Abstract:
CorDex is a program that organizes/categorizes/sorts items with a large set of attributes using a SOM (self-organizing map). This map allows the user to visually scan for similar items and the clustering of items based on their attributes. The program also allows users to visually search `attribute planes' of this SOM's attributes to find: correlations among attributes, clusters which share similar attribute values but are very different in other respects, and attributes which are statistically unrelated. CorDex can process tens of thousands of items and find `information' or `rules' from `raw data'. It has also been used to cluster items into distinct groups to allow traditional Rule Induction to find rules in large data sets that were not attainable without the CorDex pre-clustering. CorDex has been used to analyze semiconductor manufacturing yield problems and find root causes of problems that have eluded even the best process engineers for years.