利用模糊自適應共振理論建構晶圓表面缺陷點及缺陷群聚之管制程序

Abstract

就積體電路（integrated circuit，IC）產業而言，晶圓（wafer）上晶片的良率（yield）是評估晶圓品質的一項重要指標。而影響晶片良率最主要的因素即是出現在晶圓表面上缺陷點（defect）的多寡。隨著晶圓面積不斷增大，晶圓上缺陷點呈現群聚（clustering）的現象愈來愈明顯，使得缺陷點之分佈不再隨機，因此以卜瓦松（Poisson）分配為基礎之缺陷數管制圖（c-chart）會產生過多之假警報，因而不再適用於管制晶圓之缺陷點數。針對此問題，中、外文獻提出了一些修正之缺陷數管制圖，但這些修正管制圖仍各有不周延之處。由於晶圓上缺陷數太多或缺陷群聚程度太嚴重均表示製程失控，因此，本論文的主要目的即是發展一套可同時監控缺陷點及缺陷群聚程度之管制程序。本論文首先利用一個群聚指標來量化每片晶圓上缺陷點群聚之嚴重程度，進而利用類神經網路（artificial neural network）中的模糊自適應共振理論（Fuzzy Adaptive Resonance Theory，Fuzzy ART）合併晶圓上位置相近的缺陷點，使修正後之缺陷點數與晶片之良率能有一致之趨勢，最後再應用Hotelling T2多變量管制圖來同時管制晶圓缺陷點數及群聚程度。本論文所提之管制程序對於缺陷點數過多或缺陷群聚現象嚴重的異常晶圓都能夠有效地偵測出來，使製程工程師能正確且快速地辨認出造成製程異常的原因，儘早改善以提升晶片良率。本論文最後利用新竹科學園區某積體電路公司之實際晶圓資料來驗證所提之管制程序的可行性與有效性，結果顯示本論文之管制程序比現有中、外文獻所提之方法均有效。

As the size of the wafers used in the fabrication of integrated circuits（IC） increases, the clustering of defects on wafers becomes increasingly apparent. When a conventional c-chart, based on the Poisson distribution, is used to monitor the wafer defects, clustered defects cause many false alarms. Several modified control charts of wafer defects have been developed to reduce the number of false alarms associated with the clustered defects. However, these control charts still have some weaknesses. This study proposes a procedure for controlling number of defect and defect clustering simultaneously. Fuzzy Adaptive Resonance Theory（Fuzzy ART）is utilized to alter the number of defects in clusters for each wafer. A cluster index is also employed to measure the extent of defect clustering for each wafer. A Hotelling T2 multivariate control chart is then constructed to monitor the wafer defects and defect clustering simultaneously. The proposed procedure can reduce the number of false alarms caused by the clustering of defects effectively and can also monitor the number of defect and the extent of clustering thereof. A case study of an IC company in Taiwan is also presented to demonstrate the effectiveness of the proposed procedure.