Artificial neural networks for compound semiconductor device modeling and characterization

This paper reviews applications of artificial neural networks (ANNs) to several distinct problem areas that arise in compound semiconductor device modeling and characterization. Properties and corresponding benefits of ANNs for these applications are presented culminating in an accurate large signal-model of GaN HEMT transistors (with thermal and trapping effects). Smooth functional approximations of device properties and parameters are also illustrated based on unique properties of ANNs. Finally, it is suggested that ANN technology can be quite helpful as a device characterization tool, over and above the obvious utility for multi-dimensional data fitting.

I. Introduction

Artificial neural network (ANN) technology has become a versatile, powerful, and efficient tool for the general representation of transistor model multi-dimensional constitutive relations (e.g. terminal currents and charges as nonlinear functions of intrinsic terminal voltages, junction temperature, and trapped charge) from measurements, in a form useful for a wide variety of compound semiconductor technologies, and deployed in a wide range of simulations (DC, S-parameter, harmonic balance, etc) [1].