Ultrafast silicon photonic reservoir computing engine delivering over 200 TOPS

Area of Science:

• Photonics
• Machine Learning
• Computer Engineering

Background:

• Reservoir computing (RC) is a machine learning technique for information processing.
• Existing optical RC implementations face speed and scalability challenges due to complex connections and nonlinearities.

Purpose of the Study:

• To propose a streamlined photonic RC design using a next-generation RC paradigm.
• To overcome the limitations of traditional RC systems in optical implementations.

Main Methods:

• Developed a compact silicon photonic computing engine.
• Implemented a novel, streamlined RC design.
• Experimentally validated processing speeds and performance on machine learning tasks.

Main Results:

• Achieved an experimentally demonstrated processing speed exceeding 60 GHz.
• Demonstrated state-of-the-art performance in prediction, emulation, and classification tasks.
• The silicon photonic RC engine shows no speed limitations, a compact footprint, and high fabrication tolerance.

Conclusions:

• The proposed design represents a significant advancement for ultrafast on-chip photonic RC.
• This work paves the way for next-generation high-speed photonic computing and signal processing.