Network Interface Distributed DMA Engine Data Transfer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing on-chip communication systems, particularly between network-on-chip (NoC) and on-chip bus, are slow and time-consuming due to reliance on standard DMA engines, which are unavailable during data transfer and inefficient in handling multiple channels.
Innovation Solution
A method and network interface architecture that sends request packets to initiate data transfer between source and destination targets, using a distributed DMA engine to incrementally transfer data, with the network interface responsible for converting and synchronizing data formats, allowing for improved performance and processing capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If a standard DMA engine is used for data transfer between on-chip bus and network-on-chip, then data transfer can be performed, but the system becomes slow and time-consuming
Solution Approach 1:
The patent segments the data transfer function by introducing multiple independent DMA channels (at least 8 channels) that can operate simultaneously. Each channel handles a separate data transfer stream, allowing parallel processing of multiple data transfers. This segmentation eliminates the bottleneck of single-channel sequential processing and significantly improves overall data transfer speed while reducing total time consumption.
Solution Approach 2:
The patent implements continuous data transfer capability by ensuring that while one DMA channel is occupied with a transfer operation, other channels remain available to initiate new transfers. The system maintains continuous useful action by preventing idle time in the processing unit, as it can immediately assign new transfer tasks to available channels without waiting for previous transfers to complete.
2Productivity
If a standard DMA engine is used for data transfer, then data can be moved between addresses, but the processing unit becomes unavailable during the entire data transfer process
Solution Approach 1:
The patent implements self-service by enabling the DMA engine to autonomously manage multiple data transfer operations without requiring continuous intervention from the processing unit. The processing unit only needs to initialize the DMA channels and assign high-level transfer parameters. Once configured, the DMA engine independently handles address registration, data movement, and channel management, allowing the processing unit to remain available for other computational tasks throughout the transfer process.
3Productivity
If multiple DMA channels are used to improve transfer capacity, then more data can be transferred in parallel, but each channel becomes unavailable during its specific DMA transfer
Solution Approach 1:
The patent applies universality by designing the DMA engine with multiple channels that all share the same control logic and operational framework. Each channel is a universal data transfer unit capable of handling different types of transfers (memory-to-memory, memory-to-peripheral, etc.). This standardized multi-functional approach simplifies management compared to having specialized channels for different purposes, as the same control mechanisms apply to all channels regardless of their specific transfer tasks.
Data Source
AI summary
A method is for transferring data from a source target to a destination target in a network. The method includes sending at least one request packet for the destination target, with the request packet containing information relating to a first address where data are located and a second address where data are to be stored. Moreover, at least one transaction request is sent to the source target, with the read request being elaborated from information contained in the request packet. The source target transfers the data located at the first address to the second address.