Intelligible test techniques to support error-tolerance

We have developed a new digital system mode of operation, referred to as error-tolerance, the purpose of which is to increase effective yield. Error-tolerance is based on the fact that many digital systems exhibit acceptable behavior even though they contain defects and occasionally output errors. A radically new test methodology, called intelligible testing, is required to support error-tolerance. This paper addresses parts of this methodology. There are several fundamental philosophical differences between intelligible testing and classical testing, such as: intelligible testing is application oriented; it partitions die and chips into multiple categories, not just good and bad parts; and it supplies quantitative information about the effects of defects on errors, i.e. it is error based rather than fault based. We describe three types of error attributes, namely error-rate, error-accumulation (retention), and error-significance. We present test techniques for estimating quantitative values for these qualitative attributes. Testing to support error-tolerance involves new ATPG tools, new fault simulators, and new DFT and BIST techniques.