Skewness and Kurtosis Risks of Quality Control in Overlay Inspection

The population of overlay errors in layer-to-layer structures in semiconductor manufacturing often obeys non-normal distribution functions such as elliptic, rectangular, and skewed distributions. Thus, there can be many wrong diagnoses of quality control that can lead to consumer's risk and producer's risk. In a lot acceptance test, a product should be judged by variables rather than by attributes because of the small sample size. A new acceptance test by variables was introduced for a non-normal population. It included sample skewness and kurtosis as well as mean and standard deviations. Using operating characteristic (OC) curves, it was shown that sampling inspection by variables can be applied safely to a strong non-normal population with new variables. (C) 2010 The Japan Society of Applied Physics