Machine learning assisted inverse design of microresonators

Area of Science:

• Photonics and Optical Engineering
• Materials Science
• Computational Physics

Background:

• Microresonators are crucial for optical applications, requiring precise geometry control for desired properties.
• Dispersion significantly impacts optical nonlinearities and intracavity dynamics in microresonators.
• Current methods for optimizing microresonator geometry are often complex and time-consuming.

Purpose of the Study:

• To develop and validate a machine learning (ML) approach for determining microresonator geometry from dispersion profiles.
• To compare the performance of different ML algorithms for this inverse design problem.
• To assess the accuracy and experimental feasibility of the ML-based method.

Main Methods:

• Generating a dataset of approximately 460 microresonator samples using finite element simulations.
• Training and evaluating two ML algorithms, including Random Forest, with hyperparameter tuning.
• Experimentally verifying the best-performing ML model using integrated silicon nitride microresonators.

Main Results:

• The Random Forest algorithm demonstrated superior performance in predicting microresonator geometry from dispersion data.
• The ML model achieved an average error well below 15% on simulated data.
• Experimental validation confirmed the model's effectiveness in real-world integrated photonic devices.

Conclusions:

• Machine learning offers an efficient and accurate tool for the inverse design of microresonators based on their optical dispersion.
• The developed ML approach can accelerate the fabrication process of microresonators with tailored optical characteristics.
• This study highlights the potential of ML in optimizing photonic device design and performance.