How acontis Delivers Sub-Millisecond EtherCAT Control for Physical AI on Qualcomm Dragonwing™ Processors

Executive Summary

acontis technologies, the global market leader in EtherCAT MainDevice software with its flagship product EC-Master powering more than 2 million controllers, selected Dragonwing IQ-9075 as the compute foundation for its real-time EtherCAT stack for next-gen applications like humanoids or AI-empowered PLCs. After extensive benchmarking, acontis found that the Dragonwing IQ-9075 delivers outstanding EtherCAT cycle time performance and extremely low jitter while simultaneously providing the AI inference, vision processing, and sensor fusion capabilities that Physical AI applications demand.

The Challenge: Connecting AI Decisions to Deterministic Motion

Physical AI systems combine two very different workloads. The AI side ingests large volumes of camera and sensor data for perception, path planning, and vision processing. The control side sends very short messages to drives and I/O, but those messages are highly time-sensitive and must arrive at the right time, every cycle. Even a delay in ms-range can cause frame-loss and thus be the difference between a smooth correction and a collision. As robots move closer to people and operate at higher speeds, deterministic communication becomes a precision and safety requirement.

Technical requirements

• Sub-millisecond cycle times with minimal jitter for deterministic motor control across EtherCAT drive networks.
• Low CPU utilization so EtherCAT communication leaves headroom for AI inference, vision, and path planning on the same SoC.
• Multi-kHz update rates for the attached EtherCAT SubDevices.
• ROS2 integration for rapid prototyping and deployment in the robotics ecosystem.
• Industrial reliability including cable redundancy, MainDevice redundancy and hot-connect
• Reliable and proven basis for Fail Safe over EtherCAT (FSoE).
• Scalable platform family to span volume PLCs through high-end robotics without re-engineering the software stack.

Why Competing Platforms Fell Short

acontis has deployed EC-Master across every major compute architecture. Each presents trade-offs for applications requiring both AI compute and real-time control:

• A leading GPU-based platform: Superior raw AI compute, but EtherCAT real-time performance does not match its AI capabilities. Larger, more expensive, and higher power when precise motor control is equally important.
• x86-based processor: Some legacy x86 platforms remain among the fastest EtherCAT platforms acontis has benchmarked, but newer x86-based systems have not always matched that real-time latency performance. Limited on-chip AI acceleration means separate NPU hardware may be needed for inference workloads.
• Mid-range Arm processors: Cost-effective for simple PLCs, but lack the AI compute headroom for sensor fusion, vision processing, and path planning that Physical AI demands.

The Solution: Dragonwing IQ Processors as the Physical AI Foundation

acontis ported EC-Master to Dragonwing IQ-9075 and prepared it for release on the Dragonwing IQ-8275 and the Dragonwing IQ-X processors, developing optimized real-time drivers for the platform’s internal ethernet controllers. Performance- and functional testing was done using the Dragonwing IQ-9075 EVK as well as Advantech’s AFE-A503 embedded controller which provides four independent ethernet ports and many more interfaces for maximum flexibility combined with the performance of the Dragonwing IQ-9075; this solution provides a complete compute-to-control stack for Physical AI.

Architecture: From Perception to Motion on One SoC

The Dragonwing IQ9 series is designed to offload key stages of the physical AI pipeline to a dedicated hardware engine, engineered to maximize efficiency and performance per watt. These engines share memory but not execution pipelines, so AI workloads never contend with deterministic motor control timing.

Cameras and high-bandwidth sensors feed the Dragonwing IQ9 processing pipeline for perception, path planning, and vision. Torque, position, encoder, drive, and safety data move through EtherCAT. The table below maps each architecture layer to its hardware engine and the practical benefit it delivers.

EC-Master and the Software Ecosystem

acontis technologies has spent 25 years building one of the most widely deployed EtherCAT MainDevice stack in the world. EC-Master has been provided to customers since 2004 and has since been adopted by tier-1 industrial OEMs across robotics, PLCs, CNC, factory automation, medical technology, aerospace, semiconductor and many more industries. 

On the Dragonwing processors, the acontis EC-Master can be paired with additional layers:

• ROS2 integration (newly released). EC-Master fully integrated into Robot Operating System 2, allowing robotics teams to prototype with familiar tools and deploy to production without rewriting the control layer.
• Qualcomm® AI Hub. Model import, conversion, quantization, and validation for the Hexagon NPU. A library of 250+ optimized models enable AI inference to run alongside EC-Master on the same SoC.

Performance Benchmarking

acontis measures EtherCAT performance by what production deployments require: cycle time accuracy (jitter), round-trip latency, and CPU utilization under sustained operation.

EC-Master on Dragonwing IQ-9075: Measured Performance

Test: EC-Master running on Dragonwing IQ-9075 with Linux® CLOCK_MONOTONIC real-time scheduling. 1 ms target cycle time. Full EtherCAT frame processing including send, receive, and application workload.

Key takeaway: EC-Master on the Dragonwing IQ9 delivers outstanding low jitter - down to single-digit microsecond and allows continous stable round-trip times of approximately 100µs  (tested with acontis standard performance measurement bench, 7 slaves with 512 byte process data) - the control foundation needed for physical AI systems that must turn perception into motion.

Target Applications

Conclusion

acontis brought EC-Master to Dragonwing because physical AI needs both AI acceleration and deterministic control. The Dragonwing IQ9 series provides the embedded compute for perception, planning, and vision. EC-Master provides the real-time EtherCAT layer for drives, encoders, sensors, tools, and safety I/O. 

EC-Master provides the real-time EtherCAT layer for drives, encoders, sensors, tools, and safety I/O. For technical teams, the value is direct: predictable cycle timing, low round-trip latency, low CPU overhead, mature EtherCAT features, and integration paths for both industrial automation and ROS-based robotics.

Learn More

• Advantech Dragonwing IQ9-based embedded controllers: AFE-A503 Robot Control System
• acontis EC-Master EtherCAT Stack: acontis.com
• EtherCAT for ROS2 
• Dragonwing Platform: qualcomm.com/dragonwing