Exploratory Data Analysis for Semiconductor Manufacturing

This article introduces a few modern statistical techniques that are especially well suited to analysis of data obtained from semiconductor testing. Test data is particularly likely to be distributed in non-Gaussian fashion, thus data summaries and displays based on sample means and standard deviations are likely to be misleading. Engineers-test engineers in particular-should become familiar with stem-and-leaf displays, box-plots, medians, letter-value displays, and pseudosigmas. These displays and statistics can supplement and gradually replace the more traditional histogram, mean, and standard deviation in test data reporting and in testing literature over the next several years.