- Google has expanded its Coral portfolio with the addition of a chip and dev board
- Small but powerful, they are based on TensorFlow Lite
It seems to be a good start to 2020 for Google. Now, finally out of the beta phase, Google has launched two new products: Coral Accelerator Module and Coral Dev Board Mini have been launched for AI and ML applications.
High accuracy multi-chip
The Coral Accelerator Module is a multi-chip module that has an on-board Edge TPU coprocessor that is capable of performing up to 4 trillion operations per second (TOPS) while consuming 0.5 watts for two TOPS each. Based on the open-source deep learning framework called TensorFlow Lite, the chip is capable of performing high-accuracy image classification with the help of AutoML Vision Edge.
On the hardware design side, this SMT chip has a dimension of 15 x 10 x 1.5 mm with a 120-pin LGA (Land Grid Array). It has PCIe and USB serial interfaces with an optional I2C control interface.
Offers Great Connectivity
Just like the Coral Accelerator Module, the Coral Dev Board Mini has an Edge TPU coprocessor that is capable of 4 TOPS at 0.5 watts for two TOPS each. The board is essentially a computer that has connectivity for SoC, ML and wireless technologies such as Wi-Fi 5 and Bluetooth 5.0. Its CPU is based on MediaTek 8167s SoC (Quad-core Arm Cortex-A35) with a RAM of 2GB and Flash Memory of 8GB. Mendel, a lightweight derivative of Debian Linux runs on-board, thus supporting various Linux tools.
For I/O, the board has a 40-pin GPIO header and 2x USB Type-C ports. TensorFlow Lite models can be compiled to run on the Edge TPU.
Google plans to commercially make them available in the first half of 2020. The company will be displaying this launch at this year’s Consumer Electronics Show in Las Vegas, the US.