Linear Weight Update in MoS2/Graphene Memristive Synapses for Unsupervised Learning

Abstract

Memristive synaptic devices are considered one of the most promising candidates for brain-inspired neuromorphic computing, owing to their decreased complexity and nanoscale footprint compared to conventional complementary metal oxide semiconductor (CMOS) circuitry[1], [2]. However, the non-linearity and asymmetry in weight update observed in most memristive synapses due to the difference in the long-term potentiation and depression characteristics makes it difficult to use these devices for unsupervised learning applications[1]. Previously, linearity in synaptic weight update has been engineered using non-identical input voltage pulsing scheme[3]. However, not many reports on linear synaptic weight update using identical input voltage pulses exist in literature [4]. In this work, we present large-area chemical vapor-deposited (CVD) $\mathrm{MoS}_{2}$ /graphene memristive synapses which exhibit linear weight update using identical input voltage pulses. These synaptic devices also exhibit spike-timing dependent plasticity, essential for online training.