Method and apparatus for data processing using a photon-electron hybrid computing system

CN122111174APending Publication Date: 2026-05-29HANGZHOU GUANGZHIYUAN TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
HANGZHOU GUANGZHIYUAN TECH CO LTD
Filing Date
2024-11-28
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing optical computing units suffer from reduced energy efficiency, making it difficult to support floating-point calculations, especially fixed-point calculations with precision exceeding 8 bits. Furthermore, the high noise level of intermediate results affects the accuracy of the calculations.

Method used

A photonic-electronic hybrid computing system is adopted. By shifting the input data in the electronic domain to match its bit width with the photonic multiplication unit, multiplication calculation is performed in the photonic domain, and subsequent shifting and accumulation operations are performed in the electronic domain, thus realizing multiplication operations with arbitrary bit width.

Benefits of technology

It simplifies the design of photon matrix multipliers, reduces power consumption, and optimizes data refresh rates in artificial intelligence applications by pipelined computation between optical and electrical computing through decoupling the multiplication and accumulation processes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122111174A_ABST
    Figure CN122111174A_ABST
Patent Text Reader

Abstract

The present disclosure relates to a method and device for data processing using a photon-electron hybrid computing system, the method comprising: performing a first shift operation on a first multiplier and a second multiplier to be input into a photon multiplication unit, so that the bit width of the multipliers is the same as the bit width supported by the photon multiplication unit; storing the first multiplier and the second multiplier subjected to the first shift operation into a first buffer and a second buffer respectively, and having a first index value and a second index value respectively; obtaining the multipliers to be multiplied from the buffers and inputting them into the photon multiplication unit for multiplication processing to obtain a multiplication result; storing the multiplication result into a third buffer and having a third index value, which is the sum of the first index value and the second index value; and performing a second shift operation on the multiplication result, the shift direction of which is opposite to that of the first shift operation, and the shift bit number is the third index value. Thus, multiplication operation of input numbers with any bit width can be realized.
Need to check novelty before this filing date? Find Prior Art