BP-based supervised learning algorithm for multilayer photonic spiking neural network and hardware implementation.

Area of Science:

• Optoelectronics
• Artificial Intelligence
• Neuromorphic Computing

Background:

• Photonic spiking neural networks (SNNs) offer potential for low-power, high-speed computation.
• Implementing efficient learning algorithms directly on photonic hardware remains a challenge.

Purpose of the Study:

• To introduce a novel supervised learning algorithm for photonic SNNs.
• To demonstrate the algorithm's effectiveness in classification tasks both numerically and experimentally.
• To enable hardware-algorithm collaborative computing for photonic neural networks.

Main Methods:

• Developed a supervised learning algorithm based on backpropagation for photonic SNNs.
• Encoded information into spike trains with varying strengths for training.
• Utilized vertical-cavity surface-emitting laser (VCSEL) based photonic spiking neurons, functionally similar to leaky-integrate-and-fire neurons.
• Performed numerical and experimental classification tasks.

Main Results:

• Successfully demonstrated the supervised learning algorithm on a photonic SNN.
• Achieved classification task performance numerically and experimentally.
• Validated the hardware implementation of the proposed learning algorithm.

Conclusions:

• The developed supervised learning algorithm is suitable for photonic SNNs.
• Hardware implementation of photonic neural networks with efficient learning algorithms is feasible.
• This work contributes to ultra-low power consumption and delay in photonic neural network computing.