Error Containment in the Presence of Metastability

Error containment is an important concept in fault-tolerant system design, and techniques like voting are applied to mask erroneous outputs, thus preventing their propagation. In this presentation we will use the example of DARTS, a fault-tolerant distributed clock generation scheme in hardware, to demonstrate that metastability is a substantial threat to error containment. We will illustrate how metastability can originate and propagate such that a single fault may upset the system. The main conclusion is that modeling efforts on all design levels are definitely required in order to mitigate and quantify the deteriorating effect of metastability on system dependability.