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

Currently, the key challenge in designing computational multi-spectral metasurfaces in the long-wave infrared (LWIR) region mainly lies in identifying structures with minimally correlated transmission spectra. Traditionally, predefined materials and geometries have been relied on to generate spectral responses. Despite recent advancements in inverse design methodologies, the applications of hybrid deep learning combined with genetic algorithms in gallium antimonide (GaSb)-based photonic crystal structures have not been sufficiently investigated. GaSb possesses low optical loss and a high refractive index in the long-wave infrared region, and is compatible with III–V semiconductor fabrication techniques, enabling integration with existing optoelectronic components. Therefore, we propose an end-to-end inverse design framework using deep neural networks to optimize the photonic crystal design, which encodes the geometric parameters of the photonic crystal to identify metasurfaces using the desired structural features. Low-correlation spectra are selected via a genetic algorithm and fed into the inverse model to predict the corresponding geometries. The spectral response is then reconstructed, achieving a mean square error (MSE) on the order of 10 −3 . This rapid prototyping solution, especially in resource-constrained scenarios, can mark a significant step forward in the development of the compact and efficient computational multispectral metasurface system.