Processing circuit for CNN acceleration and method of operating processing circuit

By using processing circuits in convolutional neural networks to detect effective value pairs and compress the calculation, the problem of high computational complexity of convolutional neural networks is solved, and the effect of reducing power consumption and time cost is achieved.

CN119721146APending Publication Date: 2025-03-28SAMSUNG ELECTRONICS CO LTD +1
0 Cites 0 Cited by

Patent Information

Application Number
CN202411341484.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2024-05-08
Filing Date
2024-09-25
Publication Date
2025-03-28

AI Technical Summary

Technical Problem

The high computational complexity of convolutional neural networks in deep neural networks leads to high power consumption and time costs.

Method used

In the operation of the neural network model, the processing circuit is used to detect the effective value pairs between two inputs and reduce the computational amount, power consumption and time by compressing the chunk and mask.

Benefits of technology

It effectively reduces the computational amount of convolutional neural network, reduces power consumption and time cost, and improves computing efficiency.

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

Abstract

A processing circuit and a method of operating the processing circuit are provided. The method includes: generating a first compression chunk including only a first effective value; generating a first mask including a reference value at the same position as the position of the first effective value and including a plurality of first sub-masks; generating a second compression chunk including only the second effective value; generating a second mask including a reference value at the same position as the position of the second effective value, and including a plurality of second sub-masks; generating a valid pair position value for each of the current first sub-mask and the current second sub-mask; generating a first accumulated value corresponding to the number of reference values included in the first previous sub-mask; and generating a second accumulated value corresponding to the number of reference values included in the second previous sub-mask.
Need to check novelty before this filing date? Find Prior Art