Memristor-based 3D neuromorphic computing system and its application to associative memory learning

Abstract

3D integration technology offers a near term strategy for bypassing Moore's Law. Applying 3D integration to neuromorphic computing (NC) could provide a low power consumption, high-connectivity, and massively parallel processed system that can accommodate high demand computational tasks. This paper proposes a novel analog spiking nanoscale 3D NC system, wherein both neurons and synapses are stacked three-dimensionally, with monolithic inter-tier via (MIV) technology, and vertical resistive random-access memory (V-RRAM) structures. An application of the proposed system to associative memory learning is performed to demonstrate its capability in high demand computational tasks. The computational efficiency and performance improvement of the proposed architecture are demonstrated.