Machine learning enabled prediction of meta-atom robustness to fabrication error (Conference Presentation)

Silicon metasurfaces are gaining broad interest due to their compact packaging, scalable fabrication via widely-available CMOS-compatible techniques, and novel performance capabilities for communication systems, sensors, and beyond. However, practical realization of metasurfaces is challenging due to the competing demands of selecting meta-atoms that simultaneously provide the required amplitude and phase modulation while being robust to changes due to inevitable fabrication errors. To address this challenge, we use a combination of forward design principles and machine learning to characterize the fabrication sensitivity of individual meta-atoms and refine their selection. By modeling error tolerance we provide an initial assessment of manufacturing tolerance criteria for metasurface design and insights on how to integrate them into the design process.