Quantum imaging of the reconfigurable VO₂ synaptic electronics for neuromorphic computing.

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Neuromorphic computing leverages silicon-based artificial intelligence for advanced AI capabilities.
• Current research often focuses on two-terminal artificial synapses integrated with silicon circuits, posing integration challenges.
• Mott materials offer potential for optimizing functional synaptic connections in artificial intelligence.

Purpose of the Study:

• To propose a dynamic network architecture for neuromorphic computing.
• To utilize laser-controlled conducting filaments for simulating synaptic connections.
• To overcome the limitations of traditional two-terminal artificial synapses and silicon-based circuit integration.

Main Methods:

• Employing electric field-induced insulator-to-metal transition in vanadium dioxide (VO2).
• Utilizing focused laser manipulation to control conducting filament formation.
• Implementing quantum sensing for conductivity-sensitive imaging of filaments.

Main Results:

• Demonstrated laser-controlled manipulation of filament formation location.
• Successfully simulated dynamical synaptic connections between neurons.
• Achieved signal processing with both long-term and short-term potentiation.
• Observed a ~60 times on/off ratio in pathway switching.

Conclusions:

• The proposed dynamic network offers a novel approach to neuromorphic computing.
• Laser-controlled conduction pathways provide a flexible method for mimicking biological neural systems.
• This research facilitates the development of advanced, adaptable artificial neural networks.