Machine learning design of spectral-selective infrared metasurfaces based on Conway patterns

Area of Science:

• Optics and Photonics
• Materials Science
• Computational Physics

Background:

• Metasurfaces offer precise control over infrared radiation for applications like thermal management and stealth.
• Current computational methods are limited by simplistic designs and material choices, hindering performance.
• A need exists for advanced design strategies to unlock the full potential of infrared metasurfaces.

Purpose of the Study:

• To develop a machine learning-driven platform for designing band-selective infrared metasurfaces.
• To overcome limitations of existing methods by incorporating complex geometries and diverse materials.
• To enable the discovery of novel metasurface structures with enhanced optical and thermal properties.

Main Methods:

• Integration of an automated optimization framework with Conway-inspired patterns and rule-based morphologies.
• Utilizing a deep neural network for forward predictions and particle swarm optimization for inverse design.
• Employing an extensive database of 71 materials and incorporating image filtering and symmetry processing for pattern optimization.

Main Results:

• Successful generation of diverse, high-performance infrared metasurfaces with band-selective properties.
• Demonstrated superior radiative cooling and stealth capabilities within the 5-8 μm atmospheric transparency window.
• Optimized mosaic-like patterns improved fabrication feasibility and reduced polarization dependence.

Conclusions:

• The developed machine learning platform significantly broadens the design space for infrared metasurfaces.
• This approach facilitates the discovery of novel structures with multifunctional capabilities in optics and thermal management.
• The findings pave the way for next-generation infrared devices with enhanced performance and versatility.