Description The Coral M.2 Accelerator B+M Key adds a Google Edge TPU coprocessor to compatible computers and embedded systems through a PCIe M.2 slot. It delivers up to 4 TOPS of int8 machine-learning inference while using only 2 TOPS per watt, enabling low-latency AI processing without sending data to the cloud. The 22 x 80 mm M.2-2280 module is designed for TensorFlow Lite models compiled for the Edge TPU. Its B+M-keyed connector supports installation in compatible B-key or M-key sockets that expose a PCIe Gen2 x1 interface. Features Google Edge TPU coprocessor for local machine-learning inference Up to 4 TOPS of int8 performance 2 TOPS per watt power efficiency M.2 B+M key connector in the M.2-2280 form factor PCIe Gen2 x1 host interface Supports TensorFlow Lite models compiled for the Edge TPU Suitable for compatible Debian-based Linux and Windows host systems Specifications Accelerator Google Edge TPU coprocessor Peak performance 4 TOPS (int8) Power efficiency 2 TOPS per watt Host interface PCIe Gen2 x1 Connector M.2 B+M key Form factor M.2-2280-B-M-S3 Dimensions 22 x 80 x 2.35 mm Weight 5.8 g Supply voltage 3.3 V +/- 10% Operating temperature -20 to +85 degrees C, subject to thermal management Storage temperature -40 to +85 degrees C Relative humidity 0 to 90%, non-condensing Applications On-device image classification and object detection Smart cameras and machine-vision systems Robotics and autonomous equipment Industrial inspection and predictive monitoring Edge AI gateways and embedded prototypes Product List 1 x Coral M.2 Accelerator B+M Key Other Content A matching B-key or M-key socket does not by itself guarantee compatibility. The host slot must provide a PCIe connection rather than a SATA-only interface, and the host platform must support the Coral PCIe driver and runtime. Check the computer or carrier-board documentation before installation. Install the Edge TPU runtime and PCIe driver before testing a compiled TensorFlow Lite model. For sustained workloads, provide suitable airflow or heat dissipation and observe electrostatic-discharge precautions while handling the module. Resources Coral M.2 Accelerator B+M Key Datasheet Coral M.2 Setup Guide PCIe Parameters and Temperature Management Edge TPU Inference Guide Edge TPU Model Compatibility Guide Raspberry Pi 5 Installation Guide for Coral M.2 B+M Key Edge TPU Model Conversion Guide FAQ Is this the Dual Edge TPU model? No. This B+M-key module contains one Google Edge TPU and provides up to 4 TOPS of int8 inference performance. Will it work in every M.2 slot? No. The slot must accept the B+M-keyed 2280 module and expose PCIe Gen2 x1. A SATA-only M.2 slot is not compatible. Which models can run on the accelerator? It runs supported TensorFlow Lite models that have been quantized and compiled for the Edge TPU. Does the module need cooling? Cooling requirements depend on workload and ambient temperature. Sustained inference may require airflow or additional heat dissipation to remain within the operating range.
