Rescuing Memristor-based Neuromorphic Design with High Defects

TLDR

Memristor‑based synaptic networks are widely studied for fast, low‑cost neuromorphic computing, yet increasing density leads to crossbar bit failures that significantly degrade accuracy. This work proposes a defect‑rescuing design to restore computation accuracy. The design identifies significant weights in the network and applies retraining and remapping algorithms to mitigate defects. Real‑device testing on a two‑layer MNIST classifier shows the design recovers nearly full performance even with 20 % random defects, maintaining 92.64 % accuracy.

Abstract

Memristor-based synaptic network has been widely investigated and applied to neuromorphic computing systems for the fast computation and low design cost. As memristors continue to mature and achieve higher density, bit failures within crossbar arrays can become a critical issue. These can degrade the computation accuracy significantly. In this work, we propose a defect rescuing design to restore the computation accuracy. In our proposed design, significant weights in a specified network are first identified and retraining and remapping algorithms are described. For a two layer neural network with 92.64% classification accuracy on MNIST digit recognition, our evaluation based on real device testing shows that our design can recover almost its full performance when 20% random defects are present.