Sequential tests for integrated-circuit failures

摘要

We propose a sequential probability ratio test (SPRT) based on an2-parameter Weibull distribution for integrated-circuit (IC) failurenanalysis. The shape parameter of the Weibull distribution characterizesnthe decreasing, constant, or increasing failure rate regions in the bathntub model for IC. The algorithm (SD) detects the operating region of thenIC based on the observed failure times. Unlike the fixed-length tests,nthe SD, due to its sequential nature, uses the minimum average number ofndevices for the test for fixed error tolerances in the detectionnprocedure. We find that SD is, on average, 96% more statisticallynefficient than the fixed-length test. SD is highly robust to thenvariations in the model parameters, unlike other existing sequentialntests. Since the accuracy of the tests and the test length arenconflicting requirements, we also propose a truncated SD which allows anbetter control of this tradeoff. It has both the sequential nature ofnexamining measurements and the fixed-length property of guaranteeingnthat the tolerances be met approximately with a specified number ofnavailable measurements