探針卡印刷電路板電鍍銅製程關鍵參數搜尋之資料挖礦架構與實證研究

DATA MINING FRAMEWORK FOR PROBE CARD PCB CRITICAL PARAMETERS AND EMPIRICAL STUDY FOR COPPER PROCESSING TECHNOLOGY

王俊堯(Chun-Yao Wang) ； 陳暎仁(Ying-Jen Chen) ； 簡禎富(Chen-Fu Chien)

《品質學報》 25卷6期 (2018/12) Pp. 361-379

https://doi.org/10.6220/joq.201812_25(6).0001

摘要

因應半導體製程日益精進，對晶圓針測探針卡的品質要求也愈趨嚴格。探針卡印刷電路板為組成探針卡的關鍵原物料，其高孔數、高縱橫比等特性使其製程較傳統電路板複雜許多。其中，電鍍銅製程品質好壞直接影響電路板電性訊號的傳遞品質，影響晶圓針測階段的良率準確性，其良率準確性更影響著半導體下游封裝成本。電鍍銅製程的因素包含電鍍電壓、電鍍液濃度、馬達電流等眾多參數，參數間交互作用影響電鍍品質，且機臺設備表現異常亦影響品質，其不同程度影響關係使得異常真因判斷不易。本研究為探討造成鍍銅製程品質的異常真因，建立一套資料挖礦分析架構，透過資料處理轉換，透過敘述統計、統計假設檢定等方法初步篩選重要因子，並結合羅吉斯迴歸以及決策樹模型，進一步將重要因子進行模型規則萃取，將可能影響鍍銅厚度品質的真因進行分析。藉由與國內某探針卡印刷電路板製造商進行實證研究，收集線上生產履歷資料、品質檢驗資料以及機臺參數即時監控資料進行研究架構實證分析，並將結果回饋個案公司，以提供線上工程師判斷異常真實成因之依據，進而提升電鍍銅製程之品質。

關鍵字

資料挖礦與大數據分析 ； 良率提升 ； 錯誤偵測與分類 ； 探針卡印刷電路板 ； 半導體製造

並列摘要

The increasing of the semiconductor manufacturing has facilitated wafer probing test process quality. Probe card printed circuit board (probe card PCB) is a crucial component when probe-testing. It bridges the electronic signal between wafer die and testing machine to ensure whether the dies are functioning properly or not. Due to its high-density and high aspect ratio of holes, the process of probe card PCB is more complicated than general PCB products. Among the processes of probe card PCB, copper electroplating process directly affects the signal transmission quality, and the defective products will influence the accuracy when probing. Consequently, the low accuacy of probing test leads to the cost of assembly and packaging phase. The factors in copper electroplating include the elctroplating voltage, the concentration of plating solution, the voltage of auxiliary equipments and so on. However, the interactions of factors are difficult to discover. In addition, there are variations of equipments in the real situation, influencing the quality of the process. In this paper, we propose a dataming framework to extract the root cause during the process. Description statistic methods, hypothesis tests, logistics regression and decision tree algorithm are applied to denote the key factors and to analyze the interactions of the key factors which cause the defects. This study carried out a empirical research with a probe card PCB company in Taiwan. Proposed framework is used to validate the results. The results give a criterion to the decision makers in the precess to clarify the root cause in the process.

並列關鍵字

data mining and big data analytic ； yield improvement ； fault detection and classification ； probe card PCB ； semiconductor manufacturing

參考文獻

吳冠宏、吳信宏、郭廣洋(2006)。應用分群技術於交通事故資料分析。品質學報。13(3)，305-312。

林定皓(2015)。高密度印刷電路板技術。臺北:全華圖書。

簡禎富、林昀萱、鄭仁傑(2008)。建構模糊決策樹及其在有交互作用之半導體資料之資料挖礦以提昇良率之研究。品質學報。15(3)，193-210。

簡禎富、林國義、許鉅秉、吳政鴻(2016)。臺灣生產與作業管理之相關期刊文獻回顧與前瞻：從工業3.0到工業3.5。管理學報。33(1)，87-103。

簡禎富、許嘉裕(2014)。資料挖礦與大數據分析。新北:前程文化。