An Evolutionary Multilayer Perceptron-Based Large-Signal Model of GaN HEMTs Including Self-Heating and Trapping Effects

Abstract

Neural networks are widely used to build large-signal models; an appropriate network architecture is important for high model accuracy and good generalization ability. In this article, an evolutionary multilayer perceptron (EMLP)-based modeling approach is proposed to construct an accurate model with a proper architecture for GaN high electron mobility transistors (HEMTs). The multiobjective gray wolf optimizer (MOGWO) is used to optimize architectures, including the number of hidden layers and neurons. The two objective functions consist of the training error, generalization error, number of test cases, and number of parameters. An accurate EMLP-based model with a proper architecture is selected from the Pareto optimal solutions. The proposed method can be directly used to develop a large-signal model considering self-heating and trapping effects. Alternatively, to improve model accuracy and generalization ability, this approach is used to accurately model the effective trap potential and the bias- and temperature-dependent factor. The capacitance models are built with the EMLP and approximation. The large-signal model is implemented in Keysight Advanced Design System (ADS), and a good agreement is achieved between the measured and simulated results.