AI-Driven Modeling of GaN HEMT and SBD for Wireless Power Transfer Systems

This paper proposes a hybrid artificial intelligence (AI)-driven modeling framework to address the nonlinear characterization challenges of gallium nitride Schottky barrier diodes (GaN-SBDs) and high-electron-mobility transistors (GaN HEMTs) in high-frequency wireless power transfer (WPT) systems. For the GaN-SBD, a novel scaling method integrated with an artificial neural network (ANN) is developed to accurately model the current-voltage (I-V) characteristics spanning six orders of magnitude. The proposed ANN achieves a root mean square error of 0.107% across a wide bias range from -30 V to 1 V. For the GaN HEMT, a small-signal modeling approach combining the Harris Hawks optimization (HHO) algorithm with direct parameter extraction techniques is employed to optimize parasitic and intrinsic parameters. Compared to conventional optimization methods, genetic algorithms (GA), and particle swarm optimization (PSO), the HHO algorithm demonstrates faster convergence, higher accuracy, and lower parameter-tuning costs. Furthermore, a three-layer fully connected ANN architecture is adopted to model the large-signal I-V characteristics of the GaN HEMT. The proposed framework effectively overcomes the limitations of traditional modeling approaches and provides critical device-level design support for next-generation WPT systems.