Machine-Learning-Enabled p-Gate GaN HEMT Surrogate Model for Semiconductor Database

Electrical data for emerging non-silicon semiconductor devices (e.g., GaN) remain scarce due to high fabrication costs. While Technology Computer-Aided Design (TCAD) tools can simulate device behavior, TCAD licenses are generally not available to the public, and the required computational resources can be substantial. Therefore, it is desirable to have a universal automated framework to generate surrogate models for various devices. This paper demonstrates the concept using machine learning (ML) applied to TCAD-generated data for a p-gate GaN high-electron-mobility transistor (HEMT). The proposed framework enables the extraction of electrical characteristics (IDVD, IDVG, IGVG, BV) under the variations of seven device parameters. High accuracy was obtained through the use of modified autoencoders, which can learn the latent physics. The TCAD data and the surrogate models are implemented on an online semiconductor database for research and education purposes.