Built by Google for accelerating neural network, the dev board allows faster inferencing and deployment of low-power embedded systems
Coral Dev Board Micro. Image credit: www.coral.ai
Dev Boards are fascinating hardware pieces. They come with great power and save us time in building complex electronic circuits. That’s a big relief for faster time-to-market of a product. In this age of rapid advancements around machine learning (ML), every second counts in introducing a new product that can solve existing and new challenges.
With this objective, Coral has introduced a new development board called the Coral Dev Board Micro. The soon to be launched development platform combines Cortex M4 and M7 processors with the Coral Edge TPU for ML inferencing ranging from low-power to complex.
Along with a built-in camera and microphone, the Coral Dev Board Micro allows quick prototyping and deployment of low-power embedded systems built for ML inferencing. The dev board supports TensorFlow Lite and TensorFlow Lite for Microcontrollers.
MCU – NXP i.MX RT1176 (Cortex-M7 and Cortex-M4)
ML accelerator – Coral Edge TPU coprocessor: 4 TOPS (int8); 2 TOPS per watt
RAM – 512 Mbit
Flash memory – 1 Gbit
Sensors – Colour camera (324 x 324 px); PDM mono microphone
I/O – (2x) 12-pin GPIO header; (2x) 100-pin board-to-board connectors; USB Type-C (USB 2.0)
More On Coral Technology
Built by Google for accelerating neural network inferencing at a low cost, the Coral Edge TPU coprocessor is a small ASIC capable of performing 4 trillion operations (tera-operations per second or TOPS) using 0.5 watts for each TOPS (i.e. 2 TOPS per watt).
The Edge TPU supports various model architectures built with TensorFlow, including models built with Keras. So no additional APIs are required to build or run your model; only a small runtime package can execute your model to the Edge TPU.
For easing product development via the Dev Board and System-on-Module, Google has created a derivative of Debian Linux called Mendel. Besides offering a huge range of customisations, Mendel also includes various tools for building ML applications, including standard Python and C++ libraries, the Edge TPU API, and the Edge TPU runtime. The Mendel Development Tool (MDT) enables secure connection (using SSH/mDNS), file transfers and other commands from a remote computer.
Other highlights of the Coral technology include co-compiling the ML models for running them simultaneously and model pipelining for executing different segments of the same model on different Edge TPUs.
Coral by Google has made its code open-source so that everyone can collaborate and contribute. It is working with other machine learning teams to help build the next generation of neural networks for faster inferencing in low-power devices.