Brainchip’s MetaTF Software Tools ease the Transition of ML Models to the Edge
Kurt Manninen, BrainChip Senior Solutions Architect
Empowering Companies: Seamless AI Migration to the Edge
In their latest survey report, Edge Impulse highlights that the estimated market value of edge AI will be $143.6B by 2032. Companies that are early adopters of AI on the edge, particularly those that are willing to invest in the latest neural network hardware to give themselves an advantage over their competitors, will be looking to their vendors to help make the transition as painless as possible. This is why BrainChip developed our MetaTF framework: as a way for companies to effortlessly transition their existing machine learning models into a format that takes advantage of Akida’s event-based computing platform.
MetaTF is a suite of tools that can be used to convert, quantize and run existing machine learning models on an Akida event-based processor, which reduces computations by only computing on event data, not the raw sparse data as conventional NPUs must do. It was built for the purpose of providing an easy path for current and prospective BrainChip customers to validate their existing models on our advanced processor. As an experienced software and machine learning engineer who is new to event-based computing, BrainChip’s MetaTF allowed me to use tools that I already know (i.e. Python and Jupyter Notebooks) to experiment with ML on the edge. This is a valuable offering that will help engineers at developer organizations quickly migrate their processes to edge-based learning.
Embedded Devices ≠ Servers
When it comes to deploying machine learning solutions, embedded devices present unique challenges compared to traditional server environments. Unlike servers with standardized architectures and operating systems, the world of IoT devices is incredibly diverse. From smartphones to smart sensors, each device may have a different architecture, run on a different operating system, or in some cases, operate on bare metal without any OS at all. This design diversity makes software packaging and installation a complex task. Moreover, when scaling to thousands or even millions of devices, manually managing each installation becomes impractical. This is where software build tools become invaluable, offering a way to manage complexity, ensure repeatability, and create consistent device images across a wide range of hardware configurations.
Edge Impulse+MetaTF = The Easy Button for Buildroot and Yocto
Training a model, and making it small enough for low size, weight and power devices is one thing, but packaging and deploying the application to the edge device is another. BrainChip has extensive experience with major embedded build systems such as Buildroot and Yocto. Our goal is to ensure that our Akida System on Chip design can be seamlessly integrated into our customers unique embedded environments. Furthermore, our partnership with Edge Impulse is helping to revolutionize the ML-to-Edge training and deployment process, making it easier than ever to deploy advanced event-based machine learning solutions on embedded devices.
A prime example of our expertise in action is the Akida™ Edge Box, developed in collaboration with our partners VVDN and Edge Impulse. This innovative product embodies the principles discussed in this blog post, showcasing our ability to tackle complex edge AI challenges. The Edge Box runs on an NXP i.MX 8M application processor, featuring dual Akida 1000 chips on reference boards connected to a single M.2 PCI card. Our demonstration models were trained using Edge Impulse Studio and we bundled our MetaTF libraries into the target image using NXP’s i.MX Yocto Project Board Support Package. This real-world application demonstrates our commitment to integrating cutting-edge event-based computing with established embedded system build tools.
To learn more about how BrainChip’s solutions architecture team can integrate cutting-edge event-based machine learning into your embedded device toolchain, or to explore how the BrainChip Edge AI Box could benefit your AI initiatives, contact us for a demonstration. We’re here to help you take your specific ML processing needs and AI products to the edge.
