Experimental and theoretical evidence of ion engineering in nanocrystalline molybdenum disulfide memristors for non-filamentary switching actions and ultra-low-voltage synaptic features

Abstract

The human brain comprises 10¹⁵ synapses and consumes only 20 W of power, where a single synaptic function requires an activation potential of around 100 mV. These facts provide inspiration for current memristive technology to reach the ubiquitous goal of creating neuromorphic electronics. Herein, it is shown that Na-modified nanocrystalline molybdenum disulfide (MoS₂) flexible memristors can provide great feasibility in ultra-low-voltage switching operations as well as synaptic actions. Excellent switching features, including the spike-time- and spike-rate-dependent plasticity, as well as long-term potentiation/depression are demonstrated within 500 mV. The device parameters are compared with the reported values to establish its unique adaptive behavior at such a low stimulus, like biological synapses. Furthermore, an artificial network is constructed based on the device parameters and revealed a stunning learning accuracy of 94% for the MNIST handwritten datasets. Our collective experimental and DFT-based first-principles investigation unveils a novel switching mechanism driven by Na diffusion at the vdW gaps, which endorses intriguing homogenous switching features (stochastic catastrophe-free) and biomimetic actions at a notably adjusted voltage (an order of magnitude lower) compared with conventional filamentary memristors. Therefore, this work shows an encouraging avenue for future adaptable electronics to design reliable and energy-efficient cognitive hardware.