A Framework for Automatic Generation of Neural Models of Electron Devices

In this paper a framework for automatic generation of neural network models
for the dual-gate MESFET is presented. Values of the large-signal model are
extracted from S-parameter measurements at many DC-bias points. The automatic
model generation is accomplished by integrating multiple software tools, including
one developed by the author, in a “homegrown” integration environment. The
technique is used to model a 6-gate 1x100 μm dual-gate MESFET manufactured
by Nortel Networks, Ottawa, Canada. The large-signal model is then verified
through a variable gain large-signal amplifier application based on the dual-gate
MESFET. Measurements and harmonic balance simulations of the verification
circuit showed very good agreement of the first harmonic. For the second and third
harmonic, some discrepancies between the measurements and the model are
observed. This is mainly due to some model simplifications and second order
effect.