AgPS3-Based Electrical Synapses with Tunable Multilevel Conductance for Noise-Robust Image Classification

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Neuromorphic computing requires advanced artificial synaptic electronics for brain-inspired systems.
• Memristive devices with controllable conductance are crucial for neuromorphic applications.
• Novel functional materials are needed to achieve high-performance synaptic devices.

Purpose of the Study:

• To introduce layered silver phosphorus trisulfide (AgPS3) as a novel material for artificial synaptic devices.
• To investigate the neuromorphic application potential of AgPS3-based synaptic devices.
• To evaluate the performance of AgPS3 synapses in image processing tasks.

Main Methods:

• Fabrication of synaptic devices using layered AgPS3.
• Characterization of device conductance modulation under various electrical stimulation protocols.
• Design and implementation of a AgPS3 synapse-based Gaussian front-end for medical image preprocessing.
• Performance evaluation using BloodMNIST dataset and convolutional neural networks (CNNs).

Main Results:

• AgPS3 devices exhibit arbitrarily tunable multilevel conductance states.
• The devices successfully emulate biological synaptic functions, including stimulus-dependent weight update, memory transition, and frequency-selective signal processing.
• A AgPS3 synapse-based Gaussian front-end significantly enhances noise robustness in BloodMNIST classification via CNNs.

Conclusions:

• Layered AgPS3 is a promising candidate material for developing artificial synapses.
• AgPS3-based synaptic devices offer great prospects for edge-intelligent sensing and neuromorphic computing.
• The material shows potential for hardware-efficient image processing applications.