I-V MODELS WITH THE TEMPERATURE DEPENDENCE OF GAN HEMT UTILIZING FEEDFORWARD NEURAL NETWORKS, RADIAL BASIS NEURAL NETWORKS, AND GENERALIZED REGRESSION NEURAL NETWORKS

Abstract

This work contains aim to propose effective and reliable I-V models with temperature dependence to explain the characteristic of Gallium Nitride (GaN) High Electron Mobility Transistor (HEMT). Machine learning (ML) algorithms which are most useful in less time consumption by excluding multiple calculations were applied. Within algorithms of ML such as Feed Forward Neural Networks, Radial Basis Neural Networks, and Generalized Neural Networks developed by the MATLAB programming tool. During the calculations of the given dataset, network training was conducted by utilizing the Levenberg-Marquardt backpropagation algorithm. The number of hidden layers and the neurons they contain, the initialization methods for the weights and biases, and some other crucial parameters all play a significant role in determining the accuracy and effectiveness of the model. However, in order to provide the validity of the model, elapsed time and the capabilities of the proposed model are also assessed via various learning algorithms. As performance evaluation criteria, the suggested network model used means squared error (MSE), mean absolute error (MAE), Coefficient of Determination R2, and regression analysis. Therefore, between all these proposed neural networks’ architecture the comparison was analyzed to find out which of them is well fitting to model the behavior of the GaN HEMT that is grown on Silicon. It is predicted in several research works that the Artificial Neural Network ensures good generalization performance under change of the parameters of GaN HEMT.