A Physics-Informed Neural Network-Based Scalable Model for GaN HEMTs

The integration of physical knowledge into artificial intelligence (AI)-based models for gallium nitride high electron mobility transistor (GaN HEMT) modeling is gaining increasing attention, as it reduces the amount of required training data, enhances the model’s physical plausibility, and contributes to the development of explainable AI. In this article, a physics-informed neural network (PINN)-based scalable GaN HEMT model is proposed. The model is trained with channel charge densities as fitting targets, while physical consistency is ensured by incorporating the drain current and gate capacitances into the PINN’s physical constraints. The model architecture and training algorithm are described in detail. The proposed PINN-based GaN HEMT modeling method achieves a high degree of automation in model development, while ensuring excellent prediction accuracy and physical plausibility. The significant self-heating and trapping effects in GaN HEMTs are also embedded into the proposed model. These features contribute to the model’s inherent and strong scalability. The developed GaN HEMT models of various sizes have undergone extensive small-signal and large-signal validation, demonstrating the model’s high accuracy and strong scalability.