Google Launches Coral Dev Board Mini and Coral Accelerator Module

2020-01-05

Coral is a platform developed by Google for building IoT hardware with on-device AI capabilities. Google announced today, ahead of CES 2020, the launch of new hardware for Coral. The Coral Dev Board, USB Accelerator, and PCIe Accelerators were launched in 2019 and are now available in 36 countries/regions. Google notes that a wide range of applications have been built on Coral, covering industries from healthcare and agriculture to smart cities."


Last March, Google completed the packaging of Coral and launched the Beta version in selected regions, followed by a wider release in October last year. Today, ahead of the 2020 Consumer Electronics Show, Google announced new features for the Coral series, which will be available later this year.


First is the Coral Accelerator Module, which is a multi-chip package that utilizes Google's custom-designed Edge Tensor Processing Unit (TPU). The module has both PCIe and USB interfaces and can be easily integrated into custom PCB designs. Google has closely collaborated with manufacturing partner Murata to prepare the module for the first or second quarter of 2020.


The TPU is an application-specific integrated circuit (ASIC) developed specifically for neural network machine learning. The internal chip, Edge TPU, can perform multiple computer vision models at a rate of 30 frames per second or a single model, such as MobileNet V2, at a speed of over 100 fps. This is partly due to Google's Cloud IoT Edge data management and processing stack.


The Edge TPU is not the same as the chips that accelerate Google's data center algorithms - these TPUs are liquid-cooled and designed to be inserted into server racks. The size of the Edge TPU is about a quarter of a US cent and can perform offline computations to complement local microcontrollers and sensors. Additionally, they do not train machine learning models but instead use the lightweight Google TensorFlow machine learning framework called TensorFlow Lite for inference.


Today, Google also introduced the Coral Dev Board Mini, which offers a smaller form factor, lower power consumption, and lower cost compared to the Coral Dev Board. The Mini combines the new Coral Accelerator module with the MediaTek 8167s system-on-chip to create a board that is "excellent" for 720p video encoding and decoding as well as computer vision use cases. It will be available in the first half of 2020.Google also stated that, in addition to the original 1GB configuration, it will soon offer new features of the Coral System-on-Module with 2GB and 4GB RAM. Finally, it mentioned that ASUS will soon release the Tinker Edge T single-board computer powered by the Coral System-on-Module, which features a range of I/O interfaces, multiple camera connectors, programmable LEDs, and color coding.


Billy Rutledge, Director of the Coral team, wrote in a blog post, "As always, we have been looking for ways to improve the platform... Since its release, we have been excited about the various applications that have been built on Coral, covering a wide range of industries from healthcare and agriculture to smart cities. More and more industries are beginning to recognize the value of on-device AI, as the speed of local inference can save a significant amount of bandwidth and cloud computing costs, while keeping data local can protect user privacy."

RECENT AI NEWS

RECENT AI TOOLS