Deep Learning-based Inverse Design Of Microstructured Materials For Optical Optimization And Thermal Radiation Control

This article presents a novel model that takes in optical properties across a wavelength range and outputs a material and micropyramid geometry that best match it. Multiple deep learning methods are utilized in the construction of the method, including a deep neural network surrogate and an image-based surrogate. The output of the inverse neural network is put through a post-processing stage where user set geometric and material constraints are used to produce appropriate solutions. The novel generated material properties are matched to a material from a library material and the constrained output is simulated. This process enables rapid optimization of a material and geometric combination for a desired optical spectrum.