Time delay reservoir computing with a silicon microring resonator and a fiber-based optical feedback loop.

Area of Science:

• Integrated photonics
• Nonlinear optics
• Neuromorphic computing

Background:

• Silicon microring resonators are key for photonic neural networks due to their size, CMOS compatibility, and nonlinear dynamics.
• Their filtering and nonlinear properties can be used for weighting and activation functions, respectively.

Purpose of the Study:

• Investigate the linear and nonlinear dynamics of microring resonators for time delay reservoir computing using optical feedback.
• Evaluate the computational capacity and performance of this system on benchmark tasks.

Main Methods:

• Implemented an external optical feedback loop with silicon microring resonators.
• Mitigated environmental noise effects on feedback phase dependencies.
• Assessed system performance on delayed-boolean tasks, Santa Fe, and Mackey Glass prediction tasks at Mbps bit rates.

Main Results:

• Optical feedback is crucial for error-free operation in tasks requiring up to 3 bits of memory, using the microring in the linear regime with photodetection as the nonlinear activation.
• Microring nonlinearities enable solving Santa Fe and Mackey Glass prediction tasks.
• Achieved competitive results with as few as 7 virtual nodes.

Conclusions:

• Silicon microring resonators demonstrate versatile performance for time delay reservoir computing with optical feedback.
• The system can be tailored for diverse computational applications by leveraging linear or nonlinear dynamics.