BeagleBoard Founder Talks About Artificial Intelligence and Open Source Hardware

”

The BeagleBoard.org Foundation recently announced the launch of the BeagleY-AI, which is about the size of a business card, but it is open-source hardware and works without a fan, with an AI computing power of 4 TOPS, which is suitable for executing deep learning algorithms.

Recently, EmbeddedCOMPUTING interviewed Jason Kridner, co-founder and president of the board of directors of BeagleBoard, about artificial intelligence and open source.

Rich Nass: Please tell me a bit about the history of the BeagleBoard and describe the current product.

Jason Kridner: BeagleBoard.org has a long history. The organization began building small, low-power single-board computers back in 2007. From the beginning, we've been passionate about serving the open source developer ecosystem, providing education and better tools for building embedded systems. We've done this through open hardware design and community-based support, and the results have been amazing. We've changed the way people develop embedded systems.

BeagleBoard.org has established itself as a trusted and reliable SBC provider. By prioritizing open source principles and supporting education and industrial applications, BeagleBoard.org plays a key role in democratizing access to technology and empowering individuals to innovate and create. Our commitment to these principles for over 15 years is a testament to the lasting impact and importance of open source initiatives in advancing technology and fostering inclusive learning environments.

We have produced over 10 million boards and consistently offer products such as the BeagleBone Black, which was first launched in 2013. We continue to offer variants of the BeagleBone Black and industrial wide temperature applications, as well as several other designs, and have introduced four new designs over the course of a year, including our first microcontroller board, the BeagleConnect Freedom, with 1km wireless communication capability running Zephyr and Micropython support. Another product, BeaglePlay, is a flexible user interface and gateway design that supports the 1km wireless protocol and Single Pair Ethernet, as well as several other innovative features that eliminate the need for complex wiring, reducing the effort required to add a large number of sensors. We've also introduced two interesting RISC-V-based board options for those looking to explore emerging ISAs.

We are very excited to talk about our latest product, the BeagleY-AI, which is able to work with a large number of existing enclosures and additional hardware with an emerging industry-standard 40-pin connector. BeagleY-AI stands out for its open-source hardware design, fanless operation, and 4 TOPS deep learning engine for AI workloads.

Nass: How is the BeagleY-AI board different from existing boards in the industry? As I understand it, it uses the latest processors offered by Texas Instruments. Was TI involved in the development of the board? In what other ways are companies involved? Why is this important or beneficial for developers?

Kridner: The long-standing partnership between BeagleBoard.org and Texas Instruments has been instrumental in our success. The commitment to using TI processors in BeagleBoard products, including the AM67X in the BeagleY-AI board, underscores a shared vision to advance technology and provide a cutting-edge development platform for manufacturers and industry partners. The AM67X processor brings significant benefits such as energy-efficient machine learning capabilities, low-latency cores for time-critical applications, and support for standard high-speed I/O interfaces such as USB 3 and PCIe Gen 3. The low power consumption, accelerated vision processing, and production stability obtained with the AM67S SoC can help BeagleY-AI stand out from its peers.

The collaboration between BeagleBoard.org and Texas Instruments is not limited to hardware integration, with TI's hardware and software design teams actively participating in the BeagleY-AI testing and review process. Their supply chain and ongoing technological advancements underscore their commitment to supporting the open source community, ensuring the success of initiatives like BeagleBoard.org.

Nass: The new board has AI in its name, so please tell me what that means.

Kridner: The short answer is that it means we have a built-in accelerator that executes deep learning models at very high rates, which can be used in different places than larger, more power-hungry, more expensive AI accelerators.

The longer answer is that this means that we are focused on providing developers with better tools to understand the possibilities of AI, including object detection, pose estimation, and image segmentation, among others. We do this through a large number of easily accessible examples and materials on docs.beagleboard.org.

Nass: In terms of real-world examples, what are some of the interesting aspects of the Beagle deployment?

Kridner: It's incredible to see the range of applications for Beagles, from healthcare solutions such as affordable open-source real-time PCR machines for COVID-19 testing, to underwater rescue drones, AI-powered machines, and even space missions. Diversity emphasizes the robustness and adaptability of Beagles, making it the first choice for innovators in different industries looking for reliable and versatile open-source hardware solutions.

But we've only heard about a small percentage of Beagle usage, because it's up to the developers to decide what they want to share with us. So there are thousands of apps out there, and we'll support them when we need them. One that we know is laser engraving applications.

Nass: BeagleY-AI is built with "open-source hardware". What does this mean? Why is it so important? Is there anything scary about open source? Is it safe?

Kridner: "Open source hardware" refers to hardware whose design is publicly available so that anyone can research, modify, distribute, manufacture, and sell the design or hardware based on that design. This really gets to the heart of BeagleBoard.org, and to answer it in its entirety, I think we need to start by looking at history.

The history of open-source hardware is closely related to the history of open-source software. When computers were first introduced in the '50s and 1960s, it was standard practice to provide software sources, as well as hardware design information, which was critical for users to understand how to program these complex machines and even fix them if something went wrong. This level of understanding is not achievable by keeping the design closed.

High-quality audits are essential when building a safe and secure system. That's why the mission-critical systems of the New York Stock Exchange can run Linux. A large number of experts were able to give their opinions on possible vulnerabilities. Developers can choose when, where, and how to lock things down to achieve their own goals, not other people's security goals. A lot of people are looking at this code to make sure it's robust. Personally, I feel that running software that has undergone this level of scrutiny is much safer than anything made by any one company, and the same is true for hardware.

Open-source hardware means you can choose to protect what you see fits best. This means you can even build the board yourself if you need to achieve safety goals.

Nass: What programming languages and environments do you want developers to adopt when using the new BeagleY-AI, and why?

Kridner: The benefit of building on Linux is that it supports most languages. Instead of trying to build language-specific bindings to control the hardware, we're focusing on the interfaces provided by Linux and Zephyr.

With Zephyr, we can build things as small as 4 KB, which is usually small enough for any system or subsystem – especially in the prototype phase. It allows you to focus on the code and pull the supporting code as needed, saving you development time. You can even get a full POSIX environment that is very easy to run like Micropython etc.

Python, JavaScript, C, C++, Go, and Rust all get a lot of attention. We enabled the self-hosted Visual Studio Code environment, but we left it to others to provide value-added libraries and make sure Linux and Zephyr provided the interfaces that developers need.

For deep learning algorithms, there are quite a few high-level language integrations, mostly focused on Python and C++ with familiar APIs, such as TensorFlow Lite. This isn't the realm of OS integration interfaces yet, but I expect that to change.

NASS: How can developers become part of the Beagle community?

Kridner: BeagleBoard is closely aligned with the needs and feedback of its users and developers, ultimately leading to more relevant and impactful product designs. We encourage joining the BeagleBoard.org community through the forum.beagleboard.org Forum.