计算机科学
嵌入式系统
工程类
电子工程
计算机硬件
信号处理
控制系统
软件
电气工程
集成电路
可靠性(半导体)
测试系统
作者
Junkai Zhao,Kwen‐Siong Chong,Wei Shu,Joseph Sylvester Chang
标识
DOI:10.1109/tns.2026.3677548
摘要
In our envisioned “Chang’s Next Paradigm of New Space”, COTS systems – embodying multiple COTS ICs – would be employed in space missions. When applied to COTS systems with power management ICs (including multiple voltage regulators), reported ML/AI-based SEL/μ-SEL detection approaches require access to the internal sub-power rails (after regulators). This is because regulator-induced distortions significantly obscure SEL/μ-SEL characteristics on the external power rail (before regulators). However, accessing internal rails requires physical modification, which is highly undesirable. In this paper, we propose an adaptive pre-processing module that, for the first time, enables ML/AI-based SEL/μ-SEL detection in COTS systems without physical modification. Specifically, by compensating for regulator-induced distortions, our proposed module allows ML/AI-based approaches to detect SELs/μ-SELs by monitoring only the external power rail, thereby, very advantageously, without any physical modification to the COTS system. Integrated with this proposed module, our previous reported AI-based approach achieves a high average detection accuracy of 91.1%. This is a very worthy improvement of 11.1%–53.9% over reported approaches without our proposed module, while incurring modest hardware overhead.
科研通智能强力驱动
Strongly Powered by AbleSci AI