PROJECT TITLE :

Experimental Demonstration and Tolerancing of a Large-Scale Neural Network (165 000 Synapses) Using Phase-Change Memory as the Synaptic Weight Element

ABSTRACT:

Using 2 section-change memory devices per synapse, a three-layer perceptron network with 164 885 synapses is trained on a subset (5000 examples) of the MNIST database of handwritten digits employing a backpropagation variant appropriate for nonvolatile memory (NVM) + selector crossbar arrays, obtaining a coaching (generalization) accuracy of eighty two.2percent (82.ninep.c). Employing a neural network simulator matched to the experimental demonstrator, in depth tolerancing is performed with respect to NVM variability, yield, and also the stochasticity, linearity, and asymmetry of the NVM-conductance response. We show that a bidirectional NVM with a symmetric, linear conductance response of high dynamic range is capable of delivering the identical high classification accuracies on this drawback as a typical, software-primarily based implementation of this same network.