Google has launched Coral, a new platform of hardware components and software tools featuring the Edge TPU to accelerate neural networks on local devices. The Coral USB Accelerator was designed as a pluggable accessory to bring the Edge TPU chip to Raspberry Pi boards.
“Building upon our work with the original AIY kits, we’re delighted to introduce our new line of products that bring private, fast, and power-efficient neural network acceleration right into your own projects,” says Billy Rutledge, Google’s Director of AIY and Coral teams.
These on-device neural networks will enable makers to integrate fast and efficient AI features into their prototypes and projects while maintaining data privacy. Developers will use TensorFlow to create and train neural networks before compiling them to run on the Edge TPU boards using the software utilities provided. Once the compiled network is installed, all of the inference is performed locally on the Edge TPU and not sent into the cloud, increasing performance without network latency and keeping user data local and under their control.
The Coral platform also includes a collection of pre-trained, pre-compiled models to make it easy to get started. These pre-built models have been optimised for the Edge TPU, and software utilities allow developers to repurpose the models for their own use.
Using the Coral USB Accelerator with Raspberry Pi
See our tutorial on using the USB Accelerator to build a Teachable Machine in The MagPi issue 79 for an example of how Coral can bring local AI into your projects.
These products are now listed for sale at Mouser:
Coral products are welcome additions to the Raspberry Pi developer environment. Artificial intelligence remains one of the most creative parts of modern computing, and we can’t wait to see what prototypes the community comes up with.
Update: The Coral USB Accelerator is not available in France and is listed as "not available" on the mouser.fr website. Mouser.eu told us: "Currently, we are unable to offer this part in France due to availability restrictions. As soon as this changes, this will be reflected on our website. Sorry for the inconvenience." Thanks to Pierre-yves Baloche for alerting us to this.