Approximate computation based on NAND-SPIN MRAM for CNN on-chip training

Abstract

Approximate computation is a widely used method to accelerate CNN training. In this work, the stochastic switching mechanism of the NAND-SPIN MRAM is utilized to perform the approximate update and storage of the synaptic weight. By reducing the programming time of the NAND-SPIN MTJs from 3 ns to 1 ns, more than 67% speedup and nearly 70% energy saving have been achieved with less than 1% accuracy loss.