Construct an intelligent yield alert and diagnostic analysis system via data analysis: Empirical study of a semiconductor foundry

Abstract

As semiconductor manufacturing technology advances, the process becomes longer and more complex. A critical issue is to determine how to avoid yield loss at an early stage or to diagnose the cause of yield loss soon, in order to save more money. Traditional statistical regression analysis and correlation analysis are unable to quickly and easily figure out the causes of process anomalies and potential problems. This study aims to construct an intelligent yield alert and diagnostic analysis framework combined within a big data analysis architecture. Through an intelligent detection and early warning mechanism, instant detection of yield anomalies and automatic diagnostic analysis based on good/bad wafer classification, we can effectively and rapidly find out the factors that may cause process variation to help quickly clarify the causes of abnormal product yield. The case study in this paper uses real-world data from a foundry in Taiwan. We hope to provide engineers and domain experts with a reference framework for building a yield analysis system to help improve the yield of semiconductor manufacturing and enhance the competitiveness of high-tech industries.