No abstract

Software cost estimation and overruns continue to plague the software engineering community, especially in the area of safety-critical systems. We provide some preliminary models to predict the cost of adding fault detection, fault-tolerance, or fault isolation techniques to a software system or subsystem if the cost of originally developing the system or subsystem is known. Since cost is a major driver in the decision to develop new safety-critical systems, such models will be useful to requirements engineers, systems engineers, decision makers, and those intending to reuse systems and components in safety-critical environments where fault tolerance is critical.

As embedded and safety-critical applications begin to employ many-core SoCs using sophisticated on-chip networks, ensuring system quality and reliability becomes increasingly complex.Infrastructure IP has been proposed to assist system designers in meeting these requirements by providing various services such as testing and error detection, among others. One such service provided by infrastructure IP is concurrent online testing (COLT) of SoCs. COLT allows system components to be tested in-field and during normal operation of the SoC. However, COLT must be used judiciously in order to minimize excessive test costs and application intrusion. In this paper, we propose and explore the use of an Anomaly-based Test Triggering Unit (ATTU) for on-demand concurrent testing of SoCs. On-demand concurrent testing is a novel solution to satisfy the conflicting design constraints of fault-tolerance and performance. Ultimately, this ensures the necessary level of design quality for safety-critical applications. To validate this approach, we explore the behavior of the ATTU using a NoC-based SoC simulator. The test triggering unit is shown to trigger tests from test infrastructure IP within 1ms of an error occurring in the system while detecting 81% of errors, on average. Additionally, the ATTU was synthesized to determine area and power overhead.