Mapping Hebbian Learning Rules to Coupling Resistances for Oscillatory Neural Networks

Area of Science:

• Neuromorphic Engineering
• Materials Science
• Artificial Intelligence

Background:

• Oscillatory Neural Networks (ONNs) are an emerging neuromorphic architecture where oscillators represent neurons and information is encoded in their phase relations.
• Coupling elements in ONNs define network weights, crucial for massive parallel computation and preserving network functionality.
• Mapping these weights to physical coupling elements is vital for ONN performance and scalability.

Purpose of the Study:

• To propose a methodology for mapping Hebbian coefficients to ONN coupling resistances using relaxation oscillators based on VO2 material.
• To enable the design of large-scale ONNs by effectively translating synaptic weights into physical parameters.
• To analyze the performance of ONN architectures through an analytical framework for mapping weight coefficients to coupling resistor values.

Main Methods:

• Investigation of relaxation oscillators utilizing VO2 material for ONN implementation.
• Development of a methodology to map Hebbian coefficients to specific ONN coupling resistance values.
• Creation of an analytical framework to evaluate ONN architecture performance based on mapped weight coefficients and resistor values.

Main Results:

• Successful mapping of Hebbian coefficients to ONN coupling resistances.
• Demonstration of a large-scale ONN design methodology.
• Implementation and testing of a 60-oscillator fully-connected ONN capable of pattern recognition.

Conclusions:

• The proposed methodology effectively maps Hebbian coefficients to coupling resistances for large-scale ONN design.
• VO2-based relaxation oscillators are suitable for implementing ONNs.
• The developed 60-oscillator ONN successfully performs pattern recognition, showcasing its potential in neuromorphic computing applications.