Yield Evaluation of Faulty Memristive Crossbar Array-based Neural Networks with Repairability

Abstract

This paper evaluates the yield of a memristor-based crossbar array of artificial neural networks in the presence of stuck-at-faults (SAFs). A technique based on Markov chains is used to estimate the yield in the presence of stuck-at-faults. This method provides a high degree of accuracy. Another method that is used for analysis and comparison is the Poisson distribution, which uses the sum of all repairable fault patterns. A fault repair mechanism is also considered when evaluating the yield of the memristor crossbar array. The results demonstrate that the yield could be improved with redundancies and a higher repairable stuck-at-fault ratio.