Study On Prediction Model Of Key Parameters For Dynamic Scheduling Of Semiconductor Wafer Fabrication

With the fierce market competition and the rapid changes in market demand of the semiconductor industry, shorten the cycle time (CT), reduce the cost and improve product quality, is the basic solution for enterprises to adapt to changes in the market and improve the market competitiveness. In order to improve the performance and production efficiency of semiconductor production line, the prediction model of key performance indicators (KPI) in SWFS should be researched. The adjustable parameters could be found to achieve the multi-objective optimization through the analysis of the impact factors. This paper focuses on the prediction of CT, equipment utilization (EU) and yield. The prediction model of multiple key performance indicators(MKPI) is built to predict CT and EU, and defect data is used to analyze congeries characteristic of the wafer defects and predict yield. The main contribution of this paper can be summarized as follows.(1) In order to solve the problems of predicting the CT and EU simultaneously under dynamic environment, the Bayesian Neural Network (BNN) is used to establish the simultaneous prediction model for MKPI and the closed-loop prediction model structure is developed to complete model modification. In addition, the weight analysis method is proposed to identify the key factors for CT and EU, since the weights of BNN are adjusted according to the importance of input to output.(2) To solve wafer defect problem, a prediction method of yield with wafer defect data driven is researched. The Density-based Spatial Clustering of Applications with Noise (DBSCAN) is applied to analyze congeries characteristic of the wafer defects. And the fuzzy support vector machine (FSVM) is used to build the forecast model of yield.The MKPI and the FSVM prediction models are applied to predict the KPIs in the semiconductor wafer fabrication. The simulation results show that these two kinds of prediction models can achieve good prediction performance. It provides a novel way to solve the dynamic scheduling problems and lays the foundation for the realization of multi-objective optimization in SWFS.