Artificial Intelligence-Based Modeling of Microwave GaN HEMT Power Dissipation

Abstract

The GaN HEMT power dissipation is modeled as function of the drain source voltage, the gate source voltage, the drain quiescent bias voltage and the gate quiescent bias voltage in pulsed mode. The developed model is based on Artificial Neural Networks (ANN) technique. The neural network is trained with two different algorithms: the genetic and the Levenberg-Marquard algorithms. High ANN model accuracy is obtained with the Levenberg-Marquard algorithm. An optimization procedure is carried out in the purpose to determine the optimal numbers of quiescent bias voltages used in the network training. The obtained optimal data are used to train the ANN. The comparison between the device output power versus the drain source voltage curves obtained from the ANN model and those simulated from analytical equations, has shown a weak model error and validates the proposed methodology.