過程裝備與控制工程中英文對(duì)照外文翻譯文獻(xiàn)

1、 中英文對(duì)照外文翻譯文獻(xiàn)(文檔含英文原文和中文翻譯)譯文：半導(dǎo)體制造過程控制和監(jiān)測(cè)：工廠全框架摘要 半導(dǎo)體行業(yè)已經(jīng)開始了從 200 毫米到 300 毫米的晶片技術(shù)的過渡，以提高制造效率，降低制造成本。這些技術(shù)變革展出現(xiàn)了優(yōu)化設(shè)計(jì)下一代工廠控制系統(tǒng)的獨(dú)特的機(jī)會(huì)。本文首先提出為 300毫米設(shè)備和計(jì)量工具和材料處理高度自動(dòng)化的系統(tǒng)在全工廠范圍分層控制的框架?，F(xiàn)有相關(guān)運(yùn)行的技術(shù)在工廠控制范圍內(nèi)通過了審查和分析。過程和計(jì)量數(shù)據(jù)的監(jiān)測(cè)，通過是舉例來說明的。缺失的部分，作為未來研究和發(fā)展的方向而被指出。結(jié)束語附在文章末尾。2005 年由 Elsevier 有限公司出版關(guān)鍵詞 半導(dǎo)體制造 波段范圍內(nèi)的控制 電

2、氣參數(shù)控制 運(yùn)行控制 故障檢測(cè)和分類 計(jì)量數(shù)據(jù)監(jiān)控1 導(dǎo)言半導(dǎo)體行業(yè)已開始從 200 毫米的技術(shù)過渡到 300 毫米轉(zhuǎn)換，以提高生產(chǎn)效率，降低 制造成本。隨著這種轉(zhuǎn)變，300 毫米的資金的開支為 200 毫米的一倍。（一個(gè)生產(chǎn) 200毫米廠的費(fèi)用超過 10 億美元，而 300 毫米晶圓廠的費(fèi)用超過 20 億。）其他技術(shù)變革包括：?jiǎn)尉庸つ芰?，而不是批量業(yè)務(wù)營(yíng)運(yùn)能力；全自動(dòng)化物料處理系統(tǒng)（AMHS）的跨海灣和內(nèi)灣運(yùn)輸；綜合計(jì)量，以便及時(shí)控制；過程控制和故障診斷的高度自動(dòng)化。由于新時(shí)代的工廠資本高度密集，工廠的關(guān)鍵是保持高效率的運(yùn)作，減少設(shè)備停機(jī)的時(shí)間，優(yōu)化高品質(zhì)產(chǎn)品的產(chǎn)量。國(guó)際技術(shù)路線圖 明確

3、說明工廠的信息和控制系統(tǒng)是42一項(xiàng)重要的技術(shù)，是減少周期時(shí)間提高利潤(rùn)。技術(shù)變革預(yù)示著為新時(shí)代工廠優(yōu)化設(shè)計(jì)的過程控制系統(tǒng)的獨(dú)特機(jī)會(huì)。缺乏現(xiàn)場(chǎng)傳感器提供的反饋控制和優(yōu)化晶圓狀態(tài)的實(shí)時(shí)信息是半導(dǎo)體制造控制業(yè)長(zhǎng)久的挑戰(zhàn)。但幸運(yùn)的是，近期計(jì)量技術(shù) 的發(fā)展提供了改進(jìn)及時(shí)性和測(cè)量數(shù)據(jù)的作用44性的機(jī)會(huì)。通常一個(gè)現(xiàn)代化的工廠，在半導(dǎo)體制造持續(xù)的挑戰(zhàn)控制是至關(guān)重要的現(xiàn)場(chǎng)傳感器提供的缺乏，晶圓的實(shí)時(shí)狀態(tài)信息反饋的控制和優(yōu)化。最近，推進(jìn)計(jì)量技術(shù) 提供44了一個(gè)機(jī)會(huì)，改善及時(shí)性和實(shí)用性測(cè)量的數(shù)據(jù)。通常一個(gè)現(xiàn)代化的工廠具有相當(dāng)多的的測(cè)量數(shù)據(jù)可供分析和控制：（1）在工具層面，實(shí)時(shí)數(shù)據(jù)反映了設(shè)備的健康狀況和提供反饋并實(shí)時(shí)控

4、制;（2）綜合測(cè)量和在線測(cè)量數(shù)據(jù)的幾何尺寸可進(jìn)行后期處理，有輕微的計(jì)量延遲;（3）樣品和最終電氣測(cè)試（電子測(cè)試）提供數(shù)據(jù)與中期或長(zhǎng)期的電性能時(shí)間延遲，但他們最重要的信息制造業(yè)的成效。先進(jìn)的控制手段與優(yōu)化方法應(yīng)盡量在所有的信息的使用綜合等級(jí)的高效率生產(chǎn)和嚴(yán)格的產(chǎn)品質(zhì)量控制。監(jiān)測(cè)和控制的半導(dǎo)體制造 程序已經(jīng)在一些美國(guó)的大學(xué)和工業(yè)研究實(shí)驗(yàn)室研發(fā)出來。 作為代表的有 U.C.伯克利28,32對(duì)統(tǒng)計(jì)建模與控制等離子蝕刻機(jī)，密歇根實(shí)時(shí)運(yùn)行多變量控制 ，以及麻省理工學(xué)院在不同的傳感器和控制技術(shù) 。由于缺乏現(xiàn)場(chǎng)傳感227,8器大部分控制開發(fā)工作從運(yùn)行（R2R）控制策略12,41。在馬里蘭大學(xué)研究小組貢獻(xiàn)的運(yùn)行

5、區(qū)控制2,5,53。領(lǐng)先的半導(dǎo)體制造商協(xié)會(huì) SEMATECH發(fā)布了在等離子設(shè)備故障檢測(cè)和診斷的幾種基準(zhǔn)問題 。自適應(yīng)非線性 R2R 控制問題被提出。模型預(yù)測(cè)適用于 R2R 控4 制并具有額外的處理能力，可以明確的設(shè)定系統(tǒng)參數(shù) 。美國(guó)德州大學(xué)奧斯汀分校，我19術(shù)的計(jì)量數(shù)據(jù)光盤 。其他新的發(fā)展，控制和故障檢測(cè)是在最近由 Sematech 組織的 Spie14裝備水平控制涉及的工具自動(dòng)反饋控制參數(shù)和小規(guī)模運(yùn)行控制使用的綜合計(jì)量。第二個(gè)層次的運(yùn)行控制涉及在線測(cè)量的前饋使用和反饋控制。第三個(gè)層次是島嶼控制。層次結(jié)構(gòu)的頂層是工廠全控制，這是最高級(jí)別的優(yōu)化，通過重新計(jì)算所需的最優(yōu)幾何目標(biāo)而把消耗控制在劑量較

6、低的水平。該文件的組織如下。我們首先提出為 300 毫米設(shè)備和計(jì)量工具和材料處理高度自動(dòng)化的系統(tǒng)在全工廠范圍分層控制的框架?，F(xiàn)有相關(guān)運(yùn)行的技術(shù)在工廠控制范圍內(nèi)通過了審查和分析。過程和計(jì)量數(shù)據(jù)的監(jiān)測(cè)，通過是舉例來說明的。缺失的部分，作為未來研究和發(fā)展的方向而被指出。結(jié)束語附在文章末尾。化學(xué)，力學(xué)，拋光）。2 一個(gè)工廠的框架范圍內(nèi)的控制 幾乎所有現(xiàn)有的發(fā)展都是以基于裝備水平計(jì)量數(shù)據(jù)的 R2R 控制為基礎(chǔ)的。這些被稱為島嶼控制如圖 1 的下部所示。現(xiàn)有的控制策略都不能檢測(cè)協(xié)調(diào)多個(gè)制造步驟，從而提高電氣參數(shù)方面的總體質(zhì)量。R2R 控制器的漂移補(bǔ)償通過計(jì)量設(shè)備反饋，但他們無法彌補(bǔ)計(jì)量漂移和不確定性。這里

7、提出的電氣參數(shù)直接控制可以彌補(bǔ)計(jì)量漂移和幾何測(cè)量低于 SPC 限制的系統(tǒng)誤差。據(jù)認(rèn)為，控制和電氣參數(shù)優(yōu)化代表半導(dǎo)體制造控制系統(tǒng)的新一代產(chǎn)品，因?yàn)樗苯涌刂葡录?jí) R2R 控制器的使用。當(dāng)市場(chǎng)需求數(shù)量一定時(shí)，電氣參數(shù)的控制和優(yōu)化將最大限度地高檔產(chǎn)品產(chǎn)量或降低運(yùn)營(yíng)成本。如圖 1 所示工廠全控制框架提供優(yōu)化和加強(qiáng)協(xié)調(diào)，逐步減少變性，返工和廢料，從而改善整體設(shè)備效率，降低制造成本。這一框架由秦和桑德曼 在 AMD 上部署許多38R2R 控制器并分析更高層次的控制需求后提出。裝備水平的控制涉及工具參數(shù)的自動(dòng)反饋控制。下一個(gè)層次是 R2R 控制采用綜合或內(nèi)嵌計(jì)量，以達(dá)到特定的目的。第三個(gè)層次是控制島嶼從多個(gè)

8、步驟來執(zhí)行前饋和反饋控制以及工具的性能匹配。層次結(jié)構(gòu)的頂層是電氣參數(shù)控制（EPC）或工廠范圍的控制，通過重新計(jì)算對(duì)下級(jí)的最優(yōu)目標(biāo)來實(shí)現(xiàn)預(yù)期的電氣性能。設(shè)備漂移，計(jì)量漂移，和物質(zhì)補(bǔ)償?shù)淖兓窃?EPC 反饋水平，從而改善進(jìn)程，提高可用性，減少計(jì)量校準(zhǔn)器和測(cè)試晶圓的使用。這種多層次控制框架類似于已經(jīng)在煉油行業(yè) 成功的分層控制框架，但存在重大分39歧：（1）最低級(jí)的控制主要是批量操作；（2）中層 R2R 的控制，除了干擾幾乎沒有 R2R動(dòng)力學(xué)過程的動(dòng)態(tài)特性；（3）頂層的 EPC 是一個(gè)多步操作的控制，目的是彌補(bǔ)以前的步驟失誤，不考慮絕對(duì)誤差，只要分步進(jìn)行計(jì)量的測(cè)量結(jié)果是可用的。這使其與模型預(yù)測(cè)控制（

9、MPC）縮小視野的批處理不同。在化學(xué)和煉油過程中，頂級(jí)優(yōu)化是實(shí)時(shí)優(yōu)化 ，31中等水平的是全面的動(dòng)態(tài) MPC。由于 MPC 是一個(gè)強(qiáng)大的和成功的技術(shù)，它已經(jīng)在半導(dǎo)體行業(yè)擴(kuò)展到調(diào)度和生產(chǎn)規(guī)劃11,46,47。該全工廠控制框架也參考 MPC 框架，但重點(diǎn)是優(yōu)化設(shè)備的電氣參數(shù)控制。電子測(cè)試數(shù)據(jù)用來改善設(shè)備模型之間的電子測(cè)試數(shù)據(jù)和模型的參數(shù)不匹配。參數(shù)評(píng)估進(jìn)行后，估計(jì)參數(shù)被發(fā)送到工廠范圍優(yōu)化器，分配指標(biāo)，以較低級(jí)別的控制器，調(diào)節(jié)生產(chǎn)制造過程。隨著新的模型參數(shù)設(shè)置更新，該模型開始用于 EPC 控制。3 運(yùn)行控制算法近年來，運(yùn)行（R2R）控制技術(shù)已受到半導(dǎo)體制造業(yè)的巨大關(guān)注。莫恩和赫維茨（莫恩等 ）定義了

10、R2R 控制：“一個(gè)離散的過程和機(jī)械控制，其中就某一特定過程的產(chǎn)物34 易地修改，以盡量減少過程中的漂移，轉(zhuǎn)變和可變性 ”。為了修改配方，處理過程中漂移，轉(zhuǎn)移和其他變化，目前的工具和晶圓州有必要進(jìn)行估計(jì)。一類廣泛使用的運(yùn)行可以運(yùn)行的是在指數(shù)加權(quán)移動(dòng)平均（EWMA）的統(tǒng)計(jì)數(shù)字，估計(jì)過程擾動(dòng)的控制器。EWMA 已經(jīng)用于長(zhǎng)的時(shí)間質(zhì)量監(jiān)測(cè)目的 。其使用是作為 R2R 控制的近期應(yīng)用 。940如需 easurement xn,xn-1, . . .的時(shí)間序列，其中 N 表示運(yùn)行數(shù)，給出了 EWMA 的遞推公式：Xn=wxn-1+(1-w)xn最有效的一個(gè)操縱 R2R 控制變量是在諸如蝕刻時(shí)間，曝光時(shí)間，

11、處理步驟，處理時(shí)間和平整時(shí)間。在這種情況下的控制變量通常在何種程度下發(fā)展的進(jìn)程處理時(shí)間，如蝕刻和深入的關(guān)鍵方面。前面介紹的相乘模式不適合典型的線性狀態(tài)空間模型，但可以轉(zhuǎn)換為線性狀態(tài)與過程和測(cè)量空間模型，簡(jiǎn)單地由對(duì)數(shù)測(cè)量噪音。因此，本文提出的所有控制算法適用于時(shí)間控制。4 故障檢測(cè)與診斷數(shù)據(jù)處理工具例如溫度，壓力，氣體流量等將被應(yīng)用到單晶片或批量的磁盤中。比如一些典型的加工服務(wù)，包括等離子體刻蝕，薄膜沉積，快速熱退火，離子注入，化學(xué)機(jī)械研磨等。在大多數(shù)的處理步驟中，每一個(gè)感應(yīng)器都收集晶圓磁盤或那些數(shù)據(jù)處理的工具。這個(gè)數(shù)據(jù)可以制造出先進(jìn)的傳感器平臺(tái)，如光發(fā)射光譜中的實(shí)時(shí)數(shù)據(jù)，簡(jiǎn)易的統(tǒng)計(jì)數(shù)字，其數(shù)據(jù)

12、形式在每次運(yùn)行結(jié)束時(shí)都可用。數(shù)據(jù)故障檢測(cè)與診斷已成功應(yīng)用于其他行業(yè)的開發(fā)和應(yīng)用中30,49。這些驅(qū)動(dòng)的故障檢測(cè)技術(shù)是基于多元統(tǒng)計(jì)分析的基礎(chǔ)上來完成的，如主成分分析（PCA）和局部最小乘積（PLS）的數(shù)據(jù)和相關(guān)的統(tǒng)計(jì)質(zhì)量控制方法 。這些監(jiān)測(cè)方法最近的一項(xiàng)審查可參考2636。雖然半導(dǎo)體制造的批處理性質(zhì)為申請(qǐng)多路過程監(jiān)控 提供了很多機(jī)會(huì)，許多半導(dǎo)體35計(jì)量數(shù)據(jù)組織形式被分成三個(gè)方面。其中一個(gè)是 CD 計(jì)量，它的三個(gè)方面是晶圓，站點(diǎn)和參數(shù)。批量數(shù)據(jù)也通?？梢约庸わ@示出批量，時(shí)間和參數(shù)的工具（圖 2）。多路 PCA 已成功地應(yīng)用于許多不同行業(yè)的批量加工過程監(jiān)控。在半導(dǎo)體制造領(lǐng)域Yue 等人 提出了通過申

13、請(qǐng)多路 PCA 到等離子蝕刻機(jī)的光發(fā)射譜來擴(kuò)展數(shù)據(jù)的觀點(diǎn)。52對(duì)于計(jì)量和處理工具的監(jiān)測(cè)，數(shù)據(jù)可以通過站點(diǎn)或時(shí)間（每行代表一個(gè)晶片上的一個(gè)站 點(diǎn)或批處理的一個(gè)時(shí)刻）或晶片（每行代表一個(gè)晶圓）展開。在這項(xiàng)工作中，晶圓級(jí)故障檢測(cè)與識(shí)別是必需的，所以后者已被選為更好的展開方法（圖 4）。正如后面將要討論的，用站點(diǎn)或時(shí)間分析數(shù)據(jù)的優(yōu)點(diǎn)可以通過實(shí)現(xiàn)多塊做法來體現(xiàn)。晶圓批次網(wǎng)站時(shí)間參數(shù)網(wǎng)站時(shí)間/圖 4 展開的網(wǎng)站水平和批量數(shù)據(jù) 4.1 計(jì)量數(shù)據(jù)監(jiān)測(cè)雖然加工業(yè)務(wù)創(chuàng)造了結(jié)構(gòu)，但是計(jì)量業(yè)務(wù)使它們擁有了這些特點(diǎn)。計(jì)量測(cè)量的一些例子包括發(fā)展檢驗(yàn)關(guān)鍵尺寸（DICD），最后檢查關(guān)鍵尺寸（FICD）和薄膜的厚度。計(jì)量測(cè)量通

14、常在半導(dǎo)體晶片上多點(diǎn)采樣，在同一點(diǎn)檢測(cè)不同特征。故障檢測(cè)和識(shí)別應(yīng)用到站點(diǎn)級(jí)計(jì)量數(shù)據(jù)是為了驗(yàn)證整個(gè)晶圓表面建立在半導(dǎo)體晶片上的結(jié)構(gòu)是否都均勻的在他們?cè)撛诘奈恢?。作為一個(gè)例子，我們使用 PCA 進(jìn)行故障檢測(cè)并用來自得克薩斯州奧斯汀 AMD 的Fab25 DICD 進(jìn)行數(shù)據(jù)鑒定。在光阻材料發(fā)展起來之后，該 DICD 是光阻材料圖案寬度。如圖 3 所示，在各向同性發(fā)展表明各光阻底部和頂部之間的差異很小。該數(shù)據(jù)集由 700片晶圓組成，每個(gè)晶圓頂部和底部各有 9 個(gè)測(cè)點(diǎn)。圖 6 為分組的所有 9 個(gè)站點(diǎn)為兩個(gè)參數(shù)合計(jì)，圖 7 重點(diǎn)考慮到每個(gè)站點(diǎn)的參數(shù)。這9 個(gè)圖可以很容易地識(shí)別基于晶片位置的問題。由提供的

15、數(shù)據(jù)顯示，出現(xiàn)的漂移在站點(diǎn)2，3 和 4 最強(qiáng)，而它是很難在點(diǎn) 6，8 和 9 強(qiáng)烈漂移。隨著了解每個(gè)站點(diǎn)在晶圓上的位置，將有可能使用這些圖和掩蔽工具來解決可能傾斜或焦點(diǎn)問題。 水水水圖 5 DICD 使用 SPEr 故障檢測(cè)，T r 的和烏拉圭回合。 水水水水水水水網(wǎng) 站水水 雖然跟蹤塊是好的貢獻(xiàn)，影響了一大批晶圓的做法，但是也必須考慮的一個(gè)問題是如何在一個(gè)單晶片上集成，這個(gè)目標(biāo)還有待驗(yàn)證。為了演示此功能，圖示已生成晶圓 395和 450（圖 5 中用箭頭標(biāo)出）。晶圓 395 圖示如圖 8。顯而易見，測(cè)量 4（下部：Site 4），12（頂部：Site 3），13（頂部：Site 4）值得懷

16、疑。在底部和頂部的參數(shù)，也和 Site 4 一樣作為異常的站點(diǎn)貢獻(xiàn)，表示出問題。一個(gè)合乎邏輯的解釋就是到晶圓上該站點(diǎn)有需要進(jìn)一步探討的問題，可能會(huì)影響產(chǎn)品產(chǎn)量或性能。雖然 Site 3 頂端尺寸也被特別指出，但是頂部和底部都被認(rèn)為是共同時(shí)，整體站點(diǎn)的貢獻(xiàn)是正常的。5 挑戰(zhàn)與機(jī)遇5.1 電氣參數(shù)建模為貫徹落實(shí)晶圓廠的控制，發(fā)展基于物理的器件模型映射到電氣幾何參數(shù)，如振動(dòng)頻率參數(shù)，擦除閃存時(shí)間，電阻值等是很重要的。這個(gè)模型與用于 R2R 控制器的用來描述如關(guān)鍵尺寸，深度，厚度或工藝操作條件之間的關(guān)系的模型不同。由于優(yōu)化在廣泛應(yīng)用的 EPC 中涉及到了，非線性物理模型的基礎(chǔ)模型適合于優(yōu)化。適合 EP

17、C 的模型必須可以實(shí)時(shí)執(zhí)行，它不同于模擬和設(shè)計(jì)模型，如 TCAD 模型。因此，減少型號(hào)為 EPC一個(gè)重要問題。隨著半導(dǎo)體產(chǎn)業(yè)進(jìn)入 100 納米時(shí)代（目前 90 納米，并會(huì)很快發(fā)展到 65 nm），多尺度建模與仿真變得很重要。這些模型10，18可以幫助了解微觀行為并有效控制和避免潛在的缺陷。無方程仿真模型 可用于控制和約束處理使用。275.2 長(zhǎng)延遲模型更新隨著這一進(jìn)程的計(jì)量和物質(zhì)隨時(shí)間變化，需要從實(shí)際使用數(shù)據(jù)適應(yīng)模型參數(shù)。為了統(tǒng)計(jì)參數(shù)集，非線性物理模型需要非線性最小二乘法。最小二乘法的目的是 EPC 的二重目標(biāo)，即最小化之間的電子測(cè)試數(shù)據(jù)和模型的輸出成品晶圓地段或受到可能的制約因素的差異。在更

18、新模型的一個(gè)具有挑戰(zhàn)性的任務(wù)是在電子測(cè)試測(cè)量數(shù)據(jù)的延遲。更新機(jī)制應(yīng)該只響應(yīng)長(zhǎng)期持久的變化，而不是短暫的臨時(shí)錯(cuò)誤。重復(fù)學(xué)習(xí)控制和實(shí)時(shí)反饋控制框架（金等人，2003 年）是一個(gè)可執(zhí)行的解決方案，但需要進(jìn)一步努力，處理的長(zhǎng)時(shí)間延遲，多了一個(gè)新的 EPC 執(zhí)行目標(biāo)會(huì)有更長(zhǎng)的時(shí)間延遲。更新后的模型被用于 EPC 控制器，以便接收下一個(gè)目標(biāo)大量數(shù)據(jù)傳入。 5.3 FDC 與 R2R 的一體化正如在圖 1 中的說明，晶圓廠控制框架中額每個(gè)步驟具有廣泛的 R2R 控制器和 FDC模塊。FDC 的目的是監(jiān)測(cè)分析以歷史數(shù)據(jù)為基礎(chǔ)的正常情況下的偏差。整合之一就是將多路 PCA 應(yīng)用于設(shè)備監(jiān)控。通過雙方合作，F(xiàn)DC

19、和 R2R 控制向他們的一體化提出了挑戰(zhàn)。首先，F(xiàn)DC 的方法通常假定具有循環(huán)或類似批長(zhǎng)度。另一方面，R2R 模塊的目的是調(diào)整安排，如生產(chǎn)時(shí)間，以盡量減少由于正常變異過程產(chǎn)生的漂移。FDC 模塊，如果不妥善設(shè)計(jì)，可認(rèn)為正常 R2R 調(diào)整偏離正常情況下并且是錯(cuò)誤的警告。另一個(gè)挑戰(zhàn)是 R2R反饋故障診斷的影響。由于工具控制反饋的故障的根本原因可能是由一個(gè)變量轉(zhuǎn)移到另一個(gè)的反饋存在時(shí)效性。對(duì)于故障診斷，反饋信息的利用33可能是一個(gè)解決的途徑。6 結(jié)束語半導(dǎo)體產(chǎn)業(yè)正在成為一個(gè)資本最密集的高比例收入行業(yè)之一。另一方面，優(yōu)化和生產(chǎn)業(yè)務(wù)的控制最近已受到重視，并證明是必要的競(jìng)爭(zhēng)優(yōu)勢(shì)。一個(gè)設(shè)計(jì)良好的晶圓廠的控制

20、框架，給半導(dǎo)體制造商提供了競(jìng)爭(zhēng)力，因?yàn)樗麄冞^渡到300 毫米技術(shù)，并預(yù)見了未來450 毫米技術(shù)。自動(dòng)化的物料處理系統(tǒng)和自動(dòng)化 R2R 控制功能提供了實(shí)施的層次各級(jí)晶圓廠的控制和故障檢測(cè)的必要基礎(chǔ)。領(lǐng)先的設(shè)備制造商設(shè)想未來大部分的日常業(yè)務(wù)將由干凈的房間轉(zhuǎn)移到未來的中央控制室。這種轉(zhuǎn)變提供了更大的挑戰(zhàn)和機(jī)遇，過程控制的研究人員和工程師將為為這個(gè)蓬勃發(fā)展的行業(yè)訂立新標(biāo)準(zhǔn)。原文： Semiconductor manufacturing process control and monitoring: A fab-wide frameworkAbstractThe semiconductor indust

21、ry has started the technology transition from 200 mm to 300 mm wafers toimprove manufacturing efficiency and reduce manufacturing cost. These technological changes present aunique opportunity to optimally design the process control systems for the next generation fabs. In thispaper we first propose

22、a hierarchical fab-wide control framework with the integration of 300 mmequipment and metrology tools and highly automated material handling system. Relevant existingrun-to-run technology is reviewed and analyzed in the fab-wide control context. Process and metrologydata monitoring are discussed wit

23、h an example. Missing components are pointed out as opportunities forfuture research and development. Concluding remarks are given at the end of the paper.2005 Published by Elsevier Ltd.Keywords: Semiconductor manufacturing; Fab-wide control; Electrical parameter control; Run-to-runcontrol; Fault de

24、tection and classification;Metrology data monitoring1. IntroductionThe semiconductor industry has started the technology transition from 200 mm to 300 mm wafers toimprove manufacturing efficiency and reduce manufacturing cost. Along with this transition is thedoubling of capital expenditure in a 300

25、 mm fab versus a 200 mm fab. (The cost of a 200 mm fab is over$1 billion while the cost of a 300 mm fab is over $2 billion.) Other technological changes include: Single wafer processing capability instead of lot-to-lot operations; Fully automated material handling systems (AMHS) with inter-bay and i

26、ntra-bay transportation; Integrated metrology that allows for timely control; Highly automated process control and fault diagnosis.Owing to the capital intensity of the new generation fabs, it is critical to maintain highly efficientoperations,minimize downtime of equipment, and optimize the yield o

27、f high quality products. TheInternational Technology Roadmap 42 clearly identifies that factory information and control systemsare a critical enabling technology to reduce cycle-time and improve yield.These technological changespresent a unique opportunity to optimally design the process control sys

28、tems for the new generation fabs.A persistent challenge in semiconductor manufacturing control is the lack of critical in situ sensors toprovide real time information of the wafer status for feedback control and optimization. Fortunately,recent advance in metrology technology 44 provides an opportun

29、ity for improving the timeliness andusefulness of the measurement data.for analysis and control:Typically a modern fab has the following measurement data available(1) Real time data at the tool level which reflect the equipment health condition and provide feedback for realtime control;(2) Integrate

30、d metrology and in-line metrology data available for geometric dimensions after a major processingstep, with small to moderate metrology delay;(3) Sample and final electrical test (E-test) data available for electrical properties with medium or long timedelay, but they have the most important inform

31、ation about the manufacturing effectiveness.Advanced control and optimization methodology should maximize the use of all the information in an integratedhierarchy for highly efficient manufacturing and tight product quality control. Monitoring and control of semiconductor manufacturing processes hav

32、e been investigated at a number of USuniversities and industrial research laboratories. Representative work includes U.C. Berkeley 28,32 onstatistical modeling and control of plasma etchers, Michigan on real-time and run to run multivariablecontrol22, as well as MIT on different sensor and control t

33、echnologies 7,8. Due to lack of in situ sensorsmuch of the control work is developed from the run-to-run (R2R) control strategy 12,41. Research groups atUniversity of Maryland contributed in the area of run to run control 2,5,53. SEMATECH, a consortium ofleading semiconductor manufacturers, posted s

34、everal benchmark problems on plasma equipment fault detectionand diagnosis 4. Adaptive and nonlinear control for R2R operations is proposed by 16. Model predictivecontrol is applied to R2R control as well which has additional capability in handling constraints explicitly 19.At UT-Austin we have deve

35、loped (i) stability conditions and tuning guidelines for multivariable EWMA anddouble EWMA control with metrology delays 20,21, (ii) multivariate statistical monitoring of RTA and etchers52,51, and (iii) multivariate statistical control of CD metrology data from lithography 14. Other newdevelopment

36、and applications of control and fault detection are reported at recent SPIE conferences andAEC/APC Symposia organized by SEMATECH and summarized in Del Castillo and Hurwitz 15 and Moyne etal. 34. Manufacturing companies like AMD, Intel, Motorola, and TI and vendors like Applied Materials,Brooks-PRI

37、Automation, and Yield Dynamics are leaders in deploying APC technologies at the manufacturinglines.In this paper we draw the analogy between semiconductor manufacturing fabs and chemical plants and propose ahierarchical optimization and control system for semiconductor fab control. A schematic diagr

38、am is shown inFig. 1 for this analogy, which was first presented by Qin and Sonderman 38. The equipment level controlinvolves automatic feedback control of tool parameters and small scale run-to-run control using integratedmetrology. The next level run-to-run control involves the use of in-line meas

39、urement for feedforward andfeedback control. The third level is the islands of control. The top level of the hierarchy is the fab-wide controlwhich is the highest level optimization to achieve desired electrical properties by recalculating the optimalgeometric targets and dosage for the lower level.

40、The organization of the paper is given as follows. We first propose a hierarchical fab-wide control strategy withthe integration of 300 mm equipment and metrology tools and highly automated material handling system.Relevant run-to-run technology is reviewed and analyzed in the fab-wide control conte

41、xt, process and metrologydata monitoring are discussed with an example, and missing components are pointed out as opportunities forfuture research and development. Concluding remarks are given at the end of the paper.2. A framework for fab-wide controlAlmost all existing development is on R2R contro

42、l which adjusts recipes of a step based on metrology data at theequipment level. These are known as islands of control as illustrated in the lower part of Fig. 1. None of the existing control strategies examine the coordination of multiple manufacturing steps to improve the overallproduct quality in

43、 terms of electrical parameters. The R2R controllers compensate for equipment drifts throughmetrology feedback, but they cannot compensate for metrology drifts and uncertainties. The direct control ofelectrical parameters proposed here can compensate for metrology drifts and systematic errors in the

44、 geometricmeasurements that are below the metrology SPC limits. It is believed that the control and optimization ofelectrical parameters represent the next generation of semiconductor manufacturing control system as it directlycontrols the electrical properties to a desired product profile by manipu

45、lating the operation requirements forlower level R2R controllers. The electrical parametric control and optimization will maximize the yield ofhigh-grade products or reduce operational cost when a demand profile is specified by market orders.The fab-wide control framework in Fig. 1 provides optimiza

46、tion and coordination from step to step to reducevariability, reworks, and scraps, thus improving the overall equipment effectiveness and reducing manufacturingcost. This framework was first presented by Qin and Sonderman 38 after having deployed many R2Rcontrollers at AMD and analyzed the need for

47、a higher level control. The equipment level control involvesautomatic feedback control of tool parameters. The next level is run-to-run control using integrated or in-linemetrology to achieve a specified target. The third level is the islands of control that shares information frommultiple steps to

48、perform feedforward and feedback control and tool performance matching. The top level of thehierarchy is electrical parametric control (EPC) or fab-wide control to achieve desired electrical properties byrecalculating the optimal targets for the lower levels. Equipment drifts, metrology drifts, and

49、material variationsare compensated by feedback at the EPC level, leading to improved process and metrology availability andreduced use of calibration and test wafers.This multiple level control framework resembles the hierarchical control framework that has been successful inthe refinery industry 39

50、, but significant differences exist: (i) the lowest level control is mostly batch operations;(ii) the middle level R2R control has virtually no R2R process dynamics except for disturbance dynamics; and(iii) the top level EPC is a multi-step operation control that aims to compensate for errors made i

51、n prior steps,regardless of the nature of the errors as long as step-wise metrology measurement is available. This makes itdifferent from model predictive control (MPC) of batch processes with shrinking horizons. In chemical andrefinery process, the top-level optimization is real-time optimization 3

52、1 and the middle-level is the full-scaledynamic MPC.As the MPC framework is a powerful and successful technology, it has been extended to scheduling andproduction planning in the semiconductor industry 11,46,47. The fab-wide control framework proposed herealso draws analogy from the MPC framework,bu

53、t the focus is optimized control of electrical parameters of thedevices.The E-test data are used to update the device model parameters based on mismatch between the E-test data andthe model. After parameter estimation is performed, the estimated parameters are sent to a fab-wideoptimizer,which distr

54、ibutes targets to lower-level controllers that regulate steps within the manufacturingprocess.The model updated with the new set of model parameters is used for EPC control.3. Run to run control algorithmsIn recent years, run-to-run (R2R) control technology has received tremendous interest in semico

55、nductormanufacturing. Moyne and Hurwitz (Moyne et al. 34) define the run-to-run control as a form of discreteprocess and machine control in which the product recipe with respect to a particular process is modified ex situ,i.e., between machine runs, so as to minimize process drift, shift, and variab

56、ility. In order to modify the recipeto address the process drift, shift and other variability, the current tool and wafer states need to be estimated. Oneclass of widely used run-to-run controllers is based on the exponentially weighted moving average (EWMA)statistics to estimate process disturbance

57、s. The EWMA has been used for a long time for quality monitoring purposes 9. Its use as a basis for run-to-runcontrol is relatively recent 40. For a time series of easurement xn,xn-1, . . ., where n denotes the runnumber, the EWMA is given in the following recursive formula:Xn=wxn-1+(1-w)xnOne of th

58、e most effective manipulated variables in R2R control is the processing time within a processing stepsuch as etch time, exposure time, and planarization time. The controlled variables in this case are typically theextent to which the process develops under the processing time, such as depth of etch

59、and critical dimensions.This multiplicative model does not fit into the typical linear state space model presented earlier, but it can beconverted to the linear state space model with process and measurement noise by simply taking the logarithm.Therefore, all the control algorithms presented earlier

60、 in this paper are applicable to time control.4. Fault detection and diagnosisProcessing tool data such as temperatures, pressures, and gas flow rates will be used to monitor recipes appliedto single wafers or batches of wafers. Some typical processing operations include plasma etching, thin filmdep