Lithographic nanofabrication processes have enabled the realization of exciting optical metasurfaces and metalens technologies. Their ability to resolve structures at single-digit nanometer scale have placed large-scale optical metasurfaces in reach for the first time. However, as with all fabrication methods, uncertainties in the product (arising from variations in the fabrication process) limit the geometrical accuracy of the resulting devices—an effect which can be quite detrimental to metasurface performance. To counter this uncertainty from the design side, we developed a Deep Learning model to predict the change in performance expected by variations in fabrication process, allowing high-volume robustness testing of metasurface designs within an optimization procedure for the first time. Techniques like this one pave the way toward wafer-scale optical metasurfaces and metalenses, one of the holy grails of optical and nanophotonic design.

The rise and proliferation of artificial intelligence (AI) has the potential to influence and disrupt many aspects of society as we currently know it. Outside of the Antennas and Propagation Society (AP-S), the main AI applications are numerous, including image recognition, classification, and segmentation, natural language processing, drug discovery, fraud detection, and recommendation systems, to name a few. While the true utility of machine learning (ML) to the AP-S community is still being discovered, 2019 was a breakthrough year for the publication of AI-related techniques in the AP-S community and this field has only grown since then. CEARL researchers are at the forefront of applied AI and Deep Learning (DL) research in the antennas, and greater electromagnetics and optical communities.