Use of ANNs for Optimizing the Design Parameters of GaN IMPATT Sources

This paper introduces a novel approach for optimizing the design of GaN-based Terahertz (THz) IMPATT sources using Artificial Neural Networks (ANNs). Traditional optimization methods relying on the self-consistent quantum drift-diffusion (SCQDD) model are computationally intensive and time-consuming. By training ANN models on SCQDD-based simulation data, we predict key performance metrics, including RF power output and DC-to-RF conversion efficiency, for GaN IMPATT diodes with different doping profiles. The ANN models demonstrated high accuracy in comparison to experimental data. Additionally, inverse ANN models were developed to optimize diode design based on desired performance targets, significantly reducing computational cost and time. This ANN-based approach enables efficient optimization of GaN IMPATT diodes across a wide frequency range, including THz, offering a scalable solution for the rapid development of high-performance THz radiation sources.