Designing a Fault Tolerant Neural Network Computing System Based On Nanoscale Electronic Elements

Abstract

This article observes the potential for building neural network computing systems designed via use of nanoscale electronic elements. The theory of interrelation between the fault-tolerance index of such systems and its predetermining factors has been systematized. We have also developed an approach to analysis of properties of parallel computing systems including nanoscale electronic elements at the stage of designing computing systems for the purpose of providing the maximum fault-tolerance index. By means of computer-generated simulation, we have experimentally tested this approach and it has proved to be superior to the available methods for solving this particular problem.