An Adaptive AI-based Approach to Detect and Protect COTS Systems against Micro-Single-Event-Latchups (μ-SELs) and SELs

In our envisioned ‘Next Paradigm’ of ‘New Space’, commercial-off-the-shelf (COTS) systems (embodying multiple COTS ICs) would be employed as payloads in space missions. Most COTS ICs are susceptible to radiation effects, particularly Micro-Single-Event-Latchups (μ-SELs) and SELs, and their characteristics are expectedly different. Consequently, hitherto reported detection approaches require characterization of the individual COTS ICs and the entire system, thereby rendering excessive overheads when applied to different COTS systems. In this paper, we propose, for the first time, the design and implementation of an adaptive AI-based approach to detect and protect various uncharacterized COTS systems (vis-à-vis pre-characterized ones) against μ-SELs and SELs. Our proposal involves the adoption of the Long-Short-Term-Memory (LSTM) neural network with our proposed two-stage training process – ex-situ pre-training and in-situ re-training – to improve general applicability. Our FPGA-based prototype achieves high (~90%) average accuracy for four different payloads. This is a worthy improvement of 13.3%-28.5% over reported approaches, yet requiring low (~115 mW) power consumption. Collectively, our proposed approach is appropriate for resource-constrained space applications and our ‘Next Paradigm’ of ‘New Space’.