201213050 六、發明說明： 【發明所屬之技術領域】 本揭示案係關於在基板之化學機械研磨期間之光學監 控。 干典 【先前技術】 積體電路通常藉由將導電層、半導體層或絕緣層相繼 沈積於矽晶圓上而形成於基板上。—個製造步驟涉及將 填料層沈積於非平面的表面上及使填制平坦化。對於 某些應用而言，將填料層平坦化直至曝露圖案化層之頂 表面為止。舉例而言’可將導電填料層沈積於經圖案化 之絕緣層上以填充絕緣層中之溝槽或孔。纟平坦化之 後’殘留在絕緣層之凸起圖案之間的導電層部分形成通 孔、插座及接線，該等通孔、插座及接線可在基板上的 薄膜電路之間提供導電路徑。對於其他應用（諸如氧化 物研磨）而言，將填料層平坦化直至在非平面的表面上 留下肢厚度為止。此外，光微影通要將基板表面 平坦化。 化予機械研磨（Chemical mechanicai卩心也叩CMp) 是-：公認的平坦化方法。此種平坦化方法通常需要將 板安裳在承載頭上。通常將基板之曝露表面抵靠旋轉 研磨墊置放。承載頭在基板上提供可控負載以推動基板 L靠研磨墊。通常將研磨液（諸如含有磨輯之漿）供應 至研磨墊之表面。 201213050 CMP中所存在的一個問題是決定研磨製程是否完 成，亦即，決定是否已將基板層平坦化至所要平度或厚 度’或決定何時已移除所需要量之材料。基板層之初始 厚度、漿組合物、研磨墊狀況、研磨墊與基板之間的相 對速度及基板上之負載的變化可5丨起材料移除速率之變 化。此等變化引起到達研磨終點所需時間之變化。因此， 不可能僅僅將研磨終點決定為研磨時間之函數。 在一些系統中，在研磨期間例如經由研磨墊中之視窗 來原位光學監控基板。然而，現有光學監控技術無法滿 足半導體裝置製造商曰益增加之需求。 【發明内容】 在一些研磨製程中，將覆蓋層’例如，阻障層，例如， 介電材料（例如，氮化物（例如，氮化组或氣化欽））自 基板移除以使下層或層結構曝露，該下層或層結構包括 不同的材料，例如，不同的介電材料（例如，低介電常 數的材料及/或低介電常數的蓋材料）。通常期望將該下 層或層結構移除直至殘留目標厚度或已移除目標量之材 料為止。—些監控技術使函數（例如，線）與匹配參考 光镨之索引值擬合，以便決定研磨終點或改變研磨速 率’因為該下層或層結構之初始厚度並不熟知且該技術 未能解決自該覆蓋層轉移至該下層或層結構，所以該等 監控技術可能在此研磨製程中存在問題。然而，若將該 函數與資料擬合，該資料係在偵測將 仏谓叫將邊覆蓋層清除及使 201213050 該下層或層結構曝露之後所累積，則可避免此等問題。 在-個態樣中’—種控制研磨之方法包括以下步驟： 儲存=有複數個參考光狀資料庫，該複數個參考光譜 之每。考光譜具有經儲存的關聯索引值；研磨基板， 該基板具有第二層’該第二層覆蓋第一層；在研磨期間 量測來自基板之光的光譜之序列；針對光譜之該序列之 每-量測光譜’找到最佳匹配參考光譜以產生最佳匹配 參考光谱之序列；自最佳匹配參考光譜之序列決定每一 最佳匹配光譜之關聯索引值以產生索引值之序列；偵測 該第一層之曝露；使函數與索引值之序列之一部分擬 合，該等索引值之該序列對應於在偵測該第一層之曝露 之後莖測的光譜；以及基於該函數決定研磨終點或對研 磨速率之調整中的至少_者。 實施可包括一或更多以下特徵結構。偵測該第一層之 曝露之步驟可包括以下步驟：在研磨期間量測來自基板 之光的光譜群組之序列；針對每一群組，計算該群組中 之光譜之色散參數值以產生色散值之序列；以及基於色 散值之該序列偵測該第一層之曝露。計算該色散參數值 之步驟可包括以下步驟··計算該群組中每一光譜之差值 以產生複數個差值。計算該色散參數值之步驟可包括以 下步驟：計算該複數個差值之標準偏差。光譜群組之序 列可包括光S普之序列。偵測該第一層之曝露之步驟可包 括以下步驟：自基板、馬達扭矩或基板與研磨墊之間的 摩擦來監控總反射強度。量測來自基板之光的光譜序列 201213050 之步驟可包括以下步驟：使感測器越過基板進行複數次 掃描。來自光譜之序列之每一光譜可對應於該感測器之 單次掃描’該單次掃描來自該複數次掃描。使感測器越 過基板進行複數次掃描之步驟可包括以下步驟 — 旋轉，該平臺具有固定至該平臺的感測器。該第二層可 為阻障層《該第一層可為介電層’該介電層具有與該阻 障層不同之組合物。該阻障層可為氮化钽或氮化鈦，且 »亥介電層可為摻碳二氧化矽或該介電層由四乙氧基矽烷 形成。當線性函數與目標索引相匹配或線性函數超過目 私索引時，可停止研磨。可決定目標索引差；可決定該 第一層之曝露處的索引值；以及可藉由使該目標索引差 加上該第一層之曝露處的索引值來計算該目標索引。可 自使用者接收待移除之目標量’且決定該目標索引差之 步驟：包括以下步驟：自待移除之目標量及預定研磨速 率計算該目標索引差。該目標索引可為在基板研磨之前 所儲存的值。該基板可包括複數個區域，且每—區域之 研磨速率可藉由獨立可變的研磨參數而獨立可控。可儲 存每-區域之目標索引值。在研磨期間可自每一區域量 測光譜之序列。針對每— 區域之光；’曰’之序列中每一量測 级譜，可決定來自參考光譜之資料庫的最佳匹配參考光 "曰。針對每一區域之每—最 取佳匹配 > 考光譜，可決定索 以產生索引值之序列。 — 町了母 ^域’線性函數可 …之—部分擬合，該等㈣值之該序列對 ’ 測該第-層之曝露之後量測的光譜。針對至少 201213050 一個區域，可基於該線性函數決定預仕. 谓怙砰間，在該預估 時間處’該區域將達到至少一個區域之目標索引值可 調整至少一個區域之研磨參數以對至少一個基板之至少 一個區域的研磨速率進行調整，以使得該至少一個區域 在該預估時間處比不進行此調整之情況更接近於目標索 引。該研磨參數可為承載頭中之壓力。 在另一態樣中，有形地實施在機器可讀取儲存裝置中 之電腦程式產品包括執行方法的指令。 在另一態樣t，一種研磨設備包括：支撐件，該支撐 ㈣；承載頭’該承載頭用以固持基板抵 靠該研磨墊；馬達’該馬達用以在該承載頭與該支樓件 之間產生相對運動以研磨基板；光學監控系統，該光學 監控系統用以在研磨基板時，量測來自基板之光的光譜 之序列；以及控制器。該控制器經配置以：儲存具有複 數個參考光譜之資料庫，該複數個參考光譜之每一參考 光谱具有經儲存的關聯索引值；針對光譜之序列之每一 里測光谱，找到最佳匹配參考光譜以產生最佳匹配之參 考光譜之序列；自最佳匹配參考光譜之序列決定每一最 佳匹配光譜之關聯索引值以產生索引值之序列；自光譜 之序列偵測該第一層之曝露；使函數與索引值之序列之 #刀擬σ ’該等索引值之該序列對應於在偵測該第一 層之曝路之後里測的光譜；以及基於該函數決定研磨終 點或對研磨速率之調整中的至少一者。 實施可視需要包括一或更多以下優點。基於殘留厚度 8 201213050 或丄移除之里中的任一者可決定終點。終點偵測系統可 對基板之間的厚度變化不太敏感因此可改良用於偵測 所要研磨終點之終點系統的可靠度，且可減少逐個晶圓 的厚度不均勻性（wafer-to-wafer thickness non-uniformity; WTWNU)。 在以下隨附圖式及描述中闡述了 一或更多實施例之細 節。其他特徵結構、態樣及優點將由描述、圖式及申請 專利範圍當可更加明白。 【實施方式】 一種光學監控技術為：在研磨期間量測自基板反射之 光之光譜，及識別來自資料庫之匹配參考光譜。該等匹 配參考光譜提供一系列索引值及一函數（例如，線），使 該函數與該系列索引值擬合。該函數對目標值之投影可 用以決定終點或改變研磨速率。 對於研磨一些類型之基板（例如，在相同平臺處對多 個層之材料進行研磨的基板）而言，存在的一個潛在問 題在於：此方法未能解決自覆蓋層轉移至下層或層結 構，從而減少了函數與資料擬合之可靠度。此外，自基 板對基板或在一基板内部可能存在下層厚度之差異，從 而導致自表面上具有相同外層厚度之基板反射的光譜之 變化’且增加了適當的終點決定之難度。 舉例而言，參閱第1A圖，基板10可包括第一介電材 料之經圖案化之第一層1 2 (此層亦可稱為下層），該第 201213050 —介電材料例如為低介電常數材料，例如，摻碳的二氧 化矽（例如，Black DiamondTM (麟自AppUed仏如心， .)或 Coral (購自 Novellus Systems, Inc.))。安置 在第層12上的為不同的第二介電材料之第二層i叹此 層亦可稱為覆蓋層），例如，阻障層（例如，氮化物（例 如’氮化鈕或氮化鈦))。視需要安置在該第一層與該第 層之間的為另—介電材料之一或更多額外層14,該另 一介電材料（例如，低介電常數的蓋材料（例如，由四 乙氧基矽烷㈣㈣师〇rth〇silicate; TE〇s)形成之材 料））不同於該第一介電材料及第二介電材料兩者。第一 層12及一或更多額外層14共同提供在第二層下方之層 堆疊。安置在第二層上（且在溝槽中藉由第一層之圖案 所提供）的為導電材料18，例如，金屬（例如，銅）。 化予機械研磨可用以將基板平坦化直至曝露第一介電材 料之第層為止。舉例而言，參閱第1B圆，在平坦化 之後’殘留在第-I 12之凸起圖案之間的導電材料Η ^ 刀形成通孔及其類似物°此外’有時期望移除第一 介電材料直至殘留目標厚度或已移除目標量之 止。 ’ 一種研磨方法為··在第-研磨墊上研磨導電材料至少 直至曝露第二層（例如’阻障層）為止。此外，例如， 在過度研磨步驟期間，在第—研磨墊上可移除第二層之 厚度的-部分。隨後將基板傳送至第二研磨墊，在第二 研磨墊處已&全移除第二層（例如，阻障層），且亦移除 10 201213050 下層之第-層之厚度的一部分（例如，低介電常數的介 電質）此外，在相同的研磨操作中，在第二研磨墊處可 移除第-層與第二層之間的該或該等額外層（若存在） (例如，蓋層）。 然而，當將基板傳送至第二研磨墊時，第二層之初始 厚度可能是未知的。如上所述，對於光學終點偵測技術 而言，此舉可產生的問題為：識別來自資料庫之匹配參 考光譜。此外，使函數（例如，線）與一系列索引值擬 合之終點偵測技術未能解決自第二層轉移至第一層。然 而，若另一監控技術用以偵測覆蓋層之清除及下層或層 結構之曝露，且使函數與來自偵測之後所收集的光譜之 索引值擬合，則可減少此等問題。 針對一些層堆疊（諸如，覆蓋TE〇S層之阻障層），偵 測覆蓋層之清除及下層之曝露可能十分困難。然而，當 清除了覆蓋層（例如，阻障層）且曝露了下層（例如， 低介電常數層或蓋層）時，來自不同位置的光譜傾向於 發散。可分析光譜之發散且可計算偏差參數值。藉由在 偏差參數值改變時進行偵測，可偵測覆蓋層之清除。 第2圖圖示研磨設備100之實例。研磨設備1〇〇包括 可旋轉的圓盤形平臺120’研磨墊11〇位於平臺12〇上。 該平臺可操作以使該平臺圍繞軸125旋轉。舉例而言， 馬達121可轉動驅動軸124以使平臺12〇旋轉。研磨墊 11〇可為雙層研磨墊，該雙層研磨墊具有外部研磨層112 及較軟的背層114。 201213050 研磨設備100可包括璋130，埠130將研磨液體132 (諸如，漿）分配至研磨墊丨丨〇上。該研磨設備亦可包 括研磨塾調節器’該研磨墊調節器磨損研磨墊丨丨〇以將 研磨墊110維持在一致的磨損狀態。 研磨6又備100包括一或更多承載頭140。每一承載頭 14〇可操作以固持基板1〇抵靠研磨墊11()。每一承載頭 140可對與每一各別基板相關聯的研磨參數（例如，壓 力）進行獨立控制。 特定言之，每一承載頭140可包括固定環142，固定 環142將基板1〇固定在撓性膜144下方。每一承載頭 140亦包括由該膜界定的複數個獨立可控的可加壓腔室 (例如，3個腔室146a-146c )，該等腔室可將獨立可控 的壓力施加至撓性膜144上的關聯區域148a-148c，且 因此將壓力施加在基板1〇上（參見第3圖）。參閱第2 圖，中心區域148a大體上可為環形，且剩餘區域 148b-148e可為圍繞中心區域148a之同心環形區域。儘 管為便於圖示，僅三個腔室圖示於第2圖及第3圖中， 但可能存在一個或兩個腔室，或四個或四個以上腔室（例 如’五個腔室）。 返回到第2圖，每一承載頭14〇自支撐結構15〇 (例 如，回轉料架）懸垂，且每一承載頭14〇由驅動軸152 連接至承載頭旋轉馬達154’以便該承載頭可圍繞軸155 旋轉。每一承載頭140可視需要例如在回轉料架15〇上 之滑動器上側向擺動；或藉由回轉料架自身之旋轉擺動 £ 12 201213050 而進订側向擺動。在操作中，平臺圍繞平臺之中心轴i25 旋轉’且每—承載頭圍繞承載頭之中心軸155旋轉，並 將每:承載頭側向平移越過研磨塾之頂表面。 儘g僅圖不了 一個承載頭14〇，但可提供更多承載頭 以固持額外的基板，以便可有效地使用研耗no之表 面積。 因 適。於固持基板以用於同時研磨製程的承載頭 組件之數量可至少冑分地基於研磨# 110之表面積。 研磨設備亦包括原位光學監控系、统160(例如，光譜 監控系統），原位光學監控系統160可用以決定是否調整 研磨速率或決定對研磨速率進行調整（如下論述）^藉由 包括孔隙（亦即，穿過墊之孔）或立體視窗i丨8來提供 經由研磨墊的光學存取。立體視窗118可例如作為填充 研磨墊中之孔隙的插塞緊固至研磨墊110，例如，模製 或黏附性地緊固至研磨墊，然而在一些實施中，立體視 固可支撐於平臺120上且凸起至研磨塾中的孔隙中。 光學監控系統1 6 0可包括光源1 6 2、光偵測器1 6 4及 電路166’電路166用於發送及接收遠端控制器i9〇(例 如’電腦）與光源162及光偵測器164之間的信號《 — 或更多光纖可用以將光自光源162傳輸至研磨墊中之光 學存取’且一或更多光纖可用以將自基板1〇反射之光傳 輸至偵測器164。舉例而言，分又光纖1 70可用以將光 自光源162傳輸至基板10且自基板1 〇傳輸回至偵測器 164。分叉光纖（英文多個an)可包括幹線172及兩條支線 £ 13 201213050 174及176’幹線172定相认扯 疋位於接近該光學存取，該兩條支 線1 74及1 76分別連接至本 咬按至先源1 62及偵測器1 64。 在一些實施申，平喜夕 室 < 頂表面可包括凹槽128，光學 磁頭168裝配至凹槽 價128中，光學磁頭168固持分叉光 纖之幹線1 72的一個支唑土飽 幻禾化。先學磁頭168可包括一機構， 該機構用以調整幹線】7 9 — r 升深172之頂部與立體視窗118之間的 垂直距離。 電路166之輸出可為數 _ ^ 双证冤子#遽，該數位電子信號 經由驅動軸124中夕炖絲*人 之旋轉耦合器129 (例如，滑環）傳 遞至光學監控系統之控制哭1〇n # / l 投fi益1 90。類似地’回應於自控 制器190經由旋轉鉍人$ ,。Λ 士 轉耦&态129傳遞至光學監控系統16〇 的數位電子信號中的控制命令，可開啟或關閉光源。或 者’電路W可藉由無線信號與控#iJ|§19〇進行通訊。 光源1 6 2可操作以益^u 發射白光。在一個實施中，所發射201213050 VI. Description of the Invention: [Technical Field of the Invention] The present disclosure relates to optical monitoring during chemical mechanical polishing of a substrate. BACKGROUND OF THE INVENTION [Integrated] Integrated circuits are usually formed on a substrate by successively depositing a conductive layer, a semiconductor layer or an insulating layer on a germanium wafer. A manufacturing step involves depositing a layer of filler on a non-planar surface and planarizing the fill. For some applications, the filler layer is planarized until the top surface of the patterned layer is exposed. For example, a layer of conductive filler can be deposited over the patterned insulating layer to fill the trenches or holes in the insulating layer. After the planarization, the portions of the conductive layer remaining between the raised patterns of the insulating layer form vias, sockets, and wirings that provide a conductive path between the thin film circuits on the substrate. For other applications, such as oxide milling, the filler layer is planarized until the limb thickness is left on the non-planar surface. In addition, the photolithography is to flatten the surface of the substrate. Chemical mechanical polishing (Chemical mechanicai 叩 叩 CMp) Yes -: Recognized planarization method. This method of planarization typically requires the board to be placed on the carrier head. The exposed surface of the substrate is typically placed against a rotating polishing pad. The carrier head provides a controllable load on the substrate to push the substrate L against the polishing pad. A slurry, such as a slurry containing a grind, is typically supplied to the surface of the polishing pad. One problem with the 201213050 CMP is to determine if the polishing process is complete, that is, to determine whether the substrate layer has been flattened to the desired level or thickness or to determine when the required amount of material has been removed. The initial thickness of the substrate layer, the slurry composition, the condition of the polishing pad, the relative velocity between the polishing pad and the substrate, and the change in load on the substrate can cause variations in the material removal rate. These changes cause a change in the time required to reach the end of the grinding. Therefore, it is not possible to determine only the grinding end point as a function of the grinding time. In some systems, the substrate is optically monitored in situ during polishing, such as via a window in the polishing pad. However, existing optical monitoring technologies are unable to meet the increased demand from semiconductor device manufacturers. SUMMARY OF THE INVENTION In some polishing processes, a cap layer, for example, a barrier layer, for example, a dielectric material (eg, a nitride (eg, nitride group or gasification)) is removed from the substrate to enable the lower layer or The layer structure is exposed, the underlying layer structure comprising different materials, for example, different dielectric materials (eg, low dielectric constant materials and/or low dielectric constant cap materials). It is generally desirable to remove the underlying layer structure or layer structure until the target thickness is retained or the target amount of material has been removed. Some monitoring techniques fit the function (eg, line) to the index value of the matching reference pupil to determine the polishing endpoint or change the polishing rate 'because the initial thickness of the underlying layer structure is not well known and the technique fails to resolve The cover layer is transferred to the underlying layer or layer structure, so such monitoring techniques may present problems in the polishing process. However, if the function is fitted to the data, the data can be avoided by detecting that the edge layer is removed and the 201213050 layer or layer structure is exposed. In one aspect, the method of controlling the grinding comprises the following steps: storing = having a plurality of reference optical databases, each of the plurality of reference spectra. The test spectrum has a stored correlation index value; the substrate is polished, the substrate has a second layer 'the second layer covers the first layer; the sequence of the spectrum of light from the substrate is measured during the grinding; and the sequence for the spectrum is - measuring the spectrum 'finding the best matching reference spectrum to produce a sequence of best matching reference spectra; determining the associated index value of each best matching spectrum from the sequence of best matching reference spectra to produce a sequence of index values; detecting the sequence Exposure of the first layer; fitting the function to a portion of the sequence of index values corresponding to the spectrum of the stem after detecting the exposure of the first layer; and determining the endpoint of the polishing based on the function or At least the adjustment of the polishing rate. Implementations may include one or more of the following features. The step of detecting the exposure of the first layer may include the steps of: measuring a sequence of spectral groups of light from the substrate during the grinding; for each group, calculating a dispersion parameter value of the spectrum in the group to generate a sequence of dispersion values; and the sequence based on the dispersion value detects exposure of the first layer. The step of calculating the value of the dispersion parameter may comprise the step of: calculating a difference between each of the spectra in the group to produce a plurality of differences. The step of calculating the value of the dispersion parameter may include the step of calculating a standard deviation of the plurality of differences. The sequence of spectral groups can include sequences of light. The step of detecting the exposure of the first layer can include the steps of monitoring the total reflected intensity from the substrate, motor torque, or friction between the substrate and the polishing pad. The step of measuring the spectral sequence of light from the substrate 201213050 can include the step of having the sensor scan across the substrate for multiple scans. Each spectrum from the sequence of spectra may correspond to a single scan of the sensor' the single scan is from the plurality of scans. The step of scanning the sensor across the substrate for a plurality of scans may include the step of rotating, the platform having a sensor affixed to the platform. The second layer can be a barrier layer "The first layer can be a dielectric layer". The dielectric layer has a composition different from the barrier layer. The barrier layer can be tantalum nitride or titanium nitride, and the drain dielectric layer can be carbon doped germanium dioxide or the dielectric layer can be formed from tetraethoxy germane. Grinding can be stopped when the linear function matches the target index or the linear function exceeds the private index. The target index difference may be determined; an index value of the exposure of the first layer may be determined; and the target index may be calculated by adding the target index difference to the index value of the exposure of the first layer. The step of receiving the target amount to be removed from the user and determining the target index difference includes the steps of: calculating the target index difference from the target amount to be removed and the predetermined grinding rate. The target index can be a value stored prior to substrate polishing. The substrate can include a plurality of regions, and the polishing rate per region can be independently controllable by independently variable milling parameters. The target index value for each region can be stored. The sequence of spectra can be measured from each region during milling. For each spectrum of the region; the spectrum of each measurement in the sequence of '曰' determines the best matching reference light from the database of the reference spectrum "曰. For each region - the best match > test spectrum, you can determine the sequence to generate the index value. — The linear domain function of the town can be partially fitted, and the sequence of the (four) values is measured for the spectrum measured after exposure of the first layer. For at least 201213050, a region may be determined based on the linear function. At the estimated time, the region will reach the target index value of at least one region, and the grinding parameter of at least one region may be adjusted to at least one region. The polishing rate of at least one region of the substrate is adjusted such that the at least one region is closer to the target index at the estimated time than if the adjustment is not made. The grinding parameter can be the pressure in the carrier head. In another aspect, a computer program product tangibly embodied in a machine readable storage device includes instructions for performing the method. In another aspect, a grinding apparatus includes: a support member (four); a carrier head for holding the substrate against the polishing pad; and a motor for the carrier head and the branch member A relative motion is generated to grind the substrate; an optical monitoring system for measuring a sequence of spectra of light from the substrate when the substrate is being polished; and a controller. The controller is configured to: store a library having a plurality of reference spectra, each reference spectrum of the plurality of reference spectra having a stored correlation index value; finding a best match for each of the spectra of the sequence of spectra Reference spectrum to produce a sequence of best matching reference spectra; determining a correlation index value for each of the best matching spectra from a sequence of best matching reference spectra to produce a sequence of index values; detecting the first layer from a sequence of spectra Exposure; the sequence of the function and the index value of the sigma σ 'the sequence of the index values corresponding to the spectrum measured after detecting the exposure of the first layer; and determining the polishing end point or the grinding based on the function At least one of the adjustments in the rate. Implementing the visual need includes one or more of the following advantages. The end point can be determined based on any of the residual thickness 8 201213050 or 丄 removed. The endpoint detection system is less sensitive to thickness variations between the substrates, thereby improving the reliability of the endpoint system for detecting the desired end point of the polishing, and reducing wafer-to-wafer thickness unevenness (wafer-to-wafer thickness) Non-uniformity; WTWNU). The details of one or more embodiments are set forth in the <RTIgt; Other features, aspects, and advantages will be apparent from the description, drawings, and claims. [Embodiment] An optical monitoring technique is to measure the spectrum of light reflected from a substrate during grinding and to identify a matching reference spectrum from a database. The matching reference spectra provide a series of index values and a function (e. g., a line) that fits the function to the series of index values. The projection of this function to the target value can be used to determine the end point or change the grinding rate. One potential problem with grinding some types of substrates (eg, substrates that grind materials of multiple layers at the same platform) is that this method fails to address the transfer from the cover layer to the underlying layer or layer structure, thereby Reduced the reliability of function and data fitting. In addition, there may be differences in the thickness of the underlying layer from the substrate to the substrate or within a substrate, resulting in a change in the spectrum of the substrate reflection from the surface having the same outer thickness and increasing the difficulty of determining the appropriate endpoint. For example, referring to FIG. 1A, the substrate 10 may include a patterned first layer 12 of a first dielectric material (this layer may also be referred to as a lower layer), and the 201213050-dielectric material is, for example, a low dielectric. A constant material, for example, carbon doped cerium oxide (for example, Black DiamondTM (from AppUed), or Coral (available from Novellus Systems, Inc.). A second layer of different second dielectric materials disposed on the first layer 12 may also be referred to as a cover layer, for example, a barrier layer (eg, a nitride (eg, a nitride button or nitride) titanium)). One or more additional layers 14 of another dielectric material disposed between the first layer and the first layer, as desired, the other dielectric material (eg, a low dielectric constant cap material (eg, by Tetraethoxydecane (4) (4) 〇rth〇silicate; TE〇s) formed material)) is different from both the first dielectric material and the second dielectric material. The first layer 12 and one or more additional layers 14 together provide a layer stack below the second layer. Placed on the second layer (and provided in the trench by the pattern of the first layer) is a conductive material 18, such as a metal (e.g., copper). Mechanical polishing can be used to planarize the substrate until the first layer of the first dielectric material is exposed. For example, referring to the 1B circle, after the planarization, the conductive material remaining between the raised patterns of the -I 12 is formed into a via hole and the like. Further, it is sometimes desired to remove the first medium. Electrical material until the target thickness remains or the target amount has been removed. A method of polishing is to polish a conductive material on a first polishing pad at least until a second layer (e.g., a 'barrier layer) is exposed. Further, for example, during the overgrinding step, a portion of the thickness of the second layer may be removed on the first polishing pad. The substrate is then transferred to a second polishing pad where the second layer (eg, barrier layer) has been & completely removed, and a portion of the thickness of the first layer of the lower layer of 2012201250 is also removed (eg, a low dielectric constant dielectric) further, in the same polishing operation, the or the additional layer (if present) between the first layer and the second layer may be removed at the second polishing pad (eg , the cover layer). However, when the substrate is transferred to the second polishing pad, the initial thickness of the second layer may be unknown. As noted above, for optical endpoint detection techniques, the problem can be solved by identifying matching reference spectra from the database. In addition, endpoint detection techniques that combine functions (eg, lines) with a series of index values fail to resolve the transition from the second layer to the first layer. However, if another monitoring technique is used to detect the removal of the overlay and the exposure of the underlying or layer structure, and to fit the function to the index values from the spectra collected after detection, these problems can be reduced. For some layer stacks (such as barrier layers covering the TE〇S layer), it may be difficult to detect the removal of the overlay and the exposure of the underlying layer. However, when the cap layer (e.g., barrier layer) is removed and the underlying layer (e.g., low dielectric constant layer or cap layer) is exposed, the spectra from different locations tend to diverge. The divergence of the spectrum can be analyzed and the deviation parameter values can be calculated. The detection of the overlay layer can be detected by detecting when the value of the deviation parameter is changed. FIG. 2 illustrates an example of a grinding apparatus 100. The grinding apparatus 1A includes a rotatable disc-shaped platform 120' which is located on the platform 12A. The platform is operable to rotate the platform about the axis 125. For example, motor 121 can rotate drive shaft 124 to rotate platform 12A. The polishing pad 11 can be a two-layer polishing pad having an outer polishing layer 112 and a softer back layer 114. 201213050 The grinding apparatus 100 can include a crucible 130 that dispenses a grinding fluid 132, such as a slurry, onto the polishing pad. The grinding apparatus can also include a grinding tamper adjuster' that the abrasive pad adjuster abrades the abrasive pad to maintain the polishing pad 110 in a consistent wear condition. Grinding 6 is further comprised of one or more carrier heads 140. Each of the carrier heads 14 is operable to hold the substrate 1A against the polishing pad 11(). Each carrier head 140 can independently control the grinding parameters (e.g., pressure) associated with each respective substrate. In particular, each carrier head 140 can include a retaining ring 142 that secures the substrate 1〇 beneath the flexible membrane 144. Each carrier head 140 also includes a plurality of independently controllable pressurizable chambers (e.g., three chambers 146a-146c) defined by the membrane that apply independently controllable pressure to the flexible Associated regions 148a-148c on film 144, and thus pressure is applied to substrate 1 (see Figure 3). Referring to Figure 2, central region 148a can be generally annular and the remaining regions 148b-148e can be concentric annular regions surrounding central region 148a. Although only three chambers are illustrated in Figures 2 and 3 for ease of illustration, there may be one or two chambers, or four or more chambers (eg 'five chambers') . Returning to Fig. 2, each carrier head 14 is suspended from a support structure 15 (e.g., a rotary rack), and each carrier head 14 is coupled to a carrier head rotation motor 154' by a drive shaft 152 so that the carrier head can be Rotate around axis 155. Each of the carrier heads 140 can be laterally oscillated, for example, on a slider on the revolving rack 15 ;; or by lateral rotation of the revolving rack itself by swinging £ 12 201213050. In operation, the platform rotates about a central axis i25 of the platform and each of the carrier heads rotates about a central axis 155 of the carrier head and laterally translates each of the carrier heads over the top surface of the polishing pad. Only one carrier head 14 〇 can be shown, but more carrier heads can be provided to hold additional substrates so that the surface area of the no-strain can be effectively used. Because of fit. The number of carrier head assemblies for holding the substrate for the simultaneous polishing process can be based at least on the surface area of the grinding #110. The grinding apparatus also includes an in-situ optical monitoring system 160 (e.g., a spectral monitoring system) that can be used to determine whether to adjust the polishing rate or to determine the polishing rate (as discussed below) by including voids ( That is, through the aperture of the pad or the stereoscopic window i 8 to provide optical access via the polishing pad. The stereoscopic window 118 can be fastened to the polishing pad 110, for example, as a plug that fills the apertures in the polishing pad, for example, molded or adhesively secured to the polishing pad, although in some implementations, the stereoscopic viewing can be supported on the platform 120. Up and raised into the pores in the grinding crucible. The optical monitoring system 160 can include a light source 1 6 2, a light detector 1 6 4 and a circuit 166' circuit 166 for transmitting and receiving a remote controller i9 (eg, 'computer) and a light source 162 and a light detector The signal between 164 "- or more fibers can be used to transmit light from the source 162 to the optical access in the polishing pad" and one or more fibers can be used to transmit light reflected from the substrate 1 to the detector 164 . For example, the split fiber 170 can be used to transfer light from the light source 162 to the substrate 10 and back from the substrate 1 to the detector 164. The bifurcated optical fiber (multiple English an) may include a trunk line 172 and two branch lines £13 201213050 174 and 176' trunk line 172 phased identification is located close to the optical access, and the two branch lines 1 74 and 1 76 are respectively connected to The bite is pressed to the first source 1 62 and the detector 1 64. In some implementations, the flat surface < top surface may include a recess 128 in which the optical head 168 is mounted, and the optical head 168 holds a branch of the branched optical fiber 1 72. . The head 168 may include a mechanism for adjusting the vertical distance between the top of the trunk 172 and the solid window 118. The output of the circuit 166 can be a number of _ ^ 冤 冤 遽 , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , n # / l vote fi benefit 1 90. Similarly, the response to the self-controller 190 is via rotation of the monk $. The steer-to-couple & state 129 is passed to a control command in the digital electronic signal of the optical monitoring system 16A to turn the light source on or off. Or 'the circuit W can communicate with the control #iJ|§19〇 by means of a wireless signal. The light source 1 6 2 is operable to emit white light. In one implementation, launched
的白光包括具有200-800太本+ *士 m_aa I 不未之波長的光。合適的光源 為氣氣燈或氣求燈。 光偵測器164可為分光計。分光計為一光學儀器，該 光學儀器用於量測電磁波譜之一部分上的光強度。合適 的分光計為光栅分光計。分光計之典型輸出為隨著波長 (或頻率）之函數變化的光強度。 如上所述，光源162及光偵測器164可連接至計算裝 置（例如，控制器19m，〜τ -丄较壯^ π μ , j亞iyu)，该汁异裝置可操作以控制光源 162及光偵測g 164之操作及接收光源162及光偵測器 164之信號。該計算裝置可包括位於研磨設備附近的微 14 201213050 處理器（例如，可程式電腦）。就控制而言，該計算裝置 可例如使光源之啟動與平臺1 2〇之旋轉同步。 在一些實施中，原位監控系統丨60之光源1 62及偵測 器164安裝在平臺120中，且原位監控系統1 60之光源 1 62及偵測器丨64隨著平臺1 2〇旋轉。在此情況下’平 臺之運動將使感測器越過每一基板掃描。特定言之，當 平臺120進行旋轉時，控制器ι9〇可使光源ία發射一 系列閃光，該等閃光剛好在光學存取通過基板丨〇下方之 前開始’且剛好在光學存取通過基板1 〇下方之後結束。 或者’計算裝置可使光源1 62連續地發射光，該光剛好 在每一基板10通過光學存取上方之前開始，且剛好在每 一基板1 0通過光學存取上方之後結束。在任一情況下， 皆可在取樣週期内將來自偵測器之信號整合，從而以取 樣頻率產生光譜量測。 在操作中，控制器i 9〇可例如接收攜帶資訊的信號， s亥資訊描述由光偵測器接收的光之光譜，該光譜針對光 源之特定閃光或偵測器之時框。因此，此光譜為在研磨 期間經原位量測的光譜。 如第4圖中所示，若將偵測器安裝在平臺中，則歸因 於平臺之旋轉（由箭頭204所示），當視窗108在承載頭 下方行進時，以取樣頻率進行光譜量測之光學監控系統 將使光譜量測在沿橫過基板10之弧的位置2〇1處進行。 舉例而言，點201a-201k中之每一者代表由監控系統進 行的光譜量測之位置（點之數量為說明性的；取決於取 201213050 樣頻率’可進行比所圖示之量測更多或更少的量測）。取 樣頻率可經選擇以便在視窗108之每次掃描中收集介於 五個與二十個之間的光譜。舉例而言’取樣週期可介於 3毫秒與100毫秒之間。 如圖所示，經由平臺之一次旋轉，自基板上不同半 徑處獲得光譜。亦即，自較接近於基板1〇之中心的位置 處獲得些光譜’且自較接近於基板10之邊緣的位置處 獲知一些光譜。因此，對於光學監控系統越過基板進行 的任何給定的掃描而言，控制器190可基於時序、馬達 編碼器資矾及對基板之邊緣及/或固定環之光學偵測，來 根據掃描計算每一量測光譜所處的徑向位置（相對於正 、，掃描之基板之中心)。研磨系統亦可包括旋轉位置感測 器例如，凸緣，該凸緣附接於平臺之邊緣，該平臺將 穿過固疋的光學斷續器，以提供用於決定量測光譜之哪 基板及基板上位置之額外資料。因此控制器可使各種 里則光„曰與基板1〇a及基板1〇b上的可控區域1481148^ (>見第2圖）相關聯。在一些實施中，光譜量測之時 間可用作徑向位置之精確計算的替代物。 由平s之夕次旋轉，針對每一區域，可隨時間的推 移獲传光4之序列。在不限於任何特定理論的情況下， 自基板10反射之光的光譜因最外層厚度之變化而隨著 磨的進仃决變（例如，經由平臺之多次旋轉，而非在 越過，板之單次掃描期間），從而產生時變光譜序列。此 外’藉由層堆疊之牿金眉 16 yd- 201213050 在一些貫施中’控制器（例如，計算裝置）可經程式 化以將量測光譜與多個參考光譜相比較及決定哪一參考 光譜提供最佳匹配。特定言之，控制器可經程式化以將 來自於得自每一區域之量測光譜之序列的每一光譜與多 個參考光譜相比較，以產生每一區域之最佳匹配參考光 言普之序列。 如本文所使用，參考光譜為在研磨基板之前產生的預 定義光譜。假定實際研磨速率遵循預期研磨速率，則參 考光譜可與在研磨製程中代表時間的值具有預定義的關 聯（亦即，在研磨操作之前所定義的關聯），期望光譜在 該時間出現。替代或另外地，參考光譜可與基板性質（諸 如，最外層厚度）之值具有預定義的關聯。 可根據經驗產生參考光譜，例如，藉由自測試基板（例 如，具有已知初始層厚度之測試基板）量測光譜。舉例 而a，為產生複數個參考光譜，使用將在研磨裝置晶圓 期間使用之相同研磨參數來研磨裝設基板，同時收集光 譜之序列。針對每一光譜，記錄代表研磨製程中收集光 譜所用時間的值。舉例而言，該值可為經過時間或平臺 旋轉數。基板可經過度研磨（亦即，研磨超過所要厚度）， 以便可獲得當達成目標厚度時自基板反射之光的光譜。 為了使每一光譜與基板性質（例如，最外層厚度）之 值相關聯’可在測量站對具有與產物基板相同之^案的 P裝設」&板之初始光譜及性質進行預研磨#測。亦可 利用相同的測量站或不同的測量站對最終光错及性質進 17 201213050 订後研磨置測。初始光譜與 藉由内插法決定亢》曰之間的先4性質可 經過時門㈣ ’基於#測測試基板之光譜所用 •左過日f間的線性内插法。The white light includes light having a wavelength of 200-800 Taiben + *士 m_aa I. A suitable light source is a gas lamp or a gas lamp. The photodetector 164 can be a spectrometer. The spectrometer is an optical instrument for measuring the intensity of light on a portion of the electromagnetic spectrum. A suitable spectrometer is a grating spectrometer. The typical output of a spectrometer is the intensity of light as a function of wavelength (or frequency). As described above, the light source 162 and the photodetector 164 can be coupled to a computing device (e.g., controller 19m, ττ - 丄 ^ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ μ The light detecting g 164 operates and receives signals from the light source 162 and the photodetector 164. The computing device can include a micro 14 201213050 processor (eg, a programmable computer) located adjacent to the grinding device. In terms of control, the computing device can, for example, synchronize the activation of the light source with the rotation of the platform. In some implementations, the light source 1 62 and the detector 164 of the in-situ monitoring system 60 are mounted in the platform 120, and the light source 1 62 and the detector 丨 64 of the in-situ monitoring system 1 60 rotate with the platform 1 2 . In this case, the motion of the platform will cause the sensor to scan across each substrate. In particular, when the platform 120 is rotated, the controller ι9 〇 causes the light source ία to emit a series of flashes that begin just before the optical access passes under the substrate ' and just before the optical access through the substrate 1 〇 After the bottom, it ends. Alternatively, the computing device may cause the light source 1 62 to continuously emit light that begins just before each substrate 10 passes over the optical access and ends just after each substrate 10 passes over the optical access. In either case, the signals from the detector can be integrated during the sampling period to produce spectral measurements at the sampling frequency. In operation, the controller i can, for example, receive a signal carrying information that describes the spectrum of light received by the photodetector for a particular flash or detector time frame of the light source. Therefore, this spectrum is the spectrum measured in situ during the grinding. As shown in FIG. 4, if the detector is installed in the platform, due to the rotation of the platform (indicated by arrow 204), the spectral measurement is performed at the sampling frequency as the window 108 travels under the carrier head. The optical monitoring system will cause the spectral measurements to be taken at position 2 〇 1 across the arc of the substrate 10. For example, each of the points 201a-201k represents the location of the spectral measurements made by the monitoring system (the number of points is illustrative; depending on taking the 201213050 sample frequency' can be performed more than the measured measurement More or less measurement). The sampling frequency can be selected to collect a spectrum between five and twenty in each scan of window 108. For example, the sampling period can be between 3 milliseconds and 100 milliseconds. As shown, the spectra are obtained from different hemispheres on the substrate via one rotation of the platform. That is, some spectra are obtained from a position closer to the center of the substrate 1 且 and some spectra are obtained from a position closer to the edge of the substrate 10. Thus, for any given scan of the optical monitoring system across the substrate, the controller 190 can calculate each scan based on the timing, motor encoder resources, and optical detection of the edges and/or the fixed ring of the substrate. The radial position of the measured spectrum (relative to the positive, center of the scanned substrate). The grinding system can also include a rotational position sensor, such as a flange, attached to the edge of the platform that will pass through the fixed optical interrupter to provide a substrate for determining the measurement spectrum and Additional information on the location on the substrate. Therefore, the controller can associate various types of light with the substrate 1a and the controllable area 1481148 of the substrate 1b (> see Fig. 2). In some implementations, the time of the spectral measurement can be Used as an alternative to the accurate calculation of the radial position. The sequence of light 4 can be obtained over time for each region by the singular rotation of the flat s. Without limitation to any particular theory, from the substrate 10 The spectrum of the reflected light is determined by the change in the thickness of the outermost layer as the grinding progresses (eg, multiple rotations through the platform, rather than over a single scan of the plate), resulting in a time-varying spectral sequence. In addition, 'by layer stacking, the golden eyebrows 16 yd- 201213050 in some implementations' controllers (eg, computing devices) can be programmed to compare the measured spectrum to multiple reference spectra and determine which reference spectrum Providing the best match. In particular, the controller can be programmed to compare each spectrum from the sequence of measurements from each region to a plurality of reference spectra to produce the best for each region. Matching reference light As used herein, a reference spectrum is a predefined spectrum that is generated prior to grinding a substrate. Assuming that the actual polishing rate follows an expected polishing rate, the reference spectrum can have a predefined association with a value that represents time in the polishing process (ie, The desired spectrum appears at this time prior to the grinding operation. Alternatively or additionally, the reference spectrum may have a predefined association with the value of the substrate properties, such as the outermost layer thickness. The reference spectrum may be generated empirically, For example, by measuring the spectrum from a test substrate (eg, a test substrate having a known initial layer thickness), for example, a, to generate a plurality of reference spectra, using the same grinding parameters that would be used during the polishing of the wafer wafer The substrate is mounted while collecting the sequence of spectra. For each spectrum, a value representative of the time taken to collect the spectrum in the polishing process is recorded. For example, the value can be the elapsed time or the number of revolutions of the platform. The substrate can be subjected to degree grinding (ie, , grinding more than the required thickness), in order to obtain the target thickness when the substrate is reversed The spectrum of the light. In order to correlate the value of each spectrum with the properties of the substrate (for example, the thickness of the outermost layer), the initial spectrum of the P device with the same substrate as the product substrate can be used at the measurement station. The nature is pre-grinded #test. It is also possible to use the same measuring station or different measuring stations to determine the final optical error and properties. The initial spectrum and the first 4 properties between the 亢 曰 内 can be determined by interpolation (4) ‘based on the spectrum of the test substrate. • Linear interpolation between left and right f.
除根據經驗沐$ + M “ 疋之外’亦可根據理論計算-些或全邛 =可:二一之光學模型。舉例一 … 給定外層厚度D之參考光譜。例 9由假u均句研磨速率移除外層，可計算代表 磨_收集參考光谱所用時間之值。舉例而言，可藉 由假定起始厚度D〇及均勾研磨速率R，來簡單地計算 特定參考光譜之時間Ts(Ts=⑽_D)/R)。如另—實例，可 執行對應預研磨厚度D1及後研磨厚度m (或在測量站 所量測的其他厚度）之量測時間τι、τ2之間的線性内 插法（Ts=T2_T1*(Dl_D)/(D1_D2))，厚度⑴、叫係基於 用於光學模型的厚度D。 參閱第5圖及第6圖’可將量測光譜3〇〇 (參見第5 圖）與來自一或更多資料庫310(參見第6圖）之參考 光譜320相比較。如本文所使用，—參考光譜之資料庫 為一代表共同地共享性質之基板的參考光譜之集合。然 而，在單個資料庫十共同地共享之性質可在參考光譜之 多個資料庫之間變化。舉例而言，兩個不同資料庫可包 括代表具有兩個不同下層厚度之基板的參考光譜。對於 參考光譜之給定資料庫而言，並非其他因素（諸如，晶 圓圖案、下層厚度或層組合物之差異），而是上層厚度之 變化可為造成光譜強度之差異的主要原因。 201213050 如上所論述，可藉由研磨多個「裝設」基板及收集光 譜來產生不同資料庫310之參考光譜WO,該等「襄設 基板具有不同基板性質（例如’下層厚度或層組合物）: 來自一個裝設基板之光譜可提供[資料庫且來自具 有不同下層厚度之另—基板之光譜可提供第二資料庫。 =代或另外地’可根據理論計算不同資料庫之參考光 曰例如，可使用光學模型來計算第一資料庫之光譜， 該光學模型具有下層，該下層具有第—厚度，且可^用 “予模型來6十算第一貢料庫之光譜’該光學模型具有下 層’該下層具有-個不同的（英文多個咖）厚度。 、=—些實施中，對每—參考光譜320指派索引值33〇。 通吊，母一資料庫310可包括許多（例如，一或更多（例 如’精確-個））參考光譜32〇，該等參考光譜針對在基 板之預期研磨時間内之每-平臺旋轉。此索引330可為 代表研磨製程中期望觀察到參考光譜32〇所用時間之值 'S值）T對光4進行索引以便特定資料庫中的 —光譜具有唯一的索引值。可實施該索引以便按昭量 測測試基板之光譜的次序對該等索引值進行排序。索引 值可經選擇以隨著研磨之進行而單調地改變（例如，辦 加或減少)。特定言之’參考光譜之索引值可經選擇讀 该等索引值形成平臺旋轉之時間或次數之線性函數（假 定研磨速率遵循模型或測試基板之研磨速率，該模型哎 ^基板用以產生資料庫中之參考光譜）。舉例而言，索 值可與平臺旋轉數成比例（例如，相等），在該索引值 201213050 處量測測試基板之參考光譜或該參考光譜將在光學模型 中出現。因此’每一索引值可為整數。索引數可代表相 關聯光譜將出現之預期平臺旋轉。 參考光譜及參考光譜之相關聯索引值可儲存在參考資 料庫中。舉例而言，每—參考光譜32〇及參考光譜3 = 之相關聯索引值330可儲存在資料庫35〇之記錄 中。在研磨設備之計算褒置的記憶體中可實施參考光譜 之參考資料庫的資料庫350。 如上所述，對於母一基板之每—區域而言，控制器19〇 可基於量測光譜之序列或區域及基板，經程式化以產生 最佳匹配光譜之序列。可藉由將量測光譜與來自特定資 料庫之參考光譜相比較來決定最佳匹配參考光譜。 在一些實施中，可藉由為每—參考光譜計算J測光譜 ”參考光譜之間的差值平方和來衫最佳匹配參考光 谱。具有最小差值平方和的參考光譜具有最佳擬合。用 於找到最佳匹配參考光譜之其他技術（例如，最小絕對 差之和）係可能的。 一種可應用於減少電腦處理之方法為：限制於搜尋 匹配光譜之資料庫部分。資 μ k 料料常包括比在研磨基板 f獲仔的光譜更甯>出.祕 ' 光。在基板研磨期間，將資 料庫搜尋限於資料庫光譜 中，決定正經研磨U板之t 在—些實施例 土板之§刖方疋轉索引N。舉例而言， 。平1%轉中，可藉由搜尋所有資料庫之參 來決定N。針對在後續旋轉期間所獲得的光譜，在自： 20 201213050 度N之範圍内搜尋資料庫。亦即，若在一次旋轉期間發 現索引數為N，則在稍後X次旋轉之後續旋轉期間（其 中自由度為γ ) ’將搜尋到自（Ν+Χ) _γ至（Ν+χ)+γ之範 圍。 參閱第7圖，該圖圖示針對僅單個基板之單個區域的 結果’可決㈣序列中每-最佳匹配光譜之索引值以產 生索引值212之時變序列。索引值之此序列可被稱為索 引跡線21 〇。在一些實施中，藉由將每一量測光譜與來 自精破個資料庫之參考光譜相比較來產生索引跡線。 通吊“跡線210可包括基板下方的光學監控系統每 掃描一個（例如’精確一個）的索引值。 針對給定索引跡線210，索引跡線2U)中存在經量測 用於光予監控系統單次掃描中之特定區域之多個光譜 (稱為「當前光譜」)’可決定當前光譜中之每一者與一 或多（例如，精確一個）資料庫之參考光譜之間的最 佳匹配。在—此眚始由 ^ 一 中’將母一選定當前光譜與選定一 或更夕貝料庫之每-參考光譜相比較。例如，給定當前 =二…，以及參考光譜五、…，可為當前光 譜與參考光譜之以下組合t的每-者計算匹配係數、 與E、e與F、e與G、f與B、f與F、f與G、g與E、g 與F及g與g。盈綸 ，"、娜那—匹配係數指示最佳匹配（例如， 為最小），決定最佳pc 匹配參考光譜，且因此決定索引值。 或者，在一些眘始φ ，可將當前光譜組合（例如，取平 均值）且，將所馄έ人 Τ 、’、&光谱與參考光譜相比較以決定最 21 201213050 佳匹配’且因此決定索引值。 在-些實施中’針對—些基板之至少 ‘险揭^，可连 生複數個索引跡線。針對給定基板之 眭域，可為 關注之每一參考資料庫產生索引 家5丨跡線。亦即，針對於定 基板之給定區域所關注的每一參考資料庫，將量_ 之序列中的每一量測光譜與來自給定資料庫的參考光： 相比較’決定最佳匹配參考光譜之序列，且最佳匹心 考光譜之序列的索引值為給定資料庫提供索引跡線。 總之，每一索引跡線包括索引值212之序列21〇,其 中該序列之每一特定索弓（值212藉由自給定資料庫選擇 參考光譜之索引而產生，該、給^資料庫與量測光譜最緊 密擬合。索引跡線2 1 〇之每—索引之時間值可與量測光 譜的量測時間相同。 原位監控技術用以谓測第二層之清除及下層或層結構 之曝露。舉例而言，如下文更詳細地論述，可藉由馬達 扭矩或自基板反射之光之總強度的突然改變，或自所收 集光譜之色散，來偵測在時間TC處第一層之曝露。 如第8圖中所示，例如使用穩健的線擬合函數，例如， 已知階數之多項式函數（例如，一階函數（例如，線2〗4 )) 與在時間TC之後所收集的光譜之索引值之序列擬合。 田該函數與索引值之序列擬合時，忽略在時間TC之前 所收集的光譜之索引值。可使用其他錢（例如，二階 多項式函數），但是線函數提供計算之便。在終點時間 TE處可停止研磨’在該終點時間TE處線214越過目標 e *m·· 22 201213050 索引it。 第9圖圖示製造及研磨產物基板之方法的流程圖。產 物基板可具有至少與測試基板相同的層結構及相同的圖 案，該等測試基板用以產生資料庫之參考光辞。 首先，將第一層在基板上沈積且將第—層圖案化（步 驟902 )。如上所述，第一層可為介電質，例如，低介電 常數材料，例如，摻碳二氧化矽（例如，BUck Diam〇ndTM (購自 Applied Materials，Inc.)或 c〇raiTM(靖自 N〇veUus Systems, Inc.))。 取決於第一材料之組合物，另一介電材料，例如，低 介電常數的蓋材料（例如’四乙氧基石夕烧（TE〇s))之一 或更多額外層視需要在產物基板上之第一層上沈積（步 驟903 )，該另一介電材料不同於第一介電材料及第二介 電材料兩者。第一層及一或更多額外層共同提供層堆 疊。圖案化可視需要發生在一或更多額外層之沈積之後 (以便該一或更多額外層不會延伸至第一層中之溝槽 中，如第1A圖中所示）。 接著，不同的第二介電材料之第二層，例如，阻障層 (例如，氮化物（例如，氮化钽或氮化鈦））在產物基板 之第層或層堆疊上沈積（步驟904)。此外，導電層， 例如’金屬層（例如’銅）可在產物基板之第二層上（及 在由第一層之圖案提供的溝槽中）沈積（步驟906 )。第 一層之圖案化可視需要發生在第二層之沈積之後（在此 情況下’第二層不會延伸至第一層中之溝槽中 C： Va· 23 201213050 對產物基板進行研磨（㈣9〇8)。舉例而言，可研磨 導電層及第二層之一部分，且可在第一研磨站使用第一 研磨墊移除導電層及第二岸 曰之一邠分（步驟908a)。隨 後可研磨第二層及第一層之— 層之一部分，且可在第二研磨站 使用第二研磨墊移除第二層及第—層之一部分（步驟 嶋）。然而’應注意對於一些實施而言，（英文多個 the)不存在導電層，例如’當研磨開始時第二層為最外 層。當然’步驟9G2_9G6可在別處執行，以便針對於研 磨設備之特定操作者，製程開始於步驟9〇8。 原位監控技術用以偵測第二層之清除及第—層之曝露 (步驟910)。舉例而言，如以下更詳細地論述，可藉由 馬達扭矩或自基板反射之光之總強度的突然改變，或自 所收集光譜之色散，來偵測在時間Tc處第—層之曝露 (參見第8圖）。 至少開始於對第二層之清除的偵測（且可能更早，例 如’自_使用第二研磨㈣研磨產物基板），例如，使 用以上所述的原位監控系統在研磨期間獲得量測光譜之 序列（步驟912 )。 對量測光譜進行分析以產生索引值之序列，且使函數 與索引值之序列擬合。特定言之，針對量測光譜之序列 中的每一量測光譜，決定最佳擬合之參考光譜之索引值 以產生索引值之序列（步驟914)。使函數（例如，線性 函數）與在時間TC之後所收集的光譜之索引值之序列 擬合，在該時間tc處偵測第二層之清除（步驟916)。 ε：. 24 201213050 換言之，在時間tc夕& β ,, 之別所收集的光譜之索引值未用於 函數之计舁，在該時間 /才间1 L處偵測第二層之清除。 ’、I值（例如’根據與索引值之新序列擬合的線 性函數產生的計算帝卩丨 异系力值）達到目標索引，可停止研磨 (步驟918)。你用士 用t可在研磨操作之前設置且儲存目標 厚度IT。或者，伸用本 ^^ 便用者可§又置待移除之目標量，且可自 待移除之目標量計曾H炉 Τ α目標索弓丨IT。舉例而言，可自待移 除之目‘里（例如，自根據經驗決定的移除量與索引（例 如，研磨速率）In addition to the experience according to the experience of $ + M "other than 疋" can also be calculated according to the theory - some or all 邛 = can be: two optical model. Example one ... given the reference spectrum of the outer layer thickness D. Example 9 by the false u sentence The polishing rate is removed from the outer layer, and the value representing the time taken to collect the reference spectrum can be calculated. For example, the time Ts of the specific reference spectrum can be simply calculated by assuming the initial thickness D〇 and the uniform rubbing rate R ( Ts=(10)_D)/R). As another example, a linear interpolation between the measured time τι, τ2 corresponding to the pre-polished thickness D1 and the post-grinding thickness m (or other thickness measured at the measuring station) may be performed. The method (Ts=T2_T1*(Dl_D)/(D1_D2)), thickness (1), is based on the thickness D used for the optical model. See Figure 5 and Figure 6 for the measurement spectrum 3〇〇 (see section 5) Figure) is compared to a reference spectrum 320 from one or more databases 310 (see Figure 6). As used herein, the library of reference spectra is a collection of reference spectra representing substrates that collectively share properties. However, the nature of sharing in a single database can be more in the reference spectrum. Variations between databases. For example, two different repositories may include reference spectra representing substrates having two different underlying thicknesses. For a given library of reference spectra, there are no other factors (such as wafer patterns) , the thickness of the underlying layer or the difference in the layer composition), but the change in the thickness of the upper layer can be the main cause of the difference in spectral intensity. 201213050 As discussed above, the difference can be produced by grinding a plurality of "installation" substrates and collecting spectra. The reference spectrum WO of the database 310, the "substrate has different substrate properties (such as 'lower layer thickness or layer composition): the spectrum from a mounting substrate can provide [database and from another layer having different underlying thicknesses - The spectrum of the substrate provides a second database. = Alternatively or additionally 'can calculate the reference pupil of different databases according to theory. For example, an optical model can be used to calculate the spectrum of the first database, the optical model having a lower layer, the lower layer Has the first thickness, and can use the "pre-model to calculate the spectrum of the first tributary library of the six tenths. The optical model has the lower layer" The layer has a different (multiple coffee) thickness. In some implementations, an index value of 33 指派 is assigned to each reference spectrum 320. Bypassing, the parent-repository 310 can include a plurality (e.g., one or more (e.g., ' precise-)) reference spectra 32〇 that are rotated for each-platform during the expected polishing time of the substrate. This index 330 can be an index of the value of the time used to represent the reference spectrum 32 研磨 in the polishing process. The light 4 is indexed so that the spectrum in a particular library has a unique index value. The index can be implemented to order the index values in the order in which the spectra of the test substrates are measured. The index value can be selected to change monotonically (e.g., add or subtract) as the grinding progresses. In particular, the index value of the reference spectrum can be selected by reading the index values to form a linear function of the time or number of revolutions of the platform (assuming that the polishing rate follows the polishing rate of the model or test substrate, the model is used to generate a database) Reference spectrum in the middle). For example, the value can be proportional (e. g., equal) to the number of platform revolutions at which the reference spectrum of the test substrate is measured or which will appear in the optical model. Therefore 'each index value can be an integer. The number of indices can represent the expected platform rotation that will appear in the associated spectrum. The associated index values of the reference and reference spectra can be stored in a reference library. For example, the associated index value 330 for each of the reference spectrum 32〇 and the reference spectrum 3 = can be stored in the record of the database 35〇. A library 350 of reference libraries of reference spectra can be implemented in the memory of the computing device of the polishing apparatus. As described above, for each region of the mother-substrate, the controller 19 can be programmed to produce a sequence of best-matched spectra based on the sequence or region of the measurement spectrum and the substrate. The best matching reference spectrum can be determined by comparing the measured spectrum to a reference spectrum from a particular library. In some implementations, the reference spectrum can be optimally matched by calculating the sum of the squared differences between the reference spectra for each reference spectrum. The reference spectrum with the smallest sum of squares has the best fit. Other techniques for finding the best matching reference spectrum (eg, the sum of the minimum absolute differences) are possible. One method that can be used to reduce computer processing is to limit the search to the portion of the database that matches the spectrum. Often, it is better than the spectrum obtained by grinding the substrate f. The light is limited to the library spectrum during the substrate polishing, and the U plate is determined to be polished. § 刖 疋 索引 index N. For example, 1% flat, can be determined by searching all the database parameters. For the spectrum obtained during the subsequent rotation, from: Search the database within the range. That is, if the number of indexes is found to be N during one rotation, then during the subsequent rotation of X rotations later (where the degree of freedom is γ) 'will search for (Ν+Χ) _γ To (Ν+χ Range of + γ. Referring to Figure 7, the figure illustrates a time varying sequence of index values for each of the best matching spectra in the result of a single region of a single substrate to produce an index value 212. Index value This sequence may be referred to as index trace 21 〇. In some implementations, the index traces are generated by comparing each of the measured spectra to a reference spectrum from a database that has been broken. An index value may be included for each scan (eg, 'one exact') of the optical monitoring system beneath the substrate. For a given index trace 210, there are multiple spectra (called "current spectra") that are measured in a particular region of a single scan of the light to the monitoring system in the index trace 2U) to determine the current spectrum. The best match between each and one or more (eg, exactly one) reference spectra of the database. In this case, the first selected current spectrum is compared with the selected one or the other reference library by ^ a middle. For example, given current = two..., and the reference spectrum five, ..., the matching coefficient can be calculated for each of the following combinations t of the current spectrum and the reference spectrum, and E, e and F, e and G, f and B, f and F, f and G, g and E, g and F, and g and g. Yinglun, ", Nana-matching factor indicates the best match (for example, is the smallest), determines the best pc to match the reference spectrum, and therefore determines the index value. Or, at some discretion φ, the current spectrum can be combined (for example, averaged) and the Τ, ', & spectra are compared with the reference spectrum to determine the best 21 201213050 good match' and therefore decided Index value. In some implementations, for at least one of the substrates, a plurality of index traces can be generated. For a given substrate, an indexed 5 丨 trace can be generated for each reference library of interest. That is, for each reference library of interest for a given area of the substrate, each measurement spectrum in the sequence of quantities is compared to the reference light from a given library: 'Determine the best match reference The sequence of spectra, and the index of the sequence of the best core spectra, provides index traces for a given database. In summary, each index trace includes a sequence 21 of index values 212, wherein each particular bow of the sequence (value 212 is generated by selecting an index of the reference spectrum from a given database, the data pool and quantity) The measured spectrum is closely fitted. The index trace 2 1 〇 each time - the index time value can be the same as the measurement time of the measurement spectrum. The in-situ monitoring technique is used to measure the second layer of the clear and the lower layer or layer structure. Exposure. For example, as discussed in more detail below, the first layer at time TC can be detected by a sudden change in the total torque of the motor torque or light reflected from the substrate, or from the dispersion of the collected spectrum. Exposure. As shown in Fig. 8, for example, using a robust line fitting function, for example, a polynomial function of known order (for example, a first order function (eg, line 2) 4) is collected after time TC. Sequence fitting of the index value of the spectrum. When the function fits the sequence of the index value, the index value of the spectrum collected before the time TC is ignored. Other money (for example, second-order polynomial function) can be used, but the line function Provide calculation At the end time TE, the grinding can be stopped. At the end time TE, the line 214 crosses the target e*m·· 22 201213050. Index it. Figure 9 is a flow chart showing the method of manufacturing and grinding the product substrate. The method may have at least the same layer structure and the same pattern as the test substrate, and the test substrates are used to generate a reference light of the database. First, the first layer is deposited on the substrate and the first layer is patterned (step 902) As noted above, the first layer can be a dielectric, such as a low dielectric constant material, such as carbon doped cerium oxide (eg, BUck Diam〇ndTM (available from Applied Materials, Inc.) or c〇raiTM ( Jing Yu N〇veUus Systems, Inc.)) Depending on the composition of the first material, another dielectric material, for example, a low dielectric constant cap material (eg 'tetraethoxy zexis (TE〇s)) One or more additional layers are deposited on the first layer on the product substrate as needed (step 903), the other dielectric material being different from both the first dielectric material and the second dielectric material. And one or more additional layers together provide a layer stack. Patterning It may be necessary to occur after deposition of one or more additional layers (so that the one or more additional layers do not extend into the trenches in the first layer, as shown in Figure 1A). Next, a different second A second layer of dielectric material, for example, a barrier layer (eg, a nitride (eg, tantalum nitride or titanium nitride)) is deposited over the first layer or layer stack of the product substrate (step 904). Additionally, the conductive layer For example, a 'metal layer (eg, 'copper) may be deposited on the second layer of the product substrate (and in the trench provided by the pattern of the first layer) (step 906). The patterning of the first layer may be required to occur in After the deposition of the second layer (in this case, the second layer does not extend into the trench in the first layer C: Va. 23 201213050 The product substrate is ground ((4) 9〇8). For example, the conductive layer and a portion of the second layer can be ground and the first polishing pad can be used to remove the conductive layer and the second side of the second layer (step 908a). One of the layers of the second layer and the first layer may then be ground, and a second polishing pad may be used at the second polishing station to remove the second layer and a portion of the first layer (step 嶋). However, it should be noted that for some implementations, there are no conductive layers, such as 'the second layer is the outermost layer when the polishing begins. Of course, step 9G2_9G6 can be performed elsewhere so that for a particular operator of the grinding equipment, the process begins at step 9-8. The in-situ monitoring technique is used to detect the removal of the second layer and the exposure of the first layer (step 910). For example, as discussed in more detail below, the exposure of the first layer at time Tc can be detected by a sudden change in the total torque of the motor torque or light reflected from the substrate, or from the dispersion of the collected spectrum ( See Figure 8). At least beginning with the detection of the removal of the second layer (and possibly earlier, eg, 'using the second grinding (four) to grind the product substrate), for example, using the in-situ monitoring system described above to obtain the measured spectrum during the grinding process The sequence (step 912). The measurement spectrum is analyzed to produce a sequence of index values and the function is fitted to a sequence of index values. In particular, for each measurement spectrum in the sequence of measurement spectra, an index value of the best fit reference spectrum is determined to produce a sequence of index values (step 914). A function (e.g., a linear function) is fitted to a sequence of index values of the spectra collected after time TC, at which time the second layer is cleared (step 916). ε:. 24 201213050 In other words, the index value of the spectrum collected at time tc & β , is not used for the function calculation, and the second layer is detected at 1 L between the time/time. ', the value of I (e.g., the calculated emissal force value generated from the linear function fitted to the new sequence of index values) reaches the target index, and the grinding may be stopped (step 918). You can use t to set and store the target thickness IT before the grinding operation. Alternatively, the user can use § § and the target amount to be removed, and the target meter that can be removed is the H Τ α target 丨 丨 IT. For example, the item to be removed can be removed (for example, from empirically determined removals and indices (e.g., polishing rate).
<比旱）计真索引差ID，且使索引差ID 與時間T C處的帝引伯T p 4n上 系^值1C相加，在該時間TC處偵測覆 蓋層之清除（參見第8圖）。 亦有可能使用函數來調整研磨參數（例如，調整基板 上或更多區域之研磨速率以改良研磨均勻性），該函數 與來自在偵測第二層之清除之後所收集的光譜的索引值 擬合。 參閱第10 _，圖示複數個索引跡線。如以上所論述， 了產生母區域之索引跡線。舉例而言，可產生第一區 域之索引值212 (用空心圓圈圖示）之第一序列21〇，可 產生第二區域之索引值222 (用空心正方形圖示）之第 一序列220,且可產生第三區域之索引值232 (用空心三 角形圖示）之第二序列23〇。儘管圖示了三個區域，但 可存在兩個區域或四個或四個以上區域。所有區域可在 相同的基板上，或一些區域可來自於在相同平臺上正經 同時研磨的不同基板。 25 201213050 如以上所論述，原位監控技術用以偵測第二層之清除 及下層或層結構之曝露。舉例而言，如以下更詳細地論 述可藉由馬達扭矩或自基板反射之光之總強度的突然 改變，或自所收集光譜之色散，來偵測在時間Tc處第 一層之曝露。 針對每一基板索引跡線，例如使用穩健的線擬合使已 知P白數之多項式函數（例如，一階函數（例如，線））與 在夺間TC之後對相關聯區域所收集的光譜之索引值之 序列擬合。舉例而言’第-線214可與第一區域之索引 $ 212擬合’第二線224可與第二區域之索引值222擬 第一線23 4可與第三區域之索引值232擬合。線 ”索引值之擬合可包括以下步驟··計算線之斜率$及X 軸相乂時間T ’在該時間τ處該線越過起始索引值（例 如’ 〇)。該函數可用以下形式表示：I⑴=S.(t—T)，其中 1為時間。X轴相交時間T可具有負值，從而指示基板 層之起始厚度小於預期厚度。因此，第一線214可具有 第斜率S 1及第一 x軸相交時間Ή，第二線224可具 有第二斜率S2及坌- Yiu 及第一 X軸相交時間T2，且第三線234 可具有第三斜率芬楚— 手b3及第二X軸相交時間Τ3。 :在研磨製程期間的一些時間處(例如，在時間τ〇處)， 率進行調萼，、：對基板之區域的研磨逮 60 以使彳于複數個區域在研磨終點時間處，比 不進行此調整之枰、、口 s n/ 近於該等區域的目標厚度。在 一些實施例中，在終點時 隹 间處母一區域可具有近似相 26 201213050 同的厚度。 參閱第11圖，右_ a — t EM η i 二貫施令，將一個區域選定為夹去 &域，且決定預估終 、疋為參考 教Ur 夺1 在該預估级點日年門 處參考區域將達到目標索引ΪΤ。舉例而▲時間ΤΕ 中所示’將第-區域選定為參考區域，：而如第11圖 :區域及/或不同基板。使用者在研磨操不 存目標厚度ΙΤ。或者，使 叹置及錯 TR，且可自待穸咚 D又置待移除之目標量 自待移除之目標量TR計算目 而吕，可自待移除之目栌詈f仓丨1 A 舉例 移除量盘索請“ 自根據經驗決定的 丨（例如’研磨速率）之比 ID，且你会？丨莖ττλ ^ 丨井牙、51差 使索引差m與時間TC處的索引值 該時間tc處偵測覆蓋層之清除。 彳加在 為了决疋預估時間，在該預估日孝門步4 土 預估時間處參考區域將達到 #引’可計算參考區域之線（例如，、線214)與目 心索引IT之相交。假定在剩餘的研磨製程中，研磨速率 :偏離預期研磨速率，則索引值之序列應保持大體上線 級數。因此’可將預期終點時間TE計算為線對目標 ，、引1T之簡單的線性内插值，例如，IT=S.(TE-T)。因 此，在將第一區域選定為參考區域之第u圖之實例中， 該參考區域具有相關聯的第一線214，IT=S1.(TE-T1), 亦即，TE=IT/S1-T1。 除了參考區域（包括其他基板上的區域）之外的一或 更夕區域（例如’所有區域）可定義為可調整區域。可 調整區域之線與預期終點時間TE相交之處定義可調整 C； 27 201213050 第u之:：終點。每一可調整區域之線性函數（例如， 聯區域 224及線⑼）可因此用以外推將在相關 咖Γ頁期㈣時間TE處達成之索弓1(例如，EI2及 例而g，在第二區域之預期終點時間te處， =線以可用以外推預期索引ΕΙ2，且在第三區域之 終點㈣ΤΕ處’第三線234可用以外推預期索引 第11圖中所不’ s在時間TG之後不對區域之任一 者的研磨速率進行調整，則在所有區域同時達到終點的 情況下，每—區域可具有不同厚度（因為此舉可導致缺 陷及產量損失，所以此結果並不理想）。 ★右對於不同的區域，在不同的時間達到目標索引（或 等效地，在參考區域之預估終料間處，可調整區域將 具有不同的職索引）’ #可將研磨速率向上或向下調 整以使付該等區域比不進行此調整之情況更接近於同 時^例如’在近似同時處）達到目標索引（且因此達到 目標厚度)’或使得該等區域在目標時間處比不進行此調 整，情況具有更接近於相同的索引值（且因此具有相同 的厚度）例如’近似相同的索引值（且因此具有近似相 同的厚度）。 α因此在第11圖之實例中，在時間處開始，第二 區域之至V —個研磨參數經修改以便增加該區域之研磨 速率（且因此增加了索引跡線220之斜率X，在此實 例中’第二區域之至少—個研磨參數經修改以便減少第 28 201213050 三區域之研磨速率（且因此減少了索引跡線23〇之斜 率）。因此該等區域將近似同時連到目標索引（且因此達 到目標厚度）（或若該等區域之壓力同時停止，則該等區 域將以近似相同的厚度結束）。 在一些實施中，若預期終點時間TE處的預估索引指 示基板之區域處於目標厚度之預定義範圍之内，則可不 需要對該區域進行調整。續餘 Τ η ^这範圍可為目標索引之2% (例 如，1 %之内）。 終 【 一 w g正从彳文π负跑域在預期 點時間處比不進行此調整之情況更接近於目標索引。 舉“而σ #考基板之參考區域可經選擇且所有其他區 域之處理參數可經調整（英文）Tmightbe)，以使得: 有區小域將在參考基板之近似的預估時間處達到終點（英 。了 reach)。參考區域可為例如預定區域（例如 2區域M8a或緊密圍繞中心區域之區域148b)，可為任 2板上之任-區域中的最早或最遲的計畫終點時間之 ’或為具有所要的預估終點之 間等效於同時停止研磨時㈣& &早的時 時間h T止研“的最缚基板。同樣地，最遲的 广问時停止研磨時的最厚基板。參考基板可為 預疋基板，該基板具有區域 早或最遲的預估終料^最早㈣基板之最 :=最薄區域。同樣地，最遲的時間 二 止研磨時的最厚區域。 針對可調整區域中之每— 者，索引跡線之所要斜率可 29 201213050 引使：可調整區域與參考區域同時達到目標索 SD，二列而·吕，可根據叩命他⑽-叫計算所要斜率 ，/、中ί為將改變研磨參數 時間το處的索引值（根 索引值之序列擬合的線性函數計算），ιτ為目標索 在 為、·里汁算預期終點時間。在第11圖之實例令， |·對第-£域’可根據（it_I2)=SD2*(te tg)計算所要斜 ί: SD2且針對第三區域，可根據（叮七助 計算所要斜率SD^ } 5，在-些實施中，不存在參考區域，且預期終點 時間可為例如由使用者在研磨製程之前設置的預定時 間’或可根據來自-或更多基板的兩個或兩個以上區域 之預期終點時間之平均值或其他組合（如藉由使各個區 域之，投影至目標索引來計算）來計算預期終點時間。 在此^巾’大體上如上所述計算所要斜率，然而亦必 須計算第一基板之第一區域的所要斜率，例如，可根據 (IT_I1) = SD1*(TE’_T0)計算所要斜率 sm。 乂 或者’在一些實施中，丨同的區域存在不同的目標索 引。此舉允許在基板上產生精密的但可控的不均勻的厚 度刀佈。目標索引可由使用者例如使用控制器上之輸入 裝置輸入。舉例而言，第一基板之第一區域可具有第一 目標索引，第一基板之第二區域可具有第二目標索引， 第二基板之第一區域可具有第三目標索引，且第二基板 之第二區域可具有第四目標索引。 針對以上所述之方法中之任何方法，調整研磨速率以 30 201213050 使索引跡線之斜率更接近於所要斜率。研磨速率可藉由 例如增加或減少承載頭之相應腔室中之壓力來調整。可 假定研磨速率之改變與壓力之改變直接成正比（例如， 簡單Prestonian模型）。舉例而言，針對每一基板之每一 區域’其中在時間T0之前以壓力p〇id對區域進行研磨， 在時間T0之後施加新的壓力Pnew，該新的壓力pnew 可計算為Pnew=P〇ld*(SD/S)，其中S為時間TO之前之 線的斜率，且SD為所要斜率。 舉例而言’假定對第一基板之第一區域施加壓力 Poldl ’對第一基板之第二區域施加壓力p〇id2，對第二 基板之第一區域施加壓力p〇ld3，且對第二基板之第二 區域施加壓力P〇ld4,則第一基板之第一區域的新壓力 Pnewl可計算為pnewl=p〇ldl*(SD1/S1),第一基板之第 二區域的新壓力 pnew2 可計算為<Determination of the true index difference ID, and the index difference ID is added to the value 1C of the TP on the time TC, and the cover layer is detected at the time TC (see the eighth) Figure). It is also possible to use a function to adjust the grinding parameters (for example, to adjust the polishing rate on the substrate or more to improve the polishing uniformity), the function and the index value from the spectrum collected after the detection of the second layer is removed. Hehe. See section 10 _, which shows multiple index traces. As discussed above, the index traces of the parent regions are generated. For example, a first sequence 21 of index values 212 (shown by open circles) of the first region may be generated, and a first sequence 220 of index values 222 (illustrated by open squares) of the second region may be generated, and A second sequence 23 of index values 232 (illustrated by open triangles) of the third region can be generated. Although three regions are illustrated, there may be two regions or four or more regions. All regions may be on the same substrate, or some regions may be from different substrates that are being simultaneously ground on the same platform. 25 201213050 As discussed above, in-situ monitoring techniques are used to detect the removal of the second layer and the exposure of the underlying or layer structure. For example, the exposure of the first layer at time Tc can be detected as discussed in more detail below by a sudden change in the total intensity of the motor torque or light reflected from the substrate, or from the dispersion of the collected spectrum. For each substrate index trace, for example, a robust line fit is used to make a polynomial function of a known P white number (eg, a first order function (eg, a line)) and a spectrum collected for the associated region after the intervening TC Sequence fitting of the index values. For example, the 'th-line 214 can be fitted to the index of the first region $212'. The second line 224 can be compared with the index value 222 of the second region. The first line 23 4 can be fitted to the index value 232 of the third region. . The fitting of the line "index value" may include the following steps: calculating the slope of the line $ and the X-axis relative time T ' at which the line crosses the starting index value (eg ' 〇). The function may be expressed in the following form : I(1) = S.(t - T), where 1 is time. The X-axis intersection time T may have a negative value indicating that the initial thickness of the substrate layer is less than the expected thickness. Thus, the first line 214 may have a first slope S 1 And the first x-axis intersection time Ή, the second line 224 may have a second slope S2 and 坌- Yiu and a first X-axis intersection time T2, and the third line 234 may have a third slope—the hand b3 and the second X Axis intersection time Τ3: At some time during the grinding process (for example, at time τ〇), the rate is adjusted, and: the grinding of the area of the substrate is captured 60 so that the plurality of areas are at the end of the grinding time Where, the target thickness is not close to the area, and the port sn/ is close to the target thickness of the regions. In some embodiments, the parent-area region at the end point may have the same thickness as the phase 26 201213050. Figure 11, right _ a — t EM η i Second order, will be a zone Selected as the folder & field, and decided to predict the end, 疋 as the reference teach Ur to take 1 at the estimated level point, the reference area will reach the target index ΪΤ. For example, ▲ time ΤΕ shown in the 'will - The area is selected as the reference area, and as shown in Figure 11: area and / or different substrate. The user does not save the target thickness in the grinding operation. Or, make the sigh and error TR, and can be set to The target quantity to be removed is calculated from the target quantity TR to be removed. It can be removed from the target. 丨 丨 丨 A A A A A A A 举例 举例 举例 举例 举例 举例 举例 举例 举例 举例 举例 自 自 自 自 自 自 自 自 自 自 自 自Grind rate) ratio ID, and you will? The stalk ττλ ^ 丨 well tooth, 51 difference The index difference m and the index value at time TC At this time tc, the detection layer is cleared. In order to determine the estimated time, the reference area at the estimated time of the forecast is to reach the line of the reference area (for example, line 214) and the index of the index. intersect. Assuming that in the remaining grinding process, the grinding rate: deviates from the expected grinding rate, the sequence of index values should remain substantially linear. Therefore, the expected end time TE can be calculated as a line-to-target, a simple linear interpolation of 1T, for example, IT=S.(TE-T). Thus, in the example of the u-th image in which the first region is selected as the reference region, the reference region has an associated first line 214, IT = S1. (TE-T1), that is, TE = IT / S1 - T1. One or more of the area other than the reference area (including the area on the other substrate) (e.g., 'all areas') may be defined as an adjustable area. The area where the adjustable area line intersects the expected end time TE defines an adjustable C; 27 201213050 u:: End point. The linear function of each adjustable region (eg, joint region 224 and line (9)) can therefore be extrapolated to be achieved at the time of the relevant coffee page (4) time TE (eg, EI2 and example g, in the first At the expected end time te of the two regions, the = line is extrapolated to the expected index ΕΙ2, and at the end of the third region (four) ' 'the third line 234 can be extrapolated. The expected index in Fig. 11 is not s after the time TG. The polishing rate of any of the zones is adjusted, and in the case where all zones reach the end point at the same time, each zone can have a different thickness (because this can lead to defects and yield loss, so the result is not ideal). For different regions, the target index is reached at different times (or equivalently, the adjustable region will have a different job index between the predicted end regions of the reference region)' # Adjust the grinding rate up or down In order to make the regions closer to the same time than when the adjustment is not made, for example, 'at the same time, the target index (and thus the target thickness) is reached' or the regions are Than the standard time adjustment is not made, the case is closer to having the same index value (and thus has the same thickness), for example, 'approximately the same index value (and thus have approximately the same thickness). Thus, in the example of Fig. 11, starting at time, the V-th grinding parameters of the second region are modified to increase the polishing rate of the region (and thus increase the slope X of the index trace 220, in this example) At least one of the grinding parameters in the 'second region' is modified to reduce the grinding rate of the 28th 201213050 three region (and thus the slope of the index trace 23〇). Therefore, the regions will be approximately connected to the target index at the same time (and Thus reaching the target thickness) (or if the pressures in the regions are simultaneously stopped, the regions will end at approximately the same thickness). In some implementations, if the predicted index at the expected end time TE indicates that the region of the substrate is at the target Within the predefined range of thickness, the area may not need to be adjusted. Continued Τ η ^ This range can be 2% of the target index (for example, within 1%). End [a wg is from 彳 π negative The running field is closer to the target index at the expected point time than if the adjustment is not made. The reference area of the σ# test substrate can be selected and the processing parameters of all other areas. Adjusted (English) Tmightbe) so that: the zoned small domain will reach the end point at the approximate estimated time of the reference substrate. The reference zone can be, for example, a predetermined zone (eg 2 zone M8a or tightly surrounding) The area 148b) of the central area may be the earliest or latest project end time in the any-area of any two boards or may have the desired estimated end point equivalent to simultaneously stopping the grinding (4) && Early time h H stop research "the most bound substrate. Similarly, the thickest substrate at the time of polishing is stopped at the latest. The reference substrate may be a pre-baked substrate having an area that is early or latest estimated by the end of the first (four) substrate: = the thinnest area. Similarly, the latest time is the thickest area at the time of grinding. For each of the adjustable regions, the required slope of the index trace can be 29 201213050. The adjustable region and the reference region reach the target cable at the same time, and the second column can be calculated according to the commandment (10). The slope, /, ί is the index value at which the grinding parameter time το is changed (the linear function of the sequence fitting of the root index value is calculated), and ιτ is the expected end time of the target. In the example of Figure 11, the |-for the -£ field can be calculated according to (it_I2)=SD2*(te tg): SD2 and for the third region, according to the calculation of the desired slope SD ^ } 5, in some implementations, there is no reference area, and the expected end time may be, for example, a predetermined time set by the user prior to the grinding process' or may be based on two or more from - or more substrates Calculate the expected end time by the average or other combination of the expected end time of the area (eg, by projecting each area to the target index). Here, the desired slope is calculated substantially as described above, but must also The desired slope of the first region of the first substrate is calculated, for example, the desired slope sm can be calculated from (IT_I1) = SD1*(TE'_T0). 乂 or 'In some implementations, different regions have different target indices. This allows a precise but controllable uneven thickness of the knives to be produced on the substrate. The target index can be input by the user, for example using an input device on the controller. For example, the first region of the first substrate can have a first The target index, the second region of the first substrate may have a second target index, the first region of the second substrate may have a third target index, and the second region of the second substrate may have a fourth target index. In any of the methods, the polishing rate is adjusted to bring the slope of the index trace closer to the desired slope at 30 201213050. The polishing rate can be adjusted, for example, by increasing or decreasing the pressure in the corresponding chamber of the carrier head. The change is directly proportional to the change in pressure (eg, a simple Prestonian model). For example, for each region of each substrate 'where the region is ground with pressure p〇id before time T0, applied after time T0 The new pressure Pnew, the new pressure pnew can be calculated as Pnew=P〇ld*(SD/S), where S is the slope of the line before time TO, and SD is the desired slope. For example, 'assumed first Applying a pressure Pold1 to a first region of the first substrate, applying a pressure p〇id2 to a first region of the second substrate, applying a pressure p〇ld3 to the first region of the second substrate, and applying a pressure p〇ld3 to the first region of the second substrate Applying pressure P〇ld4 the second region, the first new region of the first pressure Pnewl calculated as the substrate pnewl = p〇ldl * (SD1 / S1), the new pressure pnew2 second region of the first substrate can be calculated as
Pnew2=P〇ld2*(SD2/S2)，第二基板之第一區域的新壓力 Pnew3可計算為pnew3=P〇id3*(SD3/S3)，且第二基板之 第二區域的新壓力pnew4可計算為Pnew2=P〇ld2*(SD2/S2), the new pressure Pnew3 of the first region of the second substrate can be calculated as pnew3=P〇id3*(SD3/S3), and the new pressure pnew4 of the second region of the second substrate Can be calculated as
Pnew4=Pold4*(SD4/S4) 〇 決定基板將達到目標厚度之預估時間，及對研磨速率 進行調整之製程在研磨製程期間僅可被執行一次，例 如’在指^時間’例如’預期研磨時間之4〇%至60%; 或在研磨製程期間被執行多次，例如，每三十至六十秒。Pnew4=Pold4*(SD4/S4) 〇 Determines the estimated time at which the substrate will reach the target thickness, and the process for adjusting the polishing rate can only be performed once during the polishing process, such as 'in the finger time' such as 'expected grinding 4% to 60% of the time; or performed multiple times during the grinding process, for example, every thirty to sixty seconds.
在適當時，可在研磨製鋥划#炫A π名表牲期間的後續時間再次調整速 率。在研磨製程期間，僅可 j對研磨逮率進行幾次改變， 201213050 諸如四次、二次、兩次或僅一次。可在接近研磨製程之 開始時、在研磨製程當中或趨於研磨製程結束時進行調 整0 在已調整研磨速率之後（例如，時間τ〇之後），繼續 研磨，光學監控系統繼續收集至少參考區域之光譜，且 光學監控系統繼續決定參考區域之索引值。在一些實施 中，光學監控系統繼續收集光譜且決定每一區域之索引 值…旦參考區域之索引跡線達到目標索引，則調用終 點且研磨操作停止。 舉例而έ，如第12圖中所示，在時間τ〇之後，光學 監控系統繼續收集參考區域之光譜，且光學監控系統繼 續決定參考區域之索引值312。若施加在參考區域上的 壓力不改t (例如’如第n圖之實施中），則可使用來 自T0之前（但不是TC之前）& τ〇之後兩者的資料點 來計算線性函數以提供更新後的線性函數314，且該線 性函數3Μ達到目標索引ΙΤ時的時間指示研磨終點時 間另方面若施加在參考區域上的壓力在時間το 處改變’則可根據時間τ〇之後的索引值3 i2之序列來 計算新的線性函數314 ’該新的線性函數314具有斜率 S，’且該新的線性函數3 14達到目標索引ΙΤ時的時間指 示研磨終點㈣。帛於決枝點財考d域可為與以上 所述使用相同的參考區域，以計算預期終點時間，或用 於決定終點的參考區域可為不同的區域（或若如參閱第 11圖所述對所有區域進行調整，則為達成終點決定之目 32 201213050 的，可選擇一！ Irs· u、 /考Q域）。若新的線性函數314達到目 索广T的時間稍微遲於（如第12圖中所示）或早於： 一原始：性函數214計算之預估時間，則該等區域中之 S更夕者可刀別稍微經過度研磨或研磨不足。然而， 因為預期終料間與實際研料間之間的差值應小於數 秒，所以此舉不必嚴重影響研磨均勻性。 2-些實施中’例如’對於銅研磨而言，在偵測基板 •、點之後，基板立即經受過度研磨製程，例如以移除 銅殘餘物。對於基板之所有區域而言，過度研磨製程可 在均勻壓力（例如’ i psi至i 5 ps。下進行。過度研磨 製程可具有預置持續時間（例如，1〇秒至Η秒）。 若對於特定區域產生了多個索引跡線（例如，對於特 ^區域產生所關注的每—f料庫之_個索引跡線），則該 等索引跡線中之-者可經選擇用於㈣區域之終點或壓 力控制演舁法。舉例而言，對於相同區域所產生的每一 索引跡線，控制器19〇可使線性函數與該索引跡線之索 引值擬合，且控制器190可決定該線性函數與索引值之 序列的擬合優度。所產生的索引跡線具有具最佳擬合優 度之線，該索引跡線自身的索引值可選定為特定區域及 基板之索引跡線。舉例而言，當決定如何調整可調整區 域（例如在時間T0處）之研磨速率時，具有最佳擬合 優度的線性函數可用於計算中。如另一實例，當具有最 佳擬合優度的線之計算索引（如根據與索引值之序列擬 合的線性函數計算所得之索引）與目標索引相匹配或超 33 201213050 過目標索弓丨時，可調用終點。又，可將索引值自身與目 才丁索引相比較以決定終點，而非根據線性函數計算索引 值。 決疋與光譜資料庫相關聯之索引跡線是否具有與資料 庫相關聯之線性函數的最佳擬合優度可包括：相對而 °相比於相關聯穩健線及與另一資料庫相關聯之索引 跡線之間的差值（例如，最小標準偏差、最大相關或其 他方差置測）’決定相關聯光譜資料庫之索引跡線是否具 有/、相關聯穩健線最小的差值量。在一個實施中，藉由 δ十算索引資料點與線性函數之間的差值平方和來決定擬 合優度；具有最小差值平方和的資料庫具有最佳擬合優 度。 參閱第13圖，圖示概述流程圖13〇〇。如上所述，在 研磨设備中用相同的研磨墊同時研磨基板之複數個區域 (步驟1302 )。在此研磨操作期間，每一區域具有該區 域之研磨速率，該研磨速率藉由獨立可變的研磨參數（例 如，由特定區域上方承载頭中之腔室所施加的壓力）可 控制，而與其他基板無關。在研磨操作期間，如上所述， 例如使用自每一區域獲得之量測光譜之序列對基板進行 2控（步驟1304 )。針對該序列中之每一量測光譜，決 定最佳匹配之參考光譜（步驟13〇6)。決定最佳擬合之 每一參考光譜之索引值以產生索引值之序列（少驟 1308 ) 〇 偵測第二層之清除（步驟131〇)。針對每一區域，使 34 201213050 線性函數與光譜之索引值之序列擬合，該等光譜係在偵 測第二層之清除之後收集（步驟13 12 )。在一個實施十， 例如藉由線性函數之線性内插法來決定預期終點時間， 在該預期終點時間處’參考區域之線性函數將達到目標 索引值（步驟1 3 14 )。在其他實施中’預定預期終點時 間或將預期終點時間計算為多個區域之預期終點時間之 組合。在需要時，調整其他區域之研磨參數以對基板之 研磨速率進行調整，以使得複數個區域近似同時達到目 標厚度’或使得複數個區域在目標時間處具有近似相同 的厚度（或目標厚度）（步驟1316)。在調整參數之後研 磨繼續進行，且針對每一區域，執行以下步驟：量測光 譜；決定來自資料庫之最佳匹配參考光譜；決定最佳匹 配光》普之索引值以在已調整研磨參數之後的時段產生索 引值之新的序列；以及使線性函數與索引值擬合（步驟 1 3 1 8 )。一旦參考區域之索引值（例如，根據與索引值之 新序列擬合的線性函數而產生的計算索引值）達到目標 索引’可停止研磨（步驟1330)。 在一些實施中，索引值之序列用以調整基板之一或更 多區域的研磨速率，曰是另__ gA ^ , 名疋千1一疋力原位監控糸統或技術用以 偵測研磨終點。 如以上所論述，對於一些技術及—些層堆疊而言，偵 測覆蓋層之清除及下層之曝露可能十分困難。在一些實 施中，收集光譜之群組之序列，且計算光譜之每一群纽 的色散參數值以產生色散值之序列。可自色散值之序列 C： •ter 35 201213050 ㈣覆蓋層之清除。例如’在以上所述的研磨操作之步 驟9H)或步驟131G中’此技術可用以㈣第二層之清除 及第一層之曝露。 第14圖圖示方法1400，該方法用於偵測第二層之清 除及第-層之曝露。當基板正經研磨時（步驟14〇2)， 收集光譜之群組之序列（步驟14〇4)。如第4圖中所示， 若光學監控系統緊固至旋轉平臺，則在光學監控系統越 過基板進行的單次掃描中，可自基板上之多個不同位置 201b-201j收集光譜。自單次掃描所收集的光譜提供光譜 之群組。隨著研磨的進行’光學監控系統的多次掃描提 供光譜之群組之序列。對於每一平臺旋 一個群組，例如，可以與平臺旋轉速率相等之 該等群組。通常，每—群組將包括五個至二十個光譜。 可使用與在以上論述的峰值追縱技術中用以收集光譜的 相同的光學li控系統來收集光譜。 第15A圖提供在研磨開始時（例如，當覆蓋層之顯著 的厚度殘留在下層上時）’自基板1〇反射之光之量測光 譜1500a的群組的實例。光譜15〇〇&之群組可包括光譜 202a-204a ’该等光譜2〇2a_2〇4a係在光學監控系統越過 基板進行的第一次掃描中’在基板上之不同位置處收集 的。第15B圖提供在覆蓋層之清除時或接近覆蓋層之清 除時，自基板ίο反射之光之量測光譜15〇〇b的群組的實 例。光譜1500b之群組可包括光譜2〇2b_2〇4b，該等光 譜202b-204b係在光學監控系統越過基板進行的不同的 36 201213050 掃私中，在基板上之不同位置處收集的（光错 . 5〇〇a可自基板上與光譜1500b不同之位置處收集）。 最初，如第15A圖中所示，光譜1500a相當相似。然 而如第15B圖中所示，當清除了覆蓋層（例如，阻障 層）且曝露了下層（例如，低介電常數層或蓋層）時， 來自基板上之不同位置處的光譜15〇〇b之間的差異趨向 於變得更明顯。 針對光譜之每一群組’計算該群組中的光譜之色散參 數值（步驟1406)。此舉產生色散值之序列。 / 在個實施中，為計算光譜之群組之色散參數，對強 度值（作為波長之函數）共同取平均值以提供平均光譜。 'N P Iave“) —（Ι/Ν^Σϋ-Νΐ^χ)]，其中N為該群組中光譜 的數里’且Ι;(λ)為光譜。對於該群組中每—光譜而言， 例如使用差值平方和或絕對值差之和，例如， Di = [l/(Xa-Xb).[I,=u.u[Iia).lAvE(M]2]]i/2 或 其巾 Aa_Ab 為 經總計的波長範圍，隨後可計算該光譜與平均光譜之間 的總差異。 一旦已計算光譜之群组中之每一光譜的差值，則可根 據該等差值計算料組之色散參數值。可能存在各種色 散參數，諸如標準偏差、四分位數間距、範圍（最大值 減去最小值）、平均差、絕對中位差及平均絕對偏差。 可對色散值之序列進行分析，且該色散值之序列用以 偵測覆蓋層之清除（步驟丨4〇 g )。 37 201213050 第16圖圖不光譜之標準偏差隨著研磨時間之函數變 • 化的圖表1600(其中根據光譜之群組之差值計算每一標 . 準偏差）。因此，圖表中每一標繪點1602為光譜之群組 之差值的才示準偏差，言亥等光譜係在光學監控系統之給定 掃描時收集。如所圖示，在第一時段⑹〇期間，標準偏 差值保持相當小的值。然而，在時段1610之後，標準偏 差值變仔較大且更為分散。在不限於任何特定理論的情 況下’厚的阻障層可趨向於支配反射光譜，從而掩蔽了 阻障層自身與任何下層在厚度上的差異。隨著研磨的進 行阻障層變得較薄或完全移除’且反射光譜變得對下 層厚度之變化更為敏感。因此，當清除了阻障層時，光 譜之色散將趨向於增加。 當覆蓋層清除時，各種演算法可用以憤測色散值之行 為的改變。舉例而言，色散值之序列可與臨限值相比較， 且若：散值超過臨限值’則產生信號，該信號指示覆蓋 月除如另一貫例，可計算移動視窗内部色散值之 序列之-部分的斜率’且若斜率超過臨限值，則產生信 號，該信號指示覆蓋層已清除。 作為债測色散之增加的演算法之部分，色散值之序列 可經過渡波器處理（例如，低通據波器或頻帶濾波器） 以：移除高頻雜訊。低通濾波器之實例包括移動平均渡 波器及巴特沃斯（Butterworth)濾波器。 儘管以上論述集中於對阻障層之清除的偵測，但該技 術可在其他上下文中用於對覆蓋層之清除的偵測，例 38 201213050 如使用介電層堆疊之另一類型的半導體製程中覆蓋層 ‘ ! 層間 ^丨電質（interlayer dielectric; ILD))之清 除，或介電層上的薄金屬層之清除。 除如以上所論述用作啟動特徵結構追蹤之觸發之外， 用於偵測覆蓋層之清除之此技術可用於研磨操作中的其 他目的，例如，用作終點信號自身，以觸發計時器以便 繼曝路之後研磨下層達預定持續時間，或作為修改研磨 參數之觸發（例如’在下層曝露之後改變承載頭壓力或 漿組合物）。 此外，儘官以上論述假定一旋轉平臺，該旋轉平臺具 有光學終點監控器，該光學終點監控器安裝在該平臺 中’但系統可適用於監控系統與基板之間其他類型的相 對運動。舉例而言，在一些實施中（例如’軌道運動）， 光源k穿基板上之不同位置，但光源不越過基板之邊 緣。在此等情況下，仍可對所收集的光譜進行分類，例 如，可以某一頻率收集光譜，且在一時段内所收集的光 譜可視為一群組之部分。該時段應充分長以便每一群組 收集五個至二十個光譜。 如本說明書中所使用’術語基板可包括例如產物基板 (例如，產物基板包括多個記憶體或處理器晶粒）、測試 基板、裸基板及閘控基板。基板可處於積體電路製造之 各個階段，例如，基板可為裸晶圓，或基板可包括一或 更多經沈積及/或經圖案化的層。術語基板可包括圓盤形 及矩形的薄片。 39 201213050 在本說明書中描述的本發明之實施例及所有功能性操 作可實施在數位電子電路中或電腦軟體、韌體或硬體 中’該電腦軟體、細體或硬體包括揭示於本說明書中的 結構構件及結構構件的結構均等物，或該等實施例及所 有功能性操作可實施在數位電子電路與電腦軟體、勒體 或硬體的組合中。本發明之實施例可實施為一或更多電 腦程式產品’亦即，有形地實施在機器可讀取儲存媒體 中的-或更多電腦程式，該等電腦程式由資料處理設備 可程式處理器、電腦或多個處理器或電腦）執 行或控制資料處理設備之操作。電腦程式（亦稱為程式、 軟體、軟體應用程式或程式碼）可以包括編譯語言或解 澤…之任何形式的程式語言寫人，且電腦程式可以任 何形式部署’該等形式包括作為獨立程式或作為模組、 部件、次常式或適合在計算環境中使用之其他單元。電 腦程式未必對應於㈣。程式可儲存在保存其他程式或 資料之檔案之-部分中、專用於所述程式之單個樓案中 或多個同-類型的標案（例如，儲存—或更多模組了次 程式或程式碼之部分的標案）中。電腦程式可經部署以 在-個站點或分散在多個站點處且由通訊網路互連的一 個電腦或多個電腦上執行。 本說明書中描述的製程及邏輯流程可由執行一或更多 電腦㈣之—或更多可程式處理器來執行，以藉由對輸 入資料操作及產生輸出來執行功能。製程及邏輯流程亦 可由專用邏輯電路來執行，且設備亦可實施為專用邏輯 201213050 電路，例如’現場可程式閘陣列（field pr〇grammabie卿 ⑽叮；FPGA)或特殊應用積體電路（application-specific integrated circuit; ASIC) ° 〆述研磨„又備及方法可應用於各種研磨系統中。研磨 塾或承載頭或兩者皆可移動以提供研磨纟面與基板之間 的相對運動。舉例而言，平臺並非旋轉，而是可繞軌道 運行。研磨墊可為緊固至平臺的圓形（或某一其他形狀) 塾。終點偵測系統之-些態樣可適用於線性研磨系統， 例如’其中研磨塾為線性移動之連續帶或捲盤帶。研磨 層可為標準（例如’具有或不具有填料之聚氨基甲酸酯） :磨材料、軟材料或固定研磨材料。使用相對定位之術 浯，應理解，可將研磨表面及基板固持在垂直方向或某 ~~其他方向上。 已柄述本發明之特定實施例。其他實施例係在以下申 睛專利範圍之範疇内。 【圖式簡單說明】 第1A圖及第1B圖為研磨前後基板之示意性橫截面 圖。 第2圖圖示研磨設備之實例的示意性橫截面圖。 第3圖圖示基板之示意性俯視圖，該基板具有多個區 域。 第4圖圖示研磨墊之俯視圖’且圖示在基板上進行原 位量測所處之位置。 41 201213050 第圖圖不自原位光學監控系統的量測光譜。 第6圖圖示參考光譜之資料庫。 第7圖圖示索引跡線。 第8圖圖示索引跡線，該㈣跡線具有與索引值擬合 之線I·生函數’該等索引值係在偵測覆蓋層之清除之後收 集的》 第9圖為製造基板及馈測研磨終點之示例性製程之流 程圖。 第1 〇圖圖示複數個索引跡線。 第11圖圖示基於一 數個所要斜率之計算， 達到目標索引。 時間對於複數個可調整區域之複 在該時間處參考區域之索引跡線 第12圖圖示基於一時間對於終點之計算，在該時間處 參考區域之索引跡線達到目標索引。 第13圖為示例性製程之流程圖’該示例性製程用於調 整複數個基板中之複數個區域的研磨速率，以使得該複 數個區域在目標時間處具有近似相同的厚度。 第14圖圖示偵測覆蓋層之清除的流程圖。 第15A圖圖示在研磨開始時，在單次掃描期 的光譜圖表。 β 第15Β圖圖示在接近阻障清除的單次掃描期間所 的光譜圖表。 ' 第16圖圖示光譜之標準偏差隨著 |堪η寸間之函數變 化的圖表。 42 .明I! 201213050 各種圖式中的相同元件符號及名稱指示相同元件。When appropriate, the rate can be adjusted again at a subsequent time during the grinding process. During the grinding process, the grinding rate can only be changed several times, such as four times, two times, two times or only once. Adjustments can be made near the beginning of the grinding process, during the grinding process or towards the end of the grinding process. 0 After the grinding rate has been adjusted (eg after time τ〇), the grinding continues and the optical monitoring system continues to collect at least the reference area. The spectrum, and the optical monitoring system continues to determine the index value of the reference area. In some implementations, the optical monitoring system continues to collect the spectra and determines the index value for each region. Once the index trace of the reference region reaches the target index, the endpoint is called and the grinding operation is stopped. By way of example, as shown in Figure 12, after time τ , the optical monitoring system continues to collect the spectrum of the reference area and the optical monitoring system continues to determine the index value 312 of the reference area. If the pressure applied to the reference area does not change t (eg, as in the implementation of Figure n), the data points from both before T0 (but not before TC) & τ〇 can be used to calculate the linear function. An updated linear function 314 is provided, and the time when the linear function 3Μ reaches the target index 指示 indicates the polishing end time. If the pressure applied to the reference region changes at time το, then the index value after the time τ〇 can be obtained. The sequence of 3 i2 is used to calculate a new linear function 314 'The new linear function 314 has a slope S,' and the time when the new linear function 3 14 reaches the target index 指示 indicates the grinding end point (4). The d-field may be the same reference area as described above to calculate the expected end time, or the reference area used to determine the end point may be a different area (or as described in FIG. 11) For all areas to be adjusted, in order to reach the end point of the decision 32 201213050, you can choose one! Irs·u, / test Q domain). If the new linear function 314 reaches the target time T is slightly later (as shown in FIG. 12) or earlier than: an original: the estimated time calculated by the sex function 214, then the S in the regions The knife can be slightly ground or undergrinded. However, since it is expected that the difference between the final material and the actual material should be less than a few seconds, this does not necessarily seriously affect the grinding uniformity. In some implementations, for example, for copper polishing, after detecting the substrate, the substrate, the substrate is immediately subjected to an over-grinding process, for example to remove copper residues. For all areas of the substrate, the over-grinding process can be performed at a uniform pressure (eg, 'i psi to i 5 ps.) The over-grinding process can have a preset duration (eg, 1 second to leap seconds). A certain area produces a plurality of index traces (eg, for each of the index traces of each of the f-bases of interest), then those of the index traces may be selected for the (four) region Endpoint or pressure control deduction. For example, for each index trace generated by the same region, the controller 19 can fit the linear function to the index value of the index trace, and the controller 190 can determine The goodness of fit between the linear function and the sequence of index values. The resulting index trace has a line with the best fit, and the index value of the index trace itself can be selected as the index trace of the specific region and the substrate. For example, when deciding how to adjust the grinding rate of an adjustable region (eg, at time T0), a linear function with the best fit goodness can be used in the calculation. As another example, when there is a best fit Excellent line The calculated index (such as the index calculated from the linear function fitted to the sequence of index values) matches the target index or exceeds the target. When the target is crossed, the endpoint can be called. Again, the index value itself can be The index is compared to determine the endpoint, rather than the index value based on the linear function. The best fit of the index trace associated with the spectral library to the linear function associated with the database may include: Relatively the difference between the associated robust line and the index trace associated with another database (eg, minimum standard deviation, maximum correlation, or other variance) - determines the associated spectral database Whether the index trace has the smallest difference of /, the associated robust line. In one implementation, the goodness of fit is determined by the sum of the squares of the differences between the indexed data points and the linear functions; The pool of squared values has the best fit. See Figure 13, which shows an overview of Flowchart 13〇〇. As described above, the same polishing pad is used to grind simultaneously in the grinding equipment. a plurality of regions of the substrate (step 1302). During the polishing operation, each region has a polishing rate for the region, the polishing rate being independently variable polishing parameters (eg, by a chamber in the carrier head above the particular region) The applied pressure can be controlled regardless of the other substrates. During the grinding operation, as described above, the substrate is controlled (for example, using a sequence of measurement spectra obtained from each region) (step 1304). Each measured spectrum determines the best matching reference spectrum (steps 13〇6). Determines the index of each reference spectrum of the best fit to produce a sequence of index values (less steps 1308) 〇 detects second The layer is cleared (step 131A). For each region, a linear sequence of 34 201213050 is fitted to the sequence of index values of the spectrum, which are collected after detecting the second layer of clearing (step 13 12 ). In an implementation ten, the expected end time is determined, for example, by linear interpolation of a linear function at which the linear function of the reference region will reach the target index value (step 134). In other implementations, the predetermined expected end time or the expected end time is calculated as a combination of expected end times for the plurality of regions. When necessary, the grinding parameters of the other regions are adjusted to adjust the polishing rate of the substrate such that the plurality of regions approximately reach the target thickness at the same time or such that the plurality of regions have approximately the same thickness (or target thickness) at the target time ( Step 1316). Grinding continues after the parameters are adjusted, and for each region, the following steps are performed: measuring the spectrum; determining the best matching reference spectrum from the database; determining the best matching light index value after the adjusted grinding parameters The period of time produces a new sequence of index values; and fits the linear function to the index value (step 1 3 1 8 ). Once the index value of the reference region (e.g., the calculated index value resulting from a linear function fitted to the new sequence of index values) reaches the target index', the grinding may be stopped (step 1330). In some implementations, the sequence of index values is used to adjust the polishing rate of one or more regions of the substrate, which is another __gA^, which is used to detect the polishing endpoint. . As discussed above, for some techniques and for layer stacking, it may be difficult to detect the removal of the overlay and the exposure of the underlying layer. In some implementations, a sequence of groups of spectra is collected and the dispersion parameter values for each of the spectra are calculated to produce a sequence of dispersion values. Self-dispersion sequence C: •ter 35 201213050 (4) Clearance of the overlay. For example, 'in the polishing operation step 9H) or step 131G described above, this technique can be used to remove (4) the second layer and expose the first layer. Figure 14 illustrates a method 1400 for detecting the removal of the second layer and the exposure of the first layer. When the substrate is being ground (step 14〇2), a sequence of groups of spectra is collected (step 14〇4). As shown in Figure 4, if the optical monitoring system is secured to the rotating platform, the spectra can be collected from a plurality of different locations 201b-201j on the substrate in a single scan of the optical monitoring system across the substrate. The spectra collected from a single scan provide a group of spectra. As the grinding progresses, multiple scans of the optical monitoring system provide a sequence of groups of spectra. One group is rotated for each platform, for example, those groups that can be equal to the platform rotation rate. Typically, each group will include five to twenty spectra. The spectra can be collected using the same optical control system used to collect the spectra in the peak tracking technique discussed above. Fig. 15A provides an example of a group of the photometry spectrum 1500a of the light reflected from the substrate 1 at the start of the grinding (e.g., when a significant thickness of the cover layer remains on the lower layer). The group of spectra 15 〇〇 & can include spectra 202a-204a 'the spectra 2 〇 2a 2 〇 4a are collected at different locations on the substrate during the first scan of the optical monitoring system across the substrate. Fig. 15B provides an example of a group of measurement spectra 15 〇〇 b of light reflected from the substrate ί when the cover layer is removed or close to the cover layer. The group of spectra 1500b may include spectra 2〇2b_2〇4b, which are collected at different locations on the substrate during the different monitoring of the optical monitoring system across the substrate. 5〇〇a can be collected from a different location on the substrate than the spectrum 1500b). Initially, as shown in Figure 15A, the spectra 1500a are quite similar. However, as shown in Fig. 15B, when the cap layer (e.g., barrier layer) is removed and the underlayer (e.g., low dielectric constant layer or cap layer) is exposed, the spectrum from different locations on the substrate is 15〇 The difference between 〇b tends to become more pronounced. The dispersion parameter values for the spectra in the group are calculated for each group of spectra (step 1406). This produces a sequence of dispersion values. / In one implementation, to calculate the dispersion parameters for the group of spectra, the intensity values (as a function of wavelength) are collectively averaged to provide an average spectrum. 'NP Iave“) —(Ι/Ν^Σϋ-Νΐ^χ)], where N is the number of spectra in the group 'and Ι; (λ) is the spectrum. For each spectrum in the group For example, using the sum of squared differences or absolute differences, for example, Di = [l/(Xa-Xb).[I,=uu[Iia).lAvE(M]2]]i/2 or its towel Aa_Ab For the total wavelength range, the total difference between the spectrum and the average spectrum can then be calculated. Once the difference in each of the spectra is calculated, the dispersion parameters of the stack can be calculated from the differences. Values. There may be various dispersion parameters such as standard deviation, interquartile range, range (maximum minus minimum), mean difference, absolute median difference, and mean absolute deviation. The sequence of dispersion values can be analyzed, and The sequence of dispersion values is used to detect the removal of the overlay layer (step 〇4〇g). 37 201213050 Fig. 16 is a graph showing the standard deviation of the spectrum as a function of the polishing time. The difference between the groups is calculated for each standard. The quasi-bias). Therefore, each plot point 1602 in the graph is the difference between the groups of the spectra. The quasi-bias is displayed, and the spectrum is collected at a given scan of the optical monitoring system. As shown, during the first period (6), the standard deviation value remains a relatively small value. However, after the period 1610, the standard The bias values become larger and more dispersed. Without being limited to any particular theory, a thick barrier layer can tend to dominate the reflection spectrum, thereby masking the difference in thickness between the barrier layer itself and any underlying layer. The polished barrier layer becomes thinner or completely removed' and the reflection spectrum becomes more sensitive to changes in the thickness of the underlying layer. Therefore, when the barrier layer is removed, the dispersion of the spectrum tends to increase. When the layer is cleared, various algorithms can be used to change the behavior of the dispersion value. For example, the sequence of dispersion values can be compared with the threshold value, and if the value of the dispersion exceeds the threshold value, the signal is generated. Indicating the coverage month division as another example, calculating the slope of the portion of the sequence of the dispersion values inside the moving window' and if the slope exceeds the threshold value, generating a signal indicating that the overlay layer is cleared As part of the algorithm for increasing the dispersion of dispersion, the sequence of dispersion values can be processed by a transitional wave (eg, low pass or band filter) to remove high frequency noise. Examples of low pass filters Including moving average ferrites and Butterworth filters. Although the above discussion focuses on the detection of the removal of the barrier layer, this technique can be used in other contexts to detect the removal of the overlay. 38 201213050 The removal of the overlay layer '! interlayer dielectric (ILD) in another type of semiconductor process using dielectric layer stacking, or the removal of thin metal layers on the dielectric layer. In addition to being used as a trigger for initiating feature tracking as discussed above, this technique for detecting the removal of an overlay can be used for other purposes in the polishing operation, for example, as the endpoint signal itself to trigger a timer to The lower layer is ground after exposure for a predetermined duration, or as a trigger to modify the grinding parameters (eg, 'change the carrier head pressure or slurry composition after the lower layer exposure). In addition, the above discussion assumes a rotating platform having an optical endpoint monitor in which the optical endpoint monitor is mounted' but the system is adaptable to other types of relative motion between the monitoring system and the substrate. For example, in some implementations (e.g., 'orbital motion), source k passes through different locations on the substrate, but the source does not cross the edge of the substrate. In such cases, the collected spectra can still be classified, for example, the spectra can be collected at a certain frequency, and the collected spectra over a period of time can be considered as part of a group. This period should be sufficiently long to collect five to twenty spectra per group. The term substrate as used in this specification may include, for example, a product substrate (e.g., a product substrate comprising a plurality of memory or processor dies), a test substrate, a bare substrate, and a gated substrate. The substrate can be at various stages of integrated circuit fabrication, for example, the substrate can be a bare wafer, or the substrate can include one or more deposited and/or patterned layers. The term substrate can include disc-shaped and rectangular sheets. 39 201213050 Embodiments of the invention and all functional operations described in this specification can be implemented in digital electronic circuits or in computer software, firmware or hardware. The computer software, body or hardware includes the disclosure herein. The structural members and the structural equivalents of the structural members, or the embodiments and all functional operations can be implemented in a combination of digital electronic circuits and computer software, or a hardware or a hardware. Embodiments of the present invention may be implemented as one or more computer program products 'that is, tangibly embodied in a machine readable storage medium - or more computer programs, which are processed by a data processing device , a computer or multiple processors or computers) to perform or control the operation of the data processing device. A computer program (also known as a program, software, software application or code) may include any form of programming language that compiles a language or disambiguates... and the computer program can be deployed in any form. 'These forms are included as stand-alone programs or As a module, component, subroutine, or other unit suitable for use in a computing environment. The computer program does not necessarily correspond to (4). The program may be stored in a portion of the file in which other programs or materials are stored, in a single building dedicated to the program, or in multiple identical-type documents (eg, storage - or more modules) or programs. The part of the code is in the standard). Computer programs can be deployed to execute on one site or on a single computer or multiple computers that are spread across multiple sites and interconnected by a communication network. The processes and logic flows described in this specification can be performed by one or more computer(s)- or more programmable processors to perform functions by operating on input data and generating output. The process and logic flow can also be performed by dedicated logic circuits, and the device can also be implemented as dedicated logic 201213050 circuits, such as 'field programmable gate arrays (field pr〇grammabieqing (10) 叮; FPGA) or special application integrated circuits (application- Specific integrated circuit; ASIC) ° The grinding method can be applied to a variety of grinding systems. The grinding wheel or carrier head or both can be moved to provide relative motion between the grinding surface and the substrate. The platform is not rotating, but can be orbited. The polishing pad can be a circular (or some other shape) that is fastened to the platform. Some of the endpoint detection systems can be applied to linear grinding systems, such as ' Where the abrasive crucible is a linearly moving continuous belt or reel belt. The abrasive layer can be standard (eg 'polyurethane with or without fillers'): abrasive material, soft material or fixed abrasive material.浯, it should be understood that the polishing surface and the substrate can be held in a vertical direction or in some other direction. A specific embodiment of the present invention has been described. The examples are within the scope of the following patents. [Simplified illustration of the drawings] Figures 1A and 1B are schematic cross-sectional views of the substrate before and after polishing. Figure 2 is a schematic cross-section of an example of a grinding apparatus. Fig. 3 is a schematic plan view of a substrate having a plurality of regions. Fig. 4 is a plan view of the polishing pad and showing the position where the in-situ measurement is performed on the substrate. 41 201213050 The graph does not measure the spectra of the in-situ optical monitoring system. Figure 6 illustrates the library of reference spectra. Figure 7 illustrates the index traces. Figure 8 illustrates the index traces, which have the index values Fitting the line I·sheng function 'The index values are collected after the detection of the cover layer is cleared.” Figure 9 is a flow chart of an exemplary process for manufacturing the substrate and feeding the end point of the test. A plurality of index traces. Figure 11 illustrates the calculation of the target index based on a number of desired slopes. Time for a plurality of adjustable regions. The index trace of the reference region at that time is illustrated in Figure 12. Time for the end point At this time, the index trace of the reference area reaches the target index. Figure 13 is a flow chart of an exemplary process for adjusting the polishing rate of a plurality of regions in a plurality of substrates such that the plurality of The regions have approximately the same thickness at the target time. Figure 14 illustrates a flow chart for detecting the removal of the cover layer. Figure 15A illustrates the spectral chart for a single scan period at the beginning of the grinding. A spectrum chart showing during a single scan close to the barrier clearing. ' Figure 16 shows a graph of the standard deviation of the spectrum as a function of | 42 ! ! 42 ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! The same component symbols and names indicate the same components.
43 S 201213050 【主要元件符號說明】 10 基板 10a 基板 12 第一層 14 額外層 16 第二層 18 導電材料 100 研磨設備 108 視窗 110 研磨墊 112 外部研磨層 114 背層 118 立體視窗 120 平臺 121 馬達 124 驅動轴 125 中心軸 128 凹槽 129 旋轉搞合益 130 埠 132 研磨液體 140 承載頭 142 固定環 144 撓性膜 146a 腔室 146b 腔室 146c 腔室 148a 中心區域 148b 區域 148c 區域 150 支撐結構/回轉料架 152 驅動軸 154 承載頭旋轉馬達 155 中心轴 160 原位光學監控系統 162 光源 164 光偵測器 166 電路 168 光學磁頭 170 分叉光纖 172 幹線 174 支線 176 支線 190 控制器 201 位置 201a 點 201b 點 201c 點 201d 點 201e 點 201f 點 2〇lg 點 201h 點 201i 點 2〇lj 點 44 201213050 201k 點 202b 光譜 202a 203b 光譜 203a 204a 光譜 204 210 索引跡線/索引值之戾 列/第一序列 序 204b 212 214 線/第一線/原始線性函 數 220 222 索引值 224 230 索引跡線/索引值之第 三序列 232 234 第三線 300 310 資料庫 314 320 參考光譜 330 340 記錄 350 902 步驟 903 904 步驟 906 908 步驟 908a 908b 步驟 910 912 步驟 914 916 步驟 918 1300 流程圖 1302 1304 步驟 1306 1308 步驟 1310 1312 步驟 13 14 1316 步驟 13 18 1330 步驟 1400 1402 步驟 1404 1406 步驟 1408 1500a 光譜 1500b 1600 圖表 1602 1610 第—時段/時段 EI2 EI3 預期索引 IC ID 索引差 IT S' 斜率 SD2 光譜 光譜 箭頭 光譜 索引值 第二序列/索引跡線 第二線 索引值 量測光譜 線性函數 索引值 資料庫 步驟 步驟 步驟 步驟 步驟 步驟 步驟 步驟 步驟 步驟 步驟 方法 步驟 步驟 光譜 標繪點 預期索引 索弓丨值 目標索引 所要斜率 45 201213050 SD3 所要斜率 SD4 TC 時間 T0 TE 預期終點時間/預估終 • 點時間/終點時間 所要斜率 時間 4643 S 201213050 [Major component symbol description] 10 Substrate 10a Substrate 12 First layer 14 Extra layer 16 Second layer 18 Conductive material 100 Grinding device 108 Window 110 Abrasive pad 112 External polishing layer 114 Back layer 118 Stereo window 120 Platform 121 Motor 124 Drive Shaft 125 Center Shaft 128 Groove 129 Rotate Fit 130 埠132 Grinding Liquid 140 Carrier Head 142 Retaining Ring 144 Flexible Film 146a Chamber 146b Chamber 146c Chamber 148a Center Area 148b Area 148c Area 150 Support Structure / Rotating Material Rack 152 drive shaft 154 carrier head rotation motor 155 center axis 160 in-situ optical monitoring system 162 light source 164 photodetector 166 circuit 168 optical head 170 bifurcated fiber 172 trunk 174 branch 176 branch line 190 controller 201 position 201a point 201b point 201c Point 201d Point 201e Point 201f Point 2〇lg Point 201h Point 201i Point 2〇lj Point 44 201213050 201k Point 202b Spectrum 202a 203b Spectrum 203a 204a Spectrum 204 210 Index Trace/Index Value Array/First Sequence Order 204b 212 214 Line/first line/original linear function 220 222 index 224 230 Third sequence of index traces/index values 232 234 Third line 300 310 Library 314 320 Reference spectrum 330 340 Record 350 902 Step 903 904 Step 906 908 Step 908a 908b Step 910 912 Step 914 916 Step 918 1300 Flowchart 1302 1304 Step 1306 1308 Step 1310 1312 Step 13 14 1316 Step 13 18 1330 Step 1400 1402 Step 1404 1406 Step 1408 1500a Spectrum 1500b 1600 Chart 1602 1610 Period - Period / Period EI2 EI3 Expected Index IC ID Index Difference IT S' Slope SD2 Spectral Spectrum Arrow spectral index value second sequence / index trace second line index value measurement spectrum linear function index value database step step step step step step step step step method method step step spectral plot point expected index cable bow target Slope of the index 45 201213050 SD3 Desired slope SD4 TC Time T0 TE Expected end time / estimated end • Point time / end time required slope time 46