Four Hailo-8 M.2 module options
The Hailo-8 M.2 AI Acceleration Module family adds dedicated neural network inference to compatible embedded computers, industrial PCs and edge gateways. Each module uses the Hailo-8 processor, delivering up to 26 TOPS of AI compute for applications such as object detection, image classification, machine vision and video analytics.
Select the part number according to the host’s M.2 slot and available installation space:
- HM218B1C2FAE - 2280 M Key: A 22 × 80 mm module with a PCIe Gen 3 x4 interface
- HM218B1C2HAE - 2242 M Key: A 22 × 42 mm module with a PCIe Gen 3 x4 interface
- HM218B1C2KAE - 2230 A+E Key: A 22 × 30 mm module with a PCIe Gen 3 x2 interface
- HM218B1C2LAE - 2280 B+M Key: A 22 × 80 mm module with a PCIe Gen 3 x2 interface
All four listed part numbers are extended-temperature versions, specified from −40°C to 85°C. They share the Hailo-8 AI processing platform, while their connectors, dimensions and PCIe lane configurations differ.
Edge AI and computer vision applications
Hailo-8 modules process neural network workloads on the device, close to the cameras or sensors producing the data. The architecture supports multiple models and streams, allowing developers to create pipelines with concurrent inference tasks. A deployment might combine object detection with classification, or analyze feeds from several cameras, depending on its models and host configuration.
These capabilities make the module family relevant to smart surveillance, industrial visual inspection, retail analytics, robotics and AI-enabled edge gateways. The 26 TOPS rating describes maximum processor compute capability; application frame rate and latency depend on the models, input resolution and complete system design.
Choosing a compatible host
M.2 connector type is only one part of module selection. Before choosing HM218B1C2FAE, HM218B1C2HAE, HM218B1C2KAE or HM218B1C2LAE, check the host’s PCIe electrical interface, supported lane count, mounting position, power delivery and thermal design. A slot that physically accepts a module may not provide the PCIe connection that module needs.
For a host with the required M Key PCIe x4 connection, compare the 2280 and 2242 M Key options by available board space. For a design requiring a B+M Key 2280 or A+E Key 2230 module, check the corresponding PCIe x2 connection and mechanical fit.
Software development
Hailo’s software suite supports model preparation and deployment. The Hailo Dataflow Compiler translates supported trained neural networks for the Hailo-8, while HailoRT runs inference on the host. Hailo also provides model resources and application examples for developing edge AI solutions.
The module family supports workflows using TensorFlow, TensorFlow Lite, Keras, PyTorch and ONNX. Hailo lists x86 and Arm host architectures and Linux and Windows operating systems. Confirm the current software release requirements for the exact host and deployment.
