Physics-AI-Driven Design of GaN P-MISFETs with InGaN/AlN/AlGaN Heterostructures

This paper presents a hybrid physics-AI framework for the design of CMOS-compatible p-type GaN MISFETs based on a polarization-engineered InGaN/AlN/AlGaN/AlN/GaN heterostructure. The workflow combines TCAD simulations, design of experiments (DOE), clustering analysis, and SHAP-informed feature evaluation to identify key design parameters and trade-offs. Results reveal that recess depth is the dominant factor, improving subthreshold swing (SS) and boosting I_on but shifting V_th negatively. In contrast, an AlN interlayer and higher Al/In compositions raise V_th via stronger polarization but degrade transport through reduced hole velocity and screening. SHAP analysis confirms these trade-offs, underscoring the need for co-optimization of recess depth, material composition, and interlayer design to balance I_on, SS, and V_th in GaN CMOS technology.