 11, pp. 4815–4825, 2006.
or qualitative idea of the plasma-wave is needed, higher-CN- [11] K. Willis, J. Ayubi-Moak, S. Hagness, and I. Knezevic, “Global model-
ing of carrier-field dynamics in semiconductors using EMC–FDTD,” J.
low-it case can be used, allowing small simulation times with Comput. Electron., vol. 8, no. 2, pp. 153–171, 2009.
some cost to accuracy of the solution. In other words, iterative [12] M. A. Alsunaidi, S. S. Imtiaz, and S. M. El-Ghazaly, “Electromagnetic
ADI-FDTD-HD model has provided tunability between time- wave effects on microwave transistors using a full-wave time-domain
model,” IEEE Trans. Microw. Theory Techn., vol. 44, no. 6, pp. 799–808,
cost and error by modifying the CN and the iteration count Jun. 1996.
parameters. [13] R. O. Grondin, S. M. El-Ghazaly, and S. Goodnick, “A review of global
modeling of charge transport in semiconductors and full-wave electromag-
netics,” IEEE Trans. Microw. Theory Techn., vol. 47, no. 6, pp. 817–829,
IV. CONCLUDING REMARKS Jun. 1999.
[14] T. Namiki, “A new FDTD algorithm based on alternating-direction im-
In this paper, we have proposed efficient new methods for plicit method,” IEEE Trans. Microw. Theory Techn., vol. 47, no. 10,
modeling of the plasma-wave phenomenology at terahertz fre- pp. 2003–2007, Oct. 1999.
[15] S. Wang, F. L. Teixeira, and J. Chen, “An iterative ADI-FDTD with
quencies. For the device under consideration, by the use of reduced splitting error,” IEEE Microw. Wireless Components Lett., vol. 15,
it-ADI-FDTD-HD method total simulation time was reduced no. 2, pp. 92–94, Feb. 2005.
to 42% of the original time, while a nominal error of 3% was [16] F. Zheng, Z. Chen, and J. Zhang, “A finite-difference time-domain method
without the courant stability conditions,” IEEE Microw. Guided Wave Lett.,
recorded. vol. 9, no. 11, pp. 441–443, Nov. 1999.
We find for plasmonic terahertz devices, the traditional FDTD [17] R. Mirzavand, A. Abdipour, G. Moradi, and M. Movahhedi, “Full-wave
methods are slow due to highly refined mesh. When we seek semiconductor devices simulation using ADI-FDTD method,” Progress
Electromagn. Res. M, vol. 11, pp. 191–202, 2010.
ADI-FDTD based methods, we encounter significant splitting- [18] A. Taflove and S. C. Hagness, Computational Electrodyanamics, The finite
errors (or truncation errors). This becomes especially true within Difference Time Domain Method. Norwood, MA, USA: Artech House,
regions of our interests (i.e., near 2DEG) due to the associated 2005.
[19] S. Wang and F. L. Teixeira, “An efficient PML implementation for the
strong field-derivatives in those regions. Iterative-ADI-FDTD ADI-FDTD method,” IEEE Microw. Wireless Compon. Lett., vol. 13, no. 2,
based methods, therefore, become effective tools to recover the pp. 72–74, Feb. 2003.
accuracy of the solution. Overall, we maintain the accuracy- [20] D. Vasileska, S. Goodnick, and K. G., Computational Electronics, Semi-
classical and Quantum Device Modeling and Simulation. Boca Raton, FL,
levels with significant time-cost advantages as compared to tra- USA: CRC Press, 2010.
ditional explicit-FDTD modeling. [21] A. Aste and R. Vahldieck, “Time-domain simulation of the full hy-
Further, it-ADI-FDTD-HD model provided a way of tun- drodynamic model,” Int. J. Numerical Model., Electron. Netw., De-
vices Fields, vol. 16, no. 2, pp. 161–174, 2003. [Online]. Available:
ing the time-cost and accuracy levels for different applications http://dx.doi.org/10.1002/jnm.491
within terahertz plasmonic analysis. For scenarios where high [22] S. G. Garcia, T.-W. Lee, and S. C. Hagness, “On the accuracy of the
accuracy is desired, a low-CN-low-it combination could be ADI-FDTD method,” IEEE Antennas Wireless Propag. Lett., vol. 1, no. 1,
pp. 31–34, 2002.
used, but for the cases when time-cost is a concern, we can use [23] T. Namiki and K. Ito, “Investigation of numerical errors of the two-
high-CN-low-it setup for reduced simulation times with modest dimensional adi-fdtd method [for maxwell’s equations solution],” IEEE
accuracy-levels. Trans. Microw. Theory Techn., vol. 48, no. 11, pp. 1950–1956, Nov. 2000.

Authorized licensed use limited to: SASTRA. Downloaded on July 12,2024 at 04:04:41 UTC from IEEE Xplore. Restrictions apply.
36 IEEE JOURNAL ON MULTISCALE AND MULTIPHYSICS COMPUTATIONAL TECHNIQUES, VOL. 3, 2018

Shubhendu Bhardwaj (S’12–M’17) received the John L. Volakis (S’77–M’82–SM’89–F’96) was

B. Tech degree in electronics engineering from IIT born on May 13, 1956 in Chios, Greece, and im-
(ISM)-Dhanbad, India in 2004, the M.S. degree in migrated to the USA, in 1973. He received the B.E.
electrical engineering from UCLA, Los Angeles, CA, degree (summa cum laude) from Youngstown State
USA, in 2012, and the Ph.D. degree in electrical engi- University, Youngstown, OH, USA, in 1978, and
neering from The Ohio State University, Columbus, the M.Sc. and Ph.D. degrees in electrical engineer-
OH, USA, in 2017. ing from The Ohio State University, Columbus, OH,
From 2008 to 2010, he was with Samsung In- USA, in 1979 and 1982, respectively.
dia Software Operations in Bangalore, India. Since He started his career with Rockwell International
2017, he has been an Assistant Professor with the (1982–1984), now Boeing. In 1984, he was appointed
Department of Electrical and Computer Engineering, Assistant Professor at The University of Michigan,
Florida International University, Miami, FL, USA. Ann Arbor, MI, USA, becoming a Full Professor in 1994. He was also the
Dr. Bhardwaj graduated summa cum laude from IIT (ISM)-Dhanbad. He Director of the Radiation Laboratory from 1998 to 2000. From January 2003 to
is a recipient of the best student paper awards at URSI-GASS-2017, IEEE- August 2017, he was the Roy and Lois Chope Chair Professor of Engineering
iWat-2017, and IEEE-AMTA-2015. His paper also received second place at at the Ohio State University, Columbus, Ohio, and the Director of the Electro-
student paper competition at AMTA-2014 and honorable mentions at APS- Science Laboratory from 2003 to 2016. He is currently the Dean of the College
2014 and 2015. Currently he is currently working towards the electromagnetics of Engineering and Computing and a Professor with the Department of Elec-
including computational electromagnetics, power harvesting, terahertz devices, trical and Computer Engineering, Florida International University, Miami, FL,
slow-wave structures, and sub-mm-wave/terahertz antennas. USA, from August 2017. He has authored and coauthored 8 books, 400 journal
papers, nearly 800 conference papers, 26 book chapters, and 17 patents/patent
disclosures. Among his coauthored books are Approximate Boundary Condi-
tions in Electromagnetics (IET Digital Library, 1995); Finite Element Methods
for Electromagnetics (Wiley, 1998); Antenna Engineering Handbook, 4th ed.
Fernando L. Teixeira (F’15) received the B.S.E.E. (McGraw-Hill, 2007); Small Antennas (McGraw-Hill, 2010; and Integral Equa-
and M.S.E.E. degrees in electrical engineering from tion Methods for Electromagnetics (IET Digital Library, 2011). Over the years,
Pontifical Catholic University of Rio de Janeiro, Rio he carried out research in antennas, wireless communications and propagation,
de Janeiro, RJ, Brazil, in 1991 and 1995, respec- computational methods, electromagnetic compatibility and interference, design
tively, and the Ph.D. degree in electrical engineering optimization, RF materials, multiphysics engineering, millimeter waves, tera-
from the University of Illinois at Urbana- Champaign, hertz, and medical sensing.
Champaign, IL, USA, in 1999. Dr. Volakis has graduated/mentored nearly 90 doctoral students/postdocs
From 1999 to 2000, he was a Postdoctoral As- with 41 of them receiving best paper awards at conferences. His service to
sociate with the Massachusetts Institute of Technol- Professional Societies include 2004 President of the IEEE Antennas and Prop-
ogy, Cambridge, MA, USA. In 2000, he joined the agation Society (2004), Chair of USNC/URSI Commission B (2015-2017),
Ohio State University, Columbus, OH, USA, where twice the general Chair of the IEEE Antennas and Propagation Symposium,
he is currently a Professor with the Department of Electrical and Computer IEEE APS Distinguished Lecturer, IEEE APS Fellows Committee Chair, IEEE-
Engineering and affiliated with the ElectroScience Laboratory. His research wide Fellows committee member & Associate Editor of several journals. He
interests include computational electromagnetics, metamaterials, modeling of was listed by ISI among the top 250 most referenced authors (2004), and is a
electromagnetic sensors and plasma-based devices, and ultrawideband inverse Fellow of ACES. He is a recipient of several awards: The University of Michi-
scattering. gan College of Engineering Research Excellence award (1993), the Scott award
Dr. Teixeira was a Chair of the Joint IEEE AP/MTT-S Columbus Chapter from The Ohio State University College of Engineering for Outstanding Aca-
and an Associate Editor for the IEEE ANTENNAS AND WIRELESS PROPAGATION demic Achievement (2011), the IEEE AP Society C-T. Tai Teaching Excellence
LETTERS. He is currently an Associate Editor for IET Microwaves, Antennas, and award (2011), the IEEE Henning Mentoring award (2013), the IEEE Anten-
Propagation. He was the recipient of the CAREER Award from the National nas & Propagation Distinguished Achievement award (2014), The Ohio State
Science Foundation, the triennial Booker Fellowship from the U.S. National University Distinguished Scholar Award (2016), and The Ohio State University
Committee of the International Union of Radio Science, and the Outstanding ElectroScience George Sinclair Award (2017). He is a Fellow of the Applied
Young Engineer Award from the IEEE Microwave Society. Computational Electromagnetics Society.

Authorized licensed use limited to: SASTRA. Downloaded on July 12,2024 at 04:04:41 UTC from IEEE Xplore. Restrictions apply.