Gallium Nitride Power Device Modeling using Deep Feed Forward Neural Networks

Abstract

A novel approach to modelling Gallium Nitride (GaN) power devices using Machine Learning (ML) is presented in this paper. To make it easier for the power designers to use GaN devices, this work proposes deep feed forward GaN ML device models which are highly accurate and can predict the switching behaviour of the device without having to delve into the physics and geometry of the device. The strategy in this research work is to use deep learning techniques to build a GaN based regression model using stochastic gradient algorithm by back propagation. Among the different neural network architectures trained and tested, a deep feed forward neural network with 5 hidden layers and 30 neurons, was found to be the best for prediction and optimization. The possibility of employing ML techniques for GaN can help open doors for faster commercialization of GaN power electronics.