Visualization of Massive Mixed Type Semiconductor Manufacturing Data Using Self Organizing Maps

Data modelling in semiconductor industry is often challenged with complexity of massive mixed type datasets. Our main motivation is exploratory visualization of such data, usually for yield enhancement purposes, by means of a Self-Organizing Map (SOM). SOM and many other visualization methods require the data to be numeric. We replace catecorigal variables by numerical variables such that they preserve the mutual information between all original variables in the data set. Applying this method to a semiconductor manufacturing data set demonstrates how the relations between variables, and causes behind yield problems now become explicit.