Google introduced artificial intelligence and machine learning concepts to hundreds of thousands of people with their AIY projects kit such as the AIY Voice Kit with voice recognition and the AIY Vision Kit for computer vision applications.
The company has now gone further by unveiling Edge TPU, its own purpose-built ASIC chip designed to run TensorFlow Lite ML models at the edge, as well as corresponding AIY Edge TPU development board, and AIY Edge TPU accelerator USB stick to add to any USB compatible hardware.
Google Edge TPU (Tensor Processing Unit) & Cloud IoT Edge Software
Edge TPU is a tiny chip for machine learning (ML) optimized for performance-per-watt and performance-per-dollar. It can either accelerate ML inferencing on device, or can pair with Google Cloud to create a full cloud-to-edge ML stack. In either case, local processing reduces latency, remove the needs for a persistent network connection, increases privacy, and allows for higher performance using less power.
The chip will support the new Cloud IoT Edge software that lets you execute ML models trained in Google Cloud on the Edge TPU or on GPU- and CPU-based accelerators. Cloud IoT Edge can run on Android Things or Linux OS-based devices.
It comes with three main components:
- A runtime for gateway class devices, with at least one CPU, to locally store, translate, process, and derive intelligence from data at the edge, while interoperating with the rest of Cloud IoT platform.
- The Edge IoT Core runtime that more securely connects edge devices to the cloud.
- The TensorFlow Lite-based Edge ML runtime that performs local ML inference using pre-trained models
The more you can do at the edge, the better, but for tasks like training and more powerful frameworks the Cloud is still needed.
AIY Edge TPU Development Board
The company will also offer a development board with a system-on-module (SoM) combining Edge TPU with an NXP i.MX 8M processor, and baseboard exposing ports and I/Os.
AIY Edge TPU board specifications:
- Edge TPU module specifications
- SoC – NXP i.MX 8M with a quad core Cortex-A53 processor, real-time Cortex-M4F microcontroller, and Vivante GC7000 Lite Graphics
- ML accelerator – Google Edge TPU coprocessor
- System Memory – 1 GB LPDDR4
- Storage – 8GB eMMC Flash
- Connectivity
- Wi-Fi 2×2 MIMO (802.11b/g/n/ac 2.4/5GHz)
- Bluetooth 4.1
- Dimensions – 48 mm x 40 mm
- Baseboard specifications
- Storage – MicroSD slot
- USB – USB Type-C OTG port, USB type-C power port, USB Type-A 3.0 host, and USB Micro-B serial console
- Networking – Gigabit Ethernet port
- Audio – 3.5mm audio jack (CTIA compliant), 2x digital PDM microphones, 2.54mm 4-pin terminal for stereo speakers
- Video Output – HDMI 2.0a (full size), 39-pin FFC connector for MIPI-DSI display (4-lane)
- Camera I/F – 24-pin FFC connector for MIPI-CSI2 camera (4-lane)
- Expansion – 40-pin expansion header for GPIO’s
- Power Supply – 5V DC via USB Type-C
- Dimensions – 85 mm x 56 mm
The board takes some cues from Raspberry Pi with a credit card form factor, and a 40-pin connector, but the position of the connectors won’t make it compatible with RPI accessories.
Edge TPU Dev Board will run Debian Linux or Android Things, and support TensorFlow Lite.
AIY Edge TPU Accelerator USB Stick
However, if you’re already familiar with a specific development board and environment, you may not want to buy another one, and go through the learning curve again. Just like Intel’s Movidius Neural Compute Stick, AIY Edge TPU accelerator is a USB stick designed to add machine learning acceleration to existing boards via a USB interface. But instead of relying on Myriad 2 VPU and a USB 3.0 type A port, Edge TPU accelerator comes with Edge TPU chip and a USB type- C port.
AIY Edge TPU Accelerator specifications:
- ML accelerator – Google Edge TPU coprocessor
- Connector – USB Type-C (data/power) compatible with Raspberry Pi boards at USB 2.0 speeds only
- Dimensions – 65 mm x 30 mm
The casing includes mounting holes for attachment to host boards such as a Raspberry Pi Zero, the upcoming Libre Computer AML-S805X-AC board, or your custom board.
Just like the development board, Edge TPU accelerator will work with Debian, Android Things, and TensorFlow Lite framework.
Both development platforms will be available online laster this year, startin in the US with other countries following shortly after. You can find more details about Edge TPU chip and IoT Cloud Edge software on the product page, and/or register your interest for the development board or/and USB stick on AIY Projects website.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
Pricing?
Nothing that I could find yet. But it’s supposed to be low cost and low power. Intel Neural Compute Stick is $79, so Google’s Edge TPU accelerator should be the same price or lower, unless it’s also much more powerful.
Wonder if the GPU or Goggle TPU will be able to do video encoding? the iMX.8M does not have a HW Video encoder. Or if the plan is not to store any video captured by the camera.