矽磊晶晶圓多變量製程管制

Abstract

針對矽磊晶晶圓的多品質特性，目前業界並沒有一套有效的品質管制流程。業界常用的品質管制方法是針對每一個品質特性繪製一張管制圖，繪製方法是將該批量內每一品質特性所量測的多個監測點資料用來繪製 及R管制圖。在這種情況下，如果一次只考慮一個單一變量，且將每個單一變量分開來進行管制，並據以判定製程的好壞，將可能因多個變量間的相關性而導致誤判，無法有效地偵測出製程的變異。 本研究之主要目的是使用主成份分析及Hotelling 多變量管制圖，來提高矽磊晶晶圓多品質製程管制之效率及可靠性。透過本研究建議的方法，可進一步找出矽磊晶晶圓內部多個品質特性在不同量測點之間、同批量晶圓之間、以及不同批量間的變異情形。此外，若品質成本資料可得，還可利用品質損失函數，另推導出新的管制界限。本研究最後以新竹科學園區某廠商的磊晶生長資料為例，說明及驗證本研究方法的可行性及有效性。

To control the multiple quality characteristics of epitaxial wafers, until now an effective control system has not been developed yet by industry. Conventionally, and R control charts are employed for each quality characteristic and a general conclusion is made based upon these separated control charts. However, a wrong conclusion may be obtained due to the correlations existed between quality characteristics. This study utilizes principal component analysis and Hotelling multivariate control charts to upgrade the efficiency and reliability of multivariate quality control for Epitaxy wafers. In addition, the quality loss function can be employed to adjust the control limit if quality costs can be obtained. An example of epitaxial wafer from an integrated circuits company in Taiwan is illustrated to verify the effectiveness of this proposed procedure.