Accurate and Efficient Behavioral Modeling of GaN HEMTs Using An Optimized Light Gradient Boosting Machine

Abstract

An accurate, efficient, and improved Light Gradient Boosting Machine (LightGBM) based Small-Signal Behavioral Modeling (SSBM) techniques are investigated and presented in this paper for Gallium Nitride High Electron Mobility Transistors (GaN HEMTs). GaN HEMTs grown on SiC, Si and diamond substrates of geometries 2 × 50 $\mu \mathrm{m}$, 10 × 200 $\mu \mathrm{m}$, and 4 × 125 $\mu \mathrm{m}$, respectively are used in this study. A versatile set of LightGBM's hyperparameters including learning and tree specific parameters are meticulously optimized using a modern and vigorous optimization algorithm namely Osprey Optimization Algorithm (OOA) with the objective to accomplish superior model performance. The developed OOA-LightGBM based models are validated for a wide array of operating conditions including for frequency values within a broad spectrum of 0.25 to 120 GHz, 0.1 to 26 GHz, and 0.1 to 40 GHz for GaN-on-SiC, GaN-on-Si, and GaN-on-Diamond HEMTs, respectively. The proposed SSBM techniques have demonstrated remarkable prediction ability and are impressively efficient for all the GaN HEMTs devices tested in this work.