Weโre seeing a rush of new boards touting their AI support. Googleโs Coral.ai dev board will be one to beat.
Most current AI happens in supercomputing clusters, but this year weโre seeing a push to move AI to โthe edgeโ, with the processing happening on the little devices scattered around the world, rather than being phoned home over the Internet. Google has new offerings this year in both hardware and software to bring that AI future: On the software side, their โTensorFlowโ deep learning software matures to version 2.0, and gets a โLiteโ version made for simpler processors, like those at the heart of hobby electronics boards. On the hardware side, they offer their Tensor Processing Unit, an accelerator chip to push TensorFlow performance beyond anything general-purpose chips would hope to achieve.
The Coral.ai Dev board is Googleโs single board computer offering, modeled after the classic Raspberry Pi, with a Tensor Processing Unit built in to push its deep learning performance into ranges weโve only seen before from much, much more expensive boards.
Googleโs intro tutorials for using TensorFlow on the Coral.ai are some of the friendlier Iโve seen, though youโll want to have a few things ready to hit the ground running with them: Check the parts list on the tutorial at coral.withgoogle.com/tutorials/devboard to make sure you have all the supplies on the list. I also recommend picking up the optional camera module, since many of the demos assume the board is getting video input to analyze.
With processor speed creeping up on hobby boards, and TensorFlow Lite bringing AI software in reach of more microcontrollers than ever before, we expect to see a wealth of new choices for AI electronics this year. And we expect the Coral to stand out as one of this yearโs AI leaders.