Robustness Optimization of Nanophotonic Devices Using Deep Learning

Abstract

Realizing state-of-the-art metasurfaces depends on meeting strict geometric tolerances due to their inherent sensitivity to structural variations. A design may have extremely good performance in simulation which is lost when undergoing fabrication. We present how a Deep Learning-augmented multiobjective optimization method can be used for designing metasurfaces which are robust to a common type of manufacturing defect, namely erosion and dilation.