The Edge TPU is a new hardware chip, and Cloud IoT Edge is a software stack that extends the company’s cloud AI capability to gateways and connected devices. The new offerings, says Google, let users build and train machine learning (ML) models in the cloud, then run those models on the Cloud IoT Edge device through the power of the Edge TPU hardware accelerator.
“Real-time decision-making in IoT systems is still challenging due to cost, form factor limitations, latency, power consumption, and other considerations,” says Injong Rhee, VP, IoT, Google Cloud. “We want to change that.”
The Edge TPU is a purpose-built ASIC chip designed to run TensorFlow Lite ML models at the edge. When designing it, says Rhee, the company was “hyperfocused” on optimizing for “performance per watt” and “performance per dollar” within a small footprint.
Edge TPUs are designed to complement the company’s Cloud TPU offering, so users can accelerate ML training in the cloud, then have “lightning-fast” ML inference at the edge. “Your sensors become more than data collectors – they make local, real-time, intelligent decisions,” says Rhee.
The Cloud IoT Edge software extends Google Cloud’s data processing and machine learning capabilities to gateways, cameras, and end devices, making IoT applications “smarter, more secure, and more reliable.” It lets users 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. Its key components include the following:
- 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 seamlessly interoperating with the rest of Cloud IoT platform.
- The Edge IoT Core runtime that more securely connects edge devices to the cloud, enabling software and firmware updates and managing the exchange of data with Cloud IoT Core.
- The TensorFlow Lite-based Edge ML runtime that performs local ML inference using pre-trained models, significantly reducing latency and increasing the versatility of edge devices. Because the Edge ML runtime interfaces with TensorFlow Lite, it can execute ML inference on a CPU, GPU or an Edge TPU in a gateway class device, or in an end device such as a camera.
“Cloud IoT Edge, Edge TPU and Cloud IoT Core are opening up completely new possibilities with IoT,” says Rhee. “With powerful data processing and ML capabilities at the edge, devices such as robotic arms, wind turbines, and smart cars can now act on the data from their sensors in real time and predict outcomes locally.”
The company is also introducing a development kit to help users jump-start development and testing with the Edge TPU. The kit, which will be available to developers in October, includes a system on module (SoM) that combines the Edge TPU, an NXP CPU, Wi-Fi, and Microchip Technology’s secure element in a compact form factor.