New safe reliable design methodologies examined by fault injection testing and Monte Carlo simulation: tolerating shared-memory interferences in multicore architectures

摘要

Nowadays, the automotive industry utilises multicore processors to meet autonomous driving requirements. This imposes a higher complexity on the development and the verification of software applications. Compliance to the ISO 26262 safety standard increases this complexity. In this paper, means of shared-memory interferences that affect Automotive Safety Integrity Level (ASIL)-D multicore architectures have been addressed. This work proposes new safety mechanisms to detect and react to systematic and random transient memory faults as follows: 1) an enhanced software partitioning design pattern; 2) a new methodology on the memory protection unit; 3) an improved stack monitoring mechanism. New safe and reliable design configurations are introduced. The proposed safety mechanisms have been evaluated for Aurix multicore targets with suggestions to have a fully compliant architecture followed with ISO 26262 methods and principles of tolerating memory interferences. A novel fault injection platform is presented to show the experimental results with a Monte Carlo simulation proof of concept.