Regularized regression models to predict the product quality in multistep manufacturing

Abstract

Nowadays, the semiconductor manufacturing process is becoming very complex, consisting of hundreds of steps before obtaining the final product. Continuous Yield improvement is required to meet the current competitive market needs. High yield can be achieved by monitoring the wafer quality in real time throughout the auto-correlated multivariate process. However, the missing quality measurements make the product quality control very difficult. In this paper, we present an approach to predict product quality by using both the available on-line measurements of the production equipment data and the information about correlations between product quality measurements across the different steps. For this purpose, a regularized regression technique is used to represent statistical relationships between the correlated product quality measurements and the high-dimensional sensor data.