Nanophotonics-assisted computer vision

Amid the rapid development of artificial intelligence (AI), computing units are enabling unprecedented task execution across various fields, including hyperspectral imaging and computer vision. However, the accelerated growth of AI is outpacing the current advancement of semiconductor computing units beyond Moore's Law, leading to challenges related to energy consumption and latency. In this study, we explore how meta-optics can enhance computing performance on both the lens and sensor fronts. On the lens side, we demonstrate how a meta-optical lens can replace the convolutional layers in deep neural networks for image classification tasks. On the sensor side, we illustrate how a meta-optically engineered sensor can decode wavelength and polarization information of photons with enhanced sensitivity. Our findings highlight the potential of meta-optics to significantly contribute to advancements in computational technologies.