Dual-Mode Neuromorphic Device Based on MoS₂ 2T0C DRAM for Neurons and Synapses

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Artificial neural networks (ANNs) show promise for low-power computation by mimicking biological systems.
• Hardware implementation of ANNs faces challenges in integrating synaptic and neuronal functions with high density and power efficiency.
• Bio-inspired dynamics are crucial for advanced neuromorphic computing.

Purpose of the Study:

• To propose and demonstrate wafer-scale MoS2 2T0C devices for neuromorphic computing applications.
• To enable stable switching between volatile and quasi-non-volatile modes for synaptic and neuronal emulation.
• To achieve high performance in terms of retention time, linearity, speed, and energy efficiency.

Main Methods:

• Fabrication of wafer-scale Molybdenum disulfide (MoS2) 2-transistor-0-capacitor (2T0C) devices.
• Characterization of device performance as both artificial synapses and neurons.
• Construction of a spiking neural network (SNN) using an array of these dual-mode devices.

Main Results:

• The MoS2 2T0C device demonstrated synaptic functionality with >40s retention time, high linearity, 500ns programming speed, and 6pJ write energy.
• As a neuron, the device exhibited leaky-integrate-and-fire (LIF) behavior with 15pJ spike energy consumption.
• The SNN model achieved 92.88% recognition accuracy on the MNIST dataset after 100 training epochs.

Conclusions:

• Wafer-scale MoS2 2T0C devices offer a promising platform for efficient, high-density neuromorphic hardware.
• The dual-mode capability of these devices effectively emulates both synaptic and neuronal functions.
• This technology enables the development of advanced spiking neural networks for complex computational tasks.