DATA MINING OF ELECTRICAL CHARACTERISTICS FOR PROCESS COMPARISONS

摘要

Device and integration engineers are often tasked with the quantification of differences in electrical characteristics between two process flows within the same technology. Examples of these tasks include transfer between fabs and mask changes. This quantification becomes extremely difficult because of the sheer volume of data. Literally hundreds of characteristics must be compared across many lots, wafers and sites. Most off-the-shelf analysis tools take a "one response at a time" approach that is too slow for the cycle-time required by the fast pace of the semiconductor industry. As a result, engineers frequently rely on gross summarization, which may obfuscate differences and provides little information on the nature of differences. We present a solution that consists of an appropriate data collection methodology, coupled with an analysis/graphics program customized to the data structure. The SAS~(~R)-based program automates data screening, analysis and summarization, and graphics production. For better understanding of the nature of process differences, mixed models and variance component estimation are utilized. Moreover, the built-in "smart logic" employs matching metrics designed by the statistician and engineer to mine the data for those electrical characteristics that exhibit differences. The result is shorter cycle time and more accurate information for making engineering decisions.