AI for the Edge has been a promising playing field where several players are pushing for. Cloud computing has made it possible to train complex machine learning models for various application, although this seems to be working fine, the performance or the possibility of deploying AI applications on the Edge is enormous.
AI on the Edge is expected to help reduces the latency involved in the roundtrip to the cloud, saves the bandwidth and cloud storage costs for enterprises, deploy ML models faster, and build robust, intelligent applications.
Generally, Edge devices like the Raspberry Pi, Arduinos, and other embedded boards usually can’t run powerful AI applications. They have limited resources and computing power. Fortunately, this is changing with the introduction of AI Accelerators; modern processors that help assist the edge devices by taking over the complex mathematical calculations needed for running AI models. One of such AI accelerator processor is the Kendryte K210 which has seen deployments on different development boards like the HuskyLens AI Camera, Grove AI HAT, and some others.
XaLogic is continuing on this trend with the launch of their Kendryte K210 based PI AI Hat.
Over at XaLogic, we want to bring Machine Learning to the edge computing by building a machine learning accelerator module. We used Raspberry PI quite a lot in our rapid prototyping and love the form factor of the PI Zero which allows us to make cool things in a nice small form factor.
So, we took our Machine Learning accelerator module and make a PI Hat. Now, we can accelerate the PI and especially the PI Zero which is underpowered.
Although the XaLogic board won’t be the first AI Hat for the Raspberry Pi, the Grove AI HAT is also meant for the Raspberry Pi as well, but models A/B, while XaLogic board follows a more compact Raspberry Pi Zero form factor.
The XaLogic AI Hat board called the XAPIZ3500 was built based on the AI module XAM3500. At the core of the XAM3500 is the Kendryte K210 Dual-Core 64bit RISC-V processor with FPU and dedicated CNN accelerator. The module is able to achieve 0.23TOPS under 300mW.
The XaLogic AI Hat XAPIZ3500 makes it possible to build machine learning applications with the Raspberry Pi without breaking the bank. Coupled with the Pi zero form factor and 3D printing technology, you can quickly build an AI Camera.
According to XaLogic,
Adding the XAPIZ3500 allows Yolo-Like object detection with close to 10X performance improvements. YoloV2 Performance (current state…. to be further optimized.. )
Pi Zero ≈ 3FPS
Pi 3B+ ≈ 8FPS
The Hat has been tested on both the Pi Zero and the Pi 3B+, but it is expected to work on any Pi with 40 pin connector.
The edge is becoming the perfect destination for deploying machine learning models trained in the cloud, and with the launch of the likes Nvidia Jetsons series, Pi AI Accelerator module, Neural sticks, and others, the future surely looks bright.
I enjoy writing about the latest news in the areas of embedded systems with a special focus on AI on edge, fog computing, and IoT. When not writing, I am working on some cool embedded projects or data science projects. Got a tip, freebies, launch, idea, gig, bear, hackathon (I love those), or leak? Contact me.