A liquid crystal-based dynamically programmable terahertz diffraction neural network device and method

By coordinating the design of the liquid crystal layer and the fixed-phase metasurface layer, the dynamic reconfiguration problem of diffraction neural networks in the terahertz band is solved, which improves the multi-task processing capability and low-power all-optical intelligent computing, and provides information security.

CN122883388APending Publication Date: 2026-10-09UNIV OF SHANGHAI FOR SCI & TECH
0 Cites 0 Cited by

Patent Information

Application Number
CN202611375732.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-07
Publication Date
2026-10-09

AI Technical Summary

Technical Problem

Existing diffractive neural network devices lack dynamic reconfiguration flexibility in the terahertz band, have slow response speed, high power consumption, and limited integration, making it difficult to achieve coordinated dynamic control of beam amplitude and phase.

Method used

A liquid crystal-based dynamically programmable terahertz diffraction neural network device is adopted. The amplitude dynamic control dimension is introduced through the liquid crystal layer, and the optical field is manipulated in conjunction with the fixed phase metasurface layer. The amplitude modulation is achieved by utilizing the electro-controlled birefringence effect of the liquid crystal material, and the network architecture is optimized by combining a multi-task joint training algorithm.

Benefits of technology

It has improved the device's multi-task processing capabilities, reduced system drive power consumption, ensured high-accuracy all-optical intelligent computing, and provided physical-level information security.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The application discloses a liquid crystal dynamic programmable terahertz diffraction neural network device and method, and belongs to the technical field of terahertz wave regulation and optical neural network. The device comprises a first metasurface layer, a liquid crystal dynamic diffraction layer and a second metasurface layer arranged in sequence; the two metasurface layers are composed of a rectangular silicon column unit array with consistent direction angles, a fixed phase distribution is constructed by changing the lateral size of the silicon column, and the phase distribution is determined by multi-task joint training optimization and reverse mapping based on a Rayleigh-Sommerfeld diffraction model; the patterned electrode of the liquid crystal dynamic diffraction layer is connected to a multi-channel bias voltage source, the voltage of each channel is dynamically adjusted to drive the deflection of liquid crystal molecules, and a variable amplitude distribution is constructed. The application realizes the decoupling regulation of the phase and amplitude of terahertz waves, and a plurality of all-optical computing tasks can be reconstructed on the same hardware by switching the voltage coding matrix or the propagation direction; the coding matrix can also be used as a physical layer encryption key, and has the advantages of fast response, low power consumption and high integration.
Need to check novelty before this filing date? Find Prior Art