Beam and Field Testing of Configuration Self-repair in Xilinx FPGAs

SRAM-based Field Programmable Gate Arrays play a key role in digital electronics systems for Nuclear and Subnuclear Physics experiments. They make it possible to process and transfer several gigabytes of data in real-time and they are reconfigurable in the field. There is a growing interest to exploit the flexibility of these devices also on-detector, where their usage has to take into account single event upsets in the configuration. In fact, these effects may disrupt the intended functionality and the study of solutions for tolerating them is a very active research field. This work focuses on a custom softcore-based system tailored to Xilinx SRAM-based FPGAs. The system, or scrubber, is redundant at the configuration frame level and it majority votes frames for detecting and correcting upsets. We discuss the system architecture and compare two possible triple modular redundancy implementation choices, by providing a cost versus benefit evaluation for each. We provide an insight about the details of configuration correction in real-time and of the scrubber self-repair features. We report results from 62-MeV proton irradiation tests, which prove that our core can extend the lifetime of a benchmark circuit by 3.9 times. We also present results from the field-testing of our self-correcting firmware at the SuperKEKB collider (KEK, Tsukuba, JP), during its Phase-2 operation. Finally, we comment about the portability of our core over different families of devices produced by Xilinx, both on latest-generation and older devices.