AI-Assisted Design of FISO CFP-DEMOS Devices via Transfer Learning and Multiobjective Optimization

View at publisher Summarise with VIRAT
Authors: Xiaoyun Huang, Ye Song, Hongyu Tang, Yan Pan, Yiting Ye, Boying Meng, Shuang Wang, Zenan Wang, Dawei Gao, Kai Xu
Cited by: 0
References: 24

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Drain-extended metal–oxide–semiconductor (DEMOS) devices are critical components in power management integrated circuits (PMICs), yet their optimization faces significant challenges, including prohibitive experimental costs, limited datasets, and inadequate modeling of process variations. This research presents an AI-assisted device design framework for optimizing fully isolated contact field plate DEMOS (FISO CFP-DEMOS) devices. The framework employs a modular architecture integrating two key components: a pretraining and fine-tuning neural network for performance prediction, and an adaptive multiobjective optimization algorithm. The predictive model leverages uncalibrated simulation data during pretraining to learn fundamental device physics, followed by fine-tuning with minimal experimental measurements to capture real fabrication conditions. This transfer learning approach achieves mean absolute percentage errors (MAPEs) of 3.53% for breakdown voltage ($V_{\text {BD}}$) and 3.08% for specificon-resistance ($R_{\text {on},\text {SP}}$) using only 1% of available experimental data. The sequential optimization algorithm then systematically optimizes process and structural parameters across multiple voltage ratings (9–40 V) while ensuring fabrication feasibility. Experimental validation demonstrates figure-of-merit (FOM) improvements up to 52.01% compared to baseline devices. This methodology significantly reduces development cycles and experimental costs while maintaining high prediction accuracy and practical manufacturability, establishing an efficient pathway for next-generation power semiconductor device development.

Fields of Study & Subjects

Silicon Carbide Semiconductor Technologies Advancements in Semiconductor Devices and Circuit Design Advanced Multi-Objective Optimization Algorithms

Publication details

Journal / source
IEEE Transactions on Electron Devices
Volume
73
Issue
5
Pages
2680–2688
Publication date
25 March 2026
Language
EN