Keeping Important Samples for the Model Refreshing Scheme

• Main Authors: Fan-WeiKong, 孔繁偉
• Other Authors: Fan-Tien Cheng
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/44164052017799366404

碩士 === 國立成功大學 === 製造資訊與系統研究所碩博士班 === 98 === Over the past few years, virtual metrology (VM) has been widely developed and published in several VM related literature in the semiconductor and TFT-LCD industries. VM not only can provide the quality of semi-products in real time, but also can enhance the effectiveness of Advanced Process Control (APC). In the production process, events such as preventive maintenance (PM), tool drift, or using design of experiments (DOE) to calibrate the machine tools are essential. If samples related to the above important events can be collected and included in the VM models, the performance of VM can be enhanced and more robust for real-time online prediction. In this paper, we propose to use the clustering technique to keep important samples for the model refreshing scheme. In other words, VMS (Virtual Metrology System) will remove samples with similar characteristics from the conjecture model. In this way, the features of production process in the conjecture model will still be abundant even after samples are removed. The new scheme, named Dynamic Moving Window (DWM), has the advantage of absorbing the variation information of products in the online VM model, and important samples will not be discarded as time goes on. When the same machine variation has been encountered in the past, VMS can accurately provide prediction quality. This paper have shown that DMW scheme can provide VM Model with a more significant predictive capacity while the process tools have become unstable.