Experimental control of mode-competition dynamics in a chaotic multimode semiconductor laser for decision making

Area of Science:

• Optics and Photonics
• Machine Learning
• Computational Science

Background:

• Photonic computing accelerates machine learning, with photonic decision-making showing promise for reinforcement learning problems.
• Solving multi-armed bandit problems using chaotic mode-competition dynamics in lasers is a proposed but experimentally unoptimized approach.

Purpose of the Study:

• To experimentally investigate and establish optimal conditions for chaotic mode-competition dynamics in semiconductor lasers for superior decision-making performance.
• To understand how optical feedback and injection influence mode-competition dynamics and laser behavior.

Main Methods:

• Experimentally controlling chaotic mode-competition dynamics in a multimode semiconductor laser using optical injection and feedback.
• Analyzing laser dynamics through two-dimensional bifurcation diagrams of total intensity.
• Implementing decision-making experiments to solve the multi-armed bandit problem.

Main Results:

• Positive wavelength detuning with low optical injection power efficiently concentrates laser modes.
• Complex mixed dynamics were observed under optical feedback and injection.
• Fast mode concentration at positive detunings led to rapid convergence in decision-making accuracy.

Conclusions:

• Optimizing chaotic mode-competition dynamics via optical injection, particularly with positive wavelength detuning, enhances decision-making performance.
• This research provides a pathway for accelerating decision-making in adaptive optical networks using reinforcement learning principles.