Assessment of Tiny Machine-Learning Computing Systems under Neutron-Induced Radiation Effects
Résumé
This paper compares and assesses the effectiveness of three prominent machine learning (ML) models for tiny ML computing systems in tolerating neutron-induced soft errors. Results of 14-MeV and thermal neutron radiation tests suggest that the three case-study ML algorithms implemented – without any mitigation technique integrated – retain a certain intrinsic level of effectiveness in tolerating neutron effects, although all of them have been functionally interrupted in some occasions, requiring hardware resets. Notably, the implemented case-study ML algorithm “random forest” has performed no misclassification during the different radiation testing campaigns.