SEU experiments on an Artificial Neural Network implemented by means of digital processors

Abstract

The SEU sensitivity of an Artificial Neural Network intended to be used in space to detect "protonic whistlers" is investigated. We evaluate its behaviour in the presence of SEU-like faults for a hardware implementation, associating a general purpose microprocessor to a dedicated neural processor. Experimental results (SEU simulations and heavy ion ground tests) show the robustness of this implementation.