The Impact of Proton-Induced Single Events on Image Classification in a Neuromorphic Computing Architecture

Abstract

Neuromorphic computing endeavors to imitate the way biological brains process information and problem-solves. Uses for neuromorphic computing span disciplines and include applications in image processing, audio processing, optimization, and more. This work explores the effect of proton-induced single-event upsets on a neuromorphic computing architecture engaged in image recognition. Two main results are found. One, the overall classification accuracy is unchanged although a high number of hidden, tolerable errors occurred. Additionally, single event upsets are found to alter the relative occurrence of false positives and false negatives, which occurred despite the overall classification accuracy remaining unaffected.