In TFT-LCD manufacturing process, virtual metrology (VM) model is often employed to predict product's quality variables using sensor variables. However, modern industrial processes are usually equipped with a large number of sensors that provide process variables data such as pressure, temperature, spectroscopic signals, heat or power supplied, etc. So, how to design a validated VM model is the key problem. In this paper, a novel approach is presented to overcome the problem of high dimensionality and collinearity in the process variables data. Firstly, canonical correlation analysis is used to overcome the collinearity of the variables measured or quality variables. Moving time window is also considered to resolve the process uncertainty. For the purpose of reducing computation cost, a reliance index is developed to determine the frequency of model's parameters updating. Superiority of the proposed model is also presented when it applied to an industrial sputtering process.
[1]
Fan-Tien Cheng,et al.
Application development of virtual metrology in semiconductor industry
,
2005,
31st Annual Conference of IEEE Industrial Electronics Society, 2005. IECON 2005..
[2]
Fan-Tien Cheng,et al.
NN-Based Key-Variable Selection Method for Enhancing Virtual Metrology Accuracy
,
2009
.
[3]
Weihua Li,et al.
Recursive PCA for Adaptive Process Monitoring
,
1999
.
[4]
Hyunbo Cho,et al.
Partial least square-based model predictive control for large-scale manufacturing processes
,
2002
.
[5]
Chao-Ton Su,et al.
A Neural-Network Approach for Defect Recognition in TFT-LCD Photolithography Process
,
2009
.