SMALL-SIGNAL MODELLING OF GAN HEMT USING ANN, SVR, GPR AND TREE-BASED SUPERVISED MODELS

Abstract

In this work, our objective is to devise an effective and an accurate small-signal model to elucidate the behavior of Gallium Nitride (GaN) High Electron Mobility Transistor (HEMT). At first, Support Vector Regression (SVR), Artificial Neural Network (ANN), Gaussian Process Regression (GPR) and Tree-based algorithms such as Decision Tree (DT), AdaBoost, Random Forests (RFs) are developed to exploit the bias, frequency, and temperature dependence. Thereafter, a reasonable comparison is carried out to find out the model that optimally describes the behavior of 10x200 μm GaN HEMT grown on Silicon. Furthermore, three different libraries are utilized: Keras, Scikit-learn and PyTorch to develop the ANN based models. All the developed models are evaluated in terms of mean squared error (MSE), mean absolute error (MAE) and coefficient of determination. It is observed that ANN based models guarantees the optimal prediction. Of the three models of ANN, Keras library based is the more efficient. Tree-based models are also shown to be a great alternative for the tabular based problems. GPR and SVR are the most expensive algorithms in terms of computational efficiency.