A large-signal characterization of an HEMT using a multilayered neural network

Abstract

We propose an approach to describe the large-signal behavior of a high electron-mobility transistor (HEMT) by using a multilayered neural network. To conveniently implement this in standard circuit simulators, we extracted the HEMT's bias dependent behavior in terms of conventional small-signal equivalent-circuit elements. We successfully represented seven intrinsic elements with a five-layered neural network (composed of 28 neurons) whose inputs are the gate-to source bias (V/sub gs/,) and drain-to-source bias (V/sub ds/) A well-trained neural network shows excellent accuracy and generates good extrapolations.