ADVANCED MODELLING OF GaAs HBTs AND GaN HEMTs FOR RF APPLICATIONS

Abstract

Accurate models for RF/microwave transistors are essential for circuit design using electronic design automation tools. Many empirical and artificial neural network (ANN)-based models have been developed. However, empirical models are becoming increasingly complex to capture thermal effects. ANN-based modelling is sensitive to outliers, and some ANN-based models are not charge-conservative. In addition, ANN architecture is determined by a trial-and-error process. This thesis presents the small- and large-signal modelling of gallium arsenide (GaAs) heterojunction bipolar transistors (HBTs) and gallium nitride (GaN) high electron mobility transistors (HEMTs). Measurement data are obtained for model development and verification. Firstly, a dimension reduction method is investigated for fast temperature-dependent current modelling. Secondly, hybrid extraction and outlier detection are developed to obtain reliable datasets for accurate ANN-based modelling. Furthermore, an ANN-based consistent gate charge model is proposed. Lastly, an evolutionary multilayer perceptron-based modelling approach is proposed to construct an accurate model with a proper architecture.