晶圓缺陷點群聚指標之建立

Abstract

晶圓的良率（yield）是衡量積體電路製造業者生產能力的一個重要指標，影響良率高低的因素有很多，其中晶圓上缺陷點數的多寡及缺陷點(defect)的群聚(clustering)程度是決定晶圓良率高低的兩個重要因素。近年來隨著晶圓面積不斷增大，使得晶圓上缺陷點出現群聚的現象，許多中、外文獻雖然針對晶圓上的缺陷點群聚問題提出了各種不同的群聚指標來衡量缺陷點的群聚程度，然而這些衡量指標各有不周延之處。因此本研究的主要目的是在發展一個無需任何統計假設的新群聚指標，該指標應用轉軸方法可充分描述缺陷點在晶圓上的相對位置，因此能有效地反應出缺陷點群聚嚴重的程度。本研究最後以模擬資料及實例來說明本研究之群聚指標的有效性及可行性，結果顯示本研究所發展之群聚指標比中、外文獻所提之其他缺陷點群聚指標的表現更好。此指標不但不受晶片大小的影響，且有均勻的數值範圍可量化不同嚴重程度的群聚現象，因此較其他缺陷點群聚指標更能確實地反應出真實的缺陷點群聚程度。此外，本研究之群聚指標計算相當簡單，對於無統計背景的工程人員而言，甚具實用價值。

There are two major factors affecting the yield of integrated circuits (IC) products. One is the number of defects on a wafer and the other is the degree of defect clustering. The clustering of defects on a wafer due to the complicated manufacturing process becomes more evident with increased wafer size. Therefore, many cluster indices had been developed to detect the defect clustering. However, there are still some shortcomings in employing these cluster indices. This study applies the rotation of axes and develops a new cluster index, which does not require any assumptions on the distribution of defects. Therefore, the proposed cluster index is easier to be employed by engineers with little statistical background. The value of the proposed cluster index contains more information about defect locations on a wafer and reflects the true degree of defect clustering. To verify the effectiveness of the proposed cluster index, a simulation experiment and a case study are presented. In the simulation experiment, comparing with some existing cluster indices which are widely used in IC fabrication, the value of the proposed index is not affected by chip size and can measure the true degree of defect clustering linearly within a uniform range. In the case study, a Hotelling T2 multivariate control chart incorporating the proposed index and the defect counts that can simultaneously monitor the total number of defects and the degree of defect clustering on a wafer is conducted and shows that the proposed cluster index is more powerful in detecting the defect clustering than other existing cluster indices.