用于人工神经网络的存储器控制器、处理器以及系统
By optimizing processor-memory level data access through an artificial neural network memory controller, the problems of low memory bandwidth and latency in traditional neural network models are solved, resulting in more efficient computing performance and reduced power consumption.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- DEEPX CO LTD
- Filing Date
- 2021-11-02
- Publication Date
- 2026-07-17
AI Technical Summary
Traditional artificial neural network models suffer from processor operation bottlenecks due to high power consumption, heat generation, low memory bandwidth, and memory latency. Furthermore, existing prefetching algorithms cannot effectively optimize processor-memory level data access.
By using an artificial neural network memory controller, the processor's data access requests and locality patterns are analyzed, the data to be processed is prepared in advance, the memory latency and bandwidth are optimized, the processor's starvation or idle state is reduced, and data operations are optimized at the processor-memory level by leveraging the data locality of artificial neural networks.
It improves the computational processing speed of artificial neural network models, reduces memory latency and bandwidth increase, lowers power consumption, and enhances processor performance.
Smart Images
Figure CN114861898B_ABST