Embedded World Product Showcase: BeagleBone® AI-64 Open-Hardware SBC from BeagleBoard.org

June 28, 2022


Rather than an endgame, AI has become a starting point for many developers today. Like any type of engineering, you can start out with simple AI projects that run on basic embedded systems. But, like any type of engineering, the complexity can ramp up quickly. So can the complexity of underlying platforms.

To deliver the performance required for advanced AI and machine learning in a familiar, open platform that keeps things simple, BeagleBoard.org has launched the BeagleBone® AI-64 open hardware single-board computer (SBC). Continuing the tradition of open, accessible development platforms that pair with open-source software and locally hosted, ready-to-use toolchains, the BeagleBone® AI-64 features tons of performance under the hood to suit any AI – or other workload, for that matter – you need to run to set your design apart.

It’s powered by Texas Instruments’ 64-bit Jacinto TDA4VM processor featuring:

  • Dual-core 64-bit Arm Cortex-A72 microprocessor subsystem that runs at 2GHz
  • C7x floating-point, vector DSP that operates up to 1 GHz and will get you 80 GFLOPS
  • 2x C66x floating-point VLIW DSPs as fast as 1.35 GHz that yield up to 40 GFLOPS
  • PowerVR® Rogue 8XE GE8430 3D GPU that can net another 96 GLOPS at 750 MHz
  • 8-bit Deep Learning Matrix Multiplier (MMA) with speeds of 1 GHz for another 8 TOPS
  • Depth, Motion, Video Encode/Decode, and Vision Accelerators with integrated ISPs

All these features have led to Jacinto TDA4VM deployments in ADAS and autonomous vehicle use cases, and that’s not even the SoC’s entire processing subsystem. The chip brings 2x programmable real-time units (PRUs) found in other BeagleBones back in the BeagleBone® AI-64, which facilitates low-latency control and deterministic communications with protocols like Ethernet TSN. It also integrates six Arm Cortex-R5F cores, two of which are located on the TDA4VM’s MCU Island and can be used in lockstep configuration for safety applications.

In other words, whatever your application needs, there’s a core for that. Put it all together, and what you have with the BeagleBone® AI-64 is a development platform that’s ready for intensive vision, AI, and machine learning workloads out of the box.

The BeagleBone® AI-64 SBC Action

Of course, there’s more about the BeagleBone® AI-64 that helps turn those concepts into reality, starting with memory and storage. Aside from the megabytes worth of on-chip L2 cache, RAM, and SRAM, the BeagleBone® AI-64 equips another 4 GB of RAM and 16 GB eMMC flash onboard that connect to the processor over a high-speed interface.

Speaking of interfaces, the Jacinto TDA4VM’s custom on-chip interconnect fabric delivers near-maximum processor entitlement for data being piped in from the AI-64’s numerous interfaces, which include USB Type-C, dual USB superspeed (5 Gbps) Type-A hosts, Gigabit Ethernet, and an M.2 E-key slot for adding Wi-Fi or Bluetooth capabilities. Expansion is also possible via a 16-pin MikroBus header and BeagleBone-compatible headers for adding on use case-specific click boards and capes (even though the AI-64 is a tick larger than some of its predecessors at 4² x 3.1²).

In addition, BeagleBone AI-64 contains a miniDisplayPort, 4-lane DSI, and 2x 4-lane MIPI CSI connectors that allow users to plug in displays and computer vision cameras in an instant, then move on to prototyping and development of their AI-centric applications.

That truly out-of-the-box experience begins with the zero-download Debian Linux distro that ships on the board. Once the OS is booted, open-source 3D graphics drivers from Imagination Technologies provide access to the PowerVR GPU, while DSP programming tools packaged with the AI-64 run locally on the device.

Open-source AI tools like TensorFlow Lite, the ONNX neural network interface exchange, and Apache TVM machine learning compiler are supported by the platform as well, which allow the BeagleBone AI-64 to serve as a comprehensive, native development environment for intelligent applications. In other words, all you need is a power source, network connection, and simple web browser download to use the latest Beagle as a full AI development workstation.

Getting Started with BeagleBoard.org’s BeagleBone® SBC

With open-source reference documents and design files also available, the BeagleBone® AI-64 offers a path for makers interested in learning about machine learning or skilled developers looking to bring advanced AI projects to life. At a price point of just $187.50 and with open-source software downloads freely available on BeagleBoard.org and the BeagleBoard.org GitLab, it also lowers barriers to entry for everyone.

To get a head start with your AI, machine learning, or computer vision design, you can purchase a BeagleBoard.org from leading distributors such as Digi-Key, Mouser, Farnell, or OKdo. Learn more by visiting https://beagleboard.org/ai-64, watching Embedded Computing Design’s recent interview on the AI-64 with BeagleBoard.org Chairman of the Board Jason Kridner, or checking out the resources below.