Evaluating Nine Machine Learning Algorithms for GaN HEMT Small Signal Behavioral Modeling through K-fold Cross-Validation

This paper presents an investigation into the modeling of Gallium Nitride (GaN) High Electron Mobility Transistors (HEMTs) using multiple Machine Learning (ML) algorithms. Despite the documented use of various ML techniques, a thorough comparison and performance analysis under different operating conditions were lacking. This study fills this gap by conducting a rigorous evaluation of nine ML models using TCAD-generated data of Pseudomorphic AlGaN/InGaN/GaN HEMT. The research focuses on Small Signal Behavioral Modeling and examines regression techniques such as Multiple Linear Regression (MLR), Multivariate Linear Regression (MVLR), Ridge Regression (L2), Lasso Regression (L1), Elastic Net Regression (ENR), Decision Trees (DT), Random Forest (RF), Gradient Boosting Regression(GBR), and Support Vector Regression (SVR). These methods use biases, frequency, and device geometry as independent variables, with S-parameters being the dependent variables. K-fold cross-validation was employed to ensure model reliability and accuracy across diverse operating conditions. Results reveal that the RF, coupled with 10-fold cross-validation, exhibits superior performance giving 99.7% accurate results, with a Mean Squared Error (MSE) of 4.6375×10-5, and a coefficient of determination (R2) of 0.9977. Conversely, SVR, L1, and ENR fall short of expectations. This study underscores the significance of methodological advancements in ML-based GaN HEMT modeling and provides valuable insights for future research in this domain.