Dependability evaluation of a critical system by means of fault injection mechanisms

Abstract

Due to the significant impact that a medical device can have on human lives,every aspect of its development has a significant importance and the dependability features are then essential. Based on this premise, it is worth to dedicate every reasonable effort to analyze the effect of possible faults in this kind of application.In this way, this work is aimed to evaluate and improve the fault tolerance mechanisms of a prototype of an Active Implantable Medical Device (AIMD) by means of fault injection techniques. Fault injection experiments are a powerful aid to evaluate the design of fault tolerance mechanisms, particularly when performed at early development phases. The present work is fundamentally focused on the emulation-based technique.For that reason, a model of the system under study was implemented in an FPGAhosted OpenMSP430 microcontroller and the necessary instrumentation modules ere developed and then integrated to the system. The instrumentation included on one hand, Saboteour modules which are intended to inject the desired faults in the system, and on the other, an Event Recorder module which is intended to log the system behavior during the experiments. The whole system was managed through the debug interface of the microcontroller, using the GDB debugging application. As a result, a solution for low cost fault injection was developed and implemented. Even when this solution was used with a particular system, the hardware description of the instrumentation circuits can be easily adapted to be used in other systems based in small processors. The developed solution was successfully used to carry out several fault injection campaigns on the emulated system. The experiments included SEU (Single Event Upset) and stuckat faults in both data and program memory, and also stuckat faults in the buses of those memories. The obtained results contributed with important information that allowed to understand the system behavior in presence of faults and evaluate its fault tolerance features. In addition, based on the obtained information, it was possible to make specific improvements in the firmware code in order to better hand the SEUs in data memory. Finally, the campaign including SEUs injection in data memory was repeated and its results were compared with those obtained previously, verifying in this way the effectiveness of the firmware changes