ARFT: An Approximative Redundant Technique for Fault Tolerance

Gennaro Severino Rodrigues (1) , Adria Barros Barros de Oliveira (1) , Alberto Bosio (2) , Fernanda Lima Kastensmidt (1) , Edison Pignaton de Freitas (3)

1 PGMICRO - PgMicro - Instituto de Informática [UFRGS]
2 INL - CSH - INL - Conception de Systèmes Hétérogènes
3 PPGC - Instituto de Informática, Programa de Pós-Graduação em Computação [Porto Alegre]

This paper presents a novel redundancy technique for software fault tolerance, named Approximative Redundant Fault Tolerance (ARFT). It uses approximate computing in order to provide the same error detection of a classic DWC method with less overhead. In this work ARFT was implemented to protect the ARM Cortex-A9 embedded into Zynq-7000 All Programmable SoC. An extensive fault injection campaign was performed to evaluate the proposed technique. Results show that distinct applications with different approximation methods present a big variation in terms of execution time, memory footprint and error detection capability. Performance analysis shows that ARFT can reduce the overhead in some benchmarks cases up to 40%.

Mots clés: Benchmarking; Error detection; Integrated control; Radiation hardening; Software testing; System-on-chip; Approximation methods; Detection capability; Embedded processors; Fault injection; freeRTOS; Redundancy techniques; Soft error; Software fault tolerances; Fault tolerance

Domaines: Micro et nanotechnologies/Microélectronique

Gennaro Severino Rodrigues, Adria Barros Barros de Oliveira, Alberto Bosio, Fernanda Lima Kastensmidt, Edison Pignaton de Freitas. ARFT: An Approximative Redundant Technique for Fault Tolerance. 33rd Conference on Design of Circuits and Integrated Systems, DCIS 2018, Nov 2018, Lyon, France. ⟨10.1109/DCIS.2018.8681499⟩. ⟨hal-03770905⟩