晶粒圖紋瑕疵之自動檢測

Abstract

半導體產業是台灣重要的產業之一，隨著製程技術不斷提升，產品元件設計越來越精細、幾何尺寸也越來越小。然而，晶粒(Die)在製造過程中，難免因內在或外在因素導致晶粒發生瑕疵，如：(1)微粒或汙染、(2)變色、(3)護層不良、(4)護層多開/殘留、或(5)探針痕異常，部分細微的瑕疵，以人工必須透過高倍率電子顯微鏡才能有效找出，使得品質相當難以控管。 由於客戶對於產品品質要求越來越嚴格，傳統的人工目視檢測已無法符合客戶需求，因此本研究將針對半導體製程中常出現的瑕疵，利用機器視覺(Machine Vision)技術的輔助，提出一套自動化瑕疵檢測的演算法，並藉由此演算法，有效的達成(1)全數檢測、(2)高準確率，及(3)高效率的目的。

Semiconductor industry is one of the major industries in Taiwan. The design of product becomes smaller and more sophisticated for advancement of manufacturing process. However, the diverse die defects, such as particles, contaminations, discoloration, abnormal passivation, or probe marks exception, might be caused due to the inevitable results in the manufacturing process. Unexpected minor defects that are hard to inspect make quality control more difficult. As increasing strictly of quality requirements from customers, the traditional manual visual inspection can no longer meet the customers’ needs. Therefore, this research focused on the above mentioned defects in semiconductor manufacturing, and proposed an auto-inspection algorithm by using machine vision. The proposed method provides over 98% accuracy, and the average inspection time is 1.5 seconds of one image.