Physical parameter-based data-driven modeling of small signal parameters of a metal-semiconductor field-effect transistor

Abstract

In this work, physical parameter-based modeling of small signal parameters for a metal-semiconductor field-effect transistor (MESFET) has been carried out as continuous functions of drain voltage, gate voltage, frequency, and gate width. For this purpose, a device simulator has been used to generate a big dataset of which the physical device parameters included material type, doping concentration and profile, contact type, gate length, gate width, and work function. Five state-of-the-art algorithms: multi-layer perceptron (MLP), IBk, K*, Bagging, and REPTree have been used for creating a regression model. The symbolic regression algorithm has been used to obtain analytical expressions of the real and imaginary parts of the Scattering (S) parameters using the physics-based generated dataset. The regression performances of all the benchmarks and the symbolic regression have been compared to references from the device simulator results. The results of the derived equations and the best algorithms have been then compared to the device simulator results, with case studies for validation. The DC performance characteristics of the MESFET have been also obtained. The proposed model can be used to predict the small signal parameters of new devices prior to development, and allows for both the device and circuit to be optimized for specific applications.