On-chip wave chaos for photonic extreme learning

Area of Science:

• Photonics
• Artificial Intelligence
• Optical Computing

Background:

• Growing demand for scalable, energy-efficient artificial neural networks drives research into novel hardware.
• Integrated photonics provides a compact, parallel, and ultra-fast platform suitable for extreme learning machine (ELM) architectures.

Purpose of the Study:

• To experimentally demonstrate a chip-scale photonic ELM.
• To leverage wave chaos interference in a stadium microcavity for information processing.
• To showcase the system's adaptability for different computational tasks.

Main Methods:

• Input information encoded via wavelength of a tunable laser source.
• Fabrication of a stadium microcavity using direct laser writing (SU-8 polymer on glass).
• Utilized a surrounding scattering wall as a readout layer for leaky modes.

Main Results:

• Observed uncorrelated and aperiodic speckle patterns from the scattering barrier.
• Demonstrated successful classification performance on three benchmark tasks.
• Showcased tunable output nodes by adjusting readout regions of the scattering barrier.

Conclusions:

• The photonic ELM based on wave chaos is a viable hardware solution for AI.
• The system's performance can be optimized by controlling the readout size.
• This approach offers a promising direction for energy-efficient optical computing.