采用双向移位寄存器实现的二维神经网络卷积运算系统
The two-dimensional neural network convolution operation system implemented through bidirectional shift registers solves the problem of low operation efficiency of convolutional neural networks, realizes fast and efficient convolution operation, is suitable for AI accelerator chip design, and enhances the application potential of terminal and edge computing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIAXING ZHENHE TECH CO LTD
- Filing Date
- 2022-08-09
- Publication Date
- 2026-07-17
AI Technical Summary
In the field of terminal and edge computing, existing technologies for convolutional neural networks have low computational efficiency, require a large amount of data caching and high data bandwidth, which leads to increased system power consumption and latency, thus limiting the promotion and application of AI technology.
A two-dimensional neural network convolution operation system implemented with bidirectional shift registers performs convolution operations using cascaded bidirectional shift registers through N parallel operation units, local data cache units, and parameter cache units, alternating between forward and reverse shift operations to reduce memory usage and improve computational efficiency.
It achieves fast and efficient convolution operations in real-time image and video processing scenarios, improving computational efficiency and reducing memory usage, making it suitable for AI accelerator chip design.
Smart Images
Figure CN115329946B_ABST