Optimization of metasurfaces under geometrical uncertainty using statistical learning

The performance of metasurfaces measured experimentally often discords with expected values from numerical optimization. These discrepancies are attributed to the poor tolerance metasurface building blocks respect fabrication uncertainties and nanoscale imperfections. Quantifying their efficiency drop according geometry variation crucial improve range application this technology. Here, we present a novel optimization methodology account for manufacturing errors related designs. In approach, accurate results using probabilistic surrogate models used reduce number costly simulations. We employ our procedure optimize classical beam steering made cylindrical nanopillars. Our yield design that is twice more robust compared deterministic case.