Increasing profitability and improving semiconductor manufacturing throughput using expert systems

This paper describes a new procedure for using a machine-learning classification technique coupled with an expert system to increase profitability and improve throughput in a semiconductor manufacturing environment. The authors show how to use this procedure to identify relationships between work-in-process data (information obtained during semiconductor fabrication) and potential integrated circuit yield. The relationships, in the form of IF-THEN rules, are extracted from databases of previously fabricated integrated circuits and final yield. It is further shown that these rules, when incorporated into expert systems, can advise the human operator as to which batches of circuits are likely to produce submarginal yield if processed to completion, thereby providing a basis for developing or enhancing a quality control strategy. These rules also identify the parameters and values which have historically provided the highest and lowest final wafer yields. A cost analysis is given to illustrate the cost-effectiveness of this procedure. An introduction to semiconductor manufacturing and a glossary are provided