Frequency transfer and inverse design for metasurface under multi-physics coupling by Euler latent dynamic and...

Area of Science:

• Computational physics
• Machine learning
• Materials science

Background:

• Frequency transfer is crucial for machine learning predictions beyond training data ranges.
• Traditional deep neural networks (DNNs) require specific frequency data, which is often inaccessible for multi-physics problems.
• Limitations in measurement or computation hinder data acquisition at certain frequencies.

Purpose of the Study:

• To develop a generalized deep learning framework for accurate frequency transfer in metasurface analysis.
• To enable out-of-range predictions without comprehensive frequency-specific training data.
• To incorporate an inversion method for hybrid prior information in metasurface inverse design.

Main Methods:

• A multi-physics deep learning framework (MDLF) was proposed, integrating a multi-fidelity DeepONet, a Euler latent dynamic network, and a data-analytical inversion network.
• The framework dynamically utilizes a Euler latent space and single-physics information to generalize to unseen frequency bands.
• An inversion method was introduced to incorporate hybrid a priori information for inverse design.

Main Results:

• The MDLF successfully generalized to unseen frequency bands for both parametric and free-form metasurfaces.
• The framework demonstrated effective prediction capabilities without prior knowledge of multi-physics responses.
• Numerical and experimental verification under electromagnetic-thermal coupling confirmed the MDLF's efficacy.

Conclusions:

• The proposed MDLF offers a robust solution for frequency transfer challenges in machine learning, particularly for multi-physics problems.
• This framework significantly expands the predictive capabilities of deep learning models for metasurface design.
• The study validates the MDLF's performance in real-world applications through rigorous testing.