ELECTRONICS’ 2006 20 – 22 September, Sozopol, BULGARIA
MOSFET SIMULATION USING MATLAB IMPLEMENTATION OF THE
EKV MODEL
George Vasilev Angelov, Katya Konstantinova Asparuhova
FETT, Technical University of Sofia, 8 Kliment Ohridski Str.,
1797 Sofia, BULGARIA, phone: +35929652570
gva@[Link], k_asparuhova@[Link]
The paper presents an implementation of EKV MOST model in Matlab environment.
Parameter extraction and fitting procedures are briefly [Link] source modules for
transistor characteristics and additional functions for parameter extraction and fitting are
included in a Matlab toolbox. Model parameters are user defined to provide flexibility of
design [Link] accuracy of simulation results is ensured by using Labert’s W
interpolation function in the single expression for the drain current. Validation against
conventional BSIM3v3 model for a 0.35 μm CMOS technology is performed.
Keywords: MOS transistor, open source model, EKV model, Matlab.
1. INTRODUCTION
Matlab [1] is a general-purpose mathematical computation package widely used
in engineering, science and mathematics. It is quite useful for extraction and
optimization of model parameters for device and circuit characterization and
simulation. Accurate device modeling is at the core of correct design and analysis of
electronic circuits especially analog ones.
The high speed of simulation within Matlab is essential to the optimization
process. It is clear that the combined circuit and device optimization is capable of
leading to real benefits in both device design and circuit use. Furthermore, the model
codes are open source, providing direct access to model equations and parameters.
Thus, parameters can be easily extracted and characteristics optimized for circuit
design purposes.
In this context EKV MOST model is particularly suitable for implementation in
Matlab as it provides accurate modeling with a small set of parameters. Some of these
device parameters are also very useful in circuit calculations [2].
Although the majority of EKV parameters carry similar meanings to widely used
SPICE ones and therefore could, as a first approximation, be used for design and
simulation, better fit to experimental data can be obtained with parameters extracted
for the model.
2. EKV MODEL FORMULATION
EKV model is described in detail in [4] and concise set of equations and
parameters for extraction purposes are given in [5]-[7].
167
ELECTRONICS’ 2006 20 – 22 September, Sozopol, BULGARIA
2.1. Static Model
In the EKV model the gate, source and drain voltages, VG, VS and VD, are referred
to the substrate in order to preserve the intrinsic symmetry of the device. Besides
EKV proceeds from the pinch-off voltage VP (rather than the threshold voltage VTh).
A constant current bias is used to measure the pinch-off voltage versus gate voltage
characteristic in moderate inversion. This measure allows for an efficient and simple
characterization method to be formulated for the most important model parameters as
the threshold voltage, parameters accounting for mobility effects, short-channel
effects, etc. The same principle applied for various geometries, including short- and
narrow-channel devices, and forms the major part of the complete characterization
methodology.
The EKV MOST model is hierarchically structured allowing both hand
calculations and computer simulations including second order effects. The designer
handles a concise parameter set (13 intrinsic parameters) to account for first and
second order effects (see Tab.1): 10 physical parameters (COX, VTO, GAMMA,
PHI, KP, THETA, UCRIT, XJ, DW, DL) and 3 second-order fitting coefficients
(LAMBDA, WETA, LETA). equations of interest are presented below [8].
Table 1. Selected EKV parameters accounting for first and second order effects and their
default values.
Designation Description
COX 7E-4.[F/m] Gate oxide capacitance
VTO 0.5 [V] Nominal threshold voltage
GAMMA 1 [V1/2] Body effect factor
PHI 0.7 [V] Bulk Fermi potential (2x)
KP 5E-5 [A/V2] Transconductance parameter
THETA 0 [1/V] Mobility reduction coefficient
UCRIT 2E6 [V/m] Longitudinal critical field
XJ 1E-7 [m] Junction depth
DL 0 [m] Channel length correction
DW 0 [m] Channel width correction
LAMBDA 0.5 [–] Depletion length coefficient
LETA 0.1 [–] Short channel effect coefficient
WETA 0.25 [–] Narrow channel effect coefficient
2.2. Pinch-off voltage
The pinch-off voltage VP corresponds to the value of the channel potential for
which the inversion charge becomes zero in a non-equilibrium situation. VP depends
only on the gate voltage and represents the voltage that should be applied to the
channel (source and drain) to balance the effect of VG. Hence, VP can be directly
related to VG:
⎡ ⎛ γ ′ ⎞ γ ′⎤
2
VP = VG′ − PHI − γ ′ ⎢ VG′ + ⎜ ⎟ − ⎥ (1)
⎢⎣ ⎝2⎠ 2⎥
⎦
168
ELECTRONICS’ 2006 20 – 22 September, Sozopol, BULGARIA
where VG′ = VG − VTO − PHI + GAMMA ⋅ PHI (2)
ε 0 ε Si ⎡ LETA ⎛ LETA WETA ⎞ ⎤
γ ′= GAMMA − ⎢ PHI + VD + ⎜ −3 ⎟ PHI + VS ⎥ (3)
COX ⎣ L + DL ⎝ L + DL W + DW ⎠ ⎦
The modified body effect factor γ’ (3) accounts for short- and narrow-channel
effects. For large device geometries γ’ = GAMMA (note that GAMMA is the body
effect factor, i.e. GAMMA ≡ γ = 2qε Si N SUB C ox ) and if uniform doping in the
channel is assumed then VP is a function only of the gate voltage and the three
parameters VTO, GAMMA and PHI.
2.3. Drain current
The drain current ID, is expressed as
ID = IF – I R (4)
where IF is the forward component of the current (independent of VD) and IR is the
reverse component of the current (independent of VS). The currents are obtained by
integrating the inversion charge along the channel. In weak inversion (WI) and in
strong inversion (SI)
⎛ VG − VTO − nV S ( D ) ⎞
I F ( R ) = exp⎜⎜ ⎟⎟ (WI) (5)
⎝ nU T ⎠
⎛ VG − VTO − nVS ( D ) ⎞
2
I F ( R ) = ⎜⎜ ⎟⎟ (SI) (6)
⎝ 2 nU T ⎠
with the thermodynamic voltage UT = kT/q (0.026 V at 300° K).
The EKV model is based on inversion charge linearization, which is given by the
following interpolation function [4]:
1 1
q IF ( R ) = + I F ( R) − (7)
4 2
For computational purposes equations (5) and (6) are combined with
mathematical transformations into a single expression I F ( R ) = I S (q IF2 ( R ) + q IF ( R ) )
2
⎧⎪ ⎡ ⎛ VG − VTO − nVS ( D) ⎞⎤⎫⎪
I F ( R) = I S ⋅ ⎨ln⎢1 + exp⎜⎜ ⎟⎟⎥⎬ (8)
⎪⎩ ⎣ ⎝ 2 nU T ⎠⎦⎪⎭
where the specific current being I S ≡ 2nβ U T2 . (9)
In (9) β = μ C ox′ (Weff Leff ) is the transfer parameter.
The simple mobility reduction (transconductance) model uses parameter THETA:
Weff 1
β = KP , (10)
Leff 1 + THETA ⋅ V P
169
ELECTRONICS’ 2006 20 – 22 September, Sozopol, BULGARIA
where KP = μC’ox. (Note that default value for THETA is 0, Tab.1.)
Other parameter of interest is the slope factor n of the VG vs. VP characteristic
with VP, defined as the value of the channel potential for which the inversion charge
becomes zero at a given VG (non-equilibrium case) [5]. The slope factor n is defined
as n ≡ {dV P dVG } and is linked to the weak inversion slope [6]:
−1
GAMMA
n =1+ (11)
2 V P + PHI + 4U T
2.4. Effective channel lengths
The MOSFET effective channel length Leff is a very useful parameter for circuit
design and simulation as well as for technology characterization. Although related to
the metallurgical length, shorter than the mask length owing to lateral diffusion, the
effective channel length is rather an electrical parameter to which drain current is
inversely proportional [2]. Its determination is strongly correlated with the
measurement and modeling of the I-V characteristics. In EKV, the following
equations are used to calculate the effective length and width:
Leff = L + DL , Weff = W + DW (12)
3. PARAMETER EXTRACTION
Table 2. Extraction sequence specifying device sizes, measured characteristics, modes (SI–strong,
MI–moderate, WI–weak inversion, co.–conduction, sat.–saturation).
Size Characteristic Mode Parameter
matrix W/L ID vs. VG WI DL, DW
gmg vs. VG
wide/long ID vs. VS SI sat IS
VP vs. VG MI sat VTO, GAMMA, PHI
KP, THETA
ID vs. VG SI sat.@VS
wide/short ID vs. VS SI sat IS
VP vs. VG MI sat LETA
ID vs. VD SI co.-sat. LABMDA
narrow/ ID vs. VS SI sat IS
long VP vs. VG MI sat WETA
Model parameters such as gate capacitance COX, the junction depth XJ, and
longitudinal critical field UCRIT are assumed to be known from the technology
datasheet. Rest of the data is taken from experiment and subsequent fitting.
Experiments might be performed either via measurements or via simulation based
on already calibrated model (e.g. BSIM).
The steps to follow are: “measurement” (in our case simulation) of VP,
determination of IS, extraction of VTO, GAMMA, PHI, KP, THETA, DL and DW.
The extraction procedure goes through the following steps listed in Tab. 2.
170
ELECTRONICS’ 2006 20 – 22 September, Sozopol, BULGARIA
4. MODEL IMPLEMENTATION
Above listed equations are coded into a set of m-files
to obtain the device characteristics of interest: output
characteristic (ID vs. VD @ VG), transfer characteristic
(ID/√ID [Link] @ VS), transconductance (gm vs. VD @ VG).
The proposed model implementation works with user
defined parameters (Fig. 1).
The accuracy of results given by generalized equation
(8) determines the overall better coincidence between the
experimental and theoretical data. That is why (7) is
programmed with the use of Labert’s W interpolation
function
n ⎡ ⎛ VG − VTO − nVS ( D ) ⎞⎤
q IF ( R ) = LambertW⎢exp⎜⎜ ⎟⎟⎥ (13)
2.29 ⎣ ⎝ nU T ⎠⎦
to embed the following code
Qif=(nc./2.29).*lambertw(exp((VGTO-...
nc.*VS)./nt));
Qir=(nc./2.29).*lambertw(exp((VGTO-...
nc.*VD)./nt)); Figure 1. Input
Parameters.
IF = IS.*(Qif.^2+Qif);
IR = IS.*(Qir.^2+Qir);
ID = IF - IR;
5. RESULTS
We have included our m-files in a toolbox named “EKV_MOST”. The basic
modules are EKV_out (for output characteristics), EKV_trans (for transfer
characteristics), and EKV_gm (for transconductances). Our toolbox also contains
additionalodules for extraction and fitting EKV parameters.
On the figures below, EKV_out and EKV_trans results are shown that are
validated versus BSIM3v3 PSpice model for AMS 0.35 μm CMOS technology. The
plots demonstrate a very close match between simulation results of our Matlab open
source EKV and the conventional BSIM3v3 Pspice models.
6. CONCLUSION
This paper describes a Matlab implementation of the EKV model. Modules for
output characteristics, for transfer characteristics, and for transconductances are
programmed in m-files. Very good accuracy is achieved by using LabertW function.
The program code is open source giving the designer a direct access to model
equations and parameters.
171
ELECTRONICS’ 2006 20 – 22 September, Sozopol, BULGARIA
Figure 2. Output characteristics ID vs. VD @ VS = Figure 3. Transfer characteristics ID vs. VG @ VS
0 of a short channel device (W=10μm, of a short channel device (W=10μm, L=0.35μm).
L=0.35μm). (– - BSIM3v3 PSpice simulation; (– - BSIM3v3 PSpice simulation; Ο - ID vs. VG
- Matlab EKV model). and U - √ID vs. VG model).
7. REFERENCE
[1] Matlab Online Documentation
[Link]
[2] G. Angelov, T. Takov, and St. Ristiç "MOSFET Models at the Edge of 100-nm Sizes",
Proc. of the 24th Intl. Conf. on Microelectronics (MIEL 2004), Niš, Serbia and Montenegro, Vol. 1,
pp. 295-298, May 2004.
[3] [Link], [Link], [Link], “Methodology for Extracting EKV Model
Parameters from BSIM3 Simulated Characteristics”, Proc. of the 29th IEEE Intl. Spring Seminar on
Electr. Technology (ISSE 2006), St. Marienthal, Germany, May 2006.
[4] C. C. Enz, F. Krummenacher and E. A. Vittoz, “An Analytical MOS Transistor Model Valid
in All Regions of Operation and Dedicated to Low-Voltage and Low-Current Applications”, Analog
Integrated Circuits and Systems Processing Journal, vol. 8, pp. 83–114, 1995.
[5] G.A.S. Machado, C. C. Enz, and M. Bucher, “Estimating key parameters in the EKV MOST
model for analogue design and simulation”, IEEE ISCAS’95, pp. 1588-1591, 1995.
[6] M. Bucher, C. Lallement, C. Enz and F. Krummenacher, "Accurate MOS Modelling for
Analog Circuit Simulation usingthe EKV MOST Model", IEEE ISCAS 96, pp. 703-6 vol.4, 1996.
[7] M. Bucher, C. Lallement, C. Enz “An efficient parameter extraction methodology for the
EKV MOST model”, Proceedings of IEEE Intl. Conf. on Microelectronic Test Structures (ICMTS
1996), pp. 145-50, 1996.
[8] M. Bucher, C. Lallement, C.C. Enz, F. Théodoloz, F. Krummenacher, “The EPFL-EKV
Model Equations for Simulation”, Technical Report, Rev. 2, July 1998.
172