A study on pattern classifications with MoS2-based CTF synaptic device

Abstract

Neuromorphic computing, inspired by the human brain, is a promising candidate for overcoming the von Neumann bottleneck of conventional computing systems. Biological synapses play an important role in transferring signals from pre- to post-synaptic neurons and modulating the connection strength between the two neurons according to the synaptic weight. An artificial synaptic device emulates the biological synaptic weight as the device conductance. In charge trap flash (CTF) memory, the device conductance is manipulated through a tunneling process; and therefore, good tunneling efficiency is important in mimicking the behavior of biological synapses. In this study, we fabricated a MoS2-based CTF device and achieved analog memory performance to demonstrate the biological synaptic function. The tunneling efficiency was improved by using SiO2 and HfO2 as tunneling and blocking oxides, respectively, resulting in a high coupling ratio. The top-gate dielectric engineering device exhibited repetitive synaptic weight plasticity using a voltage pulse train applied to the gate electrode with low cycle-to-cycle and cell-to-cell variations. Finally, a pattern classification accuracy of over 90% was achieved on various datasets through artificial neural network simulations using the CrossSim platform.