Fault-Tolerant Neuromorphic Computing With Memristors Using Functional ATPG for Efficient Recalibration

This article focuses on recalibration of neural networks implemented in neuromorphic in-memory computing with memristors. The primary goal of the article is to reduce the amount of data required for recalibration which makes it particularly useful in scenarios where data availability is limited or where recalibration overhead is a concern. Moreover, the proposed approach is robust against both process and temperature variations at a significantly lower overhead compared to related works. This practical method addresses an important issue that can affect the accuracy of neural networks implemented using emerging resistive nonvolatile memories.