Diffractive integrated photonics for analog computing and machine learning

Abstract

Recently, optical analog signal processing draws a lot of attentions due to the rising of deep learning, and integrated photonics is a promising platform. In this thesis, the author propose a design method combining the metasurface with the integrated photonics, and demonstrate the analog computing and machine learning meta-systems. The author demonstrate both forward and inversely design methods for different applications: spatial differentiation, pattern classification and wavelength demultiplexing. Compared with free-space optical systems, the proposed on-chip meta-system allows more compact footprint and better mechanical robustness.