NOBUGS 2026

At the European XFEL (EuXFEL), modern photon science experiments generate high-rate data streams requiring low-latency processing and precise timing. Such constraints demand novel approaches to data processing that go beyond conventional centralized architectures and enable computation closer to the data source.

In this work, an FPGA-based edge computing platform integrated into Karabo, the Supervisory Control and Data Acquisition (SCADA) system used at the EuXFEL, is presented. The platform is based on a Xilinx Zynq system-on-chip (SoC) architecture, integrating programmable logic (PL) and an embedded processing system (PS). This enables dynamic partitioning of data processing tasks between hardware and software to optimize latency and throughput, while supporting real-time interaction with experimental signals via configurable interfaces.

Integration is achieved by deploying Karabo components directly on the embedded system, enabling the platform to operate as a native element of the control and data acquisition framework. Communication with FPGA resources is performed via a Python-based environment using PYNQ, enabling direct control and data exchange between PS and PL. The modular design allows new data processing methods to be integrated with minimal effort, supporting flexible and adaptive processing workflows.

The effectiveness of the proposed approach has been demonstrated using a liquid sample droplets injection system at the SPB instrument, where a photodiode signal is digitized via an analog-to-digital converter (ADC) and processed in real time to estimate the temporal delay between the sample and the X-ray beam, enabling precise temporal alignment. The system achieves reliable real-time performance and scalable data processing, supporting high data acquisition rates in the 100 kHz range.