Reconfigurable Optical Image Processor Unit via Phase‐Programmed On‐Chip Metalens

Image processing is vital for information retrieval, pattern recognition, and knowledge extraction. Compared with digital image processing using electronics, optical image processing offers potential advantages in throughput, latency, and energy efficiency by exploiting the intrinsic parallelism of light propagation. Optical neural networks have been widely implemented for feature extraction and image enhancement, while they rely on intensive data‐driven training and parameter optimization in the digital domain. In contrast, spatial differentiation provides a direct and training‐free route for certain image processing tasks simply via optical propagation. Such operations have been widely demonstrated by wavefront shaping using free‐space optics, which are often bulky and sensitive to misalignment. Integrated nanophotonic platforms serve as promising alternatives with nano‐scale alignment precision and wafer‐scale integration capability. However, the sub‐micron‐sized single‐mode waveguides have limited functionalities for wavefront engineering. Here, we propose an optical image‐processing unit in which programmable input phases select first‐ and second‐order‐dominated spatial responses of the same passive metalens. Edge detection and local contrast enhancement of eight‐level grayscale images are demonstrated through reconstruction from experimentally characterized multi‐port responses.