賈柏轉換應用於晶圓晶片之可見瑕疵檢測

Inspecting Visual Defects of Wafer Die by Using Gabor Transform

林正偉(Cheng-Wei Lin)

指導教授 ： 葉繼豪

國立臺北科技大學/管理學院/工業工程與管理系所/碩士(2005年)

摘要

本研究係利用機器視覺(machine vision)之技術與紋路分析(texture analysis)之手法，檢測晶圓(wafer)晶片(die)之外觀可見瑕疵(如刮痕、塵粒與汙染)與墨點(ink Dot)，並透過程式之撰寫以建構出一套檢測晶圓表面瑕疵之自動化檢測系統，期望晶片封裝前之瑕疵檢測步驟可達到自動化、精確與快速之目的，甚至能夠完全取代現行之人工檢測，以解決該製程往往成為瓶頸站之困境。
本研究方法主要先擷取每顆晶片表面1/24的灰階與彩色影像，再細分為四個象限之次區域影像，之後灰階與彩色晶片影像分別進行賈柏轉換;關於灰階影像，直接進行傳統賈柏轉換法，即針對每一個次區域影像設計出最符合其紋路之賈柏過濾器(gabor filter)，使其能得到最小之賈柏能量平均值。之後利用實部與虛部之目標與測試集合建立與分析比較，之後扣除重覆以得其實部與虛部之不規則像素點集合，若該集合中像素點數目過多，則可判定其為瑕疵影像，需再進行最適橢圓分析法藉以區分刮痕、塵粒與污染等瑕疵及墨點。另外，採用灰階共變異矩陣法檢測灰階晶片影像，並與傳統賈柏轉換法之檢測結果互相比較與分析。
關於彩色影像的部分，須先利用Drg模型萃取出兩個色彩特徵值r與g分量，利用彩色賈柏轉換法，求得一組賈柏參數組合其能夠使得經過其賈柏過濾器後之r與g分量之賈柏能量平均值總和為最小。之後藉由r與g分量目標集合與測試集合之建立與分析，類似於灰階影像之檢測即可。
此外，本研究將採用真實之晶圓晶片影像為測試樣本以証明其實務應用性。而且，透過賈柏轉換之運用以濾除規則紋路以及目標集合與測試集合之應用，可將需要比對之像素點數目大幅度地縮小，以加快檢測速度與節省儲存空間。

關鍵字

機器視覺 ； 紋路分析 ； 晶圓 ； 晶片 ； 可見瑕疵 ； 賈柏過濾器 ； 最適橢圓分析法

並列摘要

The objective of this research is to develop an approach for inspecting the visual defects (e.g. scratch, particle and contamination) and ink dots on the dies of wafer based on machine vision and texture analysis, and to construct an automation inspection system for the surface defects by writing programs in order to reach the purpose of automation, accuracy and acceleration for the defect inspecting steps by dipping the dies, even to fully replace the existent artificial inspection to solve the difficulties that the processing usually becomes a constriction.
The approach of the research is to capture the gray-scaled and color image that are 1/24 from the surface of each die, to equally divide these images into four subimages for further analysis, and then to separate the gray-scaled and color images into two parts for the gabor transform processing. Concerning the gray-scaled images, it is to directly proceed the traditional gabor transform processing, which means that a specified gabor filter is designed for each subimage to get the minimal energy average; then by constructing and analyzing the golden and testing set of the real and imaginary parts, it can get the irregular sets of real and imaginary part after deducting the duplication. If the number of the pixels is too large, we can determine the gray-scaled image as a defect image. Therefore, it has to go on with the best fitting ellipse process to distinguish the defection and ink dot such as scratch, particle, and contamination defect. Besides, using gray covariance matrix method to detect gray-scaled images, and comparing the detection resules with those using gabor transform.
With regard to the color images, it must extract two color factors, ‘r’ and ‘g’, by using the Drg color model. When using color gabor transform, it can find a gabor filter, whose gabor parameter combination can minimize the sum of the r and g factor’s gabor energy average. Then by constructing and analyzing the golden and testing set of the r and g factors, just do same inspection as the gray-scaled images processing.
To meet the actual application, real dies are used as a testing sample to validate the proposed method in this research. By using gabor transform to filter out the regular texture, and applying golden set and testing set can substantially reduce the number of pixels that are needed to be compared and also can speed up the detection and save the memory space.

並列關鍵字

machine vision ； texture analysis ； wafer die ； Visual defect detection ； gabor filter ； best fitting ellipse

參考文獻

[2]楊榮華，應用灰階共變異矩陣之多重指標於瑕疵檢測，碩士論文，元智大學工業工程與管理研究所，桃園，2002

[8]曾彥馨，應用機器視覺於TFT面板之表面瑕疵檢測與分類，碩士論文，元智大學工業工程與管理研究所，桃園，2003

[10]M. Akatsuka, K. Sueoka, and T. Yamamoto, “Classification of etch pits at silicon surface wafer surface using image- processing instrument”, Journal of Crystal Gowth, 210, 2000, pp.366 – 369.

[11]P. B. Chou, R. Rao, M. C. Sturzenbecker, F. Y. Wu, and V. H. Brecher, “Automatic Defect Classification for Semiconductor Manufacturing”, Machine Vision and Application, ,9(4), 1997, pp.201 - 214.

[13]T. Kubota, P. Talekar, T. Sudarshan, X. Ma, M. Parkar, and Y. Ma, ”An automated defect detection system for silicon carbide wafers”, Southeast Conference 2002 Proceedings IEEE, 2002, pp. 42 –47.