Two-dimensional MoS2-based artificial synaptic transistor for neuromorphic computing

Abstract

Although transition metal dichalcogenide (TMDC) materials are increasingly recognized for their potential in neuromorphic computing, their use as channel materials in field-effect transistors (FETs) remains elusive. This study examined the use of 2D molybdenum disulfide (MoS₂) to fabricate an artificial synaptic transistor, leveraging its unique semiconducting properties to emulate synaptic functions. The synaptic transistor exploits gate-triggered resistive switching mechanisms inherent to 2D MoS₂. This enables the transistor to replicate key synaptic behaviors such as excitatory and inhibitory postsynaptic currents, potentiation and depression, and paired-pulse facilitation. Its ability to serve as an artificial synapse with multiple stable conductance states, outstanding linearity, and low-power consumption was confirmed by modulating the conductance states of the FET. Extensive artificial neural network (ANN) simulations were performed using binary MNIST dataset digits, with 784 input neurons corresponding to 28 × 28 pixel images and 10 output neurons. Training on 42,000 MNIST digits and updating synaptic weights with conductance values derived from 18,000 training samples achieved an impressive recognition rate on the testing data. These findings highlight the potential of 2D MoS₂-based FETs in advancing neuromorphic hardware systems, offering robust synaptic functionality and effective learning rules for complex neural network applications.