Abstract : This thesis is devoted to the study of a software methodology for detection of the errors induced by the radioactive environment: the SEU phenomenon, also called upset - which may modify the content of memory elements as the result of the silicon ionization resulting from the impact of a charged particles. The consequences of the upsets for a given application depend on both the occurrence instant and the perturbed memory element, and can go from erroneous results to system crashes which may provoke the loose of the control of a space vehicle.
The proposed software approach is based on the transformation of the programs, written in any high-level language in such a way they have capabilities to detect transient errors affecting data and code. The software modifications are achieved through the application of a set of transformation rules derived from a through analysis of an existing set of rules formerly proposed in the specialized literature. The new set of rules improved the system performances in terms of reducing the number of errors escaping the detection mechanism and the program execution time.
The evaluation of this methodology was carried out by fault injection experiments and the radiation campaigns for several processors. These experiments confirmed our expectations: the hardened version of the application allows a high the error detection rate (average of 88% of the errors which have occurred in the system).
Future works will constitute the base for the definition of a mixed technique software/hardware for the error detection.