FPGA-based multi-channel DAQ systems with external PCI express link to GPU compute servers

Abstract

Our group at KIT has been developing data acquisition (DAQ) systems for many years mainly for large physics experiments like the KATRIN neutrino experiment or the Pierre Auger cosmic ray observatory. The DAQ systems were continuously enhanced as new technologies became available. The core of the DAQ systems are field programmable gate arrays (FPGAs). Trigger functions running on the FPGAs select relevant events out of the permanent data stream of the ADCs and pass it over PCI bus to a embedded Linux computer for further analysis and storage. Modern experiments have raising requirements in both data rate and complexity of trigger and analysis function. To achieve a flexible and fast data link we developed a PCI to PCI Express (PCIe) adapter board which can be connected to any PC equipped with a standard PCIe plug-in adapter. We use this adapter to replace the embedded Linux system and to connect external GPU servers directly to the DAQ system. With this powerful data processing facility at the end of the data chain we can run complex third level trigger functions, reconstruction algorithms and analysis calculations. With PCIe as fast data link and GPU computing together with the well established FPGA unit we achieved a substantial enhancement of our DAQ system.