Comparing partial and complete test sets and test metrics

Engineers need to test the integrated circuits that they manufactured to find which ones are defective. They can test them by applying and measuring voltages or currents in parts of the circuit called inputs and outputs. Logic integrated circuits are hard to test because as technology evolves the number of components in the circuit increases faster than the number of inputs and outputs.; The most common type of test, the single-stuck fault test, only takes into account the input values applied when measuring the outputs. This thesis will explain the importance of supplementing this test by another test taking into account the previous input values applied. Among the 17,900 integrated circuits tested in our experiments, the single-stuck fault test found all of the 244 defective ones whose detection depends only on the values applied when measuring the outputs. The detection of the defective circuits that the single-stuck fault test misses in our experiments depends not only on the values applied when measuring the outputs, but also on the previous values applied and how fast these values change. Whence the importance of supplementing the single-stuck fault test by a test also taking into account the previous input values applied. Although this thesis will show that the single-stuck fault test set must be supplemented, it will also show that because of imperfections in how tests are generated the single-stuck fault test set cannot be replaced.; Finally, this thesis will also show that, because of the number of integrated circuits tested, we should expect most of the hundreds or thousands of tests applied to never detect any defective integrated circuit. Engineers have noticed this in their practice and they have reported it in the literature. This thesis will present a statistical model showing that because of the number of chips tested we should expect most tests to never detect any defective circuit. Therefore the fact that so many tests appear to be useless does not mean that they are less effective than the other tests.