New BeagleY-AI Brings Open-Source Hardware to Machine Learning Applications

Everyone seems to have an AI solution for their embedded system hardware. However, in most cases, you'll need to take that hardware as-is if you want to deploy it in your product, leaving no room to cost-optimize. However, things are different at BeagleBoard, with its open-source hardware approach that also applies to its new BeagleY-AI board.

Built using the same form factor as other credit-card-sized single-board computers (SBC), the BeagleY-AI is built around the powerful AM67A system-on-chip (SoC) from Texas Instruments (TI). Alongside the quad-core, 64-bit Arm Cortex-A53 (1.4 GHz) are two general-purpose C7x DSPs with Matrix Multiply Accelerator (MMA) capable of 4 TOPs. These are complemented by an Arm Cortex-R5 that handles real-time interfaces, a GPU, and video and vision accelerators.

AI and machine learning (ML) applications can be developed using TensorFlow and are simple to get up and running on a Linux-based operating system. Optimization of ML algorithms is simplified thanks to an Arm-native toolchain for the TI DSPs for those looking to squeeze every ounce of performance from the boards. And the path to optimized hardware is also simple. Being an open-source platform, developers can reuse the schematic and BOM to create their own, cost-optimized hardware. At embedded world 2024, Jason Kridner, CTO, also demonstrated how an entire disk image for an image recognition application could be ported from the Raspberry Pi to BeagleY-AI, overcoming one of the biggest challenges around porting applications between SBCs.

More information on BeagleY-AI can be found on the BeagleBord website here: https://docs.beagle.cc/latest/boards/...