本研究是第一篇針對台灣IC設計業的研發活動進行效率分析的論文。文中對研發的投入及產出變數作深入的探討，不僅以研發資本而非研發費用及加權研發人力為投入並且加入公司知識累積存量的概念。其中自有知識存量以自有公司前期累積專利申請數為替代變數。而產出部分也不僅考慮當年申請專利數核准數量還包括下一年度營業毛利。使用兩階段資料包絡分析法探討研發投入的運用效率。第一階段使用投入導向DEA-CCR與DEA-BCC模式評估2000年到2002年台灣上市上櫃IC設計業者運用研發資源能力所得出的研發效率值並作差額變數分析，第二階段採用Tobit迴歸分析尋找可能影響研發整體效率影響因子。最後，本研究依據研究結果提出對管理者及未來研究給予建議。 實證結果發現，〈聯發科〉是連續三年被評估相對整體效率為1的廠商，〈威盛〉與〈立錡〉為表現其次的廠商。就整體產業而言，三年的研發效率呈現低效率狀況，表示其研發資源有嚴重浪費與錯置的情形。效率分析中可看出研發資本（RK）對加權研發人力（RL）比值相對高的廠商，相對整體效率值呈現逐年負向趨勢。經過Tobit迴歸參數推估檢定本研究六大假說，人力素質、每人年約收入及研發人力密集度都與研發效率成正向關係。而員工平均年資與研發效率值呈現負向關係，與研究的預期關係不同。為第一線晶圓代工廠（聯電或台積電）轉投資的IC設計公司，研發效率並不因此網絡關係而有所影響。公司規模大小也不影響研發效率的表現。

本研究旨在探討民國88年至94年台灣IC設計產業之研發效率分析。採兩階段資料包絡分析法(Data Envelopment Analysis, DEA):第一階段以保證區域資料包絡分析法(Assurance Region DEA，AR/DEA)衡量績效，以避免傳統DEA方法可能出現產出或投入的乘數為0之狀況。第二階段採資料切齊的(censored) TOBIT迴歸模型探討影響研發效率值的效率因子。 研究發現: (1)只考量專利數量與納入專利品質考量下之統計檢定結果，雖沒有顯著的差 異，但對少數的廠商的研發效率值與排名確實有變動，納入專利品質考 量，可以掌握更多研究發展資訊，了解廠商專利真正的價值。 (2)專利被引證次數與研發效率值呈現顯著正相關。 (3)TOBIT迴歸實證結果顯示: 研發效率與公司規模無顯著關係；研發效率與研發人力密集度無顯著相 關；研發效率與研究發展費用率呈負向關係；研發人員年約收入對研發技 術效率無顯著影響；外來知識流量占自有知識存量比例會正向影響研發效 率表現。 / This study investigates R&D efficiency analysis of IC design industry in Taiwan from 1999 to 2005 by using the two-stage Data Envelopment Analysis (DEA). To avoid the situations that the multiplier of the input or output might be zero, we adapt the method of Assurance Region DEA (AR/DEA) to measure achievements in the first stage. We then use censored TOBIT regression model to study the factors that influence the efficiency value in the second stage. Through our study, we conclude that: (1) Although there is no significant difference in the results of statistical test whether we take patent quantity into account or not, both the value of R&D efficiency and ranking of a few firms do change. We can grasp more information of research and development and understand the true value of firm’s patent as we put patent quality into account. (2) The correlation coefficient between the amount of patent cited and the value of R&D efficiency is positive significantly. (3) The empirical results of TOBIT regression show that: a. There is no significant relationship between the R&D efficiency and the company’s scale. b. There is no significant relationship between the R&D efficiency and the density of R&D workers. c. There is a negative relationship between the R&D efficiency and the ratio of R&D expense. d. The yearly revenue of R&D workers does not significantly influence the R&D and technology efficiency. e. The ratio between the flow rate of external knowledge and the owned knowledge storage affects positively on R&D efficiency.

本研究主要是針對台灣LED產業上中游廠商的研發活動與營運活動進行效率分析的論文。文中採取兩階段DEA之投入導向的CCR與BCC模式進行效率評估，第一階段研究LED上中游9家廠商2004~2006年運用研發資源的能力，並將所得出的研發效率值作差額變數分析，其投入項為研發資本與研發人力，產出項為專利權數量；第二階段研究LED上中游9家廠商2006~2008年運用營運資源的能力，並將所得出的獲利效率值作差額變數分析，其投入項為專利權數量與行銷費用，產出項為營業毛利。接著利用二維矩陣分析與Pearson相關性分析試圖找出研發效率與獲利效率之間的關聯性，並藉由Tobit迴歸分析尋找可能影響效率的因子。最後，本研究根據實證結果分析對管理當局及未來研究提供建議。 實證結果發現：（1）在研發效率上，【新世紀】連續三年整體效率皆為1，是最有效率的廠商，而【璨圓】、【華上】、【洲磊】也有兩年在技術效率上是表現優異的；（2）在獲利效率上，【廣鎵】連續三年整體效率皆為1，是最有效率的廠商，【新世紀】也有兩年被評估是有效率的，而【光磊】與【洲磊】在技術效率上表現優異且穩定，且就整體產業而言，各廠商在獲利效率上的表現更勝於研發效率；（3）在研發效率的影響因子中，專利變化數量及員工教育程度與研發效率呈現正向關係，專利累積存量則呈現反向關係與預期方向不同，而發明專利比例及研發資本人力比不影響研發效率的表現；（4）在獲利效率的影響因子中，員工平均年資、行銷能力及發明專利比例與獲利效率呈現正向關係，公司規模則呈現反向關係與預期方向不同，而員工平均年收及員工教育程度不影響獲利效率的表現；（5）本研究發現台灣LED產業上中游廠商，擁有研發效率的不一定擁有獲利效率，這兩個指標並沒有明顯的相關。 / This paper presents a study which utilizes the input oriented CCR&BCC model of two-stage Data Envelopment Analysis(DEA) that separates R&D efficiency and profitability efficiency to evaluate the performance for the listed LED manufacturers in Taiwan. Research period of the first stage and second stage respectively is during 2004 to 2006 and 2006 to 2008. In order to deal with the result of DEA, this study takes the Slack Variable Analysis that could be the improving reference with inefficiency of DMUs and uses two-dimension matrix analysis and Pearson Correlation Analysis to find relationships of R&D efficiency and profitability efficiency. Furthermore, this study seeks factors which may affect the efficiencies by applying Tobit Regression Analysis. Finally, we hope these results can provide actual suggestions for management authority and future research. The major empirical findings of this study are as follow：(1) In the R&D efficiency, GPI is the best firm of global efficiency for three years, and FOREPI、AOC and UniLite also have the best performance of technical efficiency for two tears. (2) In the profitability efficiency, HUGA is the best firm of global efficiency for three years and GPI is also the best for two years. OPTO TECH and UniLite have the best performance of technical efficiency for three years. In terms of overall industry, the performance in the R&D efficiency of these firms is better than in the profitability efficiency. (3) In influence factors of R&D efficiency, difference of patent quantity、education background of employee are positive correlation and accumulated patent quantity is negative correlation. Proportion of invention patent and R&D capital per labor are no correlation. (4) In influence factors of profitability efficiency, qualification of employee、market ability、proportion of invention patent are positive correlation and firm scale is negative correlation. Revenue and education background of employee are no correlation. (5) There is no evidence to prove a relevance of R&D efficiency and profitability efficiency of LED manufacturers in Taiwan.