Hailo-8L AI acceleration for embedded systems
The Hailo-8L M.2 Entry-Level AI Acceleration Module family brings dedicated neural network processing to compatible embedded computers, industrial PCs and edge gateways. Both variants are powered by the Hailo-8L processor, which delivers up to 13 TOPS for on-device AI inference.
There are two extended-temperature part numbers in this family. HM21LB1C2KAE is the 2230 A+E Key module, measuring 22 × 30 mm. HM21LB1C2LAE is the 2280 B+M Key module, with a nominal size of 22 × 80 mm. Both use a PCIe Gen 3 x2 host interface and are specified for −40°C to 85°C operation.
Choose the right connector and board size
The A+E and B+M versions offer the same Hailo-8L processor capability in different physical formats. The 2230 A+E Key option suits a compatible host designed for a shorter module. The 2280 B+M Key option suits a compatible host with space and mounting provision for the longer board.
Check the host’s M.2 documentation before choosing either part number. The connector must provide the required PCIe connection, and the system must support the module’s mechanical dimensions, power needs, software and thermal requirements. A module fitting into a slot does not by itself confirm that the host can operate it.
Computer vision and video analytics
Hailo-8L modules are designed for neural network inference close to cameras and sensors. Potential applications include object detection, image classification, smart surveillance, retail analytics and visual quality inspection. Local processing can support systems that need timely analysis without sending every inference request to a remote service.
The architecture supports multiple streams and models, so developers can plan applications with concurrent AI tasks. For example, a system may detect objects in a video feed and use a second model to classify them. Achievable frame rates depend on the selected models, input resolution, number of streams and host configuration. The 13 TOPS figure is maximum processor compute capability, not a guaranteed performance result for every application.
Hailo software development
Hailo provides the Dataflow Compiler for preparing supported trained neural networks and HailoRT for running inference on the host. Its software ecosystem also includes model resources and application examples for developing and evaluating edge AI workloads.
Supported workflows include TensorFlow, TensorFlow Lite, Keras, PyTorch and ONNX. Hailo lists x86 and Arm host architectures and Linux and Windows operating systems for the module family. The Hailo-8L software ecosystem is compatible with the higher-capacity Hailo-8 product line, which can help teams plan future designs when more AI performance is required.
Industrial design considerations
Both listed modules support an extended operating temperature range of −40°C to 85°C. For a finished industrial device, evaluate the temperature ratings of the host and other components as well as the enclosure, power supply and cooling arrangement.
