台灣半導體產業在國家經濟發展過程中，扮演著極重要角色，如何有效 運用有限資源達到最大的效率，已成為各家廠商極力追求的目標。本文旨在 以半導體產業中型封測領域廠商為例，運用獨立成份分析法(independent component analysis, ICA)與資料包絡分析法(data envelopment analysis, DEA)， 探討其在2007 至2010 年間之經營效率。為了驗證所提方法的有效性，我們 除了使用蒙地卡羅模擬法進行模擬分析外，也使用傳統的DEA 技術進行比 較。此外，我們也針對台灣半導體產業上市櫃封測領域廠商在2007 年至 2010 年的財務相關資料進行實證研究，更以麥氏生產力指數探討封測領域廠 商在2007 年至2010 年間生產力及跨期效率的平均變動情形，希望能藉此瞭 解各廠商在各年度效率與生產力的成長及衰退。本文主要貢獻：(1)利用模擬 方式驗證ICA 方法可以有效解決DEA 模式中變項資料間存在高度相關性的 問題；(2)根據實際之台灣半導體產業上市櫃公司財務資料，有效分析出各公 司的專業技術能力與生產效率，藉以協助效率較低的企業找出應調整之投入 量。實證結果發現，部分廠商無效率的主要原因為技術效率的退步及規模報 酬遞減，這樣的結果顯示廠商在經營上應更注意市場的實際情況與企業內部 資源的配適性。

In the development of Taiwan’s economy, the semiconductor industry plays a very important role. To keep the competitive advantage, almost all the semiconductor companies are dedicating in efficiency improvement. That is, how to achieve maximum efficiency based on limited resources has become their primary target. In this study, the dataset provided by several medium-size semiconductor packaging and testing companies in Taiwan is used for analysis. In additions, two approaches, independent component analysis (ICA) and data envelopment analysis (DEA) are proposed to evaluate the efficiency of each Decision Making Unit (DMU) from 2007 to 2010. To demonstrate the performance of the proposed method, we are not only comparing the discrimination capability of ICA-DEA with traditional DEA approach but also applying Malmquist productivity index (MPI) to investigate the productivity change over time. This study contributes to the DEA literature and the semiconductor industry in two aspects. First, independent component analysis (ICA) can be applied to eliminate multicollinearity between explanatory variables in DEA model. Second, this study provides the practical contribution that the selected medium-size semiconductor packaging and testing companies will have their optimal input and output resources setups so that they can enhance their competitive advantage. The result shows that the causes of inefficiency for some companies are the reduction of their technical efficiencies and the decreasing of their returns to scale, which means that their resource allocation problems should be considered more carefully.