Fault-tolerance capabilities of a software-implemented Hopfield Neural Network

Abstract

The associative Hopfield memory is a form of recurrent Artificial Neural Network (ANN) that can be used in applications such as pattern recognition, noise removal, information retrieval, and combinatorial optimization problems. In general, ANNs are considered as intrinsically fault-tolerant. A study of the capability of this algorithm to tolerate transient faults such as bit-flips provoked by the radiation environment is presented. Two software versions of the Hopfield Neural Network (HNN), one original and one fault-tolerant were implemented and executed by a LEON3 processor. Experimental results show the efficiency of the adopted strategy to tolerate faults that were injected at hardware level.