Comb-based parallel information processing and artificial intelligence

Enhancing information processing throughput is one of the primary goals in the development of optical information processing systems. Comb-based parallel information processing offers a promising approach to achieving high throughput, particularly when combined with 2D spatial multiplexing. In this talk, I will present our latest work on hyperspectral information processing using optical frequency combs that fully leverage the inherent optical parallelism of the frequency and 2D spatial domains. Such high-throughput systems could significantly benefit high-performance computing in Artificial Intelligence (AI) data centers and become exceptionally powerful in certain applications when integrated with AI-enabled advanced algorithms. First, I will introduce a hyperspectral compute-in-memory architecture that enables high-throughput, single-shot matrix-matrix multiplication designed for future AI data center infrastructure. Second, I will present hyperspectral dual-comb compressive imaging using zero-dimensional optical front-end hardware (a single-core fiber and a single-pixel detector), enhanced by an AI-enabled image reconstruction algorithm.