Synaptic delay plasticity based on frequency-switched VCSELs for optical delay-weight spiking neural networks

Area of Science:

• Photonics and Artificial Intelligence
• Neuromorphic Engineering
• Optical Computing

Background:

• Spiking neural networks (SNNs) offer energy-efficient computation.
• Implementing delay-weighted SNNs optically presents challenges in precise delay control.
• Vertical-cavity surface-emitting lasers (VCSELs) are promising optoelectronic components.

Purpose of the Study:

• To propose a novel optical delay-weight SNN architecture.
• To investigate the synaptic delay plasticity of frequency-switched VCSELs.
• To demonstrate the SNN's capability in pattern recognition tasks.

Main Methods:

• Cascaded frequency and intensity-switched VCSELs for SNN construction.
• Numerical analysis and simulations to study synaptic delay plasticity.
• Implementation of a two-layer SNN using a delay-weight supervised learning algorithm.
• Testing on spiking sequence pattern training and Iris dataset classification.

Main Results:

• Demonstrated tunable spiking delay up to 60 ns.
• Identified key factors influencing delay manipulation in VCSELs.
• Successfully applied the optical SNN to training and classification tasks.
• Achieved delay-weighted computing without external delay lines.

Conclusions:

• The proposed optical SNN architecture is compact and cost-efficient.
• Frequency and intensity-switched VCSELs enable effective delay-weighted computation.
• This approach offers a viable solution for advanced neuromorphic hardware.