As edge systems evolve to support AI-driven perception, high-speed data acquisition, and real-time control, a single processing architecture is often no longer sufficient. Many applications demand deterministic, low-latency control and I/O handling alongside high-throughput AI inference, creating a need for heterogeneous compute architectures. To address this challenge, iWave has validated a PCIe Gen3 x4 integration of Zynq UltraScale+ MPSoC and NVIDIA Jetson Orin, an architecture that combines FPGA-based real-time processing with GPU-accelerated AI.
In this integration, Zynq™ UltraScale+™ MPSoC acts as the real-time and data-ingestion engine, handling high-speed sensor interfaces, deterministic processing, and pre-processing in programmable logic. The NVIDIA® Jetson™ Orin Nano complements this by providing high-performance GPU and Tensor Core acceleration for compute-intensive AI workloads such as deep learning inference, vision analytics, and signal classification. Connected via a high-bandwidth PCIe interface, the solution enables efficient data movement between the two domains, allowing each processor to operate where it performs best.
Jetson Orin Nano functions purely as a PCIe-attached AI accelerator, offloading >95% of DL compute from the MPSoC. The MPSoC handles deterministic control, high-speed I/O, and data pre-processing in programmable logic, while AI-intensive workloads such as deep-learning inference are offloaded to a GPU-based platform.
PCIe-Based Integration and Validation
The architecture represents, Jetson Orin Nano operating as a PCIe Root Port, while ZU11EG SoM is configured as a PCIe Endpoint. The two platforms are interconnected using an M.2 NVMe edge-card interface with appropriate board-level adaptations.
On the software side:
- The Jetson runs Ubuntu-based NVIDIA JetPack (L4T Linux)
- The MPSoC runs Yocto/PetaLinux
- Xilinx XDMA IP is implemented on the FPGA side
- The XDMA Linux driver enables high-throughput data exchange over PCIe
This PCIe link enables low-latency, high-bandwidth data movement between the FPGA fabric and the Jetson GPU, allowing compute-intensive workloads such as deep learning inference and vision processing to be offloaded efficiently to the GPU.
AI Object Detection Example: Jetson Orin Nano + ZU11EG MPSoC
To validate the heterogeneous compute architecture, iWave implemented an end-to-end object detection pipeline using the YOLOv3 deep learning model.
In this setup:
- The Zynq™ UltraScale+™ MPSoC handles sensor interfacing, frame capture, real-time pre-processing, and PCIe data transfer
- The Jetson Orin Nano executes YOLOv3 inference using CUDA and TensorRT acceleration
- Inference results, including bounding boxes and confidence scores, are sent back to the MPSoC over PCIe for system-level decision-making
This closed-loop pipeline demonstrates low-latency operation and efficient workload partitioning between deterministic FPGA processing and GPU-based AI acceleration.
MPSoC and Jetson Integration
Output Image after YOLOv3 post processing
Target Applications
This Jetson + Zynq™ UltraScale+™ MPSoC architecture is well suited for a wide range of real-time and AI-driven applications:
- Industrial Machine Vision: FPGA-based image capture and pre-processing combined with GPU-accelerated inspection and defect detection
- High-Speed Data Acquisition: Wideband signal digitization for radar, lidar, and scientific instrumentation with real-time visualization
- Robotics and Automation: Low-latency sensor fusion and motion control on FPGA with GPU-accelerated path planning and perception
- Medical Imaging and Surveillance: Real-time processing of high-resolution video streams for diagnostics, guidance, and anomaly detection
The integration of iWave’s ZU11EG-based Zynq™ UltraScale+™ MPSoC SoM with the NVIDIA Jetson Orin Nano over PCIe creates a powerful heterogeneous compute platform for next-generation edge AI systems. By combining deterministic FPGA-based processing with high-performance GPU acceleration, this solution enables scalable, low-latency, and AI-enabled data pipelines for demanding applications in vision, robotics, instrumentation, and communications.
iWave Global is a leading provider of embedded computing solutions, FPGA System on Modules, and ODM design services. With over 26 years of engineering excellence, iWave specializes in high-performance SoMs built on cutting-edge processor technologies. Through deep domain expertise in FPGA, RF, AI, and edge compute architectures, iWave partners with global OEMs to accelerate product development, reduce technical risk, and deliver reliable solutions for mission-critical applications.
For more information, reach out to us through mktg@iwave-global.com
