In addition to a slew of other machine learning developments, Google today announced a platform for building intelligent hardware with on-device AI. Coral is entering public beta today as an end-to-end solution for developers creating IoT hardware from prototyping to production.
Coral offers a complete local AI toolkit that makes it easy to grow your ideas from prototype to production. It includes hardware components, software tools, and content that help you create, train and run neural networks (NNs) locally, on your device.
The local nature is ideal for offline situations where connectivity is limited, while these low-power devices take into account embedded applications. It also has the advantage of keeping user data secure and private on-device.
Coral is powered by an Edge TPU that is specifically designed to run “at the edge.” Designed by Google, this small ASIC provides high-performance ML inferencing — like mobile vision models at 100+ FPS — for low-power devices.
At launch, Coral offers five products that range from new product development to augmenting existing designs and finally scaling for production (coming soon):
- Coral Dev Board: A single-board computer with a removable system-on-module (SOM) featuring the Edge TPU, as well as Wi-Fi, Bluetooth, RAM, and eMMC memory.
- USB Accelerator: A USB accessory featuring the Edge TPU that brings ML inferencing to existing systems. It also allows for easy integration into any Linux system (including Raspberry Pi boards) over USB 2.0 and 3.0.
- Camera: 5-megapixel compatible camera module.
- PCI-E Accelerator: PCI-E device for easy integration of Edge TPU into existing systems.
- System-on-Module (SOM): A fully integrated system in a 40mm x 40mm pluggable module.
Coral will work with other Google tools like TensorFlow and TensorFlow Lite, as well Google Cloud IoT for connected device management. It can ultimately scale from startups to large-scale enterprise uses. Coral products, documentation, sample code, datasheets, and other resources are available today.