Inverse-designed metasurfaces for multifunctional spatial frequency filtering

With the rapid rise in demand for edge computing, there is a need for low-power, compact approaches to information processing. Optical metasurfaces have emerged as a powerful platform for implementing convolution operations in the analog domain that would traditionally be done digitally. However, there has been limited work exploring the full extent to which metasurfaces can be designed to filter the spatial frequency content of incident fields. We experimentally demonstrate inverse-designed metasurfaces that perform spatial frequency filtering by precisely controlling their angular, polarization, and spectral scattering characteristics. To demonstrate the flexibility of our design approach, we show that distinct transfer functions can be implemented for orthogonal polarizations, leading to directional edge detection and blurring. Furthermore, we design metasurfaces with polarization-independent and spectrally multiplexed transfer functions. Multifunctional metasurfaces that encode several filtering functions have the potential to enable new applications in robotics, metrology, and sensing, particularly in resource-constrained scenarios.