應用EWMA管制圖構建多特性多量測點資料之管制流程

Abstract

目前管制圖已廣為業界用來做製程監控之用，以有效地偵測出影響製程的非機遇原因（assignable cause），並在生產出不良品前及早採取矯正行動，藉以提昇產品之品質。但是當應用傳統的X-bar & R管制圖於目前複雜的晶圓製造環境時，會出現即使在製程穩定下，X-bar & R管制圖仍會顯示嚴重失控情況的假警報現象。此乃因X-bar & R管制圖是針對單一變異來源的生產系統而設計，而在晶圓製造中，品質變異的來源有許多種。此外，傳統X-bar & R管制圖適用於偵測製程參數的大偏移，對於製程參數的微小偏移則較不敏感，因此有可能製程平均值已經偏離目標值，但是X-bar & R管制圖仍未發出失控訊息，讓品管人員誤以為製程仍處於管制狀態，而不斷生產出不良品。本研究之主要目的在於針對多品質特性、多量測點的精密製程（例如光罩、晶圓製程等），構建出適當的管制圖以偵測製程中的微小偏移，並考慮多種變異來源的製造系統，構建出一套完整的管制流程，線上人員只要依此管制流程即可辨認出不同的變異來源並即時改善。本研究經由主成份分析縮減製程中的品質特性後，進而減少所需繪製管制圖的數量，並以Hotelling T2及多變量指數加權移動平均（Multivariate Exponential Weighted Moving Average, MEWMA）管制圖來偵測製程的穩定狀況；若製程失控，則另繪製個別X-bar & EWMA管制圖追溯變異來源。本研究將以模擬之案例說明管制流程，並驗證所建構之管制流程確實能有效管制多特性多量測點之資料，最後並以新竹科學園區某廠商之矽磊晶資料來說明如何應用本研究之管制流程。此管制流程可提供線上人員快速判斷製程狀況並修正製程參數，進而提升產品品質。

An increasing number of wafer fabrication manufacturers use a control chart to effectively monitor the wafer manufacturing process. However, conventional control charts are designed for detecting a manufacturing process with single source of variation, and, therefore, they are incapable of detecting assignable causes for a process with several sources of variation. Moreover, a statistical monitoring procedure for a complex wafer manufacturing process usually involves multiple readings from multiple quality characteristics with several sources of variation. This study presents a competent on-line control process capable of detecting assignable causes concealed behind multiple characteristics and multiple readings in a manufacturing process with several sources of variation. Principal component analysis (PCA) is employed to form new variables, which are the key components of original multiple characteristics in a manufacturing system. Their formation decreases the number of control charts since PCA reduces the number of related features. The Hotelling T2 and multivariate exponential weight moving average (MEWMA) control charts are then used to determine whether the process is in control. Additionally, for a situation in which Hotelling T2 or MEWMA indicates that the process is out of control, three unique X-bar & EWMA control charts of different sources of variation are developed to identify the source of variation. Simulation results indicate that, in addition to detecting small shifts in a manufacturing system, the proposed procedure can accurately identify which sources of variation or characteristics are out of control. In addition, an example from a silicon epitaxy wafer process employed by a Taiwan IC fabrication manufacturer demonstrates the proposed approach’s effectiveness. Results in this study can provide a valuable reference for engineers when attempting to quickly assess the conditions of a wafer manufacturing process. By effectively responding to this information, engineers can promptly adjust the manufacturing system to enhance wafer quality.