SEU Evaluation of Hardened-by-Replication Software in RISC-V Soft Processor

The interest of the space industry around soft processors is increasing. However, the advantages in terms of costs and customizability provided by soft processors are countered by the reliability issues deriving by Single Event Effects, especially Single Event Upsets. Several techniques have been proposed to tackle these issues, both at the hardware- and software levels. Software approaches rely on replicating data and computations to cope with SEUs affecting the memory where the binary code is stored. Thanks to open licenses, RISCV solutions are steadily growing in popularity among the set of available soft processors. In this works, we present a reliability evaluation of four different benchmarks running on the RI5CY soft processor implemented on SRAM-based FPGAs. The reliability of the baseline and hardened-by-replication versions of the software benchmarks are evaluated against SEUs induced faults both at the software and hardware architecture levels through fault injection campaigns in the microprocessor memory and configuration memory, respectively. Results assess how the adoption of the hardening-by-replication technique at the software level slightly improves reliability against software related faults but degrades reliability against architectural faults, making it an inefficient solution when it is not combined with hardware robustness.