Structural colors designed by deep learning

Abstract

Structural colors can be generated by metasurfaces with the capability of spectrum manipulation at subwavelength. In general, the optimization of specific color designs and iterative geometric parameters is computationally time-consuming, so obtaining thousands of different structural colors can be challenging. Deep learning methods offer a new approach to the efficient design of nanophotonic devices, as it revolutionizes the way nanophotonic devices are designed. Here, we trained a deep learning method, which can predict the colors by random geometries in the forward modeling process. The forward design model is good at processing data, which is an effective way to design nanophotonic devices.