考量具自我相關之多變異來源製程管制

Abstract

統計製程管制(Statistical Process Control；SPC)已廣泛應用在各種產業，其中又以科技製造業使用最為頻繁。SPC中之X-BAR管制圖目前更是普遍使用在晶圓製造上作為監控製程之用，X-BAR管制圖主要是針對單一變異來源（single source of variation）所設計的品質監控手法，然而在晶圓製造環境中存在著多種製程變異如晶圓內變異、晶圓間變異以及批量間變異，因此僅使用X-BAR管制圖無法有效監控晶圓製程。Wells與Smith在1991年針對上述三種晶圓品質變異提出三階段流程，然而該管制流程主要使用X-BAR管制圖，因此無法有效監控製程可能發生的小偏移。此外，X-BAR管制圖假設製程資料值彼此獨立，當資料產生自我相關性時(autocorrelation)，將使得管制界限過窄而產生過多的假警報(false alarm)，而晶圓品質資料常出現自我相關(autocorrelation)。因此本研究針對多時點及多量測點之晶圓品質問題，考慮製程資料值可能存在自我相關特性，建立一套可同時管制小偏移以及大偏移之管制流程，本研究最後利用一個實例來說明本研究所提出之管制流程確實有效可行。

Statistical Process Control (SPC) has been widely employed in many industries, especially in high-tech manufacturing. In SPC, the X-BAR control chart is generally implemented in wafer manufacturing and it is designed to monitor the quality of product data for single source of variation. However, there are multiple variations exist in wafer-manufacturing process and this restricts the use of X-BAR control chart. Wells and Smith[20] proposed a three-step control process for monitoring the wafer production. However, their proposed process cannot detect small shift in the process. In addition, when employing chart the data must be independent. Once autocorrelation exist in data, the control limits will be underestimated and false alarms will be increased. Hence, the main objective of this study is to propose a statistical process controlling procedure for multiple-time intervals and multiple measured points wafer data with autocorrelation. Furthermore, the small shifts and large shifts can simultaneously be detected using the proposed procedure. A real case is utilized to demonstrated the effectiveness of the proposed procedure.