Inverse design of optical metamaterials using generative neural networks
Abstract
Optical metamaterials are artificial materials with structural features much smaller than the wavelength of light. These man-made materials exhibit exotic properties that have not been found in nature, such as negative refraction, perfect imaging, and ultra-high optical chirality. Inverse design of optical metamaterials based on generative neural networks accelerates discovery of optical metamaterials with specified functionalities. The system considers arbitrary input specifications and produces patterns of optical metamaterial that satisfy the input specifications. A full-wave simulator validates the output patterns and the trained model can be reused to achieve new design objectives. Experimental results show good agreement with specifications, demonstrating the feasibility of exploiting machine intelligence to assist the design of metamaterials.