Fig. 1: Schematic of design and on-chip machine learning inference with nanophotonic media.
From: High computational density nanophotonic media for machine learning inference
a Qualitative analysis for the coupling efficiency, which is related to the computational density in on-chip optical systems. b Comparison of different kinds of on-chip optical computational systems with computation-related coupling efficiency. c Schematic of the training and inference process of the nanophotonic media, where gradient information is calculated via the adjoint method and used to update the structures according to fabrication constraints. After fabrication, inference is achieved by light passing through the scattering nanophotonic media. Input information is encoded in the phase-modulated input light, and results are obtained through intensity detection at the output ports.
