Forlinx, an NXP Gold Partner, is bringing higher AI performance to industrial edge applications with the FAI-ARA240-M, an M.2 AI accelerator based on the Ara240 discrete NPU. With decades of experience in ARM architecture and deep integration within NXP’s ecosystem, Forlinx and NXP are helping advance Edge AI capabilities with the FAI-ARA240-M AI accelerator.
Multi-Modal Industrial Security Solution
Across industries like industrial IoT, edge AI, and autonomous systems, single-sensor data is rarely enough. Real-time decision-making requires fusing high-bandwidth data streams to execute complex, safety-critical tasks smoothly.Â
To address these requirements, Forlinx and NXP have developed a multimodal AI solution for industrial security based on i.MX 95 platform and the FAI-ARA240-M. The solution provides a complete processing pipeline capable of handling multiple data types, including text, images and video, enabling them to be processed and analyzed at the edge.
The solution combines the 8 eTOPS eIQ Neutron NPU integrated into the NXP i.MX 95 with the Ara240 DNPU, delivering up to 40 eTOPS of additional AI acceleration. By distributing AI workloads across the two NPUs, the architecture provides additional compute capacity for demanding inference tasks while allowing the host processor to remain available for other system functions.
This approach demonstrates how dedicated AI acceleration can be integrated into an existing embedded platform without requiring a complete redesign of the host system.
FAI-ARA240-M Built for the Industrial Edge
The FAI-ARA240-M is a purpose-built M.2 module designed to bring massive AI throughput to existing industrial platforms. It enables complex AI workloads to run independently, ensuring the host processor remains responsive for primary control tasks.
Key Technical Highlights:
Massive Throughput: With up to 40 equivalent tera operations per second (eTOPS) of dedicated AI compute performance, the FAI-ARA240-M enables real-time execution of advanced models that were previously limited to cloud platforms. This makes it ideal for demanding workloads such as high-speed computer vision, multi-stream intelligent video analytics, large-scale industrial inspection and autonomous navigation systems.
Large-Scale Model Support: Available with 8GB or 16GB of low-power double data rate 4 (LPDDR4) memory, the accelerator handles larger models and high-throughput workloads, including vision, intelligent video analysis and large language learning models (LLMs) directly at the edge.
Seamless Software Ecosystem: Built on NXP’s mature AI software ecosystem, the FAI-ARA240-M supports popular frameworks including TensorFlow, PyTorch, ONNX. Using the eIQ AI software environment, developers can rapidly deploy trained models while benefiting from support for multiple precision formats such as INT4, INT8, and mixed precision. This significantly reduces development complexity and accelerates time-to-market.
Standardized Integration: The M.2 M-Key (2280) form factor with PCIe Gen4 (and USB 3.2 Gen1 support) allows for easy integration into NXP-based platform, such as the i.MX 8M Plus or i.MX 95 series.
Hardened Security & Reliability: Features secure boot and root-of-trust capabilities, industrial temperature ratings, and a 10–15 year product longevity.
YOLOv8 Multi-Stream Video Detection Demo
To showcase the integration of the hardware and software platform, we demonstrate a YOLOv8-based video detection application that highlights the ease of SDK deployment and the flexibility of the solution.
The setup is straightforward: the FAI-ARA240-M is connected to the OK-MX9596-C development board through an M.2 adapter and PCIe interface.Â
After installing the Ara240 SDK and YOLOv8 GStreamer package, developers can run single-, four-, or eight-stream video object detection directly on the platform.
This demonstration provides a practical example of how dedicated AI acceleration can be applied to multi-stream computer vision workloads at the industrial edge.
Powering Next-Generation Applications
Industrial Automation: Real-time defect detection, machine vision, and multi-modal inspection.
Intelligent Security: Multi-stream video analytics, people/vehicle detection, and perimeter monitoring.
Robotics: On-device perception, object detection, and autonomous navigation.
Smart Transportation: Traffic monitoring and real-time video analytics.
The FAI-ARA240-M gives developers a practical way to scale AI performance without replacing the host platform. By combining dedicated AI acceleration, flexible memory configurations, and NXP’s software ecosystem, it helps bridge the gap between prototype and production for industrial Edge AI applications.
