Compact reservoir computing with a photonic integrated circuit.

Area of Science:

• Optics and Photonics
• Computational Science
• Information Technology

Background:

• Photonic reservoir computing offers efficient high-speed prediction and classification.
• Miniaturization of photonic reservoir computing requires photonic integrated circuits (PICs).
• Current challenges include increasing computational power and memory effects in PICs.

Purpose of the Study:

• To experimentally demonstrate reservoir computing using a PIC with a semiconductor laser and external cavity.
• To propose and validate a method for increasing virtual nodes in delayed feedback systems.
• To evaluate the performance of the PIC-based reservoir computing for time-series prediction and channel equalization.

Main Methods:

• Developed a photonic integrated circuit incorporating a semiconductor laser and short external cavity.
• Implemented a novel method to increase virtual nodes using short node intervals and multi-delay outputs.
• Performed time-series prediction and nonlinear channel equalization tasks, comparing systems with and without optical feedback.

Main Results:

• Successfully demonstrated reservoir computing on a PIC.
• The PIC with optical feedback significantly outperformed the system without feedback for prediction tasks.
• Incorporating past input signals into current data enhanced the memory effect, enabling successful n-step-ahead prediction.

Conclusions:

• Photonic reservoir computing is feasible and effective using compact PICs.
• The proposed method enhances computational capacity and memory effects in photonic systems.
• This work paves the way for miniaturized, high-performance photonic computing devices.