The semiconductor industry is continuously facing four main challenges in film characterization techniques: accuracy, speed, throughput and flexibility. Virtual Metrology (VM), defined as the prediction of metrology variables using process and wafer state information, is able to successfully address these four challenges. VM is understood as definition and application of predictive and corrective mathematical models to specify metrology outputs (physical measurements). These statistical models are based on metrology data and equipment parameters. The objective of this study is to develop a model predicting the CVD oxide thickness (average) for an IMD (Inter Metal Dielectric) deposition process using FDC data (Fault Detection and Classification) and metrology data. In this paper, two VM models are studied: one based on Partial Least Squares Regression (PLS) and one based on Tree ensembles. We will demonstrate that both models show good predictive strength. Finally, we will highlight that model update is key for ensuring a good model robustness over time and that an indicator of confidence of the predicted values is necessary too if the VM model has to be use on-line in a production environment.

No abstract

In order to occupy a competitive position in semiconductor industry the most important challenges a fabrication plant has to face are the reduction of manufacturing costs and the increase of production yield. Predictive maintenance is one possible way to address these challenges. In this paper we present an implementation of a universally applicable methodology based on the theory of regression trees and Random Forests to predict tool maintenance operations. We exemplarily show the application of the method by constructing a model for predictive maintenance of an ion implantation tool. To fit the problem adequately and to allow a descriptive interpretation we introduce the remaining time until next maintenance as a response variable. By using R and adequately analyzing data acquired during wafer processing a Random Forest model is constructed. We can show that under typical production conditions the model is able to predict a recurring maintenance operation sufficiently accurate. This example shows that better planning of maintenance operations allows for an increase in productivity and a reduction of downtime costs. Zusammenfassung

International audienceTool condition evaluation and prognosis has been an arduous challenge in modern semiconductor manufacturing environment. More and more embedded and external sensors are installed to capture the genuine tool status for fault identification. Therefore, tool condition analysis based on real-time equipment data becomes not only promising but also more complex with the rapidly increased number of sensors. In this paper, the idea of Generalized Moving Variance (GMV) is employed to consolidate the pure variations within tool Fault Detection and Classification (FDC) data into one single indicator. A hierarchical monitoring scheme is developed to generate an overall tool indicator that can be coherently drilled down into the GMVs within functional sensor groups. Therefore, we will be able to classify excursions found in the overall tool condition into sensor groups and make the tool fault detection and identification more efficient