On-Chip Integrated Atomically Thin 2D Material Heater as a Training Accelerator for an Electrochemical Random-Access Memory Synapse for Neuromorphic Computing Application

Abstract

An artificial synapse based on oxygen-ion-driven electrochemical random-access memory (O-ECRAM) devices is a promising candidate for building neural networks embodied in neuromorphic hardware. However, achieving commercial-level learning accuracy in O-ECRAM synapses, analog conductance tuning at fast speed, and multibit storage capacity is challenging because of the lack of Joule heating, which restricts O2– ionic transport. Here, we propose the use of an atomically thin heater of monolayer graphene as a low-power heating source for O-ECRAM to increase thermally activated O2– migration within channel-electrolyte layers. Heating from graphene manipulates the electrolyte activation energy to establish and maintain discrete analog states in the O-ECRAM channel. Benefiting from the integrated graphene heater, the O-ECRAM features long retention (>104 s), good stability (switching accuracy <98% for >103 training pulses), multilevel analog states for 6-bit analog weight storage with near-ideal linear switching, and 95% pattern-identification accuracy. The findings demonstrate the usefulness of 2D materials as integrated heating elements in artificial synapse chips to accelerate neuromorphic computation.

Graphene Nanocomposite Materials for Supercapacitor Electrodes

“…The history of graphene and its relatives, with attention to the specific properties, advantages, as well as disadvantages for various possible applications beyond electrochemical energy storage and conversion, has been reviewed repeatedly [4][5][6][7][8][14][15][16][17][18]38]. In reviews, the use of graphene in numerous applications beyond energy storage and electrochemistry has been addressed [39][40][41][42][43].…”

Learing Sampling and Reconstruction Using Bregman Iteration for CS-MRI

“…In recent years, with the rapid development of deep learning, a series of learning-based models have been proposed [24][25][26][27][28][29][30]. DAGAN [30] features a sampling and full-image denoising network framework based on generative adversarial network (GAN [18]), which improves the performance of compressed perception image reconstruction by introducing adversarial loss and perceived loss.…”

Novel Light-Matter Interactions in 2D Magnets

“…For example, optical excitation on the monolayer TMDs will generate a strongly bound electron-hole pair called an exciton, instead of fee electrons and holes as in traditional bulk semiconductors [6]. Moreover, their extraordinary mechanical property makes these atomic thin materials with breakthrough of integration technologies accelerate the on-chip device applications [7].…”