Drift-enhanced Unsupervised Learning with PCM Synapses

Neuromorphic systems using resistive memory offer a promising route for energy-efficient hardware implementation of large-scale artificial neural networks. However, for synaptic devices demonstrated so far, device non-idealities significantly degrade learning performance and classification accuracy. Here we investigate the effect of resistance drift, considered as a non-ideality for PCM devices, on unsupervised learning and demonstrate that resistance drift can be exploited to boost accuracy for online learning.