Design of Compact Meta-Crystal Slab for General Optical Convolution

Photonic structures have great potential in creating novel computing hardware due to their fast processing speed, low energy cost, and massive parallelism. Many algorithms in image processing and neural networks rely heavily on convolution and will benefit from efficient optical implementation of convolution. Here we propose meta-crystal slabs for a general implementation of optical convolution. Through an optimization approach, we design structures to perform a wide variety of convolution kernels. In particular, we numerically demonstrate several low-order differentiation kernels with a Gaussian envelope operating in transmission in the normal direction. Our structures are very compact and compute directly on the incident image fields. Our work may lead to high-performance optical computing hardware, and it also points to the possibility of creating novel optical components with volumetric metamaterials for imaging and sensing applications.