562 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 57, NO.

3, MARCH 2010

Electrothermal Monte Carlo Simulation of GaN

HEMTs Including Electron–Electron Interactions
Ashwin Ashok, Dragica Vasileska, Senior Member, IEEE, Olin L. Hartin, and Stephen M. Goodnick, Fellow, IEEE

Abstract—A Monte Carlo device simulator was developed to

investigate the electronic transport properties in AlGaN/GaN
high-electron mobility transistors (HEMTs). Electron–electron in-
teractions were included using a particle–particle–particle–mesh
coupling scheme. Quantum corrections were applied to the het-
erointerface using the effective potential approach due to Ferry.
Thermal effects were also included by coupling the particle-based
device simulator self-consistently with an energy balance solver
for the acoustic and optical phonons. The electrothermal device
simulator was used to observe the temperature profiles across the
device. Hot spots or regions of higher temperatures were found
along the channel in the gate–drain spacing. Results from elec-
trothermal simulations show self-heating degradation of perfor-
mance at high source–drain bias. More importantly, the observed
nonequilibrium phonon effects may play an important role in
determining the thermal distribution in these HEMTs, resulting Fig. 1. Simulated 2-D AlGaN/GaN HEMT structure. The source and drain
in reliability issues such as current collapse. electrodes are ohmic contacts and are doped to 1018 cm−3 . The gate electrode
is a Schottky contact, and the Schottky barrier height is calculated to be equal
Index Terms—Electromechanical coupling, GaN HEMTs, to 1.17 eV [11].
Monte Carlo particle based device simulations, self-heating
effects.
five times larger as compared to AlGaAs/GaAs structures, lead-
ing to a significant increase in the sheet carrier concentration
I. I NTRODUCTION
in the vicinity of the heterointerface [7], [8]. The discontinuity
of the spontaneous polarization, due to the lack of symmetry
G aN-BASED high-electron mobility transistors (HEMTs)
are promising devices for high power, high frequency,
and high temperature applications, owing to the material prop-
in wurtzite crystals, induces free carriers at the heterointerface.
The spontaneous and piezoelectric polarization in wurtzite GaN
erties such as wide bandgap, large breakdown electric field, and and AlN are found to be ten times larger than in conventional
high saturation drift velocity [1], [2]. Recently, improvements III–V and II–VI semiconductor compounds. The spontaneous
in material growth and device design have produced state- polarization can cause electric fields of up to 3 MV/cm in
of-the-art AlGaN/GaN HEMTs with power densities as high group III nitride crystals, and strain in pseudomorphically
as 11.2 W/mm [3], pulsed total output power of 50 W [4], grown AlGaN/GaN or lnGaN/GaN heterostructures can cause
and ft of over 100 GHz [5]. Further contributing factor to an additional piezoelectric field of about 2 MV/cm. These very
the outstanding performance of AlGaN/GaN-based HEMTs is high polarizations and resulting electric fields produce high
the ability to achieve 2-D electron gases (2DEGs) with sheet interface charge densities at group III nitride interfaces [9].
carrier concentrations of 1013 cm−2 or higher close to the Including these polarization effects is a very critical issue while
heterointerface without any intentional doping, which is well in modeling AlGaN/GaN HEMTs and is addressed in [10].
excess of those achievable in other III–V material systems [6].
In wurtzite AlGaN/GaN-based transistor structures, the
piezoelectric polarization of the strained top layer is more than II. D EVICE S TRUCTURE AND T HEORETICAL M ODEL
Fig. 1 shows the simulated AlGaN/GaN HEMT structure.
A 23-nm unintentionally doped AlGaN layer was formed on
Manuscript received June 3, 2009; revised October 23, 2009. First published 100 nm of the unintentionally doped GaN substrate. An unin-
January 22, 2010; current version published February 24, 2010. This work was tentional doping of 1017 cm−3 is assumed for both the AlGaN
supported by the National Science Foundation under Grant 0901251 (Modeling
Heating Effects in Low-Power Multi-Gate SOI Devices and High-Power GaN
and GaN layers.
HEMTs). The review of this paper was arranged by Editor M. Anwar. Bulk ensemble Monte Carlo (EMC) transport simulations
A. Ashok is with the Intel Corporation, Hillsboro, OR 97124 USA. were first performed to obtain the correct material parameters
D. Vasileska and S. M. Goodnick are with the Department of Electri-
cal Engineering, Arizona State University, Tempe, AZ 85287 USA (e-mail: that fit the theoretical and experimental results. The EMC
vasileska@asu.edu). transport kernel incorporates a three-valley nonparabolic band
O. L. Hartin is with the Freescale Semiconductors, Tempe, AZ 85284 USA. model and various scattering mechanisms such as acoustic,
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. polar optical phonon, ionized impurity, intervalley phonon,
Digital Object Identifier 10.1109/TED.2009.2038585 piezoelectric, and dislocation scattering were included. The
0018-9383/$26.00 © 2010 IEEE
ASHOK et al.: MONTE CARLO SIMULATION OF GaN HEMTs INCLUDING ELECTRON-ELECTRON INTERACTIONS 563

Fig. 3. ID –VD characteristics of the GaN HEMT with and without the
inclusion of the short range electron–electron interactions for two different gate
voltages.

Fig. 2. Flowchart of the device simulator in which changes in the piezoelectric

polarization charge due to electromechanical coupling [15] are included as
well as short-range electron–electron interactions using a real-space molecular
dynamics routine.
Fig. 4. ID –VG characteristics of the GaN HEMT with and without the
material and valley parameters chosen for these simulations are inclusion of the short range electron–electron interactions.
summarized in [12].

0.85 × 1012 cm−2 , total spontaneous and piezoelectric charge

III. E LECTRON –E LECTRON I NTERACTIONS
density at the GaN/AlGaN heterointerface = 1.23 × 1013 cm−2
AlGaN/GaN-based HEMTs have a 2DEG with sheet carrier (for the case when electromechanical coupling, quantum, and
concentrations on the order of 1013 cm−2 or higher at the Coulomb effects are included in the model), and back interface
heterointerface without any intentional doping. Devices having spontaneous polarization charge density = 2.05 × 1012 cm−2 .
these extremely high electron densities are often subject to Note that the theoretical value of the total (spontaneous +
many carrier–carrier scattering events that tend to degrade the piezoelectric) charge density at the GaN/AlGaN interface is
device performance. Hence, in these systems, it is very im- 1.38 × 1013 cm−2 .
portant to include short-range electron–electron interactions in Simulations were performed to investigate the effect of the
order to accurately model carrier transport [13]. The approach inclusion of the short-range electron–electron interactions and
used by Wordelman and Ravaioli [14] is adopted here for the the electromechanical coupling on the device characteristics.
inclusion of the short-range electron–electron interactions. Figs. 3 and 4 show the ID –VD and the ID –VG characteristics
The short-range force is calculated and stored in a lookup for the coupled formulation and with the inclusion of the short-
table in the initial stages of the simulation. Once the total force range electron–electron interactions.
associated with each particle in the ensemble is computed, The inclusion of the short-range electron–electron interac-
the particle trajectories are integrated forward in time using tions tends to degrade the device characteristics of the HEMT.
the leap-frog scheme. The change in momentum of a particle Fig. 3 clearly shows the effect of the inclusion of the short-
at each time step is determined by the total force on that range forces, which acts as an additional scattering mechanism
particle. A flowchart of the simulator that uses polarization (as to lower the drain current. This effect is more prominent for
described in [10] and [14]) and a short-range force calculation gate biases for which we have higher sheet charge density at the
is given in Fig. 2. The charges used to fit experimental data heterostructure interface. This is clearly seen from the results
placed at the top surface, GaN/AlGaN heterointerface and at presented in Fig. 3 for the two voltages used (i.e., VG = 0 V
the bottom surface are: front interface sheet charge density = and VG = −2 V). For the gate bias of VG = 0 V, we have
564 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 57, NO. 3, MARCH 2010

TABLE I
S EMICLASSICAL D EVICE S IMULATION A PPROACHES
T HAT I NCLUDE Q UANTUM C ORRECTIONS

significantly higher sheet electron density and, therefore, higher

current degradation. Fig. 5. Optimal value of a0 used in the simulations for particular gate biases.

IV. Q UANTUM –M ECHANICAL S PACE Q UANTIZATION

The inclusion of quantum effects in the description of the car-
riers in a HEMT involves solving the 1-D Schrödinger–Poisson
equation for the carriers in the triangular potential well in
slices along the channel assuming the adiabatic approximation,
namely, assuming that the confinement does not change that
rapidly along the channel. Note that the delta charges at the
front interface, at the GaN/AlGaN and at the bottom interface
are accounted for when solving the 1-D Poisson’s equation to
enable identical simulation environment. Using this approach,
one obtains bound states, which give rise to the following two
major features: 1) reduced sheet charge density and 2) charge
setback. Since the lowest bound state can be regarded as the new
bottom of the conduction band, the spacing between the Fermi
level and the conduction band edge is effectively increased,
which results in reduced sheet charge density with respect to
the case in which quantum effects are excluded.
Fig. 6. Comparison of electron density at different gate voltages between the
Since the above approach is time and memory consuming, Schrödinger–Poisson solution and the effective potential solution.
approximate schemes are being used in semiclassical transport
simulators that utilize the Bohm quantum potential [16]. When
the Bohm potential is used in drift–diffusion simulations one the charge setback and quantization of energy, besides having
arrives at the so-called density gradient method [17]. When the advantage of easy numerical implementation [21]–[23].
the Bohm potential is used in conjunction with hydrody- While implementing the effective potential approach due to
namic approaches one arrives at the quantum hydrodynamic Ferry, one of the questions that arise is the actual value of the
approach [18]. Since the Bohm potential approach involves Gaussian smoothing parameter, a0 , which describes the extent
second derivative of the log of the density, as such it is difficult of the spread of the wave packet. In addition, it is uncertain
to implement in particle-based device simulations due to the as to how much this smoothing parameter will depend upon
inherent statistical noise in the electron density calculation. the shape of the confining potential or the substrate doping
Therefore, for particle-based device simulation approaches it is density. In order to determine the optimal a0 , one needs to
more appropriate to use the idea of Feynmann and Kleinart [19] compare the electron distribution of the effective potential with
that has been utilized for the first time by Ferry [20] and arrive that of a 1-D Schrödinger–Poisson self-consistent calculation.
at an effective potential that is calculated as a convolution of the Using this 1-D Schrödinger–Poisson self-consistent calculation
Hartree potential (that is obtained as a solution of the Poisson we can compute the quantum mechanical electron density for
equation) and a Gaussian function with standard deviation that different gate voltages. This is then compared to the electron
is correlated to the size of the electron. This effective potential densities obtained through the effective potential approach for
is then used in the calculation of the electric fields that drive the corresponding gate voltages, and the optimum value of the
the carriers during the free-flight-scatter part of the particle- Gaussian smoothing parameter ao is chosen.
based device simulation scheme. A summary of the quantum Fig. 5 shows the optimal value of a0 obtained from compar-
corrected approaches to classical simulations is given in Table I. ing the Schrödinger–Poisson solution to the effective potential
Note that the effective potential approach has been applied to solution. The comparison of electron densities at different gate
various device simulations and has shown to account for both voltages between the two approaches is shown in Fig. 6. These
ASHOK et al.: MONTE CARLO SIMULATION OF GaN HEMTs INCLUDING ELECTRON-ELECTRON INTERACTIONS 565

Fig. 9. ID –VG characteristics with and without the inclusion of quantum

corrections.
Fig. 7. Conduction band profile along the depth with and without effective
potential.

Fig. 8. ID –VD characteristics with and without the inclusion of quantum

corrections. Fig. 10. Average electron drift velocity along the channel with and without
the effective potential at VGS = 0 V and VDS = 10 V.

optimal values of a0 are used in the device simulations so as to difference in the drain currents with and without the inclusion
include quantum corrections accurately. of quantum corrections for the various gate voltages. Fig. 9
Simulations were run to investigate the effect of including depicts the device ID –VG characteristics with and without
quantum corrections in the device characteristics of the HEMT the inclusion of quantum corrections. Both models tend to
device being investigated. For these simulations, the coupled have a good agreement to the experimental threshold voltage,
formulation polarization model was utilized, and short-range although there is a small shift when the quantum corrections are
interactions were also included. The net polarization charge included. The inset in Fig. 9 shows the variation of drain current
at the AlGaN/GaN interface was used to fit the on current. on a log scale which shows the small shift in the threshold
The simulations were performed with and without the effective voltage.
potential correction for different gate and drain voltages. The The average electron velocity and average electron energy
conduction band profile along the depth is shown in Fig. 7. along the channel for a gate bias (VGS ) of 0 V and drain bias
It clearly shows that the effective potential approach tends to (VDS ) of 10 V are shown in Figs. 10 and 11, respectively.
lower the electron density and to introduce a charge setback The average electron drift velocity with the effective potential
effect. correction is slightly lower than the average velocity without
Figs. 8 and 9 show the ID –VD and ID –VG characteristics, the effective potential correction. On the other hand, when one
respectively. The ID –VD characteristics show that the effect compares the average electron energy along the channel, the
of including quantum corrections via the effective potential model with effective potential correction has a slightly larger
approach is to lower the drain current. In our simulations, we energy than the one without the effective potential correction.
have used the model with the effective potential approach as our
reference in matching the drain current to experiment. Hence,
the drain currents are higher for the case when the model does
V. T HERMAL E FFECTS
not include quantum corrections.
The effective potential approach lowers the electron density Lai and Majumdar [24] developed a coupled electrothermal
in addition to the charge setback effect. There is about 5%–10% model for studying thermal nonequilibrium in submicrometer
566 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 57, NO. 3, MARCH 2010

Fig. 12. Carrier density dependant phonon lifetime, τLO [112].

Fig. 11. Average electron energy along the channel with and without the
effective potential at VGS = 0 V and VDS = 10 V.

silicon MOSFETs. Their results showed that the highest elec-

tron and lattice temperatures occur under the drain side of the
gate electrode, which also corresponds to the region where
nonequilibrium effects such as impact ionization and velocity
overshoot are maximum. Majumdar et al. [25] have analyzed
the variation of hot electrons and associated hot phonon effects
in GaAs MESFETs. These hot carriers were observed to de-
crease the output drain current by as much as 15%. Recently,
there have been numerous studies on describing thermal effects
in devices that couple the Monte Carlo/Poisson approach to
electrothermal modeling in SOI and nitride devices [26]–[30].
In these studies, most of the approaches use a simplistic model
for nonequilibrium phonons, which does not distinguish the
acoustic and the optical phonons as separate subsystems. In Fig. 13. Flowchart of the electrothermal Monte Carlo device simulator.
addition to that, none of the approaches include all the thermal
effects that come into play in GaN HEMTs operating at large simulation time has elapsed, the steady-state current through a
drain biases—hot electrons, self heating, and nonequilibrium specified terminal is calculated. To continue with the thermal
phonon effects. part of the simulation, the average electron density, drift veloc-
Raleva et al. used a similar approach to Majumdar et al. ity, and electron temperature must be calculated on the device
to solve the Boltzmann transport equation for electrons using grid. These are used in the phonon balance equations to com-
the EMC method coupled with moment expansion equations pute the acoustic and optical phonon temperature distribution
for the phonons, both acoustic and optical. In this paper, the all over the device. During the simulation, the gate contact and
acoustic and optical phonon energy balance equations were the bottom of the substrate are set to 300 K, while Neumann
solved simultaneously to model electrothermal effects in fully boundary conditions for the heat transfer are used in all other
depleted SOI devices [31]. In this work, we have utilized the outer surfaces.
same approach to include thermal effects in GaN HEMTs. When the simulation starts, all variables obtained from the
In GaN, recent experimental findings have shown that the first iteration of the EMC solver are calculated using uniform
optical phonon (LO) lifetime has carrier density dependence distribution for the acoustic and optical phonon temperatures.
[30], [32]. The lifetime was found to decrease from 2.5 ps at low This means that only one scattering table is used for all elec-
densities to 0.35 ps at higher densities in Raman spectroscopy trons, no matter where they are located in the device. When
measurements [32]. We have formulated an analytical expres- the phonon temperatures are computed from the phonon energy
sion for the optical phonon (LO) lifetime that incorporates balance equations, they are “returned” at the beginning of the
this density dependence. Fig. 12 shows the experimental data MC free-flight-scattering phase. Now, for each mesh point, we
and the analytical form of the phonon lifetime used in our have a scattering table that corresponds to the acoustic and
simulations. optical phonon temperatures at that point. In this case, the
To simulate the steady-state state behavior of a device, the electron position defines which scattering table is “valid,” and
system is started in some initial condition, with the desired then, by generating a random number, the scattering mechanism
potential applied to the contacts, and then, the simulation pro- is chosen for the given electron energy.
ceeds in a time-stepping manner until steady state is reached. The outer Gummel loop between the MC solver and the
When the system is driven into a steady-state regime and MC phonon energy balance solver ends when the steady-state
ASHOK et al.: MONTE CARLO SIMULATION OF GaN HEMTs INCLUDING ELECTRON-ELECTRON INTERACTIONS 567

Fig. 14. Drain current versus number of thermal iterations for VGS = 0 V
and VDS = 10 V.

Fig. 16. Lattice (acoustic phonon) temperature profile.

Fig. 15. Electron temperature contour for VGS = 0 V and VDS = 10 V.

conditions for the phonon temperatures and the device current

are reached (see Fig. 13). To test the overall convergence of
the coupled thermal and EMC codes, the variations of the drain
current with the number of thermal iterations are registered for Fig. 17. Optical phonons temperature profile.
every bias condition. The results of our simulations show that
only five to ten thermal iterations are necessary to obtain the respectively. Both profiles show that the peak temperatures are
steady-state solution of the current. Fig. 14 shows the observed experienced along the channel in the gate–drain spacing. The
current convergence for zero gate voltage and 10 V applied on phonon temperature profiles look fairly similar, except that the
the drain. There is a reduction of about 11% in the drain current optical phonons have a slightly larger spread around the middle
after ten thermal iterations. of the channel. This “phonon hot spot” has been observed in a
On a closer inspection, the electron temperature profile recent theoretical study by Yoder et al. done on GaN HEMTs
shows a very different temperature distribution across the de- using a dispersionless optical phonon model coupled to the
vice. This is better seen in the temperature contour shown Monte Carlo transport routine.
in Fig. 15. The hot spot occurs in the gate–drain spacing, This study reveals that the hot spot in the channel is local-
right where the gate terminates, but is restricted closer to the ized at low drain–source bias, but expands toward the drain
AlGaN/GaN interface. This means that most of the hot elec- at higher bias, significantly degrading channel mobility [27].
trons are close to the AlGaN/GaN interface. The profile also In [31], polarization charge was treated as a fixed source term
shows that there might be some high energy electrons in the in Poisson’s equation, and a constant optical phonon lifetime of
AlGaN barrier layer on the drain end. 5 ps was used.
The acoustic phonon (lattice) temperature and optical Fig. 18 shows the lattice temperature distribution changing
phonon temperature profiles are shown in Figs. 16 and 17, with the change in drain voltage for gate voltage of VGS = 0 V.
568 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 57, NO. 3, MARCH 2010

Fig. 19. ID –VD characteristics with and without thermal effects.

Fig. 20. ID –VG characteristics with and without thermal effects.

the simulated range of drain voltages. The isothermal 400 K

simulation brings about a reduction of around 8% at VD = 10 V
in comparison to the thermal simulations. The ID –VG charac-
teristics also show a similar trend—a reduction in drain current
as well as a change in the slope, which results in a change in the
transconductance. The influence of the thermal effects becomes
smaller as one increases the gate bias negatively because there
are fewer electrons present in the channel.
In summary, the thermal effects arising due to contributions
from the lattice, phonons and hot electrons were included in our
Fig. 18. Lattice (acoustic phonon) temperature for lower drain voltages. simulations and current degradation is observed in the device
characteristics. The temperature profiles show that hot spots in
As the drain voltage is reduced, one can see that the peak the device are confined close to the channel suggesting a strong
temperature reduces in the hot spot regions. dependence of the thermal distribution on the average electron
Figs. 19 and 20 show the ID –VD and ID –VG characteristics, density and energy. The high thermal conductivity of GaN and
respectively. In order to quantify the effect of these nonequi- its alloys greatly helps in the faster heat dissipation seen in these
librium thermal effects on the device performance, we started devices. The presence of some electrons in the AlGaN barrier
with isothermal simulations run at 300 and 400 K. These I–V layer introduces the possibility of these electrons being lost to
characteristics were then compared with the results obtained surface states and buffer trap sites.
from our electrothermal simulations. Several studies on GaN HEMT reliability have focused on
The ID –VD characteristics show that the inclusion of ther- these current collapses observed in short-term dc stress tests.
mal effects reduces the drain current at larger drain voltages. Most researchers have proposed that the role of these trap
This reduction in drain current increases as the drain volt- sites in trapping/detrapping electrons can lead to a permanent
age increases, thereby bringing a negative roll-off—an effect reduction in the polarization charge density available at the
typically attributed to lattice heating, as seen experimentally. AlGaN/GaN interface [33]. Our simulation results also point
The drain current reduction ranges between 5% and 10% over toward the same conclusions.
ASHOK et al.: MONTE CARLO SIMULATION OF GaN HEMTs INCLUDING ELECTRON-ELECTRON INTERACTIONS 569

VI. C ONCLUSION transistors with ft of 110 GHz,” Electron. Lett., vol. 36, no. 4, pp. 358–
359, Feb. 2000.
In conclusion, we have developed a particle-based device [6] M. B. Nardelli, K. Rapcewicz, and J. Bernholc, “Polarization
simulator that takes into consideration the gate voltage depen- field effects on the electron–hole recombination dynamics in
In0.2 Ga0.8 N/In1−x Gax N multiple quantum wells,” Appl. Phys.
dence of the piezoelectric component of the polarization charge Lett., vol. 71, no. 21, p. 3135, Nov. 1997.
due to the variation in the electromechanical coupling at the [7] A. Bykhovski, B. L. Gelmont, and M. S. Shur, “Elastic strain relaxation
interface. In terms of application in commercial simulators, and piezoeffect in GaN–AlN, GaN–AlGaN and GaN–InGaN superlat-
tices,” J. Appl. Phys., vol. 81, no. 9, p. 6332, May 1997.
the gate voltage dependence of the piezoelectric polarization [8] E. T. Yu, G. J. Sullivan, P. M. Asbeck, C. D. Wang, D. Qiao, and
charge can simply be accounted for by increase of about 10% S. S. Lau, “Measurement of piezoelectrically induced charge in
of the polarization charge due to spontaneous polarization. GaN/AlGaN heterostructure field-effect transistors,” Appl. Phys. Lett.,
vol. 71, no. 19, p. 2794, Nov. 1997.
In this paper, we also show that the situation is rather differ- [9] F. Bernardini, V. Fiorentini, and D. Vanderbilt, “Spontaneous polarization
ent when we include the short-range electron–electron interac- and piezoelectric constants of III–V nitrides,” Phys. Rev. B, Condens.
tions. In this case, for gate biases that induce larger sheet charge Matter, vol. 56, no. 16, pp. R10 024–R10 027, Oct. 1997.
[10] A. Ashok, D. Vasileska, S. M. Goodnick, and O. Hartin, “Importance
density in the channel, as expected, the electron–electron inter- of the gate-dependent polarization charge on the operation of GaN
actions play a more significant role and lead to larger current HEMTs,” IEEE Trans. Electron Devices, vol. 56, no. 5, pp. 998–1006,
degradation. May 2009.
[11] O. Ambacher, J. Smart, J. R. Shealy, N. G. Weimann, K. Chu, M. Murphy,
A self-consistently coupled thermal/EMC device simulator W. J. Schaff, L. F. Eastman, R. Dimitrov, L. Wittmer, M. Stutzmann,
was developed and used in the investigation of self-heating W. Reiger, and J. Hilsenbeck, “Two-dimensional electron gases induced
effects in GaN HEMT device. Carrier-density-dependent life- by spontaneous and piezoelectric polarization changes in N- and Ga-face
AlGaN/GaN heterostructures,” J. Appl. Phys., vol. 87, no. 1, pp. 334–344,
times were employed for the first time for the longitudi- Jan. 2000.
nal phonons to capture accurately the nonequilibrium phonon [12] A. Ashok, D. Vasileska, O. Hartin, and S. M. Goodnick, “Monte Carlo
effects. Simulations were performed on a GaN HEMT de- simulation of GaN n + nn+ diode including intercarrier interactions,” in
Proc. 7th IEEE Int. Conf. Nanotechnol., Hong Kong, Aug. 2–5, 2007,
vice with the inclusion of polarization effects, short-range p. 338.
electron–electron effects, quantum corrections, and thermal [13] W. J. Gross, D. Vasileska, and D. K. Ferry, “Ultra-small MOSFETs:
effects. All the aforementioned effects tend to bring a drain The importance of the full Coulomb interaction on device characteris-
tics,” IEEE Trans. Electron Devices, vol. 47, no. 10, pp. 1831–1837,
current degradation. This suggests that the inherent polar- Oct. 2000.
ization charge at the interface, which is responsible for the [14] C. J. Wordelman and U. Ravaioli, “Integration of a
creation of the 2DEG, may be much greater than those val- particle–particle–particle–mesh algorithm with the ensemble Monte
Carlo method for the simulation of ultra-small semiconductor devices,”
ues used in simulations. This observation suggests that the IEEE Trans. Electron Devices, vol. 47, no. 2, pp. 410–416, Feb. 2000.
AlGaN barrier layer need not undergo partial relaxation for [15] A. Ashok, D. Vasileska, S. M. Goodnick, and O. Hartin, “Importance of
lowering of the interface charge as is often assumed in de- the gate-dependent polarization charge on the operation of GaN HEMTs,”
IEEE Trans. Electron Devices, vol. 56, no. 5, pp. 998–1006, May 2009.
vice simulations. On the contrary, the interplay of the afore- [16] D. Bohm, Quantum Theory. New York: Dover, 1989.
mentioned effects—polarization, quantum, thermal effects, and [17] M. G. Ancona and G. J. Iafrate, “Quantum correction to the equation
short-range Coulomb interactions—are responsible for the ob- of state of an electron gas in a semiconductor,” Phys. Rev. B, Condens.
Matter, vol. 39, no. 13, pp. 9536–9540, May 1989.
served device characteristics, and closer agreement between the [18] G. J. Iafrate, H. L. Grubin, and D. K. Ferry, “Utilization of quantum dis-
theoretical and experimental characteristics when polarization tribution functions for ultra-submicron device transport,” J. Phys., vol. 42,
charge density very close to the theoretical value (that is calcu- pp. 307–312, 1981, (Colloq. 7).
[19] P. Feynman and H. Kleinert, “Effective classical partition functions,”
lated assuming no partial relaxation of the interface) is used. Phys. Rev. A, Gen. Phys., vol. 34, no. 6, pp. 5080–5084, Dec. 1986.
[20] D. K. Ferry, “The onset of quantization in ultra-submicron semiconductor
devices,” Superlattices Microstruct., vol. 27, no. 2, pp. 61–66, Feb. 2000.
ACKNOWLEDGMENT [21] I. Knezevic, D. Vasileska, and D. K. Ferry, “Impact of strong quantum
confinement on the performance of a highly asymmetric device structure:
The authors would like to thank Prof. D. K. Ferry for the Monte Carlo particle-based simulation of a focused-ion-beam MOSFET,”
IEEE Trans. Electron Devices, vol. 49, no. 6, pp. 1019–1026, Jun. 2002.
valuable discussions and comments while this work was being [22] L. Shifren, R. Akis, and D. K. Ferry, “Correspondence between quantum
performed. and classical motion: Comparing Bohmian mechanics with a smoothed
effective potential approach,” Phys. Lett. A, vol. 274, no. 1, pp. 75–83,
Sep. 2000.
R EFERENCES [23] D. Vasileska and S. S. Ahmed, “Narrow-width SOI devices: The role of
[1] M. J. Manfra, N. J. Weimann, Y. Baeyens, P. Roux, and quantum–mechanical size quantization effect and unintentional doping
D. M. Tennant, “Unpassivated AlGaN/GaN HEMTs with CW power on the device operation,” IEEE Trans. Electron Devices, vol. 52, no. 2,
density of 3.2 W/mm at 25 GHz grown by plasma-assisted MBE,” pp. 227–236, Feb. 2005.
Electron. Lett., vol. 39, no. 8, pp. 694–695, Apr. 2003. [24] J. Lai and A. Majumdar, “Concurrent thermal and electrical modeling of
[2] V. Kumar, A. Kuliev, R. Schwindt, M. Muir, G. Simin, J. Yang, sub-micrometer silicon devices,” J. Appl. Phys., vol. 79, no. 9, p. 7353,
M. A. Khan, and I. Adesida, “High performance 0.25 μm gate-length May 1996.
AlGaN/GaN HEMTs on sapphire with power density of over 4.5 W/mm at [25] A. Majumdar, K. Fushinobu, and K. Hijikata, “Effect of gate voltage on
20 GHz,” Solid State Electron., vol. 47, no. 9, pp. 1577–1580, Sep. 2003. hot-electron and hot phonon interaction and transport in a submicrometer
[3] L. F. Eastman, J. R. Shealy, V. Tilak, J. Smart, B. Green, and T. Prunty, transistor,” J. Appl. Phys., vol. 77, no. 12, p. 6686, Jun. 1995.
“AlGaN/GaN HEMT microwave CW power limits,” in Proc. 4th Int. Conf. [26] W. D. Hu, X. S. Chen, Z. J. Quan, C. S. Xia, W. Lub, and P. D. Ye,
Nitride Semicond., Denver, CO, 2001, p. 21. “Self-heating simulation of GaN-based metal–oxide–semiconductor high-
[4] Y. F. Wu, P. M. Chavarkar, M. Moore, P. Parikh, B. P. Keller, and electron-mobility transistors including hot electron and quantum effects,”
U. K. Mishra, “A 50-W AlGaN/GaN amplifier,” in IEDM Tech. Dig., J. Appl. Phys., vol. 100, no. 7, p. 074 501, 2006.
2000, pp. 375–376. [27] S. Sridharan, A. Venkatachalam, and P. D. Yoder, “Electrothermal analysis
[5] M. Micovic, N. X. Nguyen, P. Janke, W.-S. Wong, P. Hashimoto, of AlGaN/GaN high electron mobility transistors,” J. Comput. Electron.,
L.-M. McCray, and C. Nguyen, “GaN/AlGaN high electron mobility vol. 7, no. 3, pp. 236–239, Sep. 2008.
570 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 57, NO. 3, MARCH 2010

[28] T. Sadi, R. W. Kelsall, and N. J. Pilgrim, “Electrothermal Monte Carlo Olin L. Hartin received the B.S. and M.S. de-
simulation of submicrometer Si/SiGe MODFETs,” IEEE Trans. Electron grees in physics and electrical engineering and the
Devices, vol. 54, no. 2, pp. 332–339, Feb. 2007. Ph.D. degree from the University of Texas, Austin,
[29] T. Sadi, R. W. Kelsall, and N. J. Pilgrim, “Electrothermal Monte Carlo in 1998.
simulation of submicron wurtzite GaN/AlGaN HEMTs,” J. Comput. Elec- Since 1998, he has been with Motorola and
tron., vol. 6, no. 1–3, pp. 35–39, Sep. 2007. Freescale (formerly Motorola Semiconductor Prod-
[30] A. Matulionis, “Hot phonons in GaN channels for HEMTs,” Phys. Stat. ucts Sector), Tempe, AZ, doing device and TCAD
Sol. A, vol. 203, no. 10, pp. 2313–2325, 2006. engineering. For the last eight years, he has headed
[31] R. D. Vasileska, S. M. Goodnick, and M. Nedjalkov, “Modeling thermal the Device Physics and Simulation Group, which
effects in nanodevices,” IEEE Trans. Electron Devices, vol. 55, no. 6, does device TCAD, E&M, and signal integrity sim-
pp. 1306–1316, Jun. 2008. ulations. He is currently the Technology Computer-
[32] K. T. Tsen, J. G. Kiang, D. K. Ferry, and H. Morkoç, “Subpicosecond Aided Design Standardization Committee Chairman for Freescale and is a
time-resolved Raman studies of LO phonons in GaN: Dependence on member of the Global Research Collaboration Science Technical Advisory
photoexcited carrier density,” Appl. Phys. Lett., vol. 89, no. 11, p. 112 111, Board for Modeling and Simulation. He is the author or a coauthor of over
Sep. 2006. 25 external journal and conference publications. He is the holder of 12 patents.
[33] M. Bouya, N. Malbert, N. Labat, D. Carisetti, P. Perdu, J.C. Clément, Dr. Hartin has been awarded the Distinguished Innovator designation by
B. Lambert, and M. Bonnet, “Analysis of traps effect on AlGaN/GaN Freescale.
HEMT by luminescence techniques,” Microelectron. Reliab., vol. 48,
no. 8/9, pp. 1366–1369, Aug.–Sep. 2008.

Ashwin Ashok received the B.E. degree in electrical Stephen M. Goodnick (M’88–SM’91–F’04) re-
engineering from the University of Madras, Chennai, ceived the B.S. degree in engineering science from
India, in 2001 and the M.S. and Ph.D. degrees in Trinity University, San Antonio, TX, in 1977 and
electrical engineering from Arizona State University, the M.S. and Ph.D. degrees in electrical engineering
Tempe, in 2005 and 2008, respectively. from Colorado State University, Fort Collins, in 1979
He is currently with Intel Corporation, Hillsboro, and 1983, respectively.
OR, as a Device Engineer. His research interests He was an Alexander von Humboldt Fellow
include semiconductor device physics and semicon- with the Technical University of Munich, Munich,
ductor device modeling with strong emphasis on Germany, and the University of Modena, Modena,
Monte Carlo particle-based device simulations. Italy, in 1985 and 1986, respectively. From 1986 to
Dr. Ashok was the recipient of the Best Presenta- 1997, he was a faculty member of the Department
tion Award at the Low Dimensional Structures and Devices Conference held in of Electrical and Computer Engineering, Oregon State University, Corvallis.
Cancun, December 2004. From 1996 to 2005, he was the Chair and a Professor of electrical engineering
with Arizona State University, Tempe, where he served as Deputy Dean for the
Ira A. Fulton School of Engineering from 2005 to 2006, served as Associate
Vice President for Research from 2006 to 2008, and is currently the Director
Dragica Vasileska (SM’06) received the B.S. of both the Arizona Institute for Nanoelectronics and the Arizona Institute for
(Diploma) and M.S. degrees in electrical engineering Renewable Energy. He is a coauthor of over 180 journal articles, books, and
from the Saints Cyril and Methodius University, book chapters related to transport in semiconductor devices and nanostructures.
Skopje, Macedonia, in 1985 and 1992, respectively, Dr. Goodnick was President of the Electrical Computer Engineering Depart-
and the Ph.D. degree from Arizona State Univer- ment Heads Association in 2003 and 2004, Co-Chair of the IEEE International
sity, Tempe, in 1995. Microwave Symposium in Phoenix, 2001, and Program Chair of the Fourth
From 1995 to 1997, she held a Faculty Research IEEE Conference on Nanotechnology in Munich in August 2004 and July 2009.
Associate position within the Center of Solid State
Electronics Research, Arizona State University. In
the fall of 1997, she joined the Department of Electri-
cal Engineering, Arizona State University. In 2002,
she was promoted to Associate Professor and, in 2007, to Full Professor. Her
research interests include semiconductor device physics and semiconductor
device modeling, with strong emphasis on quantum transport and Monte Carlo
particle-based simulations. She is the author of more than 140 published journal
publications and over 80 conference proceedings refereed papers and is a
coauthor of a book on computational electronics with Prof. S. M. Goodnick.
She has also given numerous invited talks.
Dr. Vasileska is a Senior Member of the American Physical Society. She
has received many awards, including the Best Student Award from the School
of Electrical Engineering in Skopje since its existence (1985, 1990). She is
also a recipient of the 1998 National Science Foundation CAREER Award.
Her students have won the Best Paper and the Best Poster Award at the
2004 International Conference on Low Dimensional Structures and Devices in
Cancun.