A survey of methods for intermittent fault analysis

As digit~l circuits grow in complexity, and as they are used in more and more critical applications, their reliability becomes an important consideration. This directs the attention of researchers involved in this field to two basic questionsfirst, how to design more reliable systems; second, how can something be said about the reliability of a particular system? Causes leading to permanent failures can easily be examined. A class of faults called permanent or solid faults was identified and it has been extensively studied. Some solid faults in some lines in digital systems behave as if they are permanently connected to a logic 0 or a logic 1. They are called stuck at 0 or stuck at 1 faults. Significant analytical results are available to handle such faults. Another class of faults, called intermittent or transient faults, are responsible for a very significant number of failures. The effects of such faults appear and disappear randomly. It has been noticed that an overwhelming proportion (up to 90 percent) of faults can be estimated to be intermittent. As these faults are more difficult to isolate and fix, they account for most of the field maintenance costs. Ball and Hardyl in their study found that, while the test and maintenance costs due to the permanent faults tend to decrease drastically with time, the costs incurred per unit time due to intermittent faults change very little. McConnel and Siewiorek report that in an LSI-II system with 24k words of dynamic MOS memory, one would expect a transient failure every 100 hours, whereas a solid failure can be expected in every 7700 hours. Needless to say, the problems of intermittent failures are very significa,nt. In this paper, a brief survey of the significant results available today is presented. These aspects of intermittent faults will be considered. First, physical reasons for intermittent faults will be considered. Second, various suggested models for intermittent faults will be examined along with the basic assumptions. Testing and reliability analysis are considered next. Finally, design methods are examined which take intermittent faults into consideration.