Deep Learning Accelerated Multi-Objective Optimization for Highly Performant and Mechanically Robust Nanophotonic Devices

Deep Learning has proven successful in accelerating electromagnetic simulations of complex structures thus greatly reducing the computational burden of inverse-design problems. Exploiting this acceleration allows for exhaustive sensitivity analysis of candidate designs that would otherwise be intractable to perform. When combined with multiobjective optimization, this enables a framework where meta-device performance and robustness to fabrication uncertainties can be simultaneously optimized.