Add edge AI through Mini PCIe
The Hailo-8R Mini PCIe AI Acceleration Module adds dedicated neural network processing to a compatible host system. Part number HMP1RB1C2GAE identifies the extended-temperature Full-Mini PCIe module. It delivers up to 13 TOPS of AI compute for applications that analyze images, video and other supported data on the device.
This module is intended for embedded computers and industrial platforms with a suitable mPCIe Full-Mini socket. Its PCIe Gen 3 x1 interface connects the accelerator to the host processor. Mini PCIe is a different form factor from M.2, so buyers should select this module according to the host’s actual socket and electrical specifications.
Computer vision applications
Hailo-8R can support local inference tasks such as object detection, image classification, smart surveillance and industrial visual inspection. Processing camera data at the edge can help an application respond without sending every inference request to a remote server.
The module supports multiple data streams and neural network models. Developers can use this capability to plan pipelines with concurrent vision tasks, subject to the selected models and host system. The 13 TOPS rating is the module’s stated AI compute capability; application frame rate and latency depend on model complexity, input resolution and the complete deployment.
Integration requirements
The Hailo-8R module uses the PCI Express Mini Card Full-Mini F1 form factor and measures 30 × 50.95 mm. Before specifying it for a design, check that the host provides a compatible Mini PCIe socket with the required PCIe connection, mounting space, power delivery and thermal arrangement.
Part number HMP1RB1C2GAE supports an extended operating temperature range of −40°C to 85°C. For industrial equipment, the enclosure, host and surrounding components must also meet the intended environmental requirements.
Software development
Hailo’s software suite includes the Dataflow Compiler for preparing supported trained models and HailoRT for runtime inference on the host. Model resources and application examples can assist with development and evaluation.
Hailo lists support for TensorFlow, TensorFlow Lite, Keras, PyTorch and ONNX workflows, x86 and Arm host architectures, and Linux and Windows operating systems. Check the requirements of the current software release when planning a specific host integration.
