On the quality of accumulator-based compaction of test responses

The accumulator-based compaction (ABC) technique uses an accumulator to generate a composite fault signature for a circuit under test. The error coverage for this method has been previously analyzed using Markov chains. We describe an alternative technique for calculating the error coverage of ABC using the asymmetric error model. This technique relies on the central limit theorem of statistics and can be applied to other count-based compaction schemes. Our analysis shows that ABC provides very high coverage of asymmetric errors. Experiments on the actual fault coverage for the ISCAS 85 benchmark circuits show that extremely high postcompaction fault coverage (close to 100%) is obtained with ABC. They also indicate that the use of a rotate-carry adder does not always improve the fault coverage; in some cases, the fault coverage is actually reduced.