Temperature Dependent I-V Models for Microwave Transistor Using Radial Basis NNs, Generalized Regression NNs and Feedforward NN

This paper carefully examines, evaluates and compare Radial Basis Neural Networks (RBNNs), Generalized Regression Neural Networks (GRNNs) and Feedforward Neural Network (FFNN) to devise an efficient, an expeditious and an accurate IV modelling scheme for Gallium Nitride (GaN) High Electron Mobility Transistor (HEMT). The modelling schemes are employed on pulsed IV measurements of GaN HEMT device. Firstly, two variants of RBNNs namely, Exact Design (ED) and More Efficient Design (MED) networks are erected to simulate the temperature and bias dependence on the drain current. Thereafter, GRNNs and FFNN based models are developed. The hyperparameters associated with all the investigated models are tuned to improve the generalization capabilities of the models. Post tuning, the models are exploited to compute the mean squared error, mean absolute error and coefficient of determination to assess the models' performance. Lastly, the models are compared on the grounds of training and simulation time, parameters tuning time, generalization capability, computational efficiency and models' simplicity.