Accelerated Terahertz Metasurface Design with Deep Learning

Abstract

There has been a flurry of recent research on all-dielectric metasurfaces (ADMs), and demonstrations of exotic scattering responses have been shown. Dielectric metasurfaces – like metal-based metamaterials – are scalable with size, are based on geometry rather than chemical makeup, but lack some of their drawbacks. For example, ADMs do not possess Ohmic losses, can have high melting temperatures, and possess low thermal conductivities, all of which make them great candidates for potential thermal applications. <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1, 2</sup> However, ADMs possess a significant evanescent field, making it difficult to design super-unit cell combinations, owing to strong non-linear unit cell interactions. There is currently lack of a complete theory of super unit cell combinations of ADMs, and simulation combined with conventional optimization techniques are unable to efficiently identify potential solutions.