Parallel Storage Requests for Neural Network Data Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural network models suffer from low data storage efficiency due to the sequential processing of output data, leading to inefficient storage operations.
Innovation Solution
A data processing method that sends multiple storage requests in parallel to the memory, allowing simultaneous storage of output data from multiple processing elements, with the processor determining the addresses for storage based on flag parameters to optimize storage efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If sequential storage of output data is used in conventional neural network models, then the storage operation is simple and easy to implement, but the data storage efficiency is low
Solution Approach 1:
The patent divides the storage operation into multiple parallel storage requests, where each processing element sends its output data to memory simultaneously through separate storage requests. This segmentation allows concurrent storage operations across multiple processing elements, improving storage efficiency from sequential to parallel processing
Solution Approach 2:
The patent introduces a time dimension to the storage operation by implementing polling mechanisms that enable multiple storage requests to be processed concurrently across different time slots. This transforms the single-dimensional sequential storage into a multi-dimensional parallel storage system, increasing storage throughput
2Productivity
If parallel storage requests are sent to memory, then data storage efficiency is improved, but the complexity of managing storage addresses and requests increases
Solution Approach 1:
Each processing element autonomously generates storage requests for its own output data without requiring centralized coordination. The processing elements self-manage their storage operations by directly sending requests to memory with appropriate addresses, eliminating the need for complex centralized address management and reducing operational overhead
Solution Approach 2:
The patent implements preliminary address generation where each processing element pre-calculates and includes the correct memory address in its storage request. This preliminary action of address preparation eliminates the need for subsequent address resolution and coordination, simplifying the management of parallel storage requests
Data Source
AI summary
A data processing method includes: an electronic device concurrently sending N storage requests for Q times to a memory in each polling by a processor, wherein the N storage requests are used for requesting the memory to store N rows of output data generated by N processing elements having continuous identifications among M processing elements, and Q is determined according to the number M of the processing elements and the number N of the storage requests; and by means of the memory, the electronic device storing a P-th row of output data generated by each of the M processing elements in a P-th polling according to received Q×N storage requests received from the processor. The present method can solve the problem of lower data storage efficiency in existing neural network models, a plurality of pieces of data may be stored at the same time by triggering concurrent requests to improve storage efficiency.