Inverse design and spectral reconstruction of computational multispectral metasurfaces using deep learning and...

Area of Science:

• Optics and Photonics
• Materials Science
• Computational Science

Background:

• Designing computational multi-spectral metasurfaces for the long-wave infrared (LWIR) spectrum is challenging due to the difficulty in achieving minimally correlated transmission spectra.
• Traditional methods rely on predefined materials and geometries, limiting spectral response optimization.
• Gallium antimonide (GaSb) is a promising material for LWIR applications due to its low optical loss, high refractive index, and compatibility with semiconductor fabrication.

Purpose of the Study:

• To propose and validate an end-to-end inverse design framework for optimizing gallium antimonide (GaSb)-based photonic crystal metasurfaces in the LWIR region.
• To investigate the application of hybrid deep learning and genetic algorithms for designing metasurfaces with specific spectral features.
• To enable rapid prototyping of compact and efficient computational multispectral metasurface systems.

Main Methods:

• An end-to-end inverse design framework combining deep neural networks and genetic algorithms was developed.
• Deep neural networks were used to encode geometric parameters of photonic crystals for desired structural features.
• Genetic algorithms selected low-correlation spectra, which were fed into the inverse model to predict corresponding geometries.

Main Results:

• The framework successfully optimized GaSb-based photonic crystal structures for LWIR metasurfaces.
• The inverse design model achieved a mean square error (MSE) on the order of 10^-3 in spectral response reconstruction.
• The developed method offers a rapid prototyping solution for computational multispectral metasurface systems.

Conclusions:

• The proposed hybrid deep learning and genetic algorithm framework provides an efficient method for designing LWIR metasurfaces.
• This approach significantly advances the development of compact and efficient computational multispectral metasurface systems.
• The framework demonstrates potential for resource-constrained scenarios in metasurface design and fabrication.