GaN Model Accuracy with an ANN Approach
Videos
As semiconductor devices become increasingly complex and design cycles accelerate, traditional compact modeling approaches often struggle to balance accuracy, scalability, and development efficiency. To address these challenges, we present a hybrid modeling methodology that integrates physics-based compact models with Artificial Neural Networks (ANN). This Hybrid Physics-Based ANN framework leverages the predictive power of machine learning while preserving physical interpretability and ensuring simulation robustness. Using Keysight’s IC-CAP ANN Toolkit, we demonstrate how accurate Verilog-A models can be rapidly generated from measured and derivative data. The approach significantly improves fitting across IV, CV, and S-parameter domains, as illustrated through real-world GaN HEMT modeling with the ASM-HEMT model. By combining physical rigor with AI-driven adaptability, this methodology reduces model development time and enhances confidence in predictive simulations—offering a practical path forward for both emerging technologies and the refinement of established device models.
