Hierarchical self-repair in heterogeneous multi-core systems by means of a software-based reconfiguration

Abstract

This paper deals with the problem of a software-based self-repair in a statically scheduled multi-core system in the presence of permanent faults. The basic idea is to adapt the application in a way that the use of faulty components is avoided. This goal is achieved by re-compiling the program-code via an off-line repair process in the field. The repair process is organized in a hierarchical manner. At the beginning a local repair is applied considering only one core. If the local repair fails, a retry at a higher system-level is performed. For that purpose, the local repair techniques are re-used in a global context. The repair at a global system level results in some specific system constraints and properties, which are investigated in this work. Due to the use of pure software-based methods one gains the possibility to repair defects in different components (even multi errors) or defects in spare components. The presented approach is not bounded to a concrete architecture and is therefore adaptable to systems like MPSoCs or NoCs, if these systems provide some basic functionality.