Distributed DMA Architecture for Parallel Chip Data Transfer

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional DMA technologies in SoC chips face limitations in meeting increasing computing power requirements due to single DMAC processing capabilities and large hardware resource overheads from caching read-back data.

Innovation Solution

An integrated chip with a DMA management module that splits data transfer tasks into subtasks, distributed across multiple DMA agent modules, utilizing a distributed architecture for parallel execution and address translation, reducing the need for large data caches.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a single DMAC processes data transfer tasks sequentially, then the hardware resource overhead is reduced, but the data transfer efficiency cannot meet increasing computing power requirements

Engineering Contradiction:
Improvedata transfer efficiencyVSAvoidDMAC processing capability
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent divides the data transfer task into multiple subtasks and distributes them to multiple DMA agent modules. Each DMA agent module processes a portion of the subtasks in parallel, thereby improving overall data transfer efficiency without requiring a single overly complex DMAC. The DMA management module segments the task and the DMA agent modules execute them concurrently.

Inventive Principle:
Principle #1Segmentation

2Reliability

If all read-back data is stored in the DMAC, then data transfer completion is ensured, but large data cache resource is required resulting in large hardware overhead

Engineering Contradiction:
Improvedata transfer completionVSAvoiddata cache resource
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

The patent segments the data transfer process into multiple subtasks distributed across multiple DMA agent modules. Each module handles a portion of the data transfer independently, eliminating the need for a single large data cache to store all read-back data. The distributed architecture reduces the cache requirement at each individual module while maintaining overall data transfer reliability.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transitions from a centralized data cache approach to a distributed data handling approach across multiple modules. Instead of accumulating all data in a single DMAC's cache, the system distributes data processing across multiple dimensions (multiple DMA agent modules), reducing the cache burden at each dimension while maintaining completeness through coordinated execution.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP4293524B1Integrated chip and data transfer method
Publication Date: 2026.01.21 HUAWEI TECH CO LTD
  • EP4293524B1 patent drawingFigure 1~2
  • EP4293524B1 patent drawingFigure 3~4
  • EP4293524B1 patent drawingFigure 5~6

AI summary

An integrated chip and a data transfer method are provided to meet an increasing computing power requirement of a processor chip. The integrated chip includes a DMA management module and a DMA agent module. The DMA management module is configured to split a data transfer task delivered by a processor into a plurality of subtasks, and deliver the plurality of subtasks to each DMA agent module in a plurality of DMA agent modules. Each DMA agent module in the plurality of DMA agent modules is configured to execute the plurality of subtasks.