Coral Accelerator Module
A solderable multi-chip module including the Edge TPU
The Accelerator Module is a surface-mounted module that includes the Edge TPU and its own power control. It provides accelerated inferencing for TensorFlow Lite models on your custom PCB hardware.
The Edge TPU is capable of performing 4 trillion operations per second (4 TOPS), using 2 watts of power—that's 2 TOPS per watt. For example, it can execute state-of-the-art mobile vision models such as MobileNet v2 at almost 400 frames per second, in a power efficient manner.
The Coral Accelerator Module is a machine learning accelerator module used for inferencing, optimized for vision applications. It offers an excellent performance / cost tradeoff at the vision processing entry-level market, and is in mass production.
Performance: 4 TOPS peak performance (int8), 2 TOPS per Watt (2 W peak power draw)
e.g. one EdgeTPU can execute MobileNet v2 at almost 400 frames per second.
Operating temp: -40 to +85°C (recommended: -20°C - 70°C)
Dimensions: 15.0 x 10.0 x 1.5 mm
Chipset: Google Edge TPU and PMIC
Mounting type: SMT (surface mount), 120-pin LGA
Reel information: 1000 pieces per reel. approximate weight: 1.09 kg
MSL rating: MSL 3 (JEDEC Standard J-STD-020). Tape and reel must be sealed with the provided anti-humidity plastic bag.
After the packing is opened, it should be stored at ≤30 °C / ≤60% RH, and used within 168 hours.
When the color of the humidity indicator on the packing changes, the product should be baked before soldering.
Interface: PCIe Gen 2 or USB 2.0. USB 3.0 available on request (requires special design considerations)
Power supply (VIN): 3.1 V - 3.63 V (typical: 3.3 V). Average current drawn is typically less than 500 mA, but can reach up to 3A during inferencing.
Note: PMIC Digital I/O power supply can be supplied with 1.7V min, typical 1.8V.
Refer to the datasheet for more in-depth information on current transients.
Software support: Requires 64-bit versions of Debian 10 or Ubuntu 16.04 or newer, x86-64 or ARMv8 system architectures. Supports Windows 10 64-bit on x86-64 as well. Requires the Edge TPU runtime and Coral PCIe driver.
RoHS compliant: yes
Availability: Australia, European Union (except Czech Republic), Ghana, Hong Kong, India, Indonesia, Israel, Japan, Kenya, Malaysia, New Zealand, Oman, Philippines, Singapore, South Korea, Taiwan, Thailand, United States, Vietnam
Note: this is a module to be designed into your own PCB designs.
To increase performance, you can use several of the Coral accelerator modules in parallel or in a serial setup (depending on your ML model).
Google part number: G313-06329-00
Asus part number: 90AN01G0-B0XAY0
Datasheet & Benchmarks
Additional Legal Information
The products offered by Google are unrelated to the products offered under the CORAL trademarks owned by Orient Development Enterprises Ltd. in Taiwan.