Outlier screening (PAT) limit calculator

How it works

Part average testing (PAT) removes parts that pass the data sheet limits but sit far from the rest of the population, because such parts are more likely to fail early in the field. This tool sets static and dynamic PAT limits from robust statistics, flags the outliers in the data you paste and shows the yield the screen removes.

1. Estimate the centre and spread robustly. A few outliers would distort a plain mean and standard deviation, so the guideline uses the median as the robust mean and the inter-quartile range as the basis of a robust sigma. For a normal population the inter-quartile range is about 1.349 standard deviations, and the guideline rounds the divisor to 1.35.robust mean = median; robust sigma = (Q3 − Q1) / 1.35
2. Static limits from a reference population. Static limits come from a reference population of parts that pass the specification, ideally several lots, and stay fixed. They sit k robust sigmas either side of the robust mean, with k equal to 6 by default. A limit is never allowed to be wider than the specification limit. For a one sided parameter such as IDDQ only the high limit is used.static low = max( median − k x robust sigma, LSL ); static high = min( median + k x robust sigma, USL )
3. Dynamic limits for each lot. Dynamic PAT repeats the calculation on the lot being tested, using the lot's own parts that pass the specification and the static limits. A lot that has shifted gets limits centred on its own population, so an outlier inside the static limits is still found. The dynamic limits are applied together with the static limits.dynamic limits per lot = median ± k x robust sigma of the lot's passing parts; a part fails if it is outside either set of limits
4. Compare with the mean and standard deviation. For comparison the tool also sets limits at the plain mean plus or minus k sample standard deviations from the same reference parts. When outliers are present the standard deviation is inflated and these limits widen, which is why the robust method is preferred.plain limits = mean ± k x s
5. Count the outliers. An outlier is a part inside the specification but outside a set of limits. Parts outside the specification already fail and are counted apart. The yield lost to the screen is the outlier count over the parts tested; a normal population is expected to lose very little at k of 6.outlier rate (ppm) = 1,000,000 x outliers / parts tested; expected false flags = sides x Φ(−k)
6. Long tailed parameters. Leakage and IDDQ are often close to lognormal. The optional logarithm transform applies the same method to log10 of the values and reports the limits back in measured units.

Sources

The methods and default figures rest on these public sources. The wording here is our own; read the source for its full text.

• AEC-Q001 Rev-D, Guidelines for Part Average Testing, Automotive Electronics Council, Component Technical Committee, 9 December 2011. Static and dynamic PAT limits, the median and inter-quartile range based robust statistics, and the rule that limits are not wider than the specification. Free from the AEC website. This is the revision found; the AEC site lists the current one.
• AEC-Q002, Guidelines for Statistical Yield Analysis, Automotive Electronics Council, Component Technical Committee, Current revision on the AEC website. Cited by name only, for the related statistical yield and bin limit screens.
• Interquartile Range, Dataplot Reference Manual (opens in a new tab), National Institute of Standards and Technology, Information Technology Laboratory, Accessed 25 October 2026. For a normal distribution the inter-quartile range is about 1.349 times the standard deviation, which is the basis of the robust sigma.