應用統計與資料採掘技術以提昇晶圓製造良率之研究

Abstract

當晶片尺寸愈小之時，半導體製造良率的 快速提昇亦愈形重要。如何從大量的製造資料 中有效偵測並診斷出良率損失，對半導體製造 商的競爭力愈益重要。 本研究旨在提出能從製造資料中自動萃取 出製程錯誤診斷的重要知識，並提出一整合性 的參數分析與圖形輔助技術，整合統計品質管 制、資料採掘、製程知識，以提昇多階段製程 整合的良率。 本整合技術包含了變異分解、主要參數過 濾、線性模式建立、圖形輔助診斷與監控法則 修改五大步驟。藉由實際資訊的測試，驗證了 本方法能輔助工程師診斷錯誤原因，因而提昇 良率。

Ramping up yield in semiconductor manufacturing is getting more difficult as the feature size of IC device continuously shrinks down. How to efficiently detect and diagnose the yield loss from the mass volume of manufacturing data is becoming critical to enhance a manufacturer’s competition capability. In this research, we focus on techniques for automatically extracting process knowledge from production database for fault diagnosis and optimizing device performance with fixed target. An integrated parametric analysis scheme is developed with supplemented graphical methods to facilitate interpretation of results. It consists of five phases: Device Variation Partition, Key Node Screening, Linear Equipment Modeling, Graph Aided Interpretation, and Control Policy Re-evaluation. The concepts of quality control, data mining, and process knowledge are integrated in this scheme. Field data case study shows that the integrated parametric analysis scheme is able to diagnose the parametric yield problem, help engineers construct the knowledge base, predict the yield, and provide insights for yield enhancement.