High-Density Reconfigurable Synaptic Transistors Targeting a Minimalist Neural Network

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Neuromorphic computing requires numerous synaptic devices for efficient neural networks.
• Improving energy efficiency necessitates reducing the number of synaptic devices through multi-state functionalities.

Purpose of the Study:

• To introduce a novel high-density synaptic (HDS) device capable of multiple nonvolatile synaptic states.
• To demonstrate the potential of photoisomerism materials for advanced neuromorphic hardware.

Main Methods:

• Utilized diarylethene, a photoisomerism material, to construct the HDS device.
• Engineered synaptic states through UV-vis light regulation, ensuring nonvolatility and reversibility.
• Mimicked comprehensive synaptic characteristics in each state.

Main Results:

• Successfully created a powerful HDS device with intrinsically converted, nonvolatile synaptic states.
• Demonstrated reconfigurable and reversible state conversion under varying light conditions.
• Achieved a 16-fold reduction in device count for a multilayer perceptron (MLP) architecture.

Conclusions:

• The developed HDS device offers a revolutionary approach to minimalist neural computing structures.
• This innovation paves the way for highly efficient, brain-like artificial neural networks.
• Diarylethene-based HDS devices represent a significant advancement in neuromorphic engineering.