Evaluating and Mitigating Neutrons Effects on COTS EdgeAI Accelerators

EdgeAI is an emerging artificial intelligence (AI) accelerator technology, which capable of delivering improved AI performance at both a lower cost and power level. With the aim implementation in large quantities safety-critical environments, it imperative to understand how single-event effects (SEEs) affect reliability this new family devices propose efficient hardening solutions. Through neutron beam experiments fault-injection analysis commercial-off-the-shelf (COTS) device, we are able identify device's SEE failure-modes, separate error rate contributions different resources, characterize reliability. During analysis, discovered that vast majority single-bit flips have no appreciable effect on output. After solution implements triple-modular redundancy (TMR) device without changing its physical architecture. We experimentally validate show correct 96% misclassifications (critical errors) with nearly zero overhead.