Arm Community
Arm Community
  • Site
  • User
  • Site
  • Search
  • User
Arm Community blogs
Arm Community blogs
Internet of Things (IoT) blog Arm CoreLink DMA-350: Next generation Direct Memory Access for Endpoint AI
  • Blogs
  • Mentions
  • Sub-Groups
  • Tags
  • Jump...
  • Cancel
More blogs in Arm Community blogs
  • AI blog

  • Announcements

  • Architectures and Processors blog

  • Automotive blog

  • Embedded and Microcontrollers blog

  • Internet of Things (IoT) blog

  • Laptops and Desktops blog

  • Mobile, Graphics, and Gaming blog

  • Operating Systems blog

  • Servers and Cloud Computing blog

  • SoC Design and Simulation blog

  • Tools, Software and IDEs blog

Tags
  • Cortex-M55
  • Corelink
  • Endpoint AI
  • AXI4
  • Arm Flexible Access
  • Artificial Intelligence (AI)
  • TrustZone for Armv8-M
  • Corstone-300
  • CoreLink DMA-350
  • Arm Ethos-U processor
  • Internet of Things (IoT)
Actions
  • RSS
  • More
  • Cancel
Related blog posts
Related forum threads

Arm CoreLink DMA-350: Next generation Direct Memory Access for Endpoint AI

SamBruggen
SamBruggen
March 7, 2022
5 minute read time.

In recent years, we have seen a profound shift of intelligence to the edge and the endpoint. Traditionally, data is transmitted to the cloud for processing, but now this processing increasingly occurs at the place where the data is generated, which we call Endpoint AI. Arm has enabled this shift with products such as the Arm Cortex-M55, our most AI-capable Cortex-M core, the Arm Ethos-U line of Neural Processing Units (NPU), and the Arm Corstone-300 reference package which seamlessly integrates these components to help SoC designers build Endpoint AI systems faster.

With increased processing requirements come increased data consumption. As a result, moving this data quickly, securely, and efficiently around the chip becomes of paramount importance. Enter CoreLink DMA-350, Arm’s new AXI Direct Memory Access (DMA) controller with TrustZone support.

CoreLink DMA-350 Diagram
 Figure 1. Example system diagram showing CoreLink DMA-350 configured to populate the Cortex-M55 Tightly-Coupled Memory TCM with a dedicated AXI5 Manager interface

CoreLink DMA-350 augments Arm’s Endpoint AI solution, effectively offloading memory movement tasks from the CPU, resulting in improved system performance and power efficiency. CoreLink DMA-350 can be used to populate Tightly Coupled Memories (TCMs) with important data ready for processing. Corelink DMA-350 is also suitable for use with Cortex-A and Cortex-R, and additionally includes the XHB-500 bridge, which can be used to bridge the DMA controller into AHB systems. 

CoreLink DMA-350 includes a range of advanced new features, extending its value beyond what is typical of other DMA controllers on the market. 

Key benefits of CoreLink DMA-350:

Highly configurable to meet varied use-cases 

CoreLink DMA-350 offers a number of new features, many of which are configurable. This allows the IP to be optimized to meet specific use cases. Key configuration points are detailed in the following table:

Feature  Configuration options  Description 
Data interface Single/Dual AMBA AXI5 Manager Interface(s) for transferring data to and from the DMA controller
Data width  32/64/128-bit Supported data bus width on the AXI interface(s)
Address width 32/64-bit Supported address bus width on the AXI interface(s)
Number of DMA channels 1-8 Functional units of the DMA controller that generate memory transfers
TrustZone On/Off Enable/Disable TrustZone security support for all channels
FIFO Depth 1/2/4/8 per Channel Number of Data Width sized entries a channel can hold for a transfer
Extended feature set On/Off per channel Additional registers for advanced copy features (2D copy, wrap, templates)
General Purpose Output 0-32-bit per channel Stable software-controlled value useful for various purposes
AXI4 Stream interface On/Off per channel Interface to expose data for manipulation by an external engine 
Trigger input/output ports 0-32 Versatile triggers configurable in software for flow control 

Arm’s first DMA controller optimized for TrustZone for Armv8-M

TrustZone is a system-wide approach to security, CoreLink DMA-350 is Arm’s first DMA controller with native support for TrustZone, for both Cortex-M and Cortex-A. To integrate a non-TrustZone-aware DMA controller it would be necessary to instantiate two copies of the controller, one for the Secure world, and one for the Non-secure world. With CoreLink DMA-350, this is no longer necessary, with Secure and non-secure partitioning taking place seamlessly within the controller.

Support for low-power interfaces and power modes

CoreLink-DMA-350 uses low power interfaces based on the P/Q channel standards from Arm, a first for an Arm DMA controller. This allows for granular power control, with support for on, warm reset, full retention, and off states and the ability to turn the clock on/off. These low power modes help to reduce DMA controller power consumption – a key feature for the energy constrained designs that CoreLink DMA-350 is targeted at.

In-line data processing over AXI4 Stream interface

CoreLink DMA-350 channels can be optionally specified with an AXI4 Stream interface. This allows data passing through the DMA channel to be routed via custom processing logic (to be implemented by the silicon-designer) on the way to its final destination. Using this feature, simple processing tasks (for example, Color Space Conversion or Alpha Blending) could be handled in a simple and efficient manner. Users of CoreLink DMA-350 are able to add in any logic they please, provided the inputs and outputs meet the AXI4 Stream specifications. With the AXI4 Stream interface, CoreLink DMA-350 enables tiny heterogenous processing where specific tasks are handled in dedicated, optimized blocks, improving overall system efficiency.

Straightforward software programming

Software size and complexity is a key consideration when designing an SoC, particularly for the IoT. CoreLink DMA-350 is designed to be easily programmed over an AMBA 4 APB interface, avoiding the need for complex microcode commands. Furthermore, using the command-linking features, where one command can fetch the next without CPU intervention, and the highly flexible trigger interfaces, it is possible to create complex data copies with little software overhead. 

Dedicated 2D operations for efficient image copy

CoreLink DMA-350 can be easily programmed to move 2D arrays of data, such as images or video frames, using the dedicated 2D copy operations. Furthermore, simple image processing such as rotation and mirroring can be achieved using these operations. Using CoreLink DMA-350’s dedicated 2D operations helps to simplify the software overhead when transferring 2D data ready for processing. More advanced image processing could be handled using custom logic integrated via the AXI Stream Interface, making CoreLink DMA-350 a good choice for Endpoint AI use cases which handle 2D media. 

Extracting maximum performance and efficiency from constrained AI

The Arm CoreLink DMA-350 is designed to help unlock the full potential of Endpoint AI devices, improving energy efficiency and performance for highly constrained systems. Offering a range of advanced data movement features in a secure and configurable package, CoreLink DMA-350 is an important component of Arm’s complete solution for intelligence at the endpoint.  

CoreLink DMA-350 is available now in Arm Flexible Access. 

Anonymous
Internet of Things (IoT) blog
  • Building vision-enabled devices to capture the emerging wave in IoT

    Diya Soubra
    Diya Soubra
    IoT devices will drive an explosion in use cases with vision. Read more about the different use cases and what Arm technology is involved here.
    • December 9, 2024
  • The power of SystemReady for custom-built OS distributions

    Pere Garcia
    Pere Garcia
    Arm developed the SystemReady Devicetree band as part of the SystemReady program, learn more in this blog post.
    • November 22, 2024
  • Software, Tools, and Ecosystem for ML Edge Devices

    Reinhard Keil
    Reinhard Keil
    Learn how Arm and our Partners enable developers and the IoT software ecosystem to deliver smart, energy efficient ML edge devices.
    • July 17, 2024