應用監督式類神經網路於晶粒表面缺陷辨識之研究

Abstract

在晶圓的製造過程中，常會因為人為的疏忽、設備不良與環境品質不佳等因素而造成晶圓表面缺陷的產生，此缺陷的產生將影響著後續的作業。目前廠商在晶圓切割後，通常採用人工目視檢驗的方式來挑檢晶粒，以維持整體晶粒之品質。然而，使用人工目視檢驗的方式，將需要投入龐大的人力、物力與作業空間，對晶圓切割廠來說是一項負擔，且維持檢驗品質亦是一項考驗，因此，目前各廠商急於改變此檢驗方式。有鑑於此，本研究提出應用類神經網路來建構一套晶粒表面缺陷之辨識方法，辨識範圍涵蓋整個晶粒表面，以確保積體電路品質。此外，本研究舉一個實際案例，來說明所提出之應用類神經網路的辨識方法，結果顯示本研究所提出之方法具有效性與實用性。

Defects commonly occur on the surface of a wafer during production owing to carelessness of the operator, poor quality of equipment and inadequate environment. Such defects negatively impact subsequent operations. To maintain the die quality, factories normally visually inspect dies after wafer sawing, subsequently leading to a significant amount of manpower, material resources and operation area and ultimately making it extremely difficult to maintain the inspection quality. This research presents an effective procedure capable of detecting die defects by using supervised neural networks. A case study involving the wafer production demonstrates the effectiveness and practicability of the proposed approach.