Programmable electronic synapse and nonvolatile resistive switches using MoS2 quantum dots

Anna Thomas,A N Resmi,K B Jinesh,Akash Ganguly

doi:10.1038/s41598-020-68822-5

Anna Thomas, A N Resmi + Show 2 more

Open Access

https://doi.org/10.1038/s41598-020-68822-5

Copy DOI

Abstract

Brain-inspired computation that mimics the coordinated functioning of neural networks through multitudes of synaptic connections is deemed to be the future of computation to overcome the classical von Neumann bottleneck. The future artificial intelligence circuits require scalable electronic synapse (e-synapses) with very high bit densities and operational speeds. In this respect, nanostructures of two-dimensional materials serve the purpose and offer the scalability of the devices in lateral and vertical dimensions. In this work, we report the nonvolatile bipolar resistive switching and neuromorphic behavior of molybdenum disulfide (MoS2) quantum dots (QD) synthesized using liquid-phase exfoliation method. The ReRAM devices exhibit good resistive switching with an On–Off ratio of 104, with excellent endurance and data retention at a smaller read voltage as compared to the existing MoS2 based memory devices. Besides, we have demonstrated the e-synapse based on MoS2 QD. Similar to our biological synapse, Paired Pulse Facilitation / Depression of short-term memory has been observed in these MoS2 QD based e-synapse devices. This work suggests that MoS2 QD has potential applications in ultra-high-density storage as well as artificial intelligence circuitry in a cost-effective way.

Highlights

Brain-inspired computation that mimics the coordinated functioning of neural networks through multitudes of synaptic connections is deemed to be the future of computation to overcome the classical von Neumann bottleneck
As we demonstrate in this manuscript, MoS2 quantum dots (QD) act as excellent candidates for e-synapse, and their synaptic responses such as paired-pulse facilitation (PPF) and depression (PPD) have been studied here as a function of various control parameters such as duty cycle, On and Off duration and frequency of the action potentials
Scanning tunneling Microscopic (STM) analysis of the QDs has been done by dispersing the QD solution on highly oriented pyrolytic graphite (HOPG) substrate

Summary

Introduction

Brain-inspired computation that mimics the coordinated functioning of neural networks through multitudes of synaptic connections is deemed to be the future of computation to overcome the classical von Neumann bottleneck. This inherent limit imposed by the computational architecture is called von Neumann bottleneck, and it is obvious that establishing alternative approaches is essential to cope up with the computational speeds required for future technologies One such promising alternative is the computation based on artificial neural networks, which mimics how the biological brain functions. Artificial synaptic systems inspired by the unique properties of the biological neural systems offer the possibilities of parallel computations with low power consumption and self- learning ability and have the potential to overcome the von Neumann bottleneck, the classical complementary metal–oxide–semiconductor (CMOS) technology currently faces[3]. Several combinations of M oS2 have been reported for memory applications, which include MoS2-graphene ultra-thin stack that produces a hysteresis in the transistor behavior due to charge trapping, MoS2 embedded in polymethyl methacrylate (PMMA) matrix, which exhibits quantum conductance due to Coulomb blockade effect etc[29]

Methods

Results

Conclusion