A machine-learning classification approach for IC manufacturing control based on test structure measurements

A machine-learning method is presented for classifying electrical measurement results from a custom-designed test chip. These techniques are used for characterizing the performance of a 1-μm integrated circuit lithography process. Emphasis is on the development of a method for producing reliable classification rules from databases containing large samples of measurement data. The algorithm used is the Iterative Dichotomiser version 3, or ID3, originally developed by J.R. Quinlan (1983). The resultant classification rules are implemented in an expert-system shell. This combination provides a means of training and customizing a diagnostic system to be responsive to process variations experienced in a semiconductor manufacturing environment. Descriptions are given of the test chip, data-handling methods, rule-generation techniques, and statistical data reduction and parameter-extraction techniques used. An analysis of error introduced by noise in the rule formation process is presented