Statistical significance of error-corrupted IC measurements

The impact of measurement errors on the statistical significance of data collected from an IC manufacturing facility is discussed. On the assumption that the errors are not systematic, are normally distributed, and have known variances, estimators are introduced for the formal evaluation of this impact. These estimators can be used to calculate confidence intervals, to test equivalence hypotheses, and to predict the required sample size so that a desired level of confidence is maintained in the presence of these measurement errors. The aforementioned concepts are illustrated through an example in which critical decisions concerning the status of an NMOS process are based on error-corrupted measurements.<
>