Inverse design of soliton microcomb based on genetic algorithm and deep learning

Area of Science:

• Photonics and optical engineering.
• Nonlinear optics.
• Computational physics.

Background:

• Soliton microcombs offer high coherence and stability for diverse applications.
• Conventional design methods rely on slow, iterative electromagnetic simulations.
• Systematic inverse design from microcomb properties to microcavity geometry remains a challenge.

Purpose of the Study:

• To develop a high-accuracy inverse design method for soliton microcombs.
• To optimize dispersive wave position and power by determining microcavity geometry.
• To overcome limitations of conventional dispersion engineering techniques.

Main Methods:

• Combined genetic algorithm (GA) and deep neural network (DNN) for inverse design.
• Utilized Lugiato-Lefever equation and GA (LLE-GA) for dispersion extraction.
• Employed a pre-trained forward DNN with GA (FDNN-GA) for geometry optimization.

Main Results:

• Achieved precise control over dispersive wave position and power.
• Demonstrated low deviations (<0.5% for position, <5 dB for power) in MgF₂ and Si₃N₄ microresonators.
• Validated the method's versatility across different materials and structures.

Conclusions:

• The proposed inverse design method is effective and versatile for various microresonator materials.
• This approach significantly advances the design process for soliton microcombs.
• Enables efficient optimization of microcavity geometry for desired microcomb characteristics.