FPGA-based fault injector for SEU-robustness analysis of ScOSA

Abstract

The Scalable On-board Computer for Space Avionics (ScOSA) project aims to develop an on-board computer which offers both reliability and high-performance through the use of a heterogeneous distributed system of commercial-off-the-shelf and radiation-hardened processors. This system should operate without failures even in the presence of single-event upsets (SEUs), which are common occurrences for electronic systems in space. The ScOSA middleware includes several fault detection, isolation and recovery (FDIR) mechanisms for coping with faults, but their effectiveness in the presence of radiation has not yet been proven, as testing such effects on the ground is challenging. This paper presents our approach to investigate the effect of single-event upsets on the ScOSA system and the effectiveness of its error handling mechanisms in their presence. A fault injector has been instantiated in the FPGA co-processor of a commercial-off-the-shelf Xilinx system-on-chip from the Zynq 7000 family using a Microblaze soft processor, which is used to simulate the effect of SEUs by flipping bits in the main memory used by the kernel, middleware and applications. A machine-learning-based image processing algorithm will be used as an example application and run using the ScOSA middleware while the fault injector is active. The system will be executed multiple times, with faults injected into different memory locations and at different times in each run. The system will be monitored for FDIR events and unrecoverable failures. The operation of the middleware and the results of the sample application will be compared to the results of a golden run, where no faults are injected, to assess the number of unhandled errors at the middleware and application levels. The results are classified by severity, such as incorrect algorithm results, handled FDIR events and unhandled system crashes. These results will then be correlated with the fault location, such as kernel or application memory. By applying SEU simulation techniques to an on-board software system, we aim to demonstrate the usefulness of such simulations as well as guiding the further development of the ScOSA system to target further SEU mitigation efforts and improve the systems robustness, as well as characterizing the systems robustness to SEUs occurring in different locations.