Abstract
This paper explores and comprehensively evaluates high frequency Small-Signal Behavioral Models (SSBMs) for Gallium Nitride High Electron Mobility Transistors (GaN HEMTs) by leveraging Artificial Neural Network (ANN) and Adaptive Boosting (AdaBoost) with Random Forest (RF) as the base estimator. The small-signal S-parameter behavior of a 2 x 125 µm GaN HEMT based on a diamond substrate, characterized at two different operating temperatures (T = 25°C and 85°C), under varied biasing conditions and over a frequency range of 0.1 GHz to 40 GHz have been modeled and presented. Initially, predictive models using ANN and AdaBoost algorithms were developed and subsequently subjected to rigorous optimizations of their key parameters to improve predictive performance. The study covers both simulation accuracy and computational efficiency. The simulation accuracy is assessed using established regression metrics such as the coefficient of determination (R 2 ), mean squared error, mean absolute error, and Smith charts comparing actual and modeled behaviors. Computational efficiency is measured in terms of training, simulation, and tuning times. Our findings reveal that ANN-based SSBMs achieved a remarkable 0.9992 average R 2 value, surpassing the 0.9894 achieved by AdaBoost-based models on the testing set. Furthermore, we found that although AdaBoost necessitated longer hyperparameter tuning time, it can be trained faster than ANN.