Professional Documents
Culture Documents
Solid-State Electronics
journal homepage: www.elsevier.com/locate/sse
a r t i c l e i n f o a b s t r a c t
Article history: Computational methods are considered for the time- and frequency-domain simulations of a microwave
Received 7 April 2014 power amplifier with a transistor built on a diamond-like semiconductor. The first method is based only
Received in revised form 6 August 2014 on a two-dimensional model of the intrinsic transistor that includes the quasi-hydrodynamic model of
Accepted 15 August 2014
electron transport and electric field equations. The second method, which is less accurate but much more
Available online 11 September 2014
computer-time-saving, is based on the intrinsic transistor large-signal lumped-element equivalent circuit
The review of this paper was arranged by model with the parameters and spline characteristics calculated by means of the abovementioned
Prof. S. Cristoloveanu two-dimensional model. Using these two methods, the microwave parameters of the amplifier are
calculated for frequencies varying from 15 to 90 GHz. When using the equivalent circuit at frequencies
Keywords: of up to 60 GHz, the calculation error (equal to the difference between the results of the time- and
Equivalent circuit frequency-domain simulations) does not exceed 5% for the input impedance and 0.6 dB for the transducer
FET power gain.
Frequency-domain simulation Ó 2014 Elsevier Ltd. All rights reserved.
Quasi-hydrodynamic model
Spline characteristics
Time-domain simulation
Currently, field effect transistors (FETs) are widely used as the using only the 2D model of the IFET that includes the quasi-
active elements of power amplifiers in the frequency range from hydrodynamic model of electron transport and electric field
several hundred megahertz to several hundred gigahertz [1–3]. equations and
Therefore, the adequacy of the intrinsic FET (IFET) large-signal combining the 2D model with the IFET equivalent circuit.
lumped-element equivalent circuit model (Fig. 1) used for
designing these amplifiers is becoming a key issue. 2. 2D model of the intrinsic transistor
In [4–6], we developed a method for calculating the charac-
teristics and parameters of the IFET equivalent circuit that is Under the operating conditions of the transistor, we can neglect
based on the cubic spline construction [7] and the two- the impact ionization of semiconductor molecules by electrons and
dimensional (2D) quasi-hydrodynamic model of electron trans- holes. Then, the 2D quasi-hydrodynamic model of electron
port [8–11]. Below, we will calculate a number of microwave transport takes the following form:
parameters of a power amplifier based on a transistor that @n
‘‘is made’’ on an n-type semiconductor, in which the field ¼ r~jn ; ð1Þ
@t
dependence of the drift velocity of electrons is close to the
~jn ¼ l !
field dependence of the drift velocity of holes in boron-doped n E n þ rðDn nÞ; ð2Þ
diamond [1,12–14]. The physical parameters of the semicon-
@ðnuÞ ! u u0
ductor and insulator (Fig. 2) are close to the corresponding ¼ r~jnu þ q~jn E n ; ð3Þ
constants for diamond. @t sn
~jnu ¼ n l !E nu þ rðn Dn nuÞ; ð4Þ
n n n
http://dx.doi.org/10.1016/j.sse.2014.08.008
0038-1101/Ó 2014 Elsevier Ltd. All rights reserved.
116 G.Z. Garber / Solid-State Electronics 103 (2015) 115–121
Fig. 1. Large-signal equivalent circuit of the IFET without impact ionization of The field dependence of the electron drift velocity appearing in
semiconductor molecules by electrons and holes: ich is the channel current
(7) is given by
generator; Cgs and Rgs, the input capacitance and resistance; and Cgd, the transfer
(feedback) capacitance. V n ¼ v e ½1 expðle Eus =v e Þ; ð11Þ
3 2 15 3
where le = 2 10 cm /(V s) at Nd = 10 cm and 500 cm /(V s) at 2
@ @
temperature; r ¼ ; is the gradient operator; u0 = 3kT0/2 is the
@x @y
Nd = 1018 cm3 and ve = 1.1 107 cm/s. The field dependence of the
mean kinetic energy of electrons in thermodynamic equilibrium; T0 energy relaxation time in (8) has the following form: function
is the lattice temperature (in our case, T0 is set equal to 300 K); Tn(x, y, Eus) is equal to vese/Vn(x, y, Eus) for Eus > Ee and to vese/
n(x, y, t), ln(x, y, u), Dn(x, y, u) = (kTe/q)ln, sn(x, y, u), and Vn(x, y, Ee) for Eus 6 Ee, where se = 1 ps and Ee = 3ve/le. With function
nn(x, y, u) are the concentration, mobility, diffusion coefficient, Tn(x, y, Eus) taken in such a form, the abovementioned solution to
energy relaxation time, and relative Peltier coefficient of electrons, (10) at u > u0 is unique. The field dependence of the relative Peltier
respectively; q is the positive elementary charge; ~ jn ðx; y; tÞ and coefficient in (9) is given by
~jnu ðx; y; tÞ are the electron flux density and electron energy flux 2
! ! Kn ¼ expðEus =Ee Þ þ 1:
density, respectively; E ðx; y; tÞ is the electric field intensity; ln E 3
!
is the electron drift velocity; and E ¼ j E j. System (1)–(6) is considered in the transistor active region
!
Electric field intensity E ðx; y; tÞ is calculated using the electric depicted in Fig. 2. Boundary conditions are the following:
field equations
! n = Nd and Te = T0 at the source and drain contacts;
!
rðe E Þ ¼ Q ; ð5Þ the normal component of ~ jn , ~
jnu , and E is equal to zero at the
contact-free semiconductor parts of the boundary;
! !
E ¼ ru; ð6Þ the normal component of E equals zero at the contact-free
insulator parts;
where u(x, y, t) is the electric potential and e(x, y) is the absolute the electric potentials of the source, gate, and drain contacts (us,
permittivity of the semiconductor or insulator; in our case, the ug, and ud, respectively) are given constants (when we calcu-
relative permittivity is equal to 5.7 for both the semiconductor late the characteristics and parameters of the IFET equivalent
and insulator. Charge density Q is equal to zero in the insulator circuit) or are determined by the extrinsic electrical circuit
and to q(Nd n) in the semiconductor, where Nd(x, y) is the concen- (when we simulate the transistor amplifier without using the
tration of ionized donors. IFET equivalent circuit).
The kinetic coefficients for electrons are calculated by the
formulas At the insulator–semiconductor interface, we require that only
ln ðx; y; uÞ ¼ V n ½x; y; Fðx; y; uÞ=Fðx; y; uÞ; ð7Þ the normal component of ~ jn and ~
jnu be zero (because the permittiv-
ity of the semiconductor and insulator is one and the same).
sn ðx; y; uÞ ¼ T n ½x; y; Fðx; y; uÞ; ð8Þ In addition to the formulated boundary conditions, initial
conditions are necessary. They are arbitrary or are determined by
nn ðx; y; uÞ ¼ K n ½x; y; Fðx; y; uÞ; ð9Þ the 2D distributions of u, n, and Te, which result from the previous
solution of (1)–(6).
where Vn(x, y, Eus), Tn(x, y, Eus), and Kn(x, y, Eus) are the given field
For solving the above initial-boundary value problem, we solve
dependences of the drift velocity, energy relaxation time, and rela-
at each time step
tive Peltier coefficient, respectively, and Eus is the intensity of the
uniform stationary electric field.
Eqs. (5) and (6) by the method of matrix elimination [15] and
When finding the field dependences, we assume that the semi-
then
conductor is homogeneous and its chemical composition is the
Eqs. (1) and (2) and, concurrently, (3) and (4) using the locally
same as at a point with coordinates x and y (in the semiconducting
one-dimensional scheme [15] with jnx, jny, jnux, and jnuy
part of the transistor active region depicted in Fig. 2). Quantity
expressed as the electron current density in [16].
Vn(x, y, Eus) is the drift velocity of electrons in such a semiconductor
subjected to uniform stationary electric field Eus. Quantities
Then we calculate the source, gate, and drain currents:
Tn(x, y, Eus) and Kn(x, y, Eus) have the same meaning. ZZ
@
Function F(x, y, u) involved in (7)–(9) is zero at u 6 u0 and is a is ¼ q jny þ e Ey dxdz; ð12Þ
solution to the energy balance equation, source @t
ZZ
qFV n ðx; y; FÞT n ðx; y; FÞ ¼ u u0 ; ð10Þ @
ig ¼ e Ey dxdz; ð13Þ
gate @t
at u > u0. This equation follows from (2) and (3) at r = o/ot = 0.
G.Z. Garber / Solid-State Electronics 103 (2015) 115–121 117
ZZ
@
id ¼ q jny e Ey dxdz; ð14Þ Fig. 4. Output current–voltage characteristics (continuous lines) and 90-GHz cycles
drain @t calculated without using (dashed cycle) and using (dash-and-dot cycle) the IFET
equivalent circuit.
where z is the third axis directed perpendicularly to the plane of
Fig. 2. In our case, the gate width along the z axis is equal to 1 mm.
the source, gate, and drain currents – is(t), ig(t), and id(t);
the electrostatic induction vector flux to the gate – qg(t) = The dependence of qg(1) = Qg on Ugs and Uds written as
RR
gate e Ey dxdz (it has the dimension of charge). Qg(Ugs, Uds) determines the input and transfer capacitances
according to the formulas Cgs(Ugs, Uds) = oQg/oUgs oQg/oUds and
The dynamic transconductance and dynamic conductance Cgd(Ugs, Uds) = oQg/oUds, respectively.
describing the channel current generator (ich in Fig. 1) are calcu- The resistance in the input branch is given by Rgs = sgs/Cgs, where
lated as sgs is the time constant of the resistor–capacitor network. The cur-
rent in this branch is calculated by the formula based on the
sinðxsÞ
SðxÞ ¼ expðjxsÞS0 ; ð15Þ assumption that sgs is small in comparison with the oscillation
xs period:
sinðxsÞ
GðxÞ ¼ expðjxsÞG0 ; ð16Þ dcgs w0 dw0
xs iin ðtÞ ¼ w0 ðtÞ þ sgs : ð17Þ
dt cgs dt
where S0 = oId/oUgs and G0 = oId/oUds are the static transconductance
and static conductance; Id(Ugs, Uds) is the dependence of id(1) = Id Here, cgs(t) = Cgs[ugs(t), uds(t)] and w0(t) = cgs(t) dugs/dt. The deriva-
on Ugs and Uds; x is the circular frequency; j is the imaginary unit; tion of formula (17) is given in Appendix A.
s = sch/2; and sch is the time of electron transit through the chan- The current in the transfer branch is given by
nel’s nonohmic part. Note that Id = Is = is(1) since the gate current
dugd
at t = 1 is absent. itr ðtÞ ¼ cgd ðtÞ ;
Formulas (15) and (16) taking into account the electron transit dt
through the channel’s nonohmic part correspond to the following where cgd(t) = Cgd[ugs(t), uds(t)] and ugd(t) = ug(t) + ub ud(t).
time-domain expression for the channel current: We use the cubic spline construction [7] to analytically con-
Z t tinue the discrete dependences found by solving (1)–(6) at the grid
1 0
ich ðtÞ ¼ Id ½ugs ðt 0 Þ; uds ðt0 Þdt ; nodes on the Ugs Uds plane depicted in Fig. 3. Therefore, the
sch tsch
dependences Id(Ugs, Uds) and Qg(Ugs, Uds) may be called the spline
where ugs(t) = ug(t) + ub us(t) and uds(t) = ud(t) us(t). characteristics of the IFET equivalent circuit.
118 G.Z. Garber / Solid-State Electronics 103 (2015) 115–121
are required parameters sgs and sch for the node (4.5; 10 V):
sgs = 0.788 ps and sch = 1.229 ps. They are determined using the
TIFET macro for Microsoft Office Excel.
Ld(ud, ug, us), Lg(ud, ug, us), and Ls(ud, ug, us) are, respectively,
currents at points d, g, and s determined by a linear active three
port (LATP), part of the circuit in Fig. 7 that is outside the IFET
Fig. 6. Time dependences of the (a) gate and (b) drain currents (continuous lines)
unit;
and their exponential approximations (dashed lines).
Md(ud us, ug us), Mg(ud us, ug us), and Ms(ud us,
ug us) = Md + Mg are currents at points d, g, and s determined
by the IFET unit.
A method for calculating parameters sgs and sch of the equiva-
lent circuit will be considered below. When simulating the amplifier in the time domain, the solution
We developed a version of the DIHEMT program [10] to solve of the initial-boundary value problem for (1)–(6) lasts until
the initial-boundary value problem for (1)–(6) at the grid nodes. periodic oscillations (with period T) set in. At each time step, the
The continuous lines in Figs. 4 and 5 show the calculated character- quadratic discrepancy
istics Id(Ugs, Uds) and Qg(Ugs, Uds), respectively.
One execution of the DIHEMT code makes it possible to simu- Fðud ; ug ; us Þ ¼ ½Md ðud us ; ug us Þ Ld ðud ; ug ; us Þ2
late all nodes of one vertical line on the Ugs Uds plane (Fig. 3):
i = 0, 1, ..., m; j = const. Under the assumption that Cgd = 0, parame- þ ½Mg ðud us ; ug us Þ Lg ðud ; ug ; us Þ2
ters sgs and sch can be found from the time dependences of gate þ ½Ms ðud us ; ug us Þ Ls ðud ; ug ; us Þ2 ð18Þ
current ig(t) and drain current id(t) as follows.
In Figs. 6a and b, continuous lines are the ig(t) and id(t) depen- of the IFET drain, gate, and source currents is minimized, with Ms
dences calculated in going from the node (5; 10 V) to the adjacent (ud us, ug us), Mg(ud us, ug us), and Md(ud us, ug us)
node (4.5; 10 V) and dashed lines (calculated by the least-squares calculated by (12)–(14). The electric potentials at points d, g, and
method) represent the exponential approximations of ig(t) and s (relative to ground, Fig. 7) are adjustable parameters.
id(t). The time constants of these exponential curves are equal to We use the Powell method [17] to minimize (nearly to zero)
0.788 and 1.229 ps, respectively. According to [6], these values function (18) at each time step. Note that not only functions Ld
Fig. 7. Electrical circuit of the model amplifier: resistors of 0.3 and 2 X are the parasitic elements of the transistor.
G.Z. Garber / Solid-State Electronics 103 (2015) 115–121 119
Table 1
Calculation of the transistor input impedance.
Frequency Power available from the Simulation Input voltage Real part of the Imaginary part of the Accuracy of input
1/T (GHz) RF generator (mW) approach standing wave ratio input impedance (O) input impedance (O) impedance calculation (%)
15 4.9 QHM only 19.2 3.436 28.2
QHM & EC 17.45 3.864 29.44 4.6
QHM & EC0 29.34 2.301 29.57 6.3
30 15.6 QHM only 15.12 3.578 14.28
QHM & EC 14.29 3.794 14.49 2
QHM & EC0 24.04 2.261 14.75 9.5
60 62.5 QHM only 13.48 3.784 7.078
QHM & EC 13.76 3.702 6.872 2.8
QHM & EC0 22.74 2.247 7.384 19.5
90 160 QHM only 12.65 3.988 4.684
QHM & EC 14.07 3.579 4.225 10
QHM & EC0 22.47 2.247 4.932 28.6
Table 2
Calculation of the transducer power gain of the amplifier.
Frequency Power available from the L C Simulation Input voltage Output RF Transducer Accuracy of transducer power
1/T (GHz) RF generator (mW) (nH) (pF) approach standing wave ratio power (mW) power gain (dB) gain calculation (dB)
15 4.9 0.484 0.693 QHM only 1.202 658.3 21.28
QHM & EC 1.313 673.5 21.38 0.1
QHM & EC0 2.339 757.4 21.89 0.61
30 15.6 0.156 0.379 QHM only 1.065 555 15.5
QHM & EC 1.066 569.4 15.62 0.12
QHM & EC0 1.578 729.4 16.69 1.19
60 62.5 0.0572 0.191 QHM only 1.079 498.2 9.015
QHM & EC 1.051 568.7 9.59 0.57
QHM & EC0 1.636 718.6 10.61 1.6
90 160 0.0341 0.125 QHM only 1.158 464.4 4.628
QHM & EC 1.145 631.3 5.961 1.33
QHM & EC0 1.736 729 6.586 1.96
(ud, ug, us), Lg(ud, ug, us), and Ls(ud, ug, us) but also functions Md of the IFET equivalent circuit. When analyzing this equivalent cir-
(ud us, ug us), Mg(ud us, ug us), and Ms(ud us, ug us) cuit, the direct and inverse discrete Fourier transformations [18]
are linear. This assertion follows from the fact that, during minimi- are performed. The Visual C++ program HIFETA (Harmonics In
zation, we solve (5) with charge density Q taken from the previous FET Amplifier) implements the frequency-domain simulation of
time step. Since Ld(ud, ug, us), Lg(ud, ug, us), Ls(ud, ug, us), Md the model amplifier.
(ud us, ug us), Mg(ud us, ug us), and Ms(ud us, ug us) The calculated periodic time dependences of the currents and
are linear, function (18) is quadratic, which facilitates the search potentials (in both the time and frequency domains) govern the
for its minimum point. We developed a version of the DIHEMT microwave parameters of the amplifier, some of which are given
program [10] for the time-domain simulation of the amplifier. in Tables 1 and 2.
The frequency-domain simulation, which is much more effi- Below, QHM stands for 2D model (1)–(6) and EC stands for the
cient in terms of CPU time saving, is based on the IFET equivalent IFET equivalent circuit. The simulation frequencies are 1/T = 15, 30,
circuit depicted in Fig. 1. We use the harmonic balance method (in 60, and 90 GHz.
our case, for five harmonics of the input RF generator frequency,
1/T) to obtain the periodic oscillation mode. To this end, the system 5. Estimation of the accuracy of the transistor equivalent circuit
of algebraic equations using the input impedance
M d ðud us ; ug us Þ ¼ Ld ðud ; ug ; us Þ; ð19Þ
First, we will consider the model amplifier (Fig. 7) with L = C = 0,
M g ðud us ; ug us Þ ¼ Lg ðud ; ug ; us Þ; ð20Þ i.e., with the unmatched impedances at the input. In this case, the
input impedance of the amplifier can be assigned to the transistor.
M s ðud us ; ug us Þ ¼ Ls ðud ; ug ; us Þ ð21Þ The frequency of the electromotive force of the input RF gener-
ator is given in the first column of Table 1. The amplitude of the
is solved, where Md, Mg, Ms, Ld, Lg, and Ls are the complex amplitudes electromotive force is set such that the amplitude of id(t) in the
of currents at points d, g, and s determined by the IFET and LATP and periodic oscillation mode is close to 80 mA for each of the four fre-
ud, ug, and us are the complex amplitudes of potentials at points d, quencies. The corresponding values of the maximum power of the
g, and s. Here, functions Md, Mg, and Ms are nonlinear. RF generator are listed in the second column.
The system of nonlinear algebraic Eqs. (19)–(21) is solved by The values of the microwave parameters found by the
the Newton method with the Jacobian matrix including the time-domain simulation (without using the IFET equivalent cir-
derivatives of the cubic splines that describe the characteristics cuit) are given in rows ‘‘QHM only’’. The values of these parameters
120 G.Z. Garber / Solid-State Electronics 103 (2015) 115–121
6. Estimation of the accuracy of the transistor equivalent circuit where cgs(t) = Cgs[ugs(t), uds(t)]; sgs = Rgscgs is the time constant of the
using the transducer power gain resistor–capacitor network (Rgs is a function of t). Differentiating
both sides of this equation with respect to time, we have
The first two columns in Table 2 are the same as in Table 1. The dcgs dugs dcgs dugc diin
third and fourth columns show the values of L and C providing the ugs þ cgs ¼ ugc þ cgs þ sgs
dt dt dt dt dt
complex conjugate matching of the 50-ohm impedance of the RF
generator with the transistor input impedance given in rows or
‘‘QHM only’’ of Table 1. dcgs dugs dugc diin
As above, rows ‘‘QHM & EC’’ were obtained at sgs = 0.794 ps and ucs þ cgs ¼ cgs þ sgs : ð22Þ
dt dt dt dt
sch = 1.283 ps and rows ‘‘QHM & EC0’’ correspond to the assump-
tion sgs = sch = 0. In Figs. 4 and 5, dashed and dash-dotted lines Since
show cycles calculated at a frequency of 90 GHz, L = 0.0341 nH, dcgs dcgs Rgs dcgs 1
C = 0.125 pF, sgs = 0.794 ps, and sch = 1.283 ps. ucs ¼ sgs iin ¼ sgs iin
dt dt sgs dt cgs
The last column shows errors in the transducer power gain cal-
culated using the IFET equivalent circuit. Gain values listed in rows and
‘‘QHM only’’ are considered as reference ones.
dugc
cgs ¼ iin ;
dt
Eq. (22) takes the form
7. Conclusion
diin dcgs 1 dugs
According to our results, the use of the IFET large-signal equiv-
sgs þ 1 sgs iin ¼ cgs : ð23Þ
dt dt cgs dt
alent circuit (with parameters sgs = 0.794 ps and sch = 1.283 ps)
instead of the 2D model at frequencies of up to 60 GHz introduces Under the assumption that sgs is small compared with oscilla-
an error not exceeding tion period T, we represent a solution to (23) in the form [19]
sgs
iin ðtÞ ¼ w0 ðtÞ þ w1 ðtÞ; ð24Þ
5% for the input impedance (Table 1) and T
0.6 dB for the transducer power gain (Table 2).
where sgs/T is a small quantity. Substituting (24) into (23) yields the
following expressions for terms of the zero and first orders of
The computer time expenses are as follows.
smallness:
For calculating the values of each row ‘‘QHM only’’ (in Tables 1 dugs
and 2), four oscillation periods were simulated. The calculation
w0 ðtÞ ¼ cgs ;
dt
of the content of all eight rows ‘‘QHM only’’ took about 20 h of
the 3-GHz processor, Intel Core 2 Quad CPU Q9650. sgs dcgs w0 dw0
w1 ðtÞ ¼ sgs sgs :
For calculating characteristics Id(Ugs, Uds) and Qg(Ugs, Uds), the T dt cgs dt
initial-boundary value problem for (1)–(6) was solved at
11 17 grid nodes (Fig. 3). It took about 24 h of the abovemen- Thus, (24) gives expression (17) for the current in the input branch
tioned processor. of the IFET equivalent circuit.
For calculating the values of rows ‘‘QHM & EC’’ and ‘‘QHM &
EC0’’, the HIFETA code was executed 16 times. It took less than References
4 s.
[1] Kasu M, Ueda K, Kageshima H, Yamauchi Y. RF equivalent-circuit analysis of p-
type diamond field-effect transistors with hydrogen surface termination. IEICE
This article should be considered as a technique for estimating Trans Electron July 2008;E91-C:1042–9.
the accuracy of the microwave FET equivalent circuit in compari- [2] Chung JW, Hoke WE, Chumbes EM, Palacios T. AlGaN/GaN HEMT with 300-GHz
fmax. IEEE Electron Dev Lett March 2010;EDL-31:195–7.
son with the 2D model, which is higher than the equivalent circuit [3] Radisic V, Deal WR, Leong KMKH, Mei XB, Yoshida W, Liu P-H, et al. A 10-mW
in the IFET model hierarchy. We use this technique for verifying submillimeter-wave solid-state power-amplifier module. IEEE Trans
the equivalent circuit of real transistors. To this end, the model Microwave Theory Tech July 2010;MTT-58:1903–9.
[4] Garber GZ. Analysis of nonlinear microwave circuits based on Schottky-gate
amplifier (Fig. 7) is complemented by the substrate conductance
field effect transistors (in Russian). Electron Eng Ser 2 1984;5:27–30.
(between points d and s) and output matching network. [5] Garber GZ. Numerical modeling of the characteristics of nonlinear equivalent
The HIFETA program (which is efficient in terms of CPU time circuits for s.h.f. Schottky-gate GaAs field-effect transistors. Soviet
saving) is used for optimizing power amplifiers, in particular, based Microelectron March 1991;22:202–7.
[6] Garber GZ. Method for calculating a small-signal equivalent circuit of
on the boron-doped diamond structure similar to the depicted in extremely high frequency heterostructural field-effect transistors. J Commun
Fig. 2. Technol Electron July 2005;50:822–5.
G.Z. Garber / Solid-State Electronics 103 (2015) 115–121 121
[7] Ahlberg JH, Nilson EN, Walsh JL. The Theory of Splines and Their [14] Garber GZ, Dorofeev AA, Zubkov AM, Ivanov KA, Kolkovsky YuV. Design of the
Applications. New York and London: Academic Press; 1967 [ch. 2]. SHF power FETs on diamond base with assistance of computer simulation (in
[8] Stratton R. Diffusion of hot and cold electrons in semiconductor barriers. Phys Russian). Electron Eng Ser 2 2013;1:9–16.
Rev June 1962;126:2002–14. [15] Samarskii AA. The Theory of Difference Schemes. New York: Marcel Dekker;
[9] Garber GZ. Modeling of operation of field effect transistors with submicron 2001 [ch. 9,10].
Schottky gate on GaAs (in Russian). Electron Eng Ser 2 1985;2:103–7. [16] Scharfetter DL, Gummel HK. Large-signal analysis of a silicon Read diode
[10] Garber GZ. Quasi-hydrodynamic modeling of heterostructure field-effect oscillator. IEEE Trans Electron Dev Jan 1969;ED-16:64–77.
transistors. J Commun Technol Electron Jan 2003;48:114–7. [17] Powell MJD. An efficient method for finding the minimum of a function of
[11] Garber GZ. Model for simulation of AlGaAs–GaAs power heterostructure FETs. several variables without calculating derivatives. Comput J 1964;7:155–62.
In: Proc IEEE Int Conf EUROCON, Belgrade; 2005. p. 867–70. http://comjnl.oxfordjournals.org/content/7/2/155.full.pdf+html.
[12] Reggiani L, Bosi S, Canali C, Nava F, Kozlov SF. Hole-drift velocity in natural [18] Hamming RW. Numerical Methods for Scientists and Engineers. New
diamond. Phys Rev B March 1981;23:3050–7. York: McGraw-Hill; 1962 [ch. 6].
[13] Volpe P-N, Pernot J, Muret P, Omnes F. High hole mobility in boron doped [19] Cole JD. Perturbation Methods in Applied Mathematics. Waltham,
diamond for power device applications. Appl Phys Lett March MA: Blaisdell; 1968 [ch. 2].
2009;94:092102-1–2-3. http://dx.doi.org/10.1063/1.3086397.