Metaplastic and energy-efficient biocompatible graphene artificial synaptic transistors for enhanced accuracy...

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Traditional CMOS computing faces limitations in parallel processing and energy efficiency compared to the human brain.
• Existing neuromorphic devices often lack biocompatibility or use toxic materials.
• Artificial synapses are crucial for developing brain-inspired computing.

Purpose of the Study:

• To develop biocompatible artificial synaptic transistors using bilayer graphene.
• To investigate the synaptic behavior, energy efficiency, and metaplasticity of these devices.
• To assess their potential for bio-interfaced online learning and deep neural networks.

Main Methods:

• Fabrication of bilayer graphene-based artificial synaptic transistors (BLAST) utilizing a dry ion-selective membrane.
• Characterization of long-term potentiation and switching energy efficiency.
• Evaluation of metaplasticity and performance in image classification tasks.

Main Results:

• BLAST devices demonstrated synaptic behavior with high energy efficiency (~50 aJ/µm²), significantly lower than previous 2D material synapses.
• Unique metaplasticity was observed, beneficial for generalizable deep neural networks.
• Metaplastic BLASTs outperformed ideal linear synapses in image classification.

Conclusions:

• Biocompatible BLASTs offer a promising solution for energy-efficient artificial synapses.
• These devices bridge the gap between artificial and biological neural networks for bio-interfaced online learning.
• The demonstrated metaplasticity enhances their suitability for advanced AI applications.