Radiation Testing of a Multiprocessor Macrosynchronized Lockstep Architecture With FreeRTOS

Abstract

Nowadays, high performance microprocessors are demanded in many fields, including those with high reliability requirements. Commercial microprocessors present a good tradeoff between cost, size and performance, albeit they must be adapted to satisfy the reliability requirements when they are used in harsh environments. This work presents a high-end multiprocessor hardened with a macro synchronized lockstep and additional protections. A commercial dual core ARM cortex A9 has been used as case study and a complete hardened system has been developed. Evaluation of the proposed hardened system has been accomplished with exhaustive fault injection campaigns and proton irradiation. The hardening approach has been accomplished for both baremetal applications and OS-based. The hardened system has demonstrated high reliability in all performed experiments with error coverage up to 99.3% in the irradiation experiments. Experimental irradiation results demonstrate a cross-section reduction of two orders of magnitude.

Bare-Metal Redundant Multi-Threading on Multicore SoCs Under Neutron Irradiation

“…Each block was delimited by means of a verification point, which when triggered, activated customized hardware, so that program context and data were saved in a dual-port private memory where the data could be compared and analysed to detect mismatches. A modification of this approach can be found in [17], where the verification process was performed directly in software. The architecture included some customized hardware IP to support the synchronized execution of redundant blocks of code on multi-core processors.…”

“…The proposal was tested in bare-metal and with a real-time OS. Only a few of the multi-threading based techniques has been tested in accelerated radiation experiments [2], [6], [11], [17] and some of them will be studied in Section V. A more comprehensive survey on multi-threading and lockstep based mitigation techniques can be found in the surveys of both [18] and [19].…”

“…Among all the results of the software-based approaches tested under radiation there were only a few that stood comparison with our proposal. Table IV summarizes the main features and results of three of the state-of-the-art techniques [14], [6], [17]. Beams of a different nature and energy (low energy protons, atmospheric and mono-energetic neutrons) were used for testing in all of the studies, making any direct comparison of the cross-section impossible.…”

“…The Macro Lockstep approach (MLockstep) [17] incorporates a rollback/recovery mechanism to synchronize the redundant execution of two cores. Although no single-core version was evaluated in this work, the data were completed with the results reported in [28] at the same facility and using an identical setup.…”

“…These results remain true even considering two detrimental factors in our experiments: the larger sizes of the matrices employed (32 and 128 vs 30) and the protection of the external memory (not included in other techniques). Furthermore, the increase [28], where authors used the same setup and facility as in [17] * * No errors observed, so for comparison purposes this is calculated given one error (assuming the worst-case) ⋆ Calculated with data provided by [17] in the σ HAN G was similar (1.4× vs 1.8× and 1.5×) for a similar matrix size (32 vs 30).…”