A generative model for the inverse design of metamaterials

The advent of metamaterials in recent years has ushered in a revolutionary means to manipulate the behavior of light on the nanoscale, and thereby enabling diverse applications in optical imaging, beam steering, light modulation, dispersion engineering, holography, and many more. However, the design of such structures, to date, has relied on the expertise of an optical scientist to guide a progression of electromagnetic simulations that iteratively solve Maxwell’s equations until locally optimized solution can be attained. Here, we develop a framework leveraging a deep generative model to identify the photonic structures in arbitrary topology from a geometric dataset given target optical response. Furthermore, if black-box optimization methods are incorporated, the framework is able to automate the inverse design and generate patterns of photonic structure with few interventions of human. The evaluation shows that over 95% average accuracy can be achieved in less than 5 seconds for all the unit patterns of the nanostructure tested. Our work introduces a generic approach for the design of photonic and optical structures in response to the near-field and far-field requirements, with broad applications in large-scale photonic design requiring trial-and-error practices.