About Performance Faults in Microprocessor Core in-field Testing

Abstract

When microprocessor-based devices are used in safety-critical applications (e.g., in automotive systems), it is common to adopt solutions aimed at testing them in-field, so that permanent faults that may affect them are identified before they cause critical consequences. In this way, the required reliability figures can be achieved. A popular solution to perform in-field test (especially when executed concurrently to the application) is based on triggering the execution of proper procedures (composing a Self-Test Library, or STL), which are able to activate faults and make them visible when checking the produced results (e.g., in memory). Unfortunately, a special class of faults exists (named Performance Faults), which do not impact the value of the results, but only the timing behavior of the processor. This paper describes a set of experiments aimed at quantitatively evaluating the number of these faults in a simple processor core, and outlines some observation techniques that can be used for their detection.