Effectiveness of Software-Based Hardening for Radiation-Induced Soft Errors in Real-Time Operating Systems.

For decades, radiation-induced failures have been a known issue for aero-space systems, in which redundancy mechanisms are employed as a protection method. Due to the shrinking of structures and operating voltages, these failures are increasingly becoming an issue even for terrestrial applications. Unfortunately, redundancy increases costs, area usage, and power consumption, which can hinder its utilization in cost- and power-sensitive safety-critical applications, such as automotive. To overcome this limitation, multiple software-based approaches have been proposed, which assume the existence of an underlying error-free operating system. In this paper, we investigate the radiation reliability of two dependability-oriented real-time operating systems, namely, the popular eCos operating system hardened through aspect-oriented programming methods, and dOSEK, an embedded kernel designed from the ground up having reliability as a major concern. Both operating systems were evaluated through extensive neutron-beam testings on a 28 nm ARM-based state-of-the-art system-on-chip, and their fault tolerance mechanisms reached reductions in the overall cross-sections relative to their baselines up to 91% and 74%, respectively.