Assessment of a Hardware-Implemented Machine Learning Technique Under Neutron Irradiation

Abstract

Hardware-implemented intelligent systems running autonomous functions and decisions are today becoming more and more ubiquitous in many fields of applications, demanding reliable operation even under harsh conditions as in nuclear power plants and avionics altitudes. Support vector machine (SVM) is a prominent machine learning solution to optimize hardware-implemented autonomous systems. This paper is the first to assess the operation of a field-programmable gate array (FPGA)-designed SVM architecture under radiation effects. A fault emulation campaign along with radiation test experiments with a 14-MeV neutron generator has been performed, and the results show that 27% of the neutron radiation-induced errors in the target SVM architecture provoked critical failures.