Integrated photonic reservoir computing with an all-optical readout.

Area of Science:

• Photonics
• Neuromorphic Computing
• Integrated Optics

Background:

• Photonic reservoir computing leverages neural network principles for complex problem-solving.
• Its key advantage lies in a simple readout system, enabling rapid training and robust hardware implementation.
• This facilitates real-time processing with integrated photonic circuits.

Purpose of the Study:

• To develop a fully-optical, coherent photonic reservoir chip with an integrated optical readout system.
• To demonstrate an effective weight update strategy compatible with integrated optical readout hardware.
• To showcase the system's capability for real-time, high-speed data processing tasks.

Main Methods:

• Development of a coherent photonic reservoir chip.
• Integration of a straightforward optical readout system.
• Implementation of an online training scheme for weight updates.
• Real-time execution of 3-bit header recognition and delayed XOR tasks.

Main Results:

• Successful demonstration of a fully-optical photonic reservoir computing system.
• Achieved real-time processing of 3-bit header recognition and delayed XOR tasks at 20 Gbps.
• Validated the effectiveness of the online training scheme with integrated optical readout.
• Confirmed fabrication variation-insensitivity and robustness of the photonic hardware.

Conclusions:

• The developed photonic reservoir chip offers a powerful platform for real-time neuromorphic computing.
• The integrated optical readout system simplifies training and enhances hardware robustness.
• This approach enables high-speed, all-optical solutions for complex computational tasks.