Gaussian Process Regression for Small-Signal Modelling of GaN HEMTs

This paper explores and develops a biasdependent small-signal model for Gallium Nitride (GaN) High Electron Mobility Transistors (HEMTs) using a non-parametric probability based Gaussian Process Regression (GPR) approach. The characterization data for HEMT is available in S-parameters. It is observed that the determination of optimal hyper-parameters for each S-parameter is crucial in this technique. Therefore, the developed modeling technique employs 10-fold cross-validation loss model to find the optimal parameters. The model incorporates Bayesian optimization alongside with cross-validation error for the best outcome. As a first step, the model is developed using two-third of the data (training set) and subsequently one-third of the remaining data (testing set) is used to capture the predictive ability of the model. The model is tested for both training and testing sets in terms of the mean square error and fitting curves of measured and simulated outputs. Excellent performance is achieved for the entire frequency range. Lastly, the model is also compared with other leading models based on Artificial Neural Network and Support Vector Regression to demonstrate the effectiveness of the proposed GPR based model.