建構大數據工程資料分析系統-以V半導體公司為例

Abstract

每家半導體業者都追求營收的成長，為求營收的成長，必須不停地發展更新的製程技術，更多的產出，更好的良率，更佳的產品組合。在台灣的半導體產業發展歷史上，從過去0.5um 到新的20 Nm，從動態隨機存取記憶體(Dynamic Random Access Memory，DRAM)到邏輯(Logic)產品，從單一晶圓廠(FAB)到Mega FAB甚至Giga FAB。隨著半導體製程的精進與產能的持續擴充，在晶圓(Wafer)生產過程中的資料量已經呈倍數增加。既有的工程資料分析系統架構無法快速儲存與處理如此大量的資料，此將導致良率的問題無法被正確鎖定解決，也讓問題影響的時間拖長，對公司營運造成影響。在近期大數據(Big Data)引起了一個風潮，Big Data彷彿是一個解決分析問題的萬靈丹。有許多論文提到如何架構分散式的Big Data平台，但少有論文談及各種不同型態資料在過去系統架構下如何移轉到Big Data平台，與移轉至Big Data平台的挑戰與經驗。本案例透過架構一個Big Data工程資料分析系統，包含資料平台與分析系統的建構。並描述現有Big Data平台如何進行整合這些資料，確保在量(Volume，資料大小）、速（Velocity，資料輸入輸出的速度）與多變（Variety，多樣性），真實性（Veracity）四個Big Data要素都能維持其有效性，最後滿足工程資料分析系統分析品質要求。

For the semiconductor industry, the managers always wants to achieve the upgrade of manufacture technology, more output, good yield and better product mix to make business growth. Good yield means less cost and more revenue, therefore it is important to keep yield stable and make all yield problems can be solved quickly. A engineer collects data from several systems and imports it into spread sheets or statistic applications to get the correlation among variables for the objective of less engineering data analysis. When a growing firm extends capacity or merges other factory, the engineering analysis will face the different data sources and complex format and large data size. Making engineering analysis system to be with more efficiency and more accuracy is the thesis’s research target. The main target of this thesis is to take a semiconductor firm which located in Taiwan as an example. Firstly we collect all issues from data platform and engineering analysis flow and define possible action plan in each process flow. We use benchmarking and proof of concept to get a better possible solution. We develop data platform with system development life cycle method and tune the analysis algorithm with prototyping methodology. Finally we combine the advantage of Big Data, leading companies, and experts experience to develop a new engineering data analysis system on Big Data platform. It provide a developing experience to fulfill the requirement of volume, velocity, variety, veracity of Big Data and a real platform migration case from legacy system to Big Data platform. The new engineering system has good improvement on data analysis performance and accuracy. These findings of this research can help a semiconductor firm and its project team to reduce Big Data evaluation effort, and give them a good way to realize the new Big Data application.