Towards a Bispectral Meta-Cavity with Generative Models

Metasurfaces acting as mirrors in an optical cavity could not only achieve unprecedented frequency stability due to their low thermal noise, but their unique control over light properties also has the potential to enable advanced functionalities such as multiwavelength operation. The case of two wavelengths that can be stabilized in a single cavity necessitates advanced topology optimization of the individual metaatoms. We report on Wasserstein Generative Adversarial Networks (WGANs) and Diffusion models that operate on a quasi-freeform pixel space. To the best of our knowledge, this is the first study showcasing a benchmark of these two state-of-the-art generative models for metamirror optimization.