On efficient modeling of drain current for designing high-power GaN HEMT-based circuits

In this paper, different modeling approaches to the drain current, including analytical and artificial neural network (ANN) modeling, are investigated. The adopted models address the inherent self-heating and kink effects, especially in high-power GaN-based high electron mobility transistors (HEMTs). Different optimization algorithms were demonstrated for extracting the model parameters, including genetic algorithm optimization (GAO), gray wolf optimization (GWO), growth optimization (GO), and particle swarm optimization (PSO). The modeling approaches are applied to DC IV measurements of 1-mm, 4-mm, and 2-mm GaN HEMTs on SiC and Si substrates. An improved optimization procedure was applied to the analytical models to find the main parameters responsible for fitting the general nonlinear behavior of the device. Then, the thermal or self-heating parameters are tuned for best fitting in the high-power dissipation region. The kink effect has been counted by adding another factor to the analytical formula to characterize the voltage dependency of this effect. The ANN modeling provides an efficient and cost-effective solution to accurately simulate the IV characteristics with less effort. In this technique, there is no need for a predefined closed formula or a complicated fitting parameter extraction process. Also, the model training was enhanced by using a genetic algorithm augmented backpropagation technique. The investigated analytical and ANN techniques were demonstrated by modeling the IV characteristics of the considered GaN HEMTs. The results obtained confirm the advantages of using ANN modeling for solving such problems and large-signal modeling applications.