A complete photonic integrated neuron for nonlinear all-optical computing.

Area of Science:

• Photonics
• Artificial Intelligence
• Integrated Photonics

Background:

• Photonic neural networks offer potential for ultrafast AI inference and improved energy efficiency.
• Achieving nonlinearity-complete all-optical neurons remains a significant challenge, limiting current photonic neural network performance.

Purpose of the Study:

• To report a complete photonic integrated neuron (PIN) with spatiotemporal feature learning and reconfigurable structures.
• To enable nonlinear all-optical computing beyond current limitations.

Main Methods:

• Interleaving the spatiotemporal dimension of photons and utilizing the Kerr effect.
• Monolithic integration on a silicon-nitride photonic chip for high-order temporal convolution and all-optical nonlinear activation.
• Developing a PIN chip system for demonstrating capabilities.

Main Results:

• Achieved neuron completeness with weighted interconnects and nonlinearities.
• Demonstrated high-accuracy image classification and human motion generation.
• Enabled ultrafast spatiotemporal processing with latency as low as 240 ps.

Conclusions:

• The developed PIN represents a significant advancement in all-optical computing.
• This technology paves the way for machine intelligence operating in the sub-nanosecond regime.
• PIN technology addresses key challenges in photonic neural network performance and scalability.