7
Introduction to Compact Models
In 1981, Level 3  MOS model was introduced for circuit CAD using
SPICE2 [39]. Level 3 MOS model introduced many empirical parameters
to model SCEs. However, the accuracy and scalability of the model for
simulation of a wide range of channel length and width using one set of
model parameters are not entirely satisfactory to the circuit designers.
The short channel and narrow width effects are not modeled accurately
in the MOS  Level 1, 2, and 3  models and high field effects are not considered properly because of the limited understanding of the physics of
small geometry devices at the time these models were developed. Thus, to
keep parity with the continuous scaling down of MOSFETs, global effort
continued for the development of accurate and efficient compact models
for circuit CAD.
1.2.2 Recent History of Compact MOSFET Modeling
As the CMOS technology became the pervasive technology of ICs in 1970s, the
complexities of MOSFET devices continued to increase. As a result, compact
models based on simplified device physics became inadequate to analyze scaled
geometry MOSFETs. The efforts for accurate and computationally efficient
models continued using different approaches. The major modeling techniques
used can be described as threshold voltage-based, surface potential–based, and chargebased as described in Sections 1.2.2.1 through 1.2.2.3.
1.2.2.1 Threshold Voltage–Based Compact MOSFET Modeling
The major development of V th -based compact MOS model is the development
of Berkeley Short Channel IGFET Model, commonly known as BSIM, in the year
1987  [24]. It incorporated some improved understanding of the SCEs and
worked well for devices with channel length of 1 μm and above. However, it
also introduced several empirical fitting parameters just to enhance the scalability of the model. Even then, the model scalability was not totally satisfactory. Also, circuit designers did not like the use of many fitting parameters,
which do not have any physical meaning.
In order to address the shortcomings of the first generation of BSIM or
BSIM1, BSIM2  was introduced in 1990  [40]. BSIM2  improved upon BSIM1
in several aspects such as model continuity, output conductance, and subthreshold current  [40]. However, the model still could not use one set of
parameters for wide range of device sizes. Users typically need to generate a few or many sets of model parameters, each covering a limited range
of device geometries in order to obtain good accuracy over the full range
of devices used in circuits. This makes the parameter extraction difficult.
Also, it is difficult to use these parameters to perform statistical modeling
or extrapolation of the model parameters from the present technology to a
future one.
Précédent

- 28/548

Suivant