Device and circuit optimization of RRAM for neuromorphic computing

Abstract

RRAM is a promising electrical synaptic device for efficient neuromorphic computing. A human face recognition task was demonstrated on a 1k-bit 1T1R array using an online training perceptron network. The RRAM device structure and materials stack were optimized to achieve reliable bidirectional analog switching behavior. A binarized-hidden-layer (BHL) circuit architecture is proposed to minimize the needs of A/D and D/A converters between RRAM crossbars. Several RRAM non-ideal characteristics were carefully evaluated for handwritten digits' recognition task with proposed BHL architecture and modified neural network algorithm.