Oscillatory Neural Networks Using VO₂ Based Phase Encoded Logic.

Area of Science:

• Non-conventional computing paradigms
• Materials science for computing
• Nano-oscillators and their applications

Background:

• Phase-transition materials like vanadium dioxide (VO2) are key for novel computing architectures.
• Autonomous non-linear oscillators are fundamental components for oscillatory neural networks (ONNs).
• Existing ONN architectures face challenges with variability in VO2 devices.

Purpose of the Study:

• To propose a new architecture for ONNs utilizing VO2 nano-oscillators.
• To enhance the robustness and variability tolerance of ONNs.
• To implement a novel weight coding scheme for improved performance.

Main Methods:

• Exploiting sub-harmonic injection locking (SHIL) for two-phase encoding in neurons.
• Developing a new interconnection scheme for coding positive and negative weights.
• Utilizing static inverting and non-inverting logic for weight representation.

Main Results:

• Demonstrated successful implementation of ONNs with inherent two-phase value encoding.
• Showcased advantages in robustness and tolerance to VO2 device variability.
• Validated the proposed architecture's operation for pattern recognition tasks.

Conclusions:

• The proposed SHIL-based architecture offers a robust and efficient method for implementing ONNs.
• This approach mitigates variability issues inherent in VO2 devices.
• The architecture shows promise for practical applications in pattern recognition.