整合正向與負向關聯規則探勘於發掘DRAM製造之問題機台

Abstract

半導體產業競爭日益激烈，製程愈趨複雜，若能快速的找出製程中的根本問題站點或機台，便能儘早的解決問題以提升良率，比對手更具競爭力。本研究提出整合正向與負向關聯規則的資料探勘方法，以產品的良率與站點機台間的關聯規則，推估造成缺陷產品的根本問題站點機台。正向關聯規則為造成缺陷產品的問題站點機台，而負向關聯規則為產出正常產品中非該站點機台與其它站點機台的關聯。一站點機台X若在正向關聯規則X=>Y及負向關聯規則¬X=>Y皆出現，則X更能確認為問題站點機台。因此整合正向與負向關聯規則探勘，分析站點機台間關聯規則，能更有效的找出問題機台。實驗評估以某DRAM半導體廠商資料為例，由結果得知，將負向關聯規則納入考量的整合關聯規則方法較單以正向關聯規則分析的方法表現佳，佐證本研究提出方法之實際效益。

Manufacturing processes have become more and more complex in semiconductor industry. Improving yield by identifying root cause machines is the key factor of improving competitiveness. In this study, we propose to identify root cause machines by integrating positive and negative association rule mining of correlations between machines from the manufacturing processes of defective and normal products, respectively. If a machine X exists both in positive association rules (eg. X=>Y) and negative association rules (eg. ¬X=>Y), then X might be the most doubtful machine. The experimental results of real DRAM manufacturing datasets show that the proposed integrated approach is more effective than the approach with considering only positive association rules.