You are on page 1of 16

International Journal of High Speed Electronics and Systems

Vol. 18, No. 4 (2008) 825–840


 World Scientific Publishing Company

AN EFFICIENT NUMERICAL METHOD OF DC MODELING FOR POWER


MOSFET, MESFET AND AlGaN/GaN HEMT

TOUHIDUR RAHMAN, MOHAMMAD A. HUQUE, SYED K. ISLAM


Department of Electrical Engineering and Computer Science, The University of Tennessee
Knoxville, TN 37996-2100, USA
trahman@utk.edu
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

In this paper, an efficient numerical model applicable for wide varieties of long channel field-effect
transistors (MOSFET, MESFET, HEMT, etc.) is developed. A set of available data is used to
calculate the model parameters and another set of data is used to verify the accuracy of the model.
This model provides a single expression that is applicable for the entire range of device biasing and
can predict the output parameters with less than 1% error compared to the experimental results.
Lagrange polynomial, the highest degree of polynomial for any given set of data, is used to derive
the model from available data. This method is efficient in the sense that it can be derived from a
limited number of experimental data and since it uses only one equation for entire range of the
device operation hence its computational cost is also small.

Keywords: Device modeling; Numerical modeling; Lagrange polynomial.

1. Introduction
Analytical models are very important in understanding the physical operation of the
semiconductor devices and optimizing their structures for specific applications. A
number of physical phenomena, such as, low-field mobility, carrier velocity saturation,
recombination-generation, charge trapping, hot carrier effect etc. dictate the
semiconductor device characteristics. These physical phenomena are highly nonlinear in
nature and working with these complex functions is not an easy task. A closed form of
analytical expression, even in piecewise form becomes almost impossible without a
number of approximations. Moreover, computation with complex expressions is also a
time consuming task. Also these models are not fast enough to be used in most circuit
simulators.
Numerical models offer an alternative to physics based analytical models for rapid
and accurate device modeling. Though this approach do not provide any physical insight;
however, these models serve as excellent tools for quick circuit simulation [1-9]. In
general this approach uses measured data to accurately reproduce the complex nonlinear
behavior of semiconductor devices. In most cases they are equally applicable to different
flavor of transistors (MOSFET, MESFET, HEMT, etc.) fabricated from different
technologies. A large number of efforts on numerical device modeling have been
reported since 1970’s [1]. Most commonly used methods involve the use of look-up table

67
826 T. Rahman, M. A. Huque & S. K. Islam

and quadratic and higher order polynomials. Authors in [8] proposed numerical MOSFET
modeling based on multidimensional Bernstein interpolation as a means to improve
simulation efficiency. Hermite polynomials based simple bicubic surface patch
generation was presented in [9] for the evaluation of the device operating point. But this
interpolation function may exhibit bumps though the original set of data might be
monotonic and concave/convex in either direction. In [10], triode region was modeled by
quadratic fits whereas linear fits were used for the saturation region. Discontinuities in
the conductance arise as the operating point shifts from triode region to saturation region
and vice versa. To reduce this discontinuity more data points are needed which increases
the experimental cost. Basic cubic spline based multidimensional interpolation techniques
are presented in [5]. The spline parameters are optimized for monotonicity preserved
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com

interpolation..
The method of look-up table is very simple to implement [2-4]. However, it requires
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

high density of data points and consumes significant amount of computational time in the
search process. It also depends on local approximation rather than global approximation
that restrict the look-up model to be used within a limited range of operation. Another
family of numerical approach uses quadratic or higher order polynomials for
interpolation [5-7]. These methods determine the coefficients of a predetermined
polynomial from the available data. Because of the use of predetermined polynomial
these interpolation methods suffer from large truncation errors. For some special
polynomials such as Bernstein [8], Hermite [9], etc., the interpolation method involves
determination of factorials and derivatives. Those models tend to slow down significantly
where large numbers of polynomial terms are used for interpolation.
Numerical modeling approach is inherently faster than analytical model development
by reducing the computational effort to evaluate the device equations. Technology
change in device fabrication (material and structure) can be more easily absorbed in
numerical modeling just by computing the new set of parameters from experimental data.
This is much faster than developing analytical model from the physical understanding of
all device properties. Another important feature of numerical modeling is that it can be
easily incorporated into a circuit simulator. Table method requires large memory
allocation to accommodate the tabular values for different parameters. On the other hand,
polynomial methods utilize functional forms but most of them are unable of incorporate
different device parameters.
Metal Semiconductor Field-Effect Transistor (MESFET) and High Electron Mobility
Transistor (HEMT) are relatively new devices compared to the MOSFETs. Although the
modeling effort of these devices is predominantly theoretical by nature, numerical
modeling attracts significant attention as well. Most of the numerical techniques
developed for these devices are adopted from those for MOSFETs [11-19]. These models
suffer from the same problems as their MOSFET counterparts.
Regarding numerical modeling approach for almost all cases, it is true that
computationally simpler models offer the same level of simulation accuracy of models
using splines of high order [8]. The accuracy of the simulation depends heavily on both

68
DC Modeling for Power MOSFET, MESFET & AlGaN/GaN HEMT 827

the choice and the implementation of numerical integration and nonlinear equation
solution algorithms adopted for the host simulator program. Therefore, the use of high
order and computationally intensive interpolants for table look-up models does not
necessarily improve overall simulation accuracy compared to simpler models. Hence it is
always desirable to have a model that is simple enough to be formulated and easy to be
implemented for any practical application and is tested for the accuracy with the
experimental data. In this paper, numerical models of long channel MOSFET, MESFET,
and HEMT are developed from available data set (either from standard analytical
expressions or from experiment). Traditionally power devices have relatively long
channel lengths compared to low power devices to achieve larger reverse blocking
voltage. Lagrange polynomial is used to derive numerical model from the available data
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com

set. Lagrange polynomial is the highest degree of polynomial for any given set of data.
As a result, it can incorporate the effects of different phenomena in the maximum
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

possible combinations. For this reason, Lagrange polynomial is chosen for the
development of this model. In addition, the developed model uses only one expression to
predict the device characteristics over the entire region of device biasing and for all
region of operation. This means, in this method transistor forward characteristics for all
gate biasing voltages and for both triode and saturation region operation only one
equation is needed. The integration of the developed model with a circuit simulator is
also discussed.

2. Mathematical Theory
Developing an expression to model a set of data involves fitting a polynomial to a set of
data points (x0 , y0), (x1, y1), …, (xm , ym). These points might be obtained from
experiments or through some complicated computations. If the value of y at another point
x is required, the entire experiment or the computation needs to be repeated. This is an
expensive process for acquisition of any data. In order to solve this problem, traditionally
an algebraic polynomial f(x) of degree m is constructed such that

yi = f (xi ), i = 0,1,..., m (1)

where, f(x) is called the interpolation polynomial and the points xi, i = 0, 1, …, m are
called the interpolation points.
Taylor series expansion is a way of approximating general functions by polynomials.
However, this has limited usefulness. Interpolation is a more practical way of
constructing polynomial approximations. There are many ways in which interpolating
polynomials can be found. The method of evaluation of undetermined coefficients is
simple and intuitive, and gives results with minimum effort [20]. But this method is not
always suitable, especially since the system tends to become ill-conditioned quickly as m
increases. This also does not give a very explicit form of the polynomial which makes it
difficult to use for analysis. Lagrange polynomial investigated by the mathematician
Joseph-Louis Lagrange (1736-1813) overcomes some of these limitations.

69
828 T. Rahman, M. A. Huque & S. K. Islam

Lagrange interpolating polynomial, denoted by P(x), is the unique polynomial of


degree m for which P(xi) = f(xi) for i = 0, 1, …, m. This can be expressed as [21]

P( x) = ∑ L (x) f (x )
i
i i (2)

where, (x0, x1, …, xm) are the interpolating or node points and the Lagrange coefficient,
Li(x) is given by,

Li ( x ) =
( x − x0 )( x − x1 )...( x − xi −1 )( x − xi + 1 )...( x − xn ) = (x − x j )

( xi − x0 )( xi − x1 )...( xi − xi −1 )( xi − xi + 1 )...( xi − xn ) j = 0, j ≠ i (xi − x j )
(3)

These coefficients have several properties that deserve attention [22]. Lagrange
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com

coefficients formed for the m + 1 points (x0, y0), (x1, y1), …, (xm, ym) is a polynomial of
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

degree m which vanishes at x = x0, ……, x = xi-1, x = xi-1, ……., x = xm, but at x = xi it
assumes the value of 1 (one). As a result, the error is zero if the interpolating point
coincides with the data point. The form of the Lagrange coefficient as given in Eq. (3)
shows that it depends only on the given x’s and are entirely independent of y’s. Lagrange
polynomial is invariant if the variable x is replaced by a new variable through the linear
transformation. Direct use of the coordinates and not using difference tables and factorial
polynomial makes the calculation of Lagrange polynomial less expensive.
Lagrange polynomial can be extended for two independent variables as follows:

 
P( x, y) = ∑ L ( x ) ∑ L ( y ) f (x , y )
i j i j (4)
i  j 
where, Li(x) and Lj(y) are expressed as in (3). It shows that Lagrange polynomial can be
written in a recursive fashion. For more than two variables Lagrange polynomial can be
expanded in the same manner as in (4). For example, the recursive formulation for three
independent variables can be written as:

  
P ( x, y, z ) = ∑ Li ( x ) ∑ L j ( y ) ∑ Lk (z ) f x i , y j , z k  
( ) (5)
  
i  j  k 

The modeling efforts start with a set of available data either from an experiment or from
some analytical models. A very important aspect of this model formulation approach is
that it does not depend on the source of data, rather it depends on the data only. For
illustrative purpose, MOSFET and MESFET data points are generated from some well-
established long channel devices’ analytical models. Experimental results are used to
develop and validate the HEMT device model.
The available data is used to form the Lagrange coefficients using Eq. (3). Thereafter,
the values of different independent variables at the interpolating point are inserted into
the equation of P(x) (Eq. (2) or (3) or (4) depending on the number of independent

70
DC Modeling for Power MOSFET, MESFET & AlGaN/GaN HEMT 829

variables) to calculate the value of the dependent variable at that point. The same steps
are followed to obtain the values of the dependent variable at all interpolation points.

3. Model Development and Verification Using Analytical Data


This section presents the model development and verification using the data generated by
the analytical expressions of the devices under consideration. Both MOSFETs and
MESFETs are used as representative devices of their classes.

3.1. MOSFET
The modeling procedure described above is now applied to develop a numerical model
for an n-MOSFET. For simplicity, it is assumed that the Early voltage of the device is
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com

very large and the channel length modulation can be neglected. The most important
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

variables of interest are drain-source biasing (VDS), gate-source voltage (VGS), threshold
voltage (VTH), and device dimension (W and L). From standard analytical expression of
long channel MOSFET it is clear that the drain current has a linear relation to device
aspect ratio (W/L). Therefore, the drain current can be normalized to W/L. The
normalized drain current can be written as:
 VGS ≤ VTH
0

I DS   V2 
'
I DS = =  µ n C ox (VGS − VTH )VDS − DS  VGS − VTH ≥ VDS (6)
W    2 
 
 L   µ nC ox
 (VGS − VTH )2 VGS − VTH ≤ VDS
 2
Once the normalized drain current is available it is very easy to determine the actual drain
current by multiplying the normalized drain current with W/L. Therefore, the remaining
variables of interest are VDS, VGS and VTH. The threshold voltage VTH remains constant
for a given process and for a constant temperature. As a result, the attention now can be
focused on VDS and VGS. Now a set of I'DS data will be generated for different values of
VDS and VGS with the Eq. (6). The following values are used for different parameters:

µn = 1000 cm2 / (V.s)


tox = 10-7 cm
Cox = 3.36452 µF/cm2
VTH = 0.7 V

Figure 1 shows the surface plot of I'DS as a function of VDS and VGS. The dots show
the data points evaluated from Eq. (6). There are very few data points and our purpose is
to develop a model from the limited number of data points that will predict I'DS for
different values of VDS and VGS not present in Figure 1. Lagrange coefficients, L(VGS) and
L(VDS) are calculated from these data points using Eq. (3). Then drain current, I’DS for

71
830 T. Rahman, M. A. Huque & S. K. Islam

unknown drain and gate biasing condition are calculated using the Lagrange interpolating
polynomial extended for two variables (Eq. 7).


'
I DS (
= P(VGS ,VDS ) = ∑ Li (VGS ) ∑ L j (VDS )I DS VGSi ,VDS j ) (7)
i  j 

Figure 2 shows I'DS as a function of VDS and VGS predicted by the developed model
with Lagrange polynomial from the available data. The dots represent the data points
generated by the model. Figure 2 shows that the model can predict the values of I'DS for
different values of VDS and VGS which are not available in the original data set.
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com

Now to check the accuracy of the developed model another set of data is generated
from Eq. (6). These new values from Eq. (6) are compared to the values predicted by the
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

model. Figure 3 shows the comparison between the results obtained from the developed
model to those generated from Eq. (6). It shows that the values obtained from the
numerical model match with those obtained from Eq. (6) very precisely.
This model can also be used to extract the threshold voltage of a device. Figure 4
shows I'DS as a function of VGS for a fixed VDS. It is well known that the intersection of
the extended line of the linear section of this curve and the potential axis is defined as the
threshold voltage. As mentioned earlier the threshold voltage of the n-MOSFET is 0.7 V.
The threshold voltage from Figure 4 is calculated to be 0.72 V, which is very close to the
actual value. It also provides a smooth variation of I'DS in the subthreshold region.

72
DC Modeling for Power MOSFET, MESFET & AlGaN/GaN HEMT 831
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

Figure 1: First set of data for an n-MOSFET used for developing the model.

Figure 2: I'DS of n-MOSFET as a function of VDS and VGS predicted by the developed model.

73
832 T. Rahman, M. A. Huque & S. K. Islam
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

Figure 3: Comparison between results obtained from the model for n-MOSFET (black circular markers) and Eq.
(7). It also shows the original data set used to develop the model.

Figure 4: I'DS of n-MOSFET as a function of VGS for a fixed VDS (a) model (black line with circular markers)
(b) Eq. (7). It also shows the predicted threshold voltage by the developed model.

74
DC Modeling for Power MOSFET, MESFET & AlGaN/GaN HEMT 833

One of the most important performance parameters of a device is the rate of


conversion of gate-to-source voltage to drain current. The transconductance of an n-
MOSFET in saturation is given by:
'
∂I DS
gm = = µ n C ox (VGS − VTH ) (8)
∂VGS

The transconductance of the developed model is obtained from the Lagrange expression
with respect to the variable representing the gate-to-source voltage. Figure 5 shows the
comparison of transconductance between the analytical model and the developed model.
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

Figure 5: Comparison of the transconductance of an n-MOSFET between the analytical model and the
developed model.

3.2. MESFET
The numerical model for a normally-on SOI-MESFET can be developed following the
same procedure as described for MOSFETs. The early voltage of MESFET is also
assumed to be very large and the channel length modulation is neglected. The normalized
drain current for the n-channel MESFET can be written as:

 VGS ≤ VTH
0
 (9)
I   2 2ε  3 3 

I '
DS = DS = qµ n N d V DS − 2 
(Vbi − VGS + V DS ) 2 − (Vbi − VGS ) 2   VGS − VTH ≥ V DS
W  
 3 qN d a   
L 
qµ N V − 2 2ε  3 
(V − VTH ) 2 − (Vbi − VGS )2 
3
VGS − VTH ≤ V DS
 n d  DS 3 qN d a 2  bi  
 

75
834 T. Rahman, M. A. Huque & S. K. Islam

where, VDS is the drain-to-source voltage, VGS is the gate-to-source voltage, VTH is the
device threshold voltage, W is the device channel width, L is the device channel length,
µn is the mobility of electron, Nd is the channel doping density, q is the charge of an
electron, a is the thickness of the channel, ε is the electric permittivity of the channel
material, Vbi is the built-in Schottky potential between gate metal and the channel
material. The following values are used for different parameters in Eq. (9):

µn = 1000 cm2 / (V.s)


a = 0.5 µm
Nd = 118 / cm3
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com

Using the above parameters the built-in potential, Vbi and the threshold voltage, VTH are
calculated to be 0.8564 V and -1.0579 V, respectively.
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

Figure 6 and 7 show the surface plot of I'DS for initial data and results obtained from
the model, respectively. As seen in Figure 7 the model predicts the values of I'DS, for
different values of VDS and VGS, which are not available in the original data set. The
model is also verified for n-channel MESFET using another set of data generated from
Eq. (9). Figure 8 shows the comparison between the results obtained from the developed
model to those generated from Eq. (9). Figure 8 shows I'DS as a function of VGS for a
fixed VDS. As mentioned earlier the threshold voltage of the n-channel MESFET is -
1.0579 V. The threshold voltage from Figure 9 is calculated to be -1.1 V, which is very
close to the analytical value.

76
DC Modeling for Power MOSFET, MESFET & AlGaN/GaN HEMT 835
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

Figure 6: First set of data for n-channel MESFET used for developing the model.

Figure 7: I'DS of n-channel MESFET as a function of VDS and VGS predicted by the developed model.

77
836 T. Rahman, M. A. Huque & S. K. Islam
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

Figure 8: Comparison between results obtained from the model for n-channel MESFET (black circular markers)
and Eq. (9). It also shows the original data set used to develop the model.

Figure 9: I'DS of n-channel MESFET as a function of VGS for a fixed VDS (a) model (black line with circular
markers) (b) Eq. (10). It also shows the predicted threshold voltage by the developed model.

78
DC Modeling for Power MOSFET, MESFET & AlGaN/GaN HEMT 837

4. Model Development and Verification Using Experimental Data


In this section, the proposed model is developed and validated using the experimental
results of an n-channel AlGaN/GaN power HEMT on SiC substrate. Several sets of
output characteristic (IDS vs. VDS) are measured for the power HEMT at various VGS
values. Thereafter a Lagrange polynomial is derived from the first set of experimental
output characteristics such that IDS is a function of two variables, namely VDS and VGS.
The same interpolation function is used to calculate IDS for different values of VDS and
VGS to obtain a set of calculated output characteristics from the numerical model. This
calculated set of IDS vs. VDS is then compared with a second set of experimental results as
shown in Figure 10. Figure 10 also shows the first set of experimental results for
clarification. The values obtained from the numerical model match the experimental
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com

results with an average relative error of less than 1%. The accuracy of the numerical
model is expected to increase with the inclusion of greater number of experimental data
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

sets in the derivation of the interpolation function.

Figure 10: Comparison between results obtained from the model for n-channel HEMT (black circular markers)
and experimental results. It also shows the original experimental data set used to develop the model.

The most attractive feature of a Lagrange polynomial is that it can adapt to any
number of variables without changing the basic structure. Therefore the model based on
this polynomial can also represent physical phenomena provided that experimental data is
available to develop the model expression.
The developed model expresses the characteristics of a device with a single
expression. This expression can include any number of variables that may affect the
device performance. Therefore, it is very simple to include a device model in the circuit
simulator. This model provides a single expression for the entire region of device

79
838 T. Rahman, M. A. Huque & S. K. Islam

operation. The coefficients of the expression are determined and then the expression is
inserted into the circuit simulator. User of the circuit simulator needs very small memory
to store the expression. The CPU time is also expected to be very small, as it has to
evaluate only one expression with several multiplication and addition operations.

5. Conclusion
The model developed in this paper is efficient in terms of derivation and computation and
always reliable as it is derived from experimental data. Although examples were given
for the two most influential parameters (namely VDS and VGS), it can easily be extended
for any number of variables such as device length and width, operating temperature etc.
provided that the experimental results are available. Polynomial based numerical models
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com

determine finite number of coefficients. Therefore, those models suffer from large
truncation error. Moreover, most of the polynomial models use different local
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

expressions for different regions of device operation. Our model which is based on
Lagrange polynomial provides a single expression that is applicable over the entire region
of operation and can be considered as a global function. This single equation
implementation will help to model devices relatively quickly and at low cost.

80
DC Modeling for Power MOSFET, MESFET & AlGaN/GaN HEMT 839

Reference
1. V. Bourenkov, K. G. McCarthy, and A. Mathewson, “MOS Table Models for Circuit
Simulation,” IEEE Trans. Computer-Aided Design Integr. Circuits Syst, vol. 24, no. 3, March
2005.
2. M. G. Graham and J. J. Paulos, “Interpolation of MOSFET table data in width, length, and
temperature,” IEEE Trans. Computer-Aided Design Integr. Circuits Syst., vol. 12, no. 12, pp.
1880–1884, Dec. 1993.
3. P. B. Meijer, “Fast and smooth highly nonlinear multidimensional table models for device
modeling,” IEEE Trans. Circuits Syst., vol. 37, no. 3, pp. 335–346, Mar. 1990.
4. A. Rofougaran and A. A. Abidi, “A table lookup FET model for accurate analog circuit
simulation,” IEEE Trans. Computer-Aided Design Int. Circuits Syst., vol. 12, no. 2, pp. 324–
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com

335, Feb. 1993.


5. J. A. Barby, J. Vlach, and K. Singhal, “Polynomial splines for MOSFET model
approximation,” IEEE Truns. Computer-Aided Design, vol. 7, no. 5, pp. 557-566, May 1988.
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

6. C. Turchetti and M. Conti, “General-Approach for Development of CAD-Oriented Analytical


Device Models,” IEE Proceedings-G Circuits Devices and Systems vol. 138 no. 6 pp. 637-
650, 1991.
7. Gerhard Schrom, Andreas Stach, Siegfried Selberherr, “An interpolation based MOSFET
model for low-voltage applications,” Microelectronics Journal vol. 29, pp. 529-534, 1998.
8. M. Yanilmaz, and V. Eveleigh, “Numerical Device Modeling for Electronic Circuit
Simulation,” IEEE Transaction on Computer-Aided Design, vol. 10, no. 3. March 1991.
9. T. Shima, H. Tamada, L. Ryo, and R. L. M. Dang, “Table look-up MOSFET modeling system
using a 2-D device simulator and monotonic piecewise cubic interpolation,” IEEE Trans.
Computer-Aided Design, vol. CAD-2, pp. 721-726, Apr. 1983.
10. B. J. Burns, “Empirical MOSFET models for circuit simulation,” Memo M84/43, Electron.
Res. Lab., Univ. California, Berkeley, 1984.
11. Davis, R.G.; Gaskell, J.M.; Ball, G.; Allenson, M.B., “The validation of nonlinear FET
modelling and circuit simulation.” Non-Linear Modelling of Microwave Devices and Circuits,
IEE Colloquium on 25 Jun 1990 Page(s):2/1 - 2/4.
12. Ouslimani, A.; Vernet, G.; Henaux, J.C.; Crozat, P.; Adde, R., “MESFET large-signal model
based on small-signal measurements for time-domain CAD.” Electronics Letters
Volume 24, Issue 15, 21 July 1988 Page(s):973 – 974
13. Ouslimani, A.; Vernet, G.; Crozat, P.; Adde, R., “Large-signal model of picosecond FETs and
measurement of the step response.” Microwave Theory and Techniques, IEEE Transactions on
Volume 37, Issue 9, Sept. 1989 Page(s):1460 – 1465
14. Kyoungmin Koh; Hyun-Min Park; Songcheol Hong, “A spline large-signal FET model based
on bias-dependent pulsed I-V measurement.” Microwave Theory and Techniques, IEEE
Transactions on Volume 50, Issue 11, Nov. 2002 Page(s):2598 – 2603
15. Siriex, D.; Noblanc, O.; Barataud, D.; Chartier, E.; Brylinski, C.; Quere, R., “A CAD-oriented
nonlinear model of SiC MESFET based on pulsed I(V), pulsed S-parameter measurements.”
Electron Devices, IEEE Transactions on Volume 46, Issue 3, March 1999 Page(s):580 – 584
16. Schreurs, D.; Baeyens, Y.; Van der Zanden, K.; Verspecht, J.; Van Hove, M.; De Raedt, W.;
Nauwelaers, B.; Van Rossum, M., “Large-signal HEMT modelling, specifically optimized for
InP based HEMTs.” Indium Phosphide and Related Materials, 1996. IPRM '96., Eighth
International Conference on 21-25 April 1996 Page(s):638 – 641

81
840 T. Rahman, M. A. Huque & S. K. Islam

17. Angelov, I.; Rorsman, N.; Stenarson, J.; Garcia, M.; Zirath, H.; “An empirical table-based
FET model,” Microwave Theory and Techniques, IEEE Transactions on
Volume 47, Issue 12, Dec. 1999 Page(s):2350 - 2357
18. Fernandez-Barciela, M.; Tasker, P.J.; Campos-Roca, Y.; Demmler, M.; Massler, H.; Sanchez,
E.; Curras-Francos, M.C.; Schlechtweg, M.; “A simplified broad-band large-signal nonquasi-
static table-based FET model,” Microwave Theory and Techniques, IEEE Transactions on
Volume 48, Issue 3, March 2000 Page(s):395 – 405
19. Fernandez-Barciela, M.; Tasker, P.J.; Campos-Roca, Y.; Demmler, M.; Massler, H.; Sanchez,
E.; Curras-Francos, M.C.; Schlechtweg, M.; “A simplified broad-band large-signal nonquasi-
static table-based FET model,” Microwave Theory and Techniques, IEEE Transactions on
Volume 48, Issue 3, March 2000 Page(s):395 - 405
20. Peter Linz and Richard L. C. Wang, “Exploring Numerical Methods: An Introduction to
Int. J. Hi. Spe. Ele. Syst. 2008.18:825-840. Downloaded from www.worldscientific.com

Scientific Computing using MATLAB”, Jones and Bartlett Publishers, Sudbury,


Massachusetts, ISBN 0-7637-1499-2, 2003.
by RENSSELAER POLYTECHNIC on 03/05/15. For personal use only.

21. Daniel Zwillinger, “CRC Standard Mathematical Tables and Formulae,” 30th edition, CRC
Press LLC, 2000, Boca Raton, Florida, ISBN 0-8493-2479-3.
22. W. E. Milne, Numerical Calculus, Princeton University Press, Princeton, New Jersey, 1949,
pp. 83.

82

You might also like