Photonic integrated computing engine for concurrent optical computing

Le résumé fourni par la source

Optical networks with parallel processing capabilities advance high-speed computing and large-scale data processing by providing ultrawide computational bandwidth. In this paper, we present a photonic integrated processor that can be segmented into multiple functional blocks, enabling compact and reconfigurable matrix operations for parallel computational tasks. Fabricated on a silicon-on-insulator platform, the processor supports reconfigurable optical matrix operations of various sizes, offering flexibility and scalability. Specifically, it performs optical convolution operations with three-channel 1×1 and 2×2 real-valued convolution kernels implemented in distinct blocks. The multichannel 1×1 convolution is experimentally validated using a deep residual U-Net for precise segmentation of pneumonia lesions in lung computed tomography images. The 2×2 convolution is validated through an optical convolution layer integrated with an electrical fully connected layer for ten-class classification of handwritten digits. The processor features high scalability and robust parallel computing capability, positioning it as a promising candidate for optical neural networks. Photonic processors can accelerate large-scale data processing but often target a single fixed task. Here, the authors demonstrate a reconfigurable photonic processor that can be partitioned into functional blocks to perform three optical convolution tasks concurrently.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.