All-optical diffraction neural network system based on metasurface

Abstract

The invention discloses an all-optical diffraction neural network system based on a metasurface. The all-optical diffraction neural network system is composed of an incident object light beam, the metasurface and a CMOS image processor. Transmission between layers of the neural network is realized through diffraction of light; an optical transparent adhesive and the CMOS image processor are glued to achieve integration, the light intensity of different incident object beams in a detection area is observed through the CMOS image processor, and the multifunctional classification recognition function is achieved. The diffraction neural network based on the super-structure surface has a smaller size, and on-chip integration is easy to realize; and the phase, the amplitude and the polarization are simultaneously regulated and controlled by the metasurface, on one hand, the design freedom degree can be increased, the number of layers of the neural network is reduced to avoid the problem of difficulty in multi-layer alignment, on the other hand, the flexibility of the metasurface adds some complex functions for the diffraction neural network, and finally a passive multifunctional device with excellent performance is obtained.

Description

technical field [0001] The invention relates to an all-optical diffraction neural network system based on a metasurface, belonging to the technical field of optical neural networks. Background technique [0002] As one of the most popular methods in artificial intelligence and machine learning, deep learning aims to enable machines to analyze input text, pictures and other data with a learning method similar to human beings, and obtain the internal laws of sample data at the fastest speed, so as to Perform complex and difficult tasks. With its excellent algorithms, deep learning has solved many complex problems in the fields of automatic driving, speech recognition, and medical analysis, making it a significant progress. As a new two-dimensional planar structure, the metasurface is composed of subwavelength units. By changing the size, arrangement and shape of the units, the electromagnetic wave can be adjusted almost arbitrarily. Therefore, the metasurface not only has a la...

AI Technical Summary

Problems solved by technology

Compared with the problems of complex structure, many neural network layers, and difficulty in integration in the existing technical solutions, the diffractive neural network based on the metasurface has a smaller size and is easy to realize on-chip integration; and the phase and amplitude of the metasurface On the one hand, the simultaneous regulation and polarization can increase the degree of design freedom and reduce the number of layers of the neural network to avoid the difficulty of multi-layer alignment. On the other hand, the flexibility of the metasurface adds some complex functions to the diffractive neural network, and finally Obtain a passive multifunctional device with excellent performance

Images