Speaker:
Kudyshev, Zhaxylyk

Abstract: With the recent development of new plasmonic/photonic materials and nano-fabrication techniques, nanophotonic devices are now capable of providing novel solutions to global challenges, including world energy consumption, rapid and accurate chemical/biological detection, quantum computing/security, telecom information densities, and space exploration. These problems are inherently complex due to their multi-disciplinary nature, requiring a manifold of stringent constraints in conjunction with the optical performance. Optimization techniques offer powerful tools to address these multi-faceted challenges. Predominantly, topology optimization has emerged as a successful architect for the systematic design of non-intuitive photonic structures. However, this technique requires substantial computational resources that limit its applicability to highly constrained optimization problems within high dimensional parametric space. We show that hybridization of topology optimization method with adversarial autoencoders can deliver substantial improvement of optimization search by providing unparalleled control of the compact design space representations. By enabling efficient global optimization searches within complex landscapes, the proposed compact hyperparametric representations will become essential for multiconstrained problems. The proposed approach can enable a broader scope of the optimal design and materials synthesis applications that go beyond the photonics and optoelectronics.