The linear statistical model of the GaAs MESFET's current generator is
obtained by means of factor analysis. Three different MESFET determin
istic models are taken into account in the analysis: the Statz model (
ST), the Materka-type model (MT) and a new proprietary model of MESFET
with implanted channel (PLD). It is shown that statistical models obt
ained using factor analysis provide excellent generation of the multid
imensional random variable representing the drain current of MESFET. T
he method of implementation of the statistical model into the SPICE pr
ogram is presented. It is proved that for a strongly limited number of
Monte Carlo analysis runs in that program, the statistical models con
sidered in each case (ST, MT and PLD) enable good reconstruction of th
e empirical factor structure. The empirical correlation matrix of mode
l parameters is not reconstructed exactly by statistical modelling, bu
t values of correlation matrix elements obtained From simulated data a
re within the confidence intervals for the small sample. This paper pr
oves that a formal approach to statistical modelling using factor anal
ysis is the right path to follow, in spite of the fact, that CAD syste
ms (PSpice[MicroSim Corp.], Microwave Harmonica[Compact Software]) are
not designed properly for generation of the multidimensional random v
ariable. It is obvious that further progress in implementation of stat
istical methods in CAD software is required. Furthermore, a new approa
ch to the MESFET's d.c. model is presented. The separate functions, de
scribing the linear as well as the saturated region of MESFET output c
haracteristics, are combined in the single equation. This way of model
ling is particularly suitable for transistors with an implanted channe
l.