201223702 六、發明說明： 【發明所屬之技術領域】 本發明係關於如在化學機械研磨基板期間進行光學監 測。 風 【先前技術】 積體電路-般係藉由相繼沉積導電層、半導體層或絕 緣層至石夕晶圓而形成於基板上。一製造步驟涉及沉積填 料層至非平面表面及平坦化填料層。就某些應用而言， 乃持續平坦化填料層直到露出圖案化層的頂表面為止。 導電填料層例如可沉積在圖f化絕緣層上，以填充絕緣 層内的溝渠或孔洞。平坦化後，部分導電層留在絕緣層 的凸起圖案之間而構成通孔、插栓和連線，以提供基板 上之薄膜電路間的傳導路徑。就其他諸如氧化物研磨等 應用而言，乃持續平坦化填料層直㈣平面上留下預定 厚度為止。此外，光微影技術通常需要平坦化基板表面。 化學機械研磨（chemicalmechanicalp〇Hshing ; cMp) 為公5忍的平坦化法之一。平坦化法一般需將基板裝設在 承载頭上°露出的基板表面通常抵靠著旋轉研磨墊。承 載頭提供可控制負載至基板上，使基板推抵著研磨整。 研磨液（如具磨粒的研磨漿）一般供應至研磨墊表面。 CMP的一困難點在於判斷研磨製程何時完成，即是否 已將基板層平坦化成預定平坦度或厚度、或者是否已移 201223702 除預定量的材料。 磨塾條件、研料：:刀始厚度、研磨漿組成、研 載都會造成不同的材對速度、和施予基板的負 研磨終點所需的時Γ 率。該等差異將導致達到 判斷研磨終點。” ’不能只根據研磨時間 在—些系統中，可 如透過研磨塾中的窗口、/間原位光學監测基板，例 w, ^ ^ # . J …、現有光學監測技術無法滿足 丰導體裝置製造業者曰益嚴苛的要求。 【發明内容】 :—光予監測製程中，比較如在研磨製程期間原位 :光譜和參照光譜圖庫，以尋找最佳匹配參照光 °曰哥找最佳匹配者的—技術為計算測量光譜與圖庫中 ;各參照光譜⑽平方差總和；平方差總和最小的參照光 谱為最佳匹配者。然對研磨某些基板而言，如在同一平 臺上移除多個介電層，匹配演算法並不可靠。不褐限於 任何特定理論，平方差總和深受光譜的波峰位置影響， 下層厚度變化則會造成波♦位置 術(:如交又比對法)來尋找最佳匹配參照::用= 少或避免該等問題發生。 在-態樣中’控制研磨的方法包括儲存具有複數個參 照光譜的圖庫、研磨基板、在研磨期間，測量基板的一 序列光譜、利用除平方差總和外的匹配技術，就該序列 201223702 光谱中的母一測莖光譜·’尋找最佳匹配參照光譜，以產 生-.序列最佳匹配參照光譜、以及依據該序列最佳匹配 參照光譜，判斷研磨終點或調整研磨速率的至少其一。 實施方式可包括一或更多下列特徵。尋找最佳匹配參 照光譜可包括交又比對測量光譜和圖庫中複數個參照光 譜的二或更多參照光譜的每一參照光譜，以及選擇與測 量光譜有最大相關性的參照光譜做為最佳匹配參照光 譜。複數個參照光譜中的每一參照光譜可具儲存關聯指 數值，並可決定該序列最佳匹配參照光譜中各最佳匹配 參照光譜的關聯指數值，以產生一序列指數值，以及使 一函數配適該序列指數值。當線性函數匹配或超過目標 指數時，可停止研磨。基板可包括覆蓋第一層的第二層， 第一層具有不同於第二層的組成1第二層可為阻障層， 而第-層可為介電層。阻障層可為氮化组或氮化欽，而 介t:層可為碳摻雜二氧化石夕，或者介電層可由四乙氧基 石夕坑組成。函數可配適_部分的該序列指數值，該部分 序列指數值對應偵測到露出第—層後測量的光讀。尋找 最佳匹配參照光譜可包括加總測量光ϋ與圖庫中複數個 參照光譜的二或更多參照光譜的每—參照光譜間的歐式 (Euchdean )肖量距冑，以及選擇總和最小的參照光譜 U為最佳匹配參照光譜。尋找最佳匹配參照光謹可包括 加=測量光ϋ與圖庫中複數個參照光譜的二或更多參照 、'3的每|照光譜間的導數差，以及選擇總和最小的 —、、'光-曰做為最佳匹配參照光譜。測量基板的該序列光 201223702 譜可包括使感測器多次掃掠基板各處。該序列光譜中的 每-光譜可對應感測器多次掃掠中的單—掃掠。基板可 包括複數個區域’且各區域的研磨速率可由獨立變動的 研磨參數個別控制。在研磨期間，測量各區域的一序列 光譜。可就各區域的該序列光譜中的每一測量光譜，交 又比對測量光譜和圖庫中複數個參照光譜的二或更多參 照光譜的每一參照光譜’以及從圖庫中選擇與測量光講 有最大相關性的參照光譜做為最佳匹配參照光譜，以產 生序歹J最佳匹配參照光譜。5調整至少一區域的研磨 參數’以調整至少一區域的研磨速率，使複數個區域於 研磨終點時有比在無調整的情況下更小的厚度差異。 在另一態樣中，實體收錄於機器可讀取儲存裝置的電 腦程式產品包括執行此方法的指令。 在又態樣中，研磨設備包括用以支承研磨墊的支撐 件用以支承基板使基板抵著研磨塾的承載頭、用以在 承载頭與支撐件間產生相對運動而研磨基板的馬達、光 學監測系 '统，光學監測系統於研磨基板時測量基板的- 序列光°曰、以及控制器。控制器配置以儲存具有複數個 參照光譜的圖庫，複數個參照光譜中的每一參照光譜具 有儲存關聯指數值，並利用除平方差總和外的匹配技 術’就該序列光譜中的每-測量光譜’尋找最佳匹配參 照先譜，以產生一序列最佳匹配參照光譜，以及依據該 序列最佳匹配參照光譜，判斷研磨終點或調整研磨速率 的至少其一。 201223702 實施方式可選擇性包括—或更多下列優點。匹配技術 杈不受測量光譜的波峰位置影響，故可降低對下層厚度 變化的敏感度。終點系統偵測預定研磨終點的可靠度將 提升’且晶圓内和晶圓間厚度不均勻度（WIWNu和 WTWNU)將減少。 本發明的一或更多實施例將配合附圖詳述於後。本發 明的其他特徵、態樣和優點在參閱實施方式說明、圖式 與申請專利範圍後將變得更清楚易懂。 【實施方式】 光學監測技術係於研磨期間測量自基板反射的光譜及 從圖庫中識別匹配參照光譜。匹配參照光譜提供一序列 指數值，且一函數（如直線）配適該序列指數值。將函 數投影到目標值可用來判斷終點或改變研磨速率。 對研磨某些類型的基板而言，如在同一平臺上研磨基 板的多層材料時’潛在問題在於從圖庫中尋找最佳匹配 光譜的技術（如選擇平方差總和最小的參照光譜）並不 可靠。不侷限於任何特定理論，平方差總和深受光譜的 波冷位置影響’下層厚度變化則會造成波峰位置位移。 但若採用另一技術（如交叉比對法）來尋找最佳匹配參 照光譜，則可減少或避免該等問題發生。 舉例來說，參照第1 A圖，基板1〇包括由第一介電材 料組成的圖案化第一層12 (此層亦稱為下層），第一介 201223702 電材料例如為低介電常數（k)材料，例如碳摻雜二氧化 矽，如Black Diamond™ (取自應用材料公司（AppIied Materials，Inc.))或 C〇ralTM(取自諾發系統公司（201223702 VI. Description of the Invention: TECHNICAL FIELD OF THE INVENTION The present invention relates to optical monitoring as during chemical mechanical polishing of a substrate. Wind [Prior Art] The integrated circuit is formed on the substrate by successively depositing a conductive layer, a semiconductor layer or an insulating layer to the Si Xi wafer. A manufacturing step involves depositing a filler layer to the non-planar surface and planarizing the filler layer. For some applications, the layer of filler is continuously planarized until the top surface of the patterned layer is exposed. A layer of electrically conductive filler can be deposited, for example, on the insulating layer to fill the trenches or holes in the insulating layer. After planarization, a portion of the conductive layer remains between the raised patterns of the insulating layer to form vias, plugs, and wires to provide a conductive path between the thin film circuits on the substrate. For other applications such as oxide milling, it is continued to flatten the filler layer to a predetermined thickness in the straight (iv) plane. In addition, photolithography typically requires planarization of the substrate surface. Chemical mechanical grinding (chemical mechanical p〇Hshing; cMp) is one of the flattening methods of the public five. The planarization method generally requires the substrate to be mounted on the carrier head. The exposed substrate surface generally abuts the rotating polishing pad. The carrier head provides a controllable load on the substrate to push the substrate against the grinding. A slurry (such as abrasive slurry) is typically supplied to the surface of the polishing pad. A difficulty with CMP is to determine when the polishing process is completed, i.e., whether the substrate layer has been planarized to a predetermined flatness or thickness, or whether a predetermined amount of material has been removed 201223702. Grinding conditions, grinding materials:: the thickness of the knife, the composition of the slurry, and the grinding load will result in different material speeds and the time required for the negative grinding end of the substrate. These differences will result in the determination of the end of the grinding. "Can not only be based on the grinding time in some systems, such as through the window in the grinding crucible, / in-situ optical monitoring of the substrate, such as w, ^ ^ # . J ..., the existing optical monitoring technology can not meet the Feng conductor device The requirements of the manufacturers are extremely demanding. [Invention]: - In the light monitoring process, compare the in-situ: spectral and reference spectrum libraries during the grinding process to find the best matching reference light. The technique is to calculate the measured spectrum and the library; the sum of the squared differences of each reference spectrum (10); the reference spectrum with the smallest sum of square differences is the best match. However, for grinding some substrates, such as removing more on the same platform. The dielectric layer, the matching algorithm is not reliable. No brown is limited to any particular theory, the sum of the squared differences is deeply affected by the peak position of the spectrum, and the thickness variation of the lower layer will cause the wave position (: such as crossover and comparison) Finding the best matching reference:: Use less = or avoid such problems. In the - state method, 'control grinding method includes storing a library with multiple reference spectra, grinding the substrate, during grinding Measuring a sequence of spectra of the substrate, using a matching technique other than the sum of squared differences, looking for the best matching reference spectrum in the sequence of the 201223702 spectrum, to find the best matching reference spectrum, and Determining at least one of the polishing endpoint or adjusting the polishing rate based on the best matching reference spectrum of the sequence. Embodiments may include one or more of the following features. Finding the best matching reference spectrum may include cross-comparison measurement spectra and complex numbers in the library Each reference spectrum of two or more reference spectra of a reference spectrum, and a reference spectrum that has the greatest correlation with the measured spectrum is selected as the best matching reference spectrum. Each of the plurality of reference spectra may have a storage association An index value, and may determine that the sequence best matches the correlation index value of each of the best matching reference spectra in the reference spectrum to produce a sequence of index values, and to have a function fit the sequence index value. When the linear function matches or exceeds the target When indexing, the grinding can be stopped. The substrate can include a second layer covering the first layer, the first layer having a different The second layer of composition 1 may be a barrier layer, and the first layer may be a dielectric layer. The barrier layer may be a nitrided group or a nitrided layer, and the t: layer may be a carbon doped dioxide. Shi Xi, or the dielectric layer may be composed of tetraethoxy zea pit. The function may be adapted to the _ part of the sequence index value, the partial sequence index value corresponding to the optical reading measured after detecting the first layer is exposed. The matching reference spectrum may include a total of two or more reference spectra of the plurality of reference spectra in the library, and a European-style (Euchdean) distance between the reference spectra, and a reference spectrum U with the smallest sum selected is the most Preferably, the reference spectrum is matched. Finding the best matching reference light may include adding or measuring two or more references of the reference pupil and the plurality of reference spectra in the library, the derivative difference between each of the spectra of '3, and the smallest sum of the selections. —,, 'Light-曰 as the best matching reference spectrum. Measuring the sequence of light of the substrate 201223702 spectrum can include having the sensor sweep across the substrate multiple times. Each-spectrum in the sequence spectrum corresponds to a single-sweep in multiple sweeps of the sensor. The substrate can include a plurality of regions' and the polishing rate of each region can be individually controlled by independently varying grinding parameters. During the grinding, a sequence of spectra of each region was measured. For each measurement spectrum in the sequence spectrum of each region, compare and compare each reference spectrum of two or more reference spectra of the plurality of reference spectra in the measurement spectrum and the library, and select and measure light from the library. The reference spectrum with the greatest correlation is used as the best matching reference spectrum to produce the sequence 最佳J best matching reference spectrum. 5 adjusting the polishing parameter of at least one region to adjust the polishing rate of at least one region such that the plurality of regions have a smaller thickness difference at the end of the polishing than in the absence of adjustment. In another aspect, the computer program product embodied by the machine readable storage device includes instructions to perform the method. In another aspect, the polishing apparatus includes a support member for supporting the polishing pad for supporting the substrate to support the substrate against the grinding head, a motor for grinding the substrate between the carrier head and the support member, and optical The monitoring system, the optical monitoring system measures the substrate-sequence light when the substrate is polished, and the 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 and utilizing a matching technique other than the sum of the square differences' for each measurement spectrum in the sequence spectrum 'Finding the best match reference profiling to produce a sequence of best matching reference spectra, and determining the polishing endpoint or adjusting the polishing rate based on the best matching reference spectrum for the sequence. The 201223702 embodiment may optionally include - or more of the following advantages. The matching technique is not affected by the peak position of the measured spectrum, so it reduces the sensitivity to changes in the thickness of the underlying layer. The reliability of the end point system to detect the intended end of the grinding will increase' and the thickness unevenness (WIWNu and WTWNU) within and between wafers will decrease. One or more embodiments of the present invention will be described in detail later with reference to the accompanying drawings. Other features, aspects, and advantages of the present invention will become more apparent from the description and appended claims. [Embodiment] The optical monitoring technique measures the spectrum reflected from the substrate during the grinding and identifies the matching reference spectrum from the library. The matching reference spectrum provides a sequence of index values, and a function (such as a straight line) fits the sequence index value. Projecting the function to the target value can be used to determine the end point or change the grinding rate. For grinding certain types of substrates, such as when polishing multiple layers of substrate on the same platform, the potential problem is that the technique of finding the best matching spectrum from the library (such as selecting the reference spectrum with the smallest sum of squared differences) is not reliable. Not limited to any particular theory, the sum of the squared differences is deeply affected by the wave-cooling position of the spectrum. The change in the thickness of the lower layer causes the position of the peak to shift. However, if another technique (such as cross-matching) is used to find the best matching reference spectrum, then these problems can be reduced or avoided. For example, referring to FIG. 1A, the substrate 1A includes a patterned first layer 12 (this layer is also referred to as a lower layer) composed of a first dielectric material, and the first dielectric material 201223702 is, for example, a low dielectric constant ( k) Materials such as carbon doped cerium oxide, such as Black DiamondTM (from AppIied Materials, Inc.) or C〇ralTM (taken from Novo Systems)
Systems，Inc·))。第—層12上為由不同的第二介電材料 組成的第二層16 (此層亦稱為上層），第二介電材料例 如為阻障層，例如氮化物，如氮化组或氮化鈦。由不同 於第一與第二介電材料的另一介電材料組成的一或更多 附加層U選擇性設在第一層與第二層間，另一介電材料 例如為低k覆蓋材料’例如由四乙氧基石夕燒（te〇s )组 成的材料。第一層12和一或更多附加層U —起在第二 層底下提供層堆疊結構。第二層（和第_層圖案提供的 溝槽）上為導電材肖18 ’導電材料例如為金屬，如銅。 化學機械研磨可用來平坦化基板，直到露出由第一介 電材料組成的第一層為止。例如，參照第ib圖.，平坦化 後，部分導電材料18留在第一層12的凸起圖案之間而 構成通孔等。此外，有時期望移除第-介電材料，直到 留下目標厚度或移除目標材料量為止。 -研磨方法係在第一研磨墊上研磨導電材料，至少到 露出第二層（如阻障層）為止。此外，可在如過度研磨 步驟期間’以第一研磨墊移除第二層的部分厚度。接著 將基板傳送到第二研磨塾，在此完全移除第二層（如阻 障層)，及移除底下第一層(如低k介電質）的部分厚产。 此外，可在相同研磨操作下，以第二研磨墊移除第j 與第二層間的—或更多附加層（若有）（如覆蓋層）。 201223702 第2圖圖示研磨設備100的實例。研磨設備1 〇〇包括 旋轉式盤狀平臺I20，平臺12〇上設置研磨墊no。平臺 可操作以繞著軸125轉動。例如，馬達121可轉動傳動 軸124，進而轉動平臺120。研磨墊110可為具外部研磨 層112與軟背層114的雙層研磨墊。 研磨設備100可包括埠口 130，用以分配研磨液132 (如研磨漿）至研磨墊110上。研磨設備還可包括研磨 墊調理器，用以摩擦研磨墊110’使研磨墊11〇維持呈一 致的研磨狀態。 研磨設備100包括一或更多承載頭140。各承載頭140 可操作以支承基板10,使基板抵著研磨墊110〇每一承 載頭140可個別控制各基板相關的研磨參數，例如壓力。 特別地，每一承載頭140可包括把基板1 〇保持在彈性 膜144下方的定位環142。各承載頭140亦包括由膜定 義的複數個個別控制加壓腔室，例如三個腔室 146a-146C，腔室146a-146c可個別施加控制壓力至彈性 膜144和基板1〇上的相關區域148a_148《參見第3圖）。 參照第3圖，中心區域148a為實質圓形，且其餘區域 148b-148c為圍繞中心區域148a的同心環狀區域。雖為 便於說明’第2及3圖僅繪示三個腔室，然其當可具一 或二個腔室、或四或更多腔室，例如五個腔室。 回溯第2圖，各承載頭140懸吊於如旋轉料架的支撐 結構150 ’且承栽頭14〇由傳動軸152連接至承載頭旋 轉馬達154,讓承載頭繞著車由155轉動。視情況而定， 10 201223702 各承载頭14〇可於旋轉料架15〇的滑件上橫向擺動、或 由方疋轉料架自轉振盪。操作時，平臺繞著平臺中心軸1 25 轉動，各承載頭繞著承载頭中心軸丨5 5轉動並橫向移動 越過研磨墊的頂表面。 雖然圖式僅繪示—個承載頭140，但其當可設置更多 承載碩來支承附加基板，以更有效利用研磨墊】丨〇的表 面積。故適於在同時研磨製程中支承基板的承載頭組件 數量至少部分取決於研磨墊11〇的表面積。 研磨设備還包括原位光學監測系統丨6〇，例如光譜監 測系統，光學監測系統丨6〇可用於判斷是否需調整研磨 速率或研磨速率調節器，此將說明於後。藉由設置口孔 (即穿過研磨墊的孔洞）或實心窗口 i丨8可提供通過研 磨墊的光學入口。實心窗口 118可固定於研磨墊11〇，例 如像填充研磨墊之口孔的插栓般鑄造或黏接固定於研磨 墊，但在一些實施方式中，實心窗口可托在平臺12〇上 並伸進研磨墊的口孔。 光學監測系統160可包括光源162、光偵測器164和 電路166’電路166發送及接收遠端控制器19〇(如電腦） 與光源162和光偵測器164間的訊號。一或更多光纖用 來把光源162的光傳遞到研磨墊的光學入口，及將自基 板10反射的光傳遞到偵測器！ 64。例如，雙又光纖i 7〇 可用來把光源162的光傳遞到基板1〇及傳回到偵測器 164。雙叉光纖包括鄰近光學入口設置的主幹172、和分 別連接至光源162與偵測器丨64的分支1 74、1 76。 201223702 在一些實施方式中，平臺的頂表面可包括凹槽128， 凹槽128内配設光學頭168，用以支承雙叉光纖的主幹 172 —端。光學頭168可包括調整主幹172頂部與實心 窗口 118間的垂直距離的機構。 電路166 出可為數位電子訊號，訊號通過傳動轴 124的旋轉式聯結器129 (如爽環）而至光學監測系統的 控制器19G。同樣地’可響應數位電子訊號的控制指令 而打開或關閉光源，訊號從控制@ 19G經由旋轉式聯結 器129而至光學監測系統16〇。或者，電路166可利用 無線訊號與控制器19〇通信。 光源162可操作以發射白光。在一實施方式中，發射 的白光包括波長為2〇〇至刚奈米（nm)的光。適合的 光源為氤氣燈或氙汞燈。 光谓測器164可為光譜儀。光譜儀為測量部分電磁波 谱的光強度的光學儀器。適合的光譜儀為光柵光胃H 光譜儀的典型輸出為隨波長（或頻率）變化的光強度。 如上所述，光源162和光谓測_ 164連接至運算 (如控制器m)，以控制光源162和光偵測器16 作及接收光源162和光偵測器164的訊號。運算裝 包括設於研磨設備附近的微處理器，例如可程式電腦。 至於控制方面，運|歩罢ml π ^ u 延-裝置例如可同步化光源開啟和 120轉動。 室 162和 °在此 在-些實施方式中，原位監測系統16〇的 偵測器164安裝於平臺12〇且偕同平臺 12 201223702 情況下，平臺的動作將促使感測器掃過各基板。特別地， 隨著平臺120旋轉，控制器ι9〇將促使光源162恰在光 學入口通過基板1〇下方前開始發射一連串閃光且緊接 在通過後結束。或者，運算裝置可促使光源162恰在各 基板1G通過光學人口前開始連續發光且緊接在通過後 結束。在任一情況下，可求得採樣週期出自偵測器的訊 號積分’而以採樣頻率產生光譜測量。 紅作時，控制器i 9〇例如可接收在光源的特定閃光或 偵測态的時段内傳達光偵測器接收的光譜資訊的訊號。 故此光譜係研磨期間的原位測量光譜。 如第4圖所示，若偵測器安裝於平臺，則當窗口 1⑽ 因平臺旋轉（如箭頭2G4所示）而於承載頭下方移動時， 以採樣頻率進行光谱測量的光學監測系統料弧形越過 基板1〇的位置201進行光譜測量。例如，各點20 la_2〇lk 代表監測系統進行光譜測量的位S (點數僅為舉例說 月，其δ可視採樣頻率進行比所繪點數更多或更少次測 量）抓樣頻率可選擇以於每次掃掠窗口 i 〇8時，收集5 至2〇個光譜。例如’採樣週期可為3至1〇〇毫秒。 如圖所示’平臺旋轉-周’即可取得基板10上不同徑 向位置的光譜。即’ -些光譜係取自較靠近基板10中心 的4置一些光譜則取自較靠近邊緣的位置。故在光學 監測系統進行的任何特定掃描基板各處方φ，依據時 序、馬達編碼器資訊和基板邊緣及/或定位環的光谓測， 幻器190可計算掃描所得各測量光譜的徑向位置（相 13 201223702 。描基板的中’"）。研磨系統還可包括旋轉式位置感測 :，例如附接平臺邊緣的凸緣’凸緣將通過固定式光遮 斷ηχ提供額外資料來判斷測量光譜係出自哪個基板 和基板位置。是以控制器可使各種測量光譜與基板1〇上 的控制區域mb-mc (參見第2圖）產生關聯。在一些 貫施方式中’光譜測量時間可代替精確計算徑向位置。 /各區域而言’平臺經多次旋轉後，將隨時間推移取 :一·序列光譜。不減於任何特定理論，隨著研磨進行 平臺經多次旋轉、非單-掃掠基板各處時），因最外 層厚度改變，將逐步形成自基板1〇反射的光譜而產生一 時間變化的光譜。再者’特定層堆疊結構厚度將 呈現特定光譜。 在一些實施方式令，控制器（如運算裝置）經程式化 =比較測量光譜和多個參照光譜，及決㈣個參照光譜 =最佳匹配者。特別地，控制器可經程式化以比較各 、域的一序列測量光譜中的每一光譜和多個參照光譜， 、産生各區域的一序列最佳匹配參照光譜。 .在此’參照光譜係研磨基板前產生的預定光譜。假設 實際研磨料㈣料率，料㈣譜與代表 :磨製程期間預期出現光譜的時間數值有預定關聯性， 操作前所定義者。或者或此外，參照光譜可與基 質數值（如最外層厚度）有預定關聯性。 參照光譜可憑經驗產生，例如藉由測量測試基板（如 一初始層厚度的測試基板）的光譜。例如，為產生 14 201223702 複數個參照㈣，可使用與將用於研磨裝置晶圓—樣的 研磨參數來研磨設置基板，同時收集-序列光譜。就各 先譜記錄代表研磨製程期間收集光譜的時間數值。例 如’此數值可為經過時間吱平喜鉍 卞间4十$轉數。可過度研磨基板， 即研磨超過預定厚度，以於揸 於達目標厚度時取得自基板反 射的光譜。 為使各光譜與基板性質數值（如最外層厚度）產生關 聯’可在研磨前，於測量站測量「設置」基板的最初光 譜和性質，設置基板具有和產品基板—樣的圖案。研磨 後，亦可利用同-測量站或不同測量站測量最終光譜和 性質。最初光譜與最線弁4並卩卩 瑕、九°曰間的先譜性質可依據測量測 試基板光譜的經過時間，由内插決定’例如線性内插。 除了憑經驗決定外，部分或所有參照光譜可依理論計 算’例如利用基板層的光學模型。例如，光學模型可用 於計算特定外層厚度D的參照光譜。如假設以均一研磨 速率移除外層，則可計算代表研磨製程期間收集參照光 4的時間數值。例如，可單純假料始厚度D0和均-研 磨速率R’以計算特定參照光譜的時間Ts (η =⑽_ 。又例如，依據光學模型的厚度D，研磨前與研磨 後厚度D1、D2(或測量站測量的其他厚度）的測量時間 ❹間的線性内指可表示成Μ]·D卿 —D2) 〇 參照第5及6圖，比較測量光譜參見第5圖） 和一或更多圖ho的參照光譜32〇(參見第㈣）。在 15 201223702 此 / ”、、光圖庫係參照光譜集合，參照光譜集合代表 有共同性質的基板。然單—圖庫的共同性質在多個參照 光4圖庫中可不盡相同。例如，兩個不同圖庫可包括代 表具兩種不同下層厚度的基板的參照光譜。就特定參照 光譜圖庫而言，上層厚度變化為光譜強度差異的主因， 而非其他因素（如晶圓圖案差異、下層厚度或層組成) 所致。 精由研磨多個具不同基板性f(如下層厚度或層組成） 的「設置」基板及收集上述光譜，可產生用於不同圖庫 310的參照光譜32〇;得自某一設置基板的光譜可提供第 :圖庫，得自具不同下層厚度的另—設置基板的光譜可 提供第二圖庫。或者或此外，不同圖庫的參照光譜可依 理論計算’例如第一圖庫的光譜可利用具第一下層厚度 的先學模型計算，第二圖庫的光譜可利用具不同下層厚 度的光學模型計算， 在一些實施方式令，每一參照光譜32〇分配到一指數 值33〇°通常，每一圖庫310可包括許多參照光譜320， 例如在基板的預期研磨時間内平臺旋轉一周而得的一或 更多（如僅一個）參照光譜。指數33〇可為代表研磨製 程期間預期發現參照光譜32〇的時間數值，例如數字。 可編定光譜指數’使特定圖庫中的光譜各具獨特的指數 值。編定指數可使指數值依測量測試基板的光譜順序排 序《指數值可選擇隨研磨過程單調改變，例如增加或降 低1別地’參照光譜的指數值可選擇使指數值構成時 16 201223702 間或平臺轉數的線性函數（假設研磨速率係按照模型或 測試基板的研磨速率，測試基板用於產生圖庫中的參照 光4 )。例如，指數值可與測量測試基板的參照光譜或光 學模型中將4現參照光_的平臺轉數呈比例關係（如等 於）”故指妻丈值可為整數。缝數字可代表預期出現相關 光谱的平臺轉數。 參照光譜和參照光譜相關指數值可儲存於參照圖庫。 例如，各參照光譜32〇和各參照光譜相關指數值别可 儲存於資料庫350的記錄340。參照光譜的參照圖庫資 料庫35G可建立在研磨設備的運算裝置的記憶體中。 如上所述，就各基板的每一區域，依據該區域和基板 的序列測量光譜，控制$ 19G可程式化以產生一序列最 佳匹配參照光谱。ϋ由比較測i光譜和特《圖庫的參照 光譜’可決定最佳匹配參照光譜。 在一些實施方式中，計算測量光譜與參照光譜間的平 方差總和，以決定各參照光譜的最佳匹配參照光譜。平 方差總和最小的參照光譜為最適配者。也可採用其他尋 找最佳匹配參照光譜的技術，例如最小絕對差總和。 在一些實施方式中，可利用除平方差總和外的匹配技 術’決定最佳匹配參照光譜。在—實施方式中，就各參 照光譜’計算測量光譜與參照光譜間的交叉相關性，並 選擇有最大相關性的參照光譜做為匹配參照光譜。交又 比對的潛在好處在於較不受光譜橫向位移影響，因而較 不受下層厚度變化影響。為進行交又比對，測量光譜的 201223702 前端和尾端可填入「 如同參照光譜相對…：：提供資料與參照光譜相比， 端可填入等於測。或者’測量光譜的前 端埴 、'置螬前端數值的數值，測量先譜的尾 鳊可填入等於測量光m % ,r 阳尾k數值的數值。快速傅立荦 C Fourier )轉換可 μ '、 4 Η 、用於加快即時應用匹配技術計算交又 相關性的速度。 貫包方式中，計算歐式向量距離總和，例如Ε)= ’如 Xb) [Σλ=λ“ u|Im(x)2 _ Σ“咖，其中 u 至九& 為 波長總數，ImW為測量光譜，叫為參照光譜。在又一 實施方式中’就各參照光譜’計算導數差總和，例如D = 1/(λα λ!?) [Σλ=λ“ u|dlMW/(a_ 叫⑴/叫]，及選擇總和 最小的參照光譜做為匹配參照光譜。 第17圖圖示就具不同TE〇s層厚度的基板，利用交又 比對法和平方差總和法匹配光譜而得指數軌跡（最佳匹 配參照光譜的指數為平臺轉數的函數）的比較結果。資 料係針對具堆疊結構的產品基板產生，堆疊結構包括厚 度1500埃（A)的黑金剛石（Black Dia_d)層、厚度 130A 的 Bl〇k 層和厚度 52〇〇A' 51〇〇A 或 5〇〇〇人的 τε〇§ 層。參照圖庫係針對參照基板產生，參照基板具有厚度 5200A的TEOS層。如軌跡1702所示，其中產品基板和 參照基板具有相同厚度（即5200A)的TEOS層，兩個 指數軌跡互相重疊而無明顯差異。然當產品基板具有厚 度：51 00A的TEOS層且參照基板具有厚度5200A的TEOS 層時’利用平方差總和產生的指數軌跡1 7〇4將稍微偏離 201223702 線性行為。反之，利用交叉比對產生的指數軌跡將與指 數軌跡1702重疊（故無法從圖上看出）。最後，當產品 基板具有厚度5000A的TE0S層且參照基板具有厚度 5200A的TEOS層時，利用平方差總和產生的指數執跡 1 706將明顯偏離線性行為和轨跡【7〇2，利用交叉比對產 生的指數軌跡1708仍大致呈線性且更接近軌跡17〇2。 總之’此結果顯示若下層厚度有異，則利用交又比對來 決疋最佳匹配光譜將產生更匹配理想的轨跡。 可應用來減少電腦處理的方法為限制搜尋部分的匹配 光譜圖庫。圖庫-般包括比研磨基板獲得的光譜還多的 光譜。基板研磨期間，圖庫搜尋宜限制在預定圖庫光譜 範匱卜在-些實施例中，決定研磨基板的當前旋轉指數 N。例如，平臺開始旋轉時，可搜尋圖庫的所有參照光譜 而決定N。故就下次旋轉取得的光譜而言’圖庫搜尋的Systems, Inc.)). The first layer 12 is a second layer 16 composed of a different second dielectric material (this layer is also referred to as an upper layer), and the second dielectric material is, for example, a barrier layer such as a nitride such as a nitride group or nitrogen. Titanium. One or more additional layers U composed of another dielectric material different from the first and second dielectric materials are selectively disposed between the first layer and the second layer, and another dielectric material such as a low-k covering material For example, a material composed of tetraethoxy sulphur (te〇s). The first layer 12 and one or more additional layers U together provide a layer stack structure under the second layer. The second layer (and the trench provided by the _ layer pattern) is a conductive material. The conductive material is, for example, a metal such as copper. Chemical mechanical polishing can be used to planarize the substrate until a first layer of first dielectric material is exposed. For example, referring to the ib diagram, after planarization, a portion of the conductive material 18 remains between the convex patterns of the first layer 12 to constitute a via or the like. In addition, it is sometimes desirable to remove the first dielectric material until the target thickness is left or the target material amount is removed. The grinding method grinds the conductive material on the first polishing pad at least until the second layer (e.g., barrier layer) is exposed. Additionally, a portion of the thickness of the second layer can be removed with the first polishing pad during steps such as over-grinding. The substrate is then transferred to a second polishing pad where the second layer (e.g., barrier layer) is completely removed and a portion of the underlying first layer (e.g., low-k dielectric) is removed. In addition, - or more additional layers (if any) (eg, a cover layer) between the jth and second layers may be removed with a second polishing pad under the same polishing operation. 201223702 FIG. 2 illustrates an example of a grinding apparatus 100. The grinding apparatus 1 〇〇 includes a rotary disk platform I20 on which the polishing pad no is placed. The platform is operable to rotate about the axis 125. For example, the motor 121 can rotate the drive shaft 124 to rotate the platform 120. The polishing pad 110 can be a two-layer polishing pad having an outer abrasive layer 112 and a soft backing layer 114. The grinding apparatus 100 can include a spout 130 for dispensing a slurry 132 (e.g., a slurry) onto the polishing pad 110. The polishing apparatus can also include a polishing pad conditioner for rubbing the polishing pad 110' to maintain the polishing pad 11 in a consistently ground state. The grinding apparatus 100 includes one or more carrier heads 140. Each carrier head 140 is operable to support the substrate 10 such that the substrate abuts the polishing pad 110. Each carrier head 140 can individually control the polishing parameters associated with each substrate, such as pressure. In particular, each carrier head 140 can include a retaining ring 142 that retains the substrate 1〇 beneath the elastomeric film 144. Each carrier head 140 also includes a plurality of individually controlled pressurization chambers defined by a membrane, such as three chambers 146a-146C, which can individually apply control pressure to the elastomeric membrane 144 and associated regions on the substrate 1〇. 148a_148 "See Figure 3). Referring to Figure 3, central region 148a is substantially circular and the remaining regions 148b-148c are concentric annular regions surrounding central region 148a. Although for ease of illustration, Figures 2 and 3 show only three chambers, but they may have one or two chambers, or four or more chambers, for example five chambers. Referring back to Fig. 2, each carrier head 140 is suspended from a support structure 150' such as a rotating rack and the carrier head 14 is coupled to the carrier head rotary motor 154 by a drive shaft 152 for rotation of the carrier head about the vehicle 155. Depending on the situation, 10 201223702 each carrier head 14〇 can be oscillated laterally on the sliding member of the rotating rack 15〇 or by the square turret. In operation, the platform rotates about the platform center axis 1 25 and each carrier head rotates about the carrier head central axis 丨 5 5 and laterally across the top surface of the polishing pad. Although the drawing only shows a carrier head 140, it can be provided with more load-bearing supports to support the additional substrate to more effectively utilize the surface area of the polishing pad. Therefore, the number of carrier head assemblies suitable for supporting the substrate during the simultaneous polishing process depends, at least in part, on the surface area of the polishing pad 11〇. The grinding apparatus also includes an in-situ optical monitoring system 〇6〇, such as a spectral monitoring system, which can be used to determine if the grinding rate or grinding rate adjuster needs to be adjusted, as will be explained later. An optical inlet through the polishing pad can be provided by providing a port (i.e., a hole through the pad) or a solid window i. The solid window 118 can be fixed to the polishing pad 11 , for example, cast or bonded to the polishing pad like a plug filling the opening of the polishing pad, but in some embodiments, the solid window can be supported on the platform 12 并Into the mouth of the polishing pad. The optical monitoring system 160 can include a light source 162, a photodetector 164, and a circuit 166' circuit 166 for transmitting and receiving signals between the remote controller 19 (e.g., a computer) and the light source 162 and the photodetector 164. One or more optical fibers are used to transfer light from source 162 to the optical entrance of the polishing pad and to transmit light reflected from substrate 10 to the detector! 64. For example, the dual fiber optic i 7 can be used to transfer light from the source 162 to the substrate 1 and back to the detector 164. The bifurcated fiber includes a stem 172 disposed adjacent the optical inlet, and branches 1 74, 1 76 coupled to the light source 162 and the detector 丨 64, respectively. 201223702 In some embodiments, the top surface of the platform can include a recess 128 in which is disposed an optical head 168 for supporting the stem 172 end of the bifurcated fiber. The optical head 168 can include a mechanism that adjusts the vertical distance between the top of the stem 172 and the solid window 118. Circuit 166 can be a digital electronic signal that passes through a rotary coupler 129 (e.g., a cool ring) of drive shaft 124 to controller 19G of the optical monitoring system. Similarly, the light source can be turned on or off in response to a control command for the digital electronic signal from control @ 19G to the optical monitoring system 16 via the rotary coupler 129. Alternatively, circuit 166 can communicate with controller 19 using a wireless signal. Light source 162 is operable to emit white light. In one embodiment, the emitted white light comprises light having a wavelength of from 2 Å to just nanometer (nm). Suitable light sources are xenon lamps or mercury lamps. Light detector 164 can be a spectrometer. A spectrometer is an optical instrument that measures the light intensity of a portion of an electromagnetic spectrum. A suitable spectrometer is a light output of a grating light stomach H spectrometer that varies with wavelength (or frequency). As described above, the light source 162 and the optical reference 164 are connected to an operation (e.g., controller m) to control the light source 162 and the photodetector 16 to receive and receive signals from the light source 162 and the photodetector 164. The computing device includes a microprocessor disposed adjacent to the grinding device, such as a programmable computer. As far as the control is concerned, the device can synchronize the light source on and the 120 rotation, for example. Chambers 162 and ° Here, in some embodiments, the detector 164 of the in-situ monitoring system 16A is mounted to the platform 12 and, in the case of the platform 12 201223702, the action of the platform will cause the sensor to sweep across the substrates. In particular, as the platform 120 rotates, the controller ι 9 促使 will cause the light source 162 to begin emitting a series of flashes just before the optical entrance passes under the substrate 1 and immediately after the passage. Alternatively, the arithmetic means may cause the light source 162 to start continuous illumination just before the respective substrates 1G pass through the optical population and end immediately after the passage. In either case, the signal integration from the detector can be determined for the sampling period and the spectral measurement is produced at the sampling frequency. In the case of red, the controller i 9 can, for example, receive a signal conveying the spectral information received by the photodetector during a specific flash or detection period of the light source. Therefore, the spectrum is an in-situ measurement spectrum during grinding. As shown in Fig. 4, if the detector is mounted on the platform, when the window 1 (10) moves under the carrier head due to the rotation of the platform (as indicated by the arrow 2G4), the optical monitoring system for measuring the spectrum at the sampling frequency is curved. Spectral measurements were taken across position 201 of substrate 1〇. For example, each point 20 la_2〇lk represents the position S of the spectrum measurement by the monitoring system (the number of points is only for example, the δ visual sampling frequency is more or less than the number of points drawn). So that each time the window i 〇8 is swept, 5 to 2 spectra are collected. For example, the sampling period can be 3 to 1 millisecond. The spectrum at different radial positions on the substrate 10 can be obtained as shown in the 'platform rotation-circle'. That is, some of the spectra are taken from a position closer to the center of the substrate 10, and some spectra are taken from a position closer to the edge. Therefore, in any specific scanning substrate φ of the optical monitoring system, according to the timing, the motor encoder information and the optical edge of the substrate edge and/or the positioning ring, the magic 190 can calculate the radial position of each measured spectrum of the scanning ( Phase 13 201223702. The middle of the substrate is '"). The grinding system can also include rotational position sensing: for example, the flange' flange attached to the edge of the platform will provide additional information by fixed light interruptions to determine which substrate and substrate position the measurement spectrum is from. It is the controller that correlates the various measurement spectra with the control region mb-mc (see Figure 2) on the substrate 1〇. In some implementations, the spectral measurement time can be used instead of accurately calculating the radial position. / For each region, after the platform has been rotated many times, it will take over time: a sequence spectrum. Without detracting from any particular theory, as the grinding progresses the platform multiple times, non-single-sweeping across the substrate, as the thickness of the outermost layer changes, the spectrum reflected from the substrate 1 will gradually form a time-varying spectrum. Furthermore, the thickness of a particular layer stack structure will exhibit a particular spectrum. In some embodiments, the controller (such as an arithmetic device) is programmed to compare the measured spectrum with a plurality of reference spectra, and to determine (four) reference spectra = best match. In particular, the controller can be programmed to compare each of a series of measured spectra and a plurality of reference spectra for each domain to produce a sequence of best matching reference spectra for each region. Here, the reference spectrum is a predetermined spectrum generated before the substrate is polished. Assume the actual abrasive (4) material rate, material (four) spectrum and representative: the time value of the spectrum expected to appear during the grinding process has a predetermined correlation, as defined before the operation. Alternatively or additionally, the reference spectrum may have a predetermined correlation with the matrix value (e.g., the outermost layer thickness). The reference spectrum can be empirically generated, for example, by measuring the spectrum of a test substrate (e.g., a test substrate of an initial layer thickness). For example, to generate 14 201223702 multiple references (4), the setup substrate can be ground using the polishing parameters that will be used to polish the wafer of the device while collecting the sequence spectra. The pre-spectral records represent the time values of the spectra collected during the grinding process. For example, 'this value can be 40 rpm for the elapsed time. The substrate may be overgrinded, i.e., ground beyond a predetermined thickness to obtain a spectrum of reflection from the substrate when the target thickness is reached. In order to correlate the spectra with substrate property values (e.g., outermost layer thickness), the initial spectrum and properties of the "set" substrate can be measured at the measurement station prior to grinding, and the substrate is provided with a pattern similar to the product substrate. After grinding, the final spectrum and properties can also be measured using a homo-measuring station or a different measuring station. The first spectral property between the initial spectrum and the most 弁4 卩卩 九, 九°曰 can be determined by interpolation based on the elapsed time of the measurement of the spectrum of the test substrate, e.g., linear interpolation. In addition to being empirically determined, some or all of the reference spectra can be calculated theoretically, e.g., using an optical model of the substrate layer. For example, an optical model can be used to calculate a reference spectrum for a particular outer layer thickness D. If it is assumed that the outer layer is removed at a uniform polishing rate, the time value representative of the collection of the reference light 4 during the polishing process can be calculated. For example, the initial thickness D0 and the homo-grinding rate R' can be simply calculated to calculate the time Ts of a specific reference spectrum (η = (10)_. For another example, according to the thickness D of the optical model, the thickness D1, D2 before and after the grinding (or The measurement time of the other thickness measured by the measuring station) can be expressed as Μ··D Qing—D2) 〇 Refer to Figures 5 and 6, compare the measured spectra (see Figure 5) and one or more maps ho The reference spectrum is 32 〇 (see (iv)). At 15 201223702, this is a reference set of spectra, and the reference set of spectra represents substrates with common properties. However, the common properties of the library can be different in multiple reference light 4 libraries. For example, two different libraries Reference spectra may be included representing substrates having two different underlying thicknesses. For a particular reference spectral library, the upper layer thickness variation is the primary cause of spectral intensity differences, rather than other factors (eg, wafer pattern differences, underlayer thickness, or layer composition). Resulting from the grinding of a plurality of "set" substrates having different substrate properties f (such as layer thickness or layer composition) and collecting the above spectra, a reference spectrum 32 〇 for different libraries 310 can be generated; The spectrum can be provided with a :: Gallery, the spectrum of the other set substrate with different underlying thicknesses can provide a second library. Alternatively or in addition, the reference spectra of different libraries may be calculated theoretically', for example, the spectrum of the first library may be calculated using a prior learning model having a first lower layer thickness, and the spectrum of the second library may be calculated using an optical model having a different lower layer thickness, In some embodiments, each reference spectrum 32 〇 is assigned an index value of 33 〇. Typically, each library 310 can include a plurality of reference spectra 320, such as one or more of a week of rotation of the platform during the expected polishing time of the substrate. Multiple (eg only one) reference spectra. The index 33 〇 can represent a time value, such as a number, that is expected to find the reference spectrum 32 期间 during the polishing process. The spectral index can be programmed to make the spectra in a particular library each uniquely indexable. The index can be indexed so that the index value is sorted according to the spectral order of the measurement test substrate. The index value can be changed monotonically with the grinding process, for example, increasing or decreasing. The index value of the reference spectrum can be selected to make the index value constitute 16 201223702 or A linear function of the number of revolutions of the platform (assuming the polishing rate is based on the polishing rate of the model or test substrate, the test substrate is used to generate reference light 4 in the library). For example, the index value may be proportional to the reference number of the reference light or the optical model of the test substrate (eg, equal to). Therefore, the value of the wife may be an integer. The number of seams may represent an expected correlation. The reference index of the reference spectrum and the reference spectrum can be stored in the reference library. For example, each reference spectrum 32〇 and each reference spectral correlation index value can be stored in the record 340 of the database 350. Reference frame of the reference spectrum The database 35G can be built in the memory of the arithmetic device of the polishing apparatus. As described above, for each region of each substrate, the spectrum is measured according to the sequence of the region and the substrate, and the control of 19 19G can be programmed to produce a sequence optimal. Matching the reference spectrum. The best matching reference spectrum can be determined by comparing the measured i-spectrum and the reference spectrum of the library. In some embodiments, the sum of the squared differences between the measured and reference spectra is calculated to determine the respective reference spectra. The best matching reference spectrum. The reference spectrum with the smallest sum of squared differences is the most suitable. Other search for the best matching reference can also be used. Spectral techniques, such as a sum of minimum absolute differences. In some embodiments, a matching technique other than the sum of squared differences can be used to determine the best matching reference spectrum. In an embodiment, the measurement spectrum and reference are calculated for each reference spectrum. The cross-correlation between the spectra is selected, and the reference spectrum with the largest correlation is selected as the matching reference spectrum. The potential benefit of the cross-matching is that it is less affected by the lateral displacement of the spectrum, and thus is less affected by the thickness variation of the lower layer. In addition, the 201223702 front-end and tail-end of the measured spectrum can be filled in. “As the reference spectrum is relative...:: the data is compared with the reference spectrum, the end can be filled in equal to the measurement. Or the front end of the measurement spectrum is 埴, The value of the value, the measured tail 鳊 can be filled with the value equal to the measured light m %, r positive tail k value. Fast Fourier C Fourier ) can be used for μ ', 4 Η, to speed up the application of instant matching technology The speed of correlation. In the per-packet mode, calculate the sum of the Euclidean vector distances, for example Ε) = '如Xb) [Σλ=λ" u|Im(x)2 _ Σ "Caf, where u to 九& is the total number of wavelengths, and ImW is the measurement spectrum, which is called the reference spectrum. In still another embodiment, the sum of the derivative differences is calculated for each reference spectrum, for example, D = 1/(λα λ!?) [Σλ= λ “ u|dlMW/(a_ is called (1)/call], and the reference spectrum with the smallest sum is selected as the matching reference spectrum. Figure 17 shows the substrate with different TE〇s layer thickness, using the cross-match method and peace. The variance sum method matches the spectrum to obtain a comparison of the exponential trajectory (the best matching reference spectrum index is a function of the number of revolutions of the platform). The data is generated for a product substrate with a stacked structure including a black layer having a thickness of 1500 angstroms (A). A layer of diamond (Black Dia_d), a layer of B1〇k with a thickness of 130A, and a layer of 〇〇ε's with a thickness of 52〇〇A' 51〇〇A or 5〇〇〇. The reference library was produced for the reference substrate, and the reference substrate had a TEOS layer having a thickness of 5,200 Å. As shown by the trace 1702, in which the product substrate and the reference substrate have TEOS layers of the same thickness (i.e., 5200 A), the two exponential trajectories overlap each other without significant difference. However, when the product substrate has a thickness of: 00 Å TEOS layer and the reference substrate has a thickness of 5200 A TEOS layer, the exponential trajectory 1 7 〇 4 produced by the sum of the square differences will slightly deviate from the 201223702 linear behavior. Conversely, the exponential trajectory produced by the cross-comparison will overlap with the index trajectory 1702 (so it cannot be seen from the figure). Finally, when the product substrate has a TEOS layer with a thickness of 5000A and the reference substrate has a TEOS layer with a thickness of 5200A, the exponential trace 1 706 generated by the sum of squared differences will deviate significantly from the linear behavior and trajectory [7〇2, using cross-comparison The resulting exponential trajectory 1708 is still substantially linear and closer to the trajectory 17〇2. In summary, this result shows that if the thickness of the lower layer is different, then using the cross-matching to determine the best matching spectrum will produce a more matching ideal trajectory. A method that can be applied to reduce computer processing is to limit the matching spectral map of the search portion. The gallery generally includes more spectra than the spectra obtained by grinding the substrate. During substrate polishing, the library search should be limited to the predetermined library spectrum. In some embodiments, the current rotation index N of the substrate is determined. For example, when the platform begins to rotate, it can determine N for all reference spectra of the library. Therefore, in terms of the spectrum obtained by the next rotation,
自由度為N ^即，若旋轉一周時的指數為N且後續晚X 個轉數旋轉，其中自由度為γ，則搜尋範圍為（Ν+χ)_γ 至（Ν+Χ) + γ。 參照第7圖’第7圖圖示只用於單一基板的單一區域 的結果’可決定序列中各最佳匹配光譜的指數值，以產 生隨時間變化的序列指數值212。此序列指數值稱為指 數轨跡210。在一些實施方式中，指數軌跡係藉由比較 各測量光譜和單—圖庫的參照光譜而產生。通常，光學 監測系統每次掃掠基板下方時’指數軌跡21〇可包括一 指數值（如僅一指數值）。 19 201223702 對特定指數軌跡210來說，其中光 -掃掠特定區域時有多個：…、糸統進行單 ^ , 則里光譜（稱為「當前光罐、 可決定各當前光譜與一或f 】先h」）’ 光譜間的最佳匹配者。在— 之 >‘系 ^ ^ , 二貫施方式中’比較各撰宁 虽月'J光譜和選定圖庫的每一 選疋 e、f Λ办 參照先谱°假定有當前光碰 e、f、g 和參照光譜 ε、ρ、r β, π， ^ 〇a # 人 則可計算下列當前與參昭 光譜組合的匹配係數： 筝… 版ν 。 與F、e與G、f金E、f 與F、f與G、g與E、g鱼 呂〇 F、和g與G。任 匹配者的匹配係數（如最 /最佳 雄和沪螌#七土 者）將決疋最佳匹配參照光 »曰和礼數值。或者，在—此 .,二貫轭方式中，可結合（如平 句）备月|J光譜，及比較所得6士 中县I 于，·口 σ先譜和參照光譜，以決 疋被佳匹配者和指數值。 在一些實施方式中， ,# 立、 』砘一些基板的至少一些區域，The degree of freedom is N ^, that is, if the index is N when the rotation is one and the X rotations are followed by the subsequent rotation, wherein the degree of freedom is γ, the search range is (Ν+χ)_γ to (Ν+Χ) + γ. Referring to Fig. 7', Fig. 7, shows the result of a single region for only a single substrate', the index value of each of the best matching spectra in the sequence can be determined to produce a sequence index value 212 that varies with time. This sequence index value is referred to as index trajectory 210. In some embodiments, the exponential trajectory is generated by comparing the respective measured spectra to a reference spectrum of a single-gallery. Typically, the optical tracking system may include an index value (e.g., only one index value) each time the substrate is swept below the index. 19 201223702 For a specific exponential trajectory 210, where there are multiple light-swept specific regions: ..., the system performs a single ^, then the spectrum (called "current light tank, can determine each current spectrum with one or f 】First h”)' The best match between spectra. In the -> &^; ^ ^ ^, two-way method of 'comparing each of the 宁 虽 ' ' ' ' ' ' ' J J J J J J J J J J J J J J J J J J J J J 选定 J J J J ° ° ° ° , g and the reference spectrum ε, ρ, r β, π, ^ 〇a # person can calculate the following matching coefficient with the reference spectrum: 筝... version ν. And F, e and G, f gold E, f and F, f and G, g and E, g fish Lv〇 F, and g and G. The matching coefficient of any match (such as the most / best male and the Hu Jin # seven soils) will determine the best match reference light » 曰 and ritual values. Or, in the - yoke method, it is possible to combine (such as a flat sentence) the moon | J spectrum, and compare the resulting 6 士中县 I, 口 σ syllabus and reference spectrum to Matcher and index value. In some embodiments, , at least some areas of the substrate,
產生複數個指數執跡。對:牲中I f特疋基板的特定區域來說，可 產生指數軌跡用於各關注表 …、、、圖庫。即，就用於特定基 板的特定區域的各關注參照圖庫，比較—序列測量光譜 令的每-測量光譜和特定圖庫的參照光譜，以決定一序 列最佳匹配參照光譜，且嗜庄 且遺序列最佳匹配參照光譜的指 數值提供指數軌跡用於特定圖庫。 /言之’各指數執跡包括—序列21〇的指數值川， 〆片列的特疋々曰數值2 ! 2係藉由從特定圖庫選擇最適配 測量光譜的參照光譜指數而產生。指數軌跡2Π)的各指 數時間數值可與測量測量光譜的時間相同。 原位監測技術用來横測是否清除第二層及露出下層或 20 201223702 層、.’η構。例如，可從馬達轉矩或自基板反射的總體光強 度突然改變、或從收集光譜分散，偵測露出第一層的時 間丁c，此將詳述於後。 如第8圖所示，如利用穩健線性配適，使—函數（如 已知階數的多項式函數，例如一階函數（如直線2 Μ )) 配適時間TC後收集的光譜序列指數值。若函數配適該 序列指數值，則可忽略時間TC前收集的光譜指數值。 也'「採用其他函數’例如二階多項式函數，直線更易 運算。可於直線214與目標指數IT相交的終點時間te , 停止研磨。 第9圖為製造及研磨產品基板的方法流程圖。產品基 板至少具有和測試基板一樣的層結構與圖案’測試基板 係用於產生圖庫的參照光譜。 取初，沉積第一層至基板上及圖案化第一層（步驟 9〇2)。如上所述，第一層可為介電質，例如低让材料， 例如碳摻雜二氧化矽，如Black Diamond™ (取自廡用材 料公司）或CoralTM (取自諾發系統公司）。 視第一材料的組成而定，由不同於第 . 介電材 枓的另一介電材料組成的—或更多附加層卿性沉積於 產品基板^第-層上（步驟9G3)，另—介電材料例如為 低k覆蓋材料’例如四乙氧基矽烷（teo —. /示—增和 或更夕附加層一起提供層堆疊結構。在沉積—或更多 附加層後，選擇性進行圖案化，使—或更多附加層不二 伸到第一層的溝槽内（如第1A圖所示）。 21 201223702 接著，由不同的第二介電材料組 品基板的第一層或層堆疊結 ⑲-層’儿積於產 材料例如為阻障層，例如氮化物：=叫第二介電 ^ ^ a 如氮化鈕或氮化鈦。 此外，導電層沉積於產品基板 ,a , J弟一層（和第一層圖案 提供的溝槽)上，導電層例如為金屬層i * ^ ^私叫 隻屬層，如銅（步驟906)。 在/儿積第一層後，選擇性 延伸到第-層的溝槽内。 層’其中第二層不 研磨產品基板（步驟9〇8)。例如， J在第一研磨站中’ 使用第一研磨墊，研磨及移除導 夕rf、导電層和部分第二層（步 驟908a)。接著，可在第_ 隹弟—研磨站中，使用第二研磨墊， 研磨及移除第二層和部分第—声 布增C步驟908b)。然應注 意在一些實施方式中沒有導 電層，例如開始研磨時，第 二層係最外層。當然，可在沿丨步/ 在別處進行步驟902至步驟 906，如此研磨設備的特定操 林丨F表扛將從步驟9 〇 8開始。 原位監測技術用來债測是否、'主^资 个丨只N疋古π除第二層及露出第一層 (步驟910 )。例如，可從黾奩結^ J從馬達轉矩或自基板反射的總體 光強度突然改變、或從收隹4： ^•並人也 乂攸吹果先5普分散，偵測露出第一層 的時間TC (第8圖），此將詳述於後。 至少在偵測是否清除第二層時開始（可能更早，例如從 以第二研磨墊研磨產品基板開始)，如利用上述原位監測 系統，取得研磨期間的一序列測.量光譜（步驟912)。 分析測量光譜以產生一序列指數值，並使一函數配適 該序列指數值。特別地，就序列測量光譜中的每一測量 光譜，決定最適配的參照光譜指數值，以產生一序列指 22 201223702 數值（步驟914 )。使一函數 、如線性函數）配谪 到清除第二層的時間Tc 適在偵測 ϋ μ 後收集的光譜序列指數值（步 驟91 6 )。換言之，在楨彳 、7 “ 4除第二層的時間TC前你 集的光譜序列指數值不用於函數計算。 二指數值（如使線性函數配適新序列指數 的=指數幻達目標指數，即可停止研磨（步驟918)。 目各、厚度IT可由使用者於研磨操作前^定及儲存。或 者，目標移除量可由使用者設 ^ 目知才日數IT可從目標 移除量計算而得。例如，指數差異ID可從目標移除量計 异而得’例如憑經驗決定移除量與指數（如研磨速率） 的比’及把指數“ ID力…貞測到清除上層㈣間TC 時的指數值1C (參見第8圖）。 亦可使用配適在偵測到清除第二層後收集的光譜指數 值的函數，以調整研磨參數，例如調整基板的—或更多 區域的研磨速率，進而改善研磨均勻度。 备照第10圖，第1 〇圖圖示複數個指數軌跡。如上所 述’可就各區域產生一指數軌跡。例如，可就第一區域 產生第一序列210的指數值212 (以圓圈表示），就第二 區域產生第二序列220的指數值222 (以方格表示），及 就第三區域產生第三序列230的指數值232 (以三角形 表示）》儘管圖顯示三個區域，然其也可為兩個區域、或 四或更多區域。所有區域可位於同一基板上，或者一些 區域可源自在同一平臺上研磨的不同基板。 如上所述’原位監測技術用來偵測是否清除第二層及 23 201223702 露出下層或層結構。例如，可從馬達轉矩或自基板反射 的總體光強度突然改變、或從收集光譜分散，偵測露出 第一層的時間TC，此將詳述於後。 就各基板指數執跡’如利用穩健線性配適，使一已知 階數的多項式函數（如-階函數，例如直線）配適時間 TC後收集相關區域的光譜序列指數值。例如，第一直線 214可配適第一區域的指數值212，第二直線224可配適 第二區域的指數值222,第三直線234可配適第三區域 ‘的指數值232。使直線配適指數值可包括計算直線的斜 率S和直線與起始指數值（如〇)相交的χ軸交會時間丁。 函數可表示成1⑴HT)，其巾！為時間。X軸交會 時間T可為負值，負值表示基板層的起始厚度比預期 薄。故第-直線214可具有第一斜率S1和第一 χ轴交會 時間τ卜第二直線224可具有第二斜率s2和第二乂抽 交會時間T2,第三直線234可具有第三斜㈣和第三χ 軸交會時間T3。 、在研磨製程期間的某一時候，例如時間丁〇,可調整至 A域的研磨參數，以調整該基板區域的研磨速率， 使複數個區域於研磨終點時間比在無調整的情況下更接 近该專區域的目押尸疮 ‘厗度。在一些實施例中，各區域於终 點時間有近乎相同的厚度。 參照第1 1圖，A . 在一些實施方式中，選擇一區域做 照區域，及決定Ir 项1文為參 、h、、、區域將達目標指數IT的預計終點時 間TE。例如，1， 第1圖所示’選擇第一區域做為參照 24 201223702Produce a number of index executions. For the specific area of the I f-specific substrate, an exponential trajectory can be generated for each of the attention tables ..., , and the library. That is, for each of the reference reference libraries for a particular region of a particular substrate, the per-measurement spectrum of the sequence-measured spectral order and the reference spectrum of the particular library are compared to determine a sequence of best-matched reference spectra, and The index value of the best matching reference spectrum provides an exponential trajectory for a particular library. / The words 'destruction' of each index includes - the index value of the sequence of 21 川, the characteristic value of the 〆 film column 2 ! 2 is generated by selecting the reference spectral index of the most suitable measurement spectrum from a specific library. The index time value of the exponential trajectory 2Π) can be the same as the time for measuring the measured spectrum. In-situ monitoring techniques are used to cross-test whether the second layer is removed and the lower layer or 20 201223702 layer, .'η structure. For example, the time of exposure of the first layer can be detected from a sudden change in motor torque or overall light intensity reflected from the substrate, or from the collection spectrum, as will be described in detail later. As shown in Fig. 8, if a robust linear fit is used, a function (such as a polynomial function of a known order, such as a first-order function (such as a straight line 2 Μ)) is applied to the spectral sequence index value collected after the time TC. If the function fits the sequence index value, the spectral index value collected before the time TC can be ignored. Also, 'using other functions' such as a second-order polynomial function, the line is easier to calculate. The grinding can be stopped at the end time te where the line 214 intersects the target index IT. Figure 9 is a flow chart of the method of manufacturing and polishing the product substrate. The same layer structure and pattern as the test substrate is used to generate a reference spectrum of the library. Initially, the first layer is deposited onto the substrate and the first layer is patterned (step 9〇2). As described above, One layer can be a dielectric, such as a low-conducting material such as carbon doped cerium oxide, such as Black DiamondTM (from Applied Materials) or CoralTM (from Novartis Systems). Alternatively, another layer of dielectric material different from the dielectric material — is deposited on the product substrate layer (step 9G3), and the other dielectric material is, for example, low. The k-cladding material 'e.g., tetraethoxy decane (teo-. / shows - add-on or add-on layers together to provide a layer stack structure. After deposition - or more additional layers, selectively patterned to make - or Multiple additional layers Do not extend into the trench of the first layer (as shown in Figure 1A). 21 201223702 Next, the first layer or layer of the different second dielectric material assembly substrate is stacked 19-layer The material to be produced is, for example, a barrier layer such as nitride: = a second dielectric such as a nitride button or titanium nitride. Further, a conductive layer is deposited on the product substrate, a, a layer of the first layer (and a first layer pattern). On the trench provided, the conductive layer is, for example, a metal layer i * ^ ^ a private layer, such as copper (step 906). After the first layer is implanted, selectively extends into the trench of the first layer Layer 'where the second layer does not grind the product substrate (steps 9-8). For example, J uses the first polishing pad in the first polishing station, grinding and removing the etched rf, the conductive layer and a portion of the second layer (Step 908a). Next, a second polishing pad may be used to polish and remove the second layer and a portion of the first sound-sounding step C 908b) in the first 隹--- grinding station. However, in some embodiments There is no conductive layer in it, for example, when the grinding starts, the second layer is the outermost layer. Of course, step 902 can be performed along the step / elsewhere At 906, the specific operation of the grinding device will start from step 9 〇 8. The in-situ monitoring technique is used to measure whether or not the main element is only N, the second layer and the first layer. a layer (step 910). For example, the overall light intensity that can be reflected from the motor torque or from the substrate suddenly changes from the 黾奁 junction, or from the 隹4: ^• Detecting the time TC exposing the first layer (Fig. 8), which will be detailed later. At least when detecting whether to clear the second layer (possibly earlier, for example, starting from grinding the product substrate with the second polishing pad) And, using the in-situ monitoring system described above, a sequence of measurements during the grinding is obtained (step 912). The measured spectrum is analyzed to produce a sequence of index values and a function is adapted to the sequence index value. In particular, for each measurement spectrum in the sequence measurement spectrum, the most adapted reference spectral index value is determined to produce a sequence of indices 22 201223702 (step 914). The time Tc at which a function such as a linear function is assigned to clear the second layer is adapted to the spectral sequence index value collected after detecting ϋ μ (step 91 6 ). In other words, the spectral sequence index value of your set before 时间, 7 “4 except the time TC of the second layer is not used for function calculation. The second index value (such as the linear function adapts the new sequence index = index magic target index, The grinding can be stopped (step 918). The thickness and thickness IT can be determined and stored by the user before the grinding operation. Alternatively, the target removal amount can be calculated by the user. For example, the index difference ID can be obtained from the target removal amount, for example, by empirically determining the ratio of the removal amount to the index (such as the grinding rate) and the index "ID force" to detect the clearance between the upper layer (four) The index value of TC is 1C (see Figure 8). It is also possible to use a function that adjusts the spectral index value collected after the second layer is detected to adjust the grinding parameters, such as adjusting the substrate - or more The polishing rate, in turn, improves the uniformity of the polishing. In Fig. 10, the first diagram illustrates a plurality of exponential trajectories. As described above, an index trajectory can be generated for each region. For example, a first sequence can be generated for the first region. 210 index value 212 (indicated by a circle), an index value 222 (in squares) of the second sequence 220 is generated for the second region, and an index value 232 (indicated by a triangle) of the third sequence 230 is generated for the third region, although the figure shows three Areas, which may also be two areas, or four or more areas. All areas may be on the same substrate, or some areas may originate from different substrates ground on the same platform. As described above, 'in situ monitoring technology Used to detect whether the second layer is removed and 23 201223702 reveals the underlying layer structure. For example, the overall light intensity reflected from the motor torque or from the substrate may be suddenly changed, or the spectrum may be dispersed to detect the time of exposing the first layer. TC, which will be described in detail later. For each substrate index, 'with a robust linear fit, a polynomial function of a known order (such as a -order function, such as a straight line) is matched with time TC to collect the relevant region. The spectral sequence index value. For example, the first line 214 may be adapted to the index value 212 of the first region, the second line 224 may be adapted to the index value 222 of the second region, and the third line 234 may be adapted to the third region' The fit of the linear value 232. χ axis rendezvous time index value D S and may comprise a straight line with slope index value calculation starting line (e.g., square) intersect. Can be expressed as a function of 1⑴HT), which towel! For time. The X-axis intersection time T can be a negative value, and a negative value indicates that the initial thickness of the substrate layer is thinner than expected. Therefore, the first line 214 may have a first slope S1 and a first yaw intersection time τ. The second line 224 may have a second slope s2 and a second 乂 stroke time T2, and the third line 234 may have a third slant (four) and The third axis is the intersection time T3. At some point during the polishing process, such as time D, the grinding parameters of the A domain can be adjusted to adjust the polishing rate of the substrate region so that the plurality of regions are closer to the polishing end time than without adjustment. The special area of the target corpse's phlegm. In some embodiments, the regions have nearly the same thickness at the end time. Referring to Figure 11, A. In some embodiments, selecting a region to perform the region and determining the Ir term 1 as the reference, h, , and region will reach the expected end time TE of the target index IT. For example, 1, as shown in Figure 1, 'select the first area as a reference 24 201223702
區域，然也可選擇不同區域及/或不同基板。目標厚度IT 由使用者於研磨操作前設定及儲存。或者，目標移除量 TR可由使用者設定，目標指數ΙΤ可從目標移除量tr 計算而得。例如，指數差異ID可從目標移除量計算而 得’例如憑經驗決定移除量與指數（如研磨速率）的比， 及把指數差異mm貞測到清除上層的時間tc時的指 數值1C。 為决疋參區域將達目標指數的預計時間，可計算參 照區域的直線（如吉& Λ 、如直線214)與目標指數IT的交點。假 設研磨速率在剩餘研磨锄由 W德表私中不偏離預期研磨速率，則Areas, however, different areas and/or different substrates can also be selected. The target thickness IT is set and stored by the user prior to the grinding operation. Alternatively, the target removal amount TR can be set by the user, and the target index 计算 can be calculated from the target removal amount tr. For example, the index difference ID can be calculated from the target removal amount by, for example, empirically determining the ratio of the removal amount to the index (such as the polishing rate), and the index difference mm is measured to the index value 1C when the upper layer time tc is cleared. . In order to determine the estimated time of the target index, the intersection of the straight line of the reference area (such as Kyrgyzstan & 、, such as line 214) and the target index IT can be calculated. Assuming that the grinding rate does not deviate from the expected grinding rate in the remaining grinding 锄, then
序列指數值應保持呈眚暂妙W 貫質線性。故預期終點時間TE可依 直線與目標指數IT沾雜„ 的簡早線性内插計算，例如IT = S’（TE - T)。在第u圖實例中 域，相關第一直線214表示成 =IT/S1 - T1。 ’第一區域被選做參照區 IT = S 卜（TE - τι)，即 TE -或更多區域（如除參照區域外的所有區域，包括其 他基板上的區域）可定義為調整區域。調整區域的線會 合預期終點時間TE虛—塞* 、 々義為調整區域的預計終點。久胡 整區域的線性函數（如笛^ 一各調 数I如第11圖的直線224、234)可用 於外推相關區域將於祐 EI2 Fn , 於預期終點時間的指數 ΕΙ2、ΕΙ3。例如，筮一 ^ 示—直線224可用於外推第-F祕认 預期終點時間τ 雅弟一 &域於 用於外推第：巴域於的預期指數ΕΙ2,第三直線以可 ΕΙ3〇 —;預期終點時間ΤΕ達到的預期指數 201223702 如第11圖所示，若時間Τ〇 Λ 磨速率，Pll '又有5周整任何區域的研 不同的厚谇r L 适點，致使各區域有 損失為如此會造成缺陷和產量 若不同區域在不同時間達 區域於灸昭〜 Ί違目“數（或相當於，調整 可上佟i τ Α 吁门有不同的預期指數），則 J丄U或下修研磨速率， 接心^ |&域比在無調整的情況下更 筏近冋時（如近乎同時）達目 是於目標時間比在益 σ目軚厚度）’或 Η比在無調整的情況下有更接近相同的指數 值（和相同厚度），例如近乎 的厚度）。 十相门的才曰數值（和近乎相同 故在第11圖實例中’從時間Τ0開始，修改第二區域 的至少-研磨參數，以加快該區域的研磨速率（因而辦 加指數軌跡220的斜率）。又在此實例中，修改第三區域 的至少·研磨參數’以減慢該區域的研磨速率（因而減 少指數軌跡230的斜率）。如此，該等區域將近乎同時達 目標指數（和目㈣度）；或者’若同時停止對該等區域 施壓，則該等區域將有近乎相同的厚度。 在一些實施方式中，若預期終點時間TE時的預計指數 代表-基板區域落在預定目標厚度範圍内，則不需調整 該區域。範圍可為目標指數的2%以内，例如1%以内。 可調整調整區域的研磨速率，使所有區域於預期終點 時間比在無調整的情況下更接近目標指數。例如，可選 擇參照基板的一參照區域及調整所有其他區域的處理參 26 201223702 數’使所有區域近乎於參照基板的預計時間達到線點。 參照區域例如為預定區域，例如中心區域⑽或緊鄰 心區域周圍的區域^ ^ 、⑹中 及148b、任何基板的任何區域中有最早 或最晚預計終點相的區域、或具預定預計終點的基板 區域。若同時停止研磨’則最早時間相當於最薄基板。 同樣地同時停止研磨，則最晚時間相當於最厚基板。 參照基板例如為預定基板、具最早或最晚預計终點土時間 之區域的基板。若同時停止研磨，則最早時間相當於最 薄區域。同樣地，若同時停止研磨，則最晚時間相當於 最厚區域。 ' 巧就各調整區域計算指數轨跡的預定斜率，使調整區 域和參照區域同時達目標指數。例如，預定斜率sd可 按似-^^⑽—叫計算’其中^時間別時研 磨參數將改變的指數值（依配適序列指數值的線性函數 計算）’ IT係目標指數，TE係計算的預期終點時間。在 第il圖實例中’可就第二區域，按（IT _ I2) = SD2x(te _ T0)計算預定斜率SD2，及就第三區域，按（IT _ 13)= SD_3x(TE - Τ0)計算預定斜率SD3。 或者’在一些實施方式中，並無參照區域，而預期終 點時間可為預定時間，例如由使用者於研磨製程前設 定、或從一或更多基板中二或更多區域的預期終點時間 平均或其他組合計算（如將不同區域的直線投影到目標 指數來計算）。在此實施方式中，預定斜率實質如同上述 計算，但亦須就第一基板的第一區域計算預定斜率，例 27 201223702 如按町-⑴：灿仰’-以㈣算預定斜率奶】。 或者，在-些實施方式中，不同區域有不同的目標指 數藉此可於基板上形成蓄意、但可控制的不均勻厚度 輪廓。目標指數可由使用者鍵人，例如利用控制器上: 輸入裝置。例如’第—基板的第—區域可具有第_目標 指數，第-基板的第二區域可具有第二目標指數，第二 基板的第-區域可具有第三目標指數，第二基板的第： 區域可具有第四目標指數。 在上述任何方法中，可調整研磨速率，使指數執跡的 斜率更接近預定斜率。研磨速率例如可藉由提高或降低 承載頭的對應腔室内的壓力而調整。研磨速率變化可假 定與壓力變化呈正比，例如簡單Prest〇nian模型。例如， 就各基板的每一區域而言，若時間τ〇前以壓力p *研磨 區域，則時間το後施加的新壓力ρ断可按ρ新=ρ * x(SD/S)計算’其中s為時間τ〇前的直線斜率，SD為預 定余:率。 例如，假設壓力Ps〗施加於第一基板的第一區域，壓 力】5 «2施加於第一基板的第二區域，壓力ρβ3施加於 第二基板的第一區域，壓力Ρβ4施加於第二基板的第二 區域，則對第一基板的第一區域的新壓力p新〗可按p新, =P « ^(SDl/Sl)計算，對第一基板的第二區域的新壓力 P新2可按PW2 = p« 2X(SD2/S2)計算，對第二基板的第一 區域的新壓力P新3可按pM3 = p* 3X(SD3/S3)計算，對 第二基板的第二區域的新壓力p新4可按p w p舊 28 201223702 x(SD4/S4)計算。 研磨製程期鬥 , 調整研磨速=製Γ基板將達目標厚度的預計時間和 I的製料只進行—次，例如在特 :例如經過預期研磨時間的40% 進 磨製程時進行客4 v 傻次可於研 在研磨製程期間:人了 :？每30至60秒進行一次。隨後 研磨速率可口 調整速率。研磨製程期間， 次 隻4次’例如四次、三次、兩次或僅— 整'可於研磨製程開始左右、中間或即將結束時進行調 學===::隼：：τ°後’繼續進行研磨，光 域的指數值。在-::實域的光譜及決定參照區 集"及二霄施方式中，光學監測系統繼續收 跡、「、疋各區域的指數值。-旦參照區域的指數執 跡達目標指鉍 叫Μ日双軌 ’ ρ達所謂 '終‘點’並且停止研磨操作。 例如，如第！ 9阁 _ 續收隼泉日S「 不’時間T〇後’光學監測系統繼 、、只>11茱參照區域的氺摄 A的以及決定參照區域的指數值312。 二：區域的壓力沒有變化（如第"圖 可利用το前（但# ΤΓ &、3 ,則）和το後的資料點計算線性 Μ 供最新線性函數314 ’線性函a314達目標 間代表研磨終點時間。另-方面，若參照區 ==於時間T°時改變，則可從時間T。後的序列指 3數14值32計算具斜率S，的新線性函數314，新線性函數 ⑼的失1τ的時間代表研磨終點時間。用於判斷 終點的參照區域可與上述用以計算_終料間的參照 29 201223702 區域相同’或者用於判斷終點的參照區域可為不同區域 (^‘者右所有區域依上述第n圖調整，則可為終點判 斷k擇參知區域）。若新線性函數3 ^ 4略比原線性函數 214計算的預計時間晚（如帛12圖所示）或早達目標指 貝丨可刀別使一或更多區域稍微過度研磨或研磨不 足U預期終點時間與實際研磨時間應相差幾秒内， 故此不會嚴重影響研磨均勻度。 在-些實施方式中，以銅研磨為例，偵測到基板的終 點後，基板立即遭過度研磨製程處理，以如移除銅殘留 物。過度研磨製程可對基板的所有區域施予均—壓力， 例如1至1.5碑/平方忖。過度研磨製程可有預設持續時 間，例如10至15秒。 就特定區域產生多個指數軌跡時，例如就各關注圖庫 針對特^區域產生-指數轨跡，可選擇—指數軌跡用於 特定區域的終點或壓力控制演算法。例如，對同一區域 產生的各指數執跡而言，控制器19〇可使—線性函數配 適該指數執跡的指數值，及決定該線性函數與序列指數 值的配適契合度。由與其自身指數值有最佳配適契合度 的直線產生的指數軌跡可選做特定區域和基板的指數二 跡。例如’當決定如何調整調整區域的研磨速率時，如 於時間T0時’可採用具最佳配適契合度的線性函數來叶 算。又例如，當具最佳配適契合度的直線的計算指數（如 由配適序列指數值的線性函數計算）匹配或超過目標指 數時，即達所謂終點。又，可不從線性函數計算指數:曰， 30 201223702 而疋將指數值本身與目標指數相比來判斷終點。 決定光譜圖庫相關的指數執跡是否與圖庫相關的線性 函數有最佳配適契合度可包括相較於相關穩健線與另— 圖庫相關之指數軌跡間的差異，決定相關光譜圖庫的指 數軌跡是否與相關穩健線有最小差異，例如最小標準 差、最大相關性或其他測量變量。在一實施方式中，配 適契合度係藉由計算指數資料點與線性函數間的平方差 總和而決定；平方差總和最小的圖庫為最適配者。 參知第13圖，第13圖圖示概括流程圖13仰。如上所 述’在研磨設傷中，利用同一研磨塾同時研磨一基板的 複數個區域（步驟13〇2)。在此研磨操作期間，各區域 的研磨速率與其他基板無關且由獨立變動的研磨參數個 別控制，例如承載頭的腔室施加至特定區域的壓力。如 上所述，在此研磨操作期間’如利用得自各區域的一序 列測量光镨，監測基板（步驟13〇4)。就該序列中的每 、ί里光„日，决疋最佳匹配參照光譜（步驟Η% )。決 定各參照光譜的最適配指數值，以產生序列指數值（步 驟].3 08 ) 〇 债測是否清除第二層（步驟131〇)。就各區域，使一 線性函數配適備測免卜、杳& '巧除第二層後收集的光譜序列指數 值（步驟1312)。在-實施方式中，衫參照區域的線 性函數將達目標指數值的預期終點時間，例如利用線性 函數的線性内插（步驟1314)。在其他實施方式中，預 月、點時間係、預先決^ '或依多個區域的預期終點時間 31 201223702 組合計算。若有需要，可調整其他區域的研磨參數，以 調整該基板的研磨速率，使複數個區域近乎同時達目標 厚度、或使複數個區域於目標時間有近乎相同的厚度（或 目標厚度）（步驟1316)。調整參數後，繼續進行研磨， . 並就各區域，測量光譜、從圖庫中決定最佳匹配參昭光 ‘譜、在調整研磨參數後的時期，決定最佳匹配光譜的指 數值，以產生新序列指數值、以及使—線性函數配適指 數值（V驟13 1 8 )。-旦參照區域的指數值（如將線性 函數配適新序列指數值而得的計算指數值）達目標指 數’即可停止研磨（步驟Π3〇)。 在些實施方式中，序列指數值係用於調整一基板的 -或更多區域的研磨速率，但利用另一原位監測系統或 技術來偵測研磨終點。 如上所述，對一些技術和一些層堆疊結構而言，偵測 疋否凊除上層及露出下層十分困難。在一些實施方式 中，收集一序列光譜群組，及計算每一光譜群組的分散 參數值，以產生一序列分散值。可從該序列分散值，偵 測是否清除上層。此技術可用於偵測是否清除第二層及 •露出第一層，例如用於上述研磨操作的步驟91〇或131〇。 第14圖顯示偵測是否清除第二層及露出第一層的方 法1400。研磨基板時（步驟14〇2)，收集一序列光譜群 組（、步驟1404 )。如第4圖所示，若光學監測系統固定 於旋轉平臺，則光學監測系統單一掃掠基板各處後，將 可收集基板上多個不同位置201b至201j的光譜。單一 32 201223702 掃掠收集的光譜提供一光譜群組。隨著研磨進行，光學 監測系統多次掃掠後，將提供—序列光譜群組。平臺旋 轉周可收集-光譜群組，例如以相當於平臺旋轉速率 的頻率收集群組。通常’每-群組包括…0個光譜。 可利用與上述波峰追蹤技術收集光譜相同的光學監測系 統來收集光譜。 第15A圖提供研磨之初（如上層仍有明顯厚度留在下 層上時），自基板ίο反射的一組測量光譜15〇〇3的實例。 該組光譜15_可包括在光學監㈣、統第—次掃掠基板 各處時，於基板上不同位置收集的光譜2〇2a至2〇4a。第 15B圖提供於清除或幾乎清除上層日夺，自基板1〇反射的 組測里光谱1 500b的實例。該組光譜丨5〇〇b可包括在 光學監測系統施行不同的第二次掃掠基板各處時，於基 板上不同位置收集的光譜202]3至2〇4b(收集光譜i5〇〇a 的基板位置可不同於光譜15〇〇b)。 最初，如第15A圖所示，光譜i 500a頗為相似。然如 第15B圖所示，當清除上層（如阻障層）而露出下層（如 低k層或覆蓋層）時，取自不同基板位置的光譜i5〇〇b 間差異變得更加顯著。 就各光譜群組，計算該光譜群組的分散參數值（步驟 1406 )。依此將產生一序列分散值。 在一實施方式中’為計算光譜群組的分散參數，可平 均強度值（如同波長函數）而提供一平均光譜。即j平均（λ) -(1/Ν).[Σί = 1至ν Ιί(λ)]’其中Ν係該群組的光譜數量，ι(λ) 33 201223702 為光譜。就该群組的每一光譜，如利用平方差總和或絕 對值差總和，計算該光譜與平均光譜間的總差值，例如The value of the sequence index should be kept linear. Therefore, the expected end time TE can be calculated by a simple linear interpolation of the line and the target index IT, such as IT = S' (TE - T). In the field of the u-th example, the relevant first line 214 is expressed as = IT /S1 - T1. 'The first area is selected as the reference area IT = S (TE - τι), ie TE - or more areas (such as all areas except the reference area, including areas on other substrates) can be defined In order to adjust the area, the line of the adjustment area meets the expected end time TE virtual-plug*, and the meaning is the estimated end point of the adjustment area. The linear function of the long-haul area (such as the flute), the modulation number I, such as the line 224 of Figure 11 234) can be used to extrapolate the relevant area will be EI2 Fn, the index at the expected end time ΕΙ 2, ΕΙ 3. For example, 筮一^示—Line 224 can be used to extrapolate the first-F secret expected end time τ Yadiyi The & domain is used for extrapolation: Barrier's expected index ΕΙ2, the third straight line is ΕΙ3〇-; expected end point ΤΕ reached expected index 201223702 as shown in Figure 11, if time honing rate , Pll 'has another 5 weeks to study the thickness of any area r r Appropriate, causing losses in various regions as such will cause defects and yields if different regions reach the region at different times in the moxibustion 〜 Ί Ί Ί “ 数 或 或 或 或 或 或 或 或 ( ( ( ( ( ( ( ( ( 调整 调整Index), then J丄U or lowering the grinding rate, the centering ^ |& field is closer to the 无 when there is no adjustment (such as near simultaneous), the target is the target time than the thickness of the target ) ' or Η is closer to the same index value (and the same thickness), such as near thickness, without adjustment. The value of the ten-phase gate (and nearly the same, in the example of Fig. 11 'starts from time Τ 0, modifies at least the grinding parameter of the second region to speed up the grinding rate of the region (thus increasing the slope of the exponential trajectory 220 In this example, the at least "grinding parameter" of the third region is modified to slow down the polishing rate of the region (thus reducing the slope of the exponential trajectory 230). Thus, the regions will nearly simultaneously reach the target index (and (4) degrees); or 'If the pressure is applied to the regions at the same time, the regions will have nearly the same thickness. In some embodiments, if the expected end time TE is expected, the predicted index represents that the substrate region falls on the predetermined target. Within the thickness range, the area does not need to be adjusted. The range can be within 2% of the target index, for example within 1%. The grinding rate of the adjustment area can be adjusted so that all areas are closer to the expected end time than without adjustment. Target index. For example, you can select a reference area of the reference substrate and adjust the processing parameters of all other areas. 26 201223702 number 'make all areas close to the reference The estimated time of the board reaches the line point. The reference area is, for example, a predetermined area, such as a central area (10) or an area immediately adjacent to the heart area ^^, (6) and 148b, and any area of any substrate having the earliest or latest predicted end phase Or the substrate area with the intended end point. If the polishing is stopped at the same time, the earliest time corresponds to the thinnest substrate. Similarly, the polishing is stopped at the same time, the latest time corresponds to the thickest substrate. The reference substrate is, for example, a predetermined substrate, with the earliest or The substrate at the end of the soil time is expected at the latest. If the polishing is stopped at the same time, the earliest time corresponds to the thinnest area. Similarly, if the polishing is stopped at the same time, the latest time corresponds to the thickest area. Calculating the predetermined slope of the exponential trajectory so that the adjustment region and the reference region reach the target index at the same time. For example, the predetermined slope sd can be calculated as -^^(10)-called "the index value at which the grinding parameter will change when the time is not The linear function of the appropriate sequence index value is calculated] 'IT system target index, TE system calculates the expected end time. In the example of the il diagram' For the second region, the predetermined slope SD2 is calculated as (IT _ I2) = SD2x (te _ T0), and for the third region, the predetermined slope SD3 is calculated as (IT _ 13) = SD_3x (TE - Τ 0). In some embodiments, there is no reference area, and the expected end time may be a predetermined time, such as a user set before the polishing process, or an expected end time average or other combination of two or more regions from one or more substrates. The calculation (such as calculating the straight line of different regions to the target index). In this embodiment, the predetermined slope is substantially the same as the above calculation, but the predetermined slope must also be calculated for the first region of the first substrate, Example 27 201223702 - (1): Can't calculate the predetermined slope milk by (d). Alternatively, in some embodiments, different regions have different target indices whereby a deliberate, but controllable, non-uniform thickness profile can be formed on the substrate. The target index can be keyed by the user, for example using the controller: input device. For example, the first region of the first substrate may have a first target index, the second region of the first substrate may have a second target index, and the first region of the second substrate may have a third target index, the first substrate of the second substrate: The region may have a fourth target index. In any of the above methods, the polishing rate can be adjusted such that the slope of the exponential trace is closer to the predetermined slope. The rate of polishing can be adjusted, for example, by increasing or decreasing the pressure within the corresponding chamber of the carrier head. The change in the grinding rate can be assumed to be proportional to the pressure change, such as the simple Prest〇nian model. For example, for each region of each substrate, if the region is pressed with a pressure p* before time τ〇, the new pressure ρ applied after the time το can be calculated as ρ new = ρ * x (SD / S) s is the slope of the line before time τ〇, and SD is the predetermined remainder: rate. For example, assume that the pressure Ps is applied to the first region of the first substrate, the pressure 5 5 «2 is applied to the second region of the first substrate, the pressure ρβ3 is applied to the first region of the second substrate, and the pressure Ρβ4 is applied to the second substrate. The second region, the new pressure p new to the first region of the first substrate can be calculated as p new, =P « ^ (SDl / Sl), the new pressure on the second region of the first substrate P new 2 According to PW2 = p« 2X(SD2/S2), the new pressure P3 for the first region of the second substrate can be calculated as pM3 = p* 3X(SD3/S3) for the second region of the second substrate. The new pressure p new 4 can be calculated according to pwp old 28 201223702 x (SD4/S4). Grinding process bucket, adjusting the grinding speed = the expected time for the substrate to reach the target thickness and the material of I to be processed only once, for example, in special: for example, 40% of the expected grinding time into the grinding process, 4 4 silly The time can be studied during the grinding process: people:? It is done every 30 to 60 seconds. The grinding rate is then adjusted to a good rate. During the grinding process, only 4 times 'for example, four times, three times, two times or only - whole' can be transferred at the beginning, middle or near the end of the grinding process ===::隼::τ° after 'continue Perform the grinding, the index value of the light domain. In the -:: real-spectrum spectrum and the decision-based zone set " and the second implementation method, the optical monitoring system continues to collect, ", the index value of each region. - The reference index of the reference area reaches the target index Called the double track ' 达 达 ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' 11茱 The reference area of the shot A and the reference area index value 312. Second: the area of the pressure has not changed (such as the first "quote can use το before (but # ΤΓ &, 3, then) and το The data point calculates the linear Μ for the latest linear function 314 'The linear function a314 reaches the end of the grinding time between the targets. On the other hand, if the reference area == changes at time T°, then the sequence from the time T can be counted as 3 The 14 value 32 calculates a new linear function 314 with a slope S, and the time of the new linear function (9) missing 1 τ represents the polishing end time. The reference area for determining the end point can be used to calculate the _finished reference 29 201223702 area. Same 'or reference for determining the end point The domain can be a different region (where the right region of ^' is adjusted according to the above nth map, then the endpoint can be judged to select the region.) If the new linear function 3^4 is slightly later than the estimated time calculated by the original linear function 214 ( As shown in Fig. 12) or as early as the target, the shellfish can be used to make one or more areas slightly overgrind or undergrind. The expected end time should be within a few seconds of the actual grinding time, so it will not seriously affect the grinding evenly. In some embodiments, in the case of copper polishing, after detecting the end point of the substrate, the substrate is immediately subjected to an over-polishing process to remove copper residues, etc. The over-grinding process can be applied to all areas of the substrate. Mean-pressure, for example 1 to 1.5 monuments per square inch. The overgrinding process can have a preset duration, for example 10 to 15 seconds. When generating a plurality of exponential trajectories for a particular region, for example, each of the attention libraries is generated for the specific region - Exponential trajectory, selectable—exponential trajectory for the end point or pressure control algorithm for a particular region. For example, for each index trajectory produced in the same region, the controller 19 can adapt the linear function The exponential value of the index and the matching degree of the linear function and the sequence index value. The exponential trajectory generated by the straight line with the best fit of its own index value can be selected as the index of the specific region and the substrate. For example, 'when deciding how to adjust the grinding rate of the adjustment zone, as in time T0', a linear function with the best fit can be used for the calculation. For example, when the optimal fit fits The calculated index of the straight line (as calculated by the linear function of the fitted sequence index value) matches or exceeds the target index, which is the so-called end point. Again, the index cannot be calculated from the linear function: 曰, 30 201223702 and the index value itself is The target index is compared to the end point. Determining whether the spectral library-related index profiling has the best fit for the linear function associated with the library may include determining the exponential trajectory of the relevant spectral library compared to the difference between the relevant robust lines and the index trajectories associated with the other There is minimal difference from the associated robust line, such as minimum standard deviation, maximum correlation, or other measured variable. In one embodiment, the fitness fit is determined by calculating the sum of the squared differences between the index data points and the linear function; the library with the smallest sum of square differences is the most suitable. Referring to Figure 13, Figure 13 illustrates a summary of the flowchart 13 . As described above, in the polishing process, a plurality of regions of a substrate are simultaneously polished by the same polishing crucible (step 13〇2). During this grinding operation, the polishing rate of each zone is independent of the other substrates and is controlled by independently varying grinding parameters, such as the pressure applied to the particular zone by the chamber of the carrier head. As described above, the substrate is monitored during the grinding operation as measured by a sequence of measurements from the respective regions (step 13〇4). For each sequence in the sequence, the best matching reference spectrum (step Η%) determines the most suitable index value of each reference spectrum to generate the sequence index value (step).3 08) It is determined whether to clear the second layer (step 131〇). For each region, a linear function is fitted to measure the spectral sequence index value collected after the second layer is removed (step 1312). In an embodiment, the linear function of the shirt reference area will reach the expected end time of the target index value, such as linear interpolation using a linear function (step 1314). In other embodiments, the pre-month, point time, pre-determination Or calculate according to the expected end time of multiple regions 31 201223702. If necessary, adjust the grinding parameters of other regions to adjust the polishing rate of the substrate so that the multiple regions reach the target thickness at the same time, or make multiple regions The target time has nearly the same thickness (or target thickness) (step 1316). After adjusting the parameters, continue to grind, and measure the spectrum for each region, and determine the best match from the library. 'Spectrum, the period after adjusting the grinding parameters, determines the index value of the best matching spectrum to generate a new sequence index value, and the linear function matching index value (V 13 13 8). The value (such as calculating the exponential value obtained by fitting the linear function to the new sequence index value) to the target index 'can stop grinding (step Π 3 〇). In some embodiments, the sequence index value is used to adjust a substrate - Or more areas of the grinding rate, but using another in-situ monitoring system or technique to detect the grinding end. As mentioned above, for some techniques and some layer stacking structures, the detection will remove the upper layer and expose the lower layer. Difficulties. In some embodiments, a sequence of spectral groups is collected, and the values of the dispersion parameters of each spectral group are calculated to generate a sequence of dispersion values. From the sequence dispersion values, it is detected whether the upper layer is cleared. In order to detect whether to clear the second layer and to expose the first layer, for example, the step 91〇 or 131〇 for the above grinding operation. Figure 14 shows whether the second layer is removed and the first layer is exposed. Method 1400. When the substrate is polished (step 14〇2), a sequence of spectral groups is collected (step 1404). As shown in Fig. 4, if the optical monitoring system is fixed to the rotating platform, the optical monitoring system single sweeps the substrate After that, the spectra of a plurality of different locations 201b to 201j on the substrate will be collected. A single 32 201223702 sweep collected spectrum provides a spectral group. As the polishing proceeds, the optical monitoring system will provide a sequence after multiple sweeps. Spectral group. The platform rotation week can collect - spectral groups, for example, to collect groups at a frequency equivalent to the rate of platform rotation. Usually 'every-group includes... 0 spectra. The same spectrum can be collected using the above-mentioned peak tracking technique. An optical monitoring system is used to collect the spectra. Figure 15A provides an example of a set of measured spectra 15〇〇3 reflected from the substrate ίο at the beginning of the grinding (when the upper layer still has significant thickness remaining on the lower layer). The set of spectra 15_ may include spectra 2〇2a to 2〇4a collected at different locations on the substrate during optical monitoring (four), system-sweeping of the substrate. Figure 15B provides an example of a 1500 bp spectrum of the spectrum measured from the substrate 1 清除 to remove or nearly remove the upper layer. The set of spectra 丨5〇〇b may include spectra 202]3 to 2〇4b collected at different locations on the substrate when the optical monitoring system performs different second sweeps of the substrate (collecting the spectrum i5〇〇a) The substrate position can be different from the spectrum 15 〇〇 b). Initially, as shown in Figure 15A, the spectrum i 500a is quite similar. However, as shown in Fig. 15B, when the upper layer (e.g., barrier layer) is removed to expose the lower layer (e.g., the low-k layer or the overlying layer), the difference between the spectra i5〇〇b taken from the different substrate positions becomes more significant. For each spectral group, the dispersion parameter values for the spectral group are calculated (step 1406). This will result in a sequence of discrete values. In one embodiment, 'to calculate the dispersion parameters of the spectral group, an average intensity value (as a function of wavelength) can be provided to provide an average spectrum. That is, j is average (λ) - (1/Ν). [Σί = 1 to ν Ιί(λ)]' where lanthanum is the number of spectra of the group, and ι(λ) 33 201223702 is the spectrum. For each spectrum of the group, such as using the sum of the squared differences or the absolute difference, the total difference between the spectrum and the average spectrum is calculated, for example
Di = ["(u 一 λ1))·[Σλ=λ“ xb [ΙΚλ) _【平均(λ)]2]]"2 或 〇丨= [l/(Ia - λΐ»).[Σλ^3_ u |1;(人）_ I 平均（九）|]]，其中入a 至入乜 為加總的波長範圍。 一旦計算該光譜群組中的每一光譜的差值，即可從差 值計算該群組的分散參數值。種種分散參數都有可能， 例如標準差、四分位數範圍、變化幅度（最大值減去最 小值）、平均差、中位絕對差和平均絕對差。 該序列分散值可加以分析及用於偵測是否清除上層 (步驟 1408 )。 第16圖顯示隨研磨時間變化的光譜標準偏差圖16〇〇 (其中各標準差係從光譜群組的差值計算而得）。故圖中 的每一繪製點1602係在光學監測系統進行特定掃掠時 收隽的光譜群組差值的標準差。如圖所示，第一時期“Μ 的標準差值仍相當小。然在時期161〇後，標準差值越來 越大且越來越分冑。不揭限於任何特定理論，厚阻障層 有控制反射光譜的傾向，導致阻障層本身與任何下層的 =度差異不易察覺。隨著研磨進行，阻障層越來越二或 兀全移除，反射光譜則變得更易受下層厚度變化影響。 因此，光譜分散性將隨著阻障層清除而趨增。 各種演算法可用於偵測當上層清除時的分散值行為變 化。例如，比較序列分散值和聞值；若分散值超過聞值， 則產生訊號指示已清除上層。又例如，計算移動視窗内 34 201223702 的部分序列分散值的斜率 號指示已清除上層。 若斜率超過閾值，則產生訊 做為偵測分散性增加的部分演算法，序列分散值可經 濾波器（如低通濾波器或帶通濾波器）處理，以移除高 頻雜訊。低通濾波器的實例包括移動平均與巴特沃斯 (Butterworth )濾波器。 雖然以上敘述係專注於偵測是否清除阻障層，但此技 術亦可用於偵測是否清除其他方面的上層，例如在另一 種半導體製程中1測是否清除介電層堆疊結構的上層 (如層間介電WILD))、或偵測是否清除介電層上的薄 金屬層。 除了上述用於觸發初始化特徵結構追縱外，此用於债 測是否清除上層的技術還可用料磨操作的其他用途， 例如做為終點訊號本身來觸發計時器’以於露出下層 後’研磨下層-段預定時間，或者引發修改研磨參數: 以如在露出下層後，立即改變承載頭壓力或研磨漿組成。 此外，雖然以上敘述係採用具光學終點監測器安裝於 平臺的旋轉平臺，但系統亦可應用到監測系統與基板間 有其他類型的相對運動。例如，在―些實施方式中，如 軌道運動，光源橫跨基板上的不同位置、但不越過基板 邊緣。在此情況下，仍可將收集的光譜分組，例如以特 定頻率收集光譜’-定時間内收集的光譜則視為群組的 4刀。收集時間應夠長，使各群組有5至個光譜。 本說明書所用的「基板」一詞例如可包括產品基板（如 35 201223702 基板包括多個記憶體或處理器晶粒）、測試基板、裸基板 和閘閂基板。基板可處於各種積體電路製造階段，例如 基板可為裸晶圓，或者基板可包括一或更多沉積層及/或 圖案化層。「基板」一詞可包括圓盤和矩形片。 本發明的實施例和說明書所述 於數位電子電路、或電腦軟體、韌體或硬體，包括本說 明書提及的結構裝置、和上述結構裝置的結構均等物或 上述結構裝置的組合物。本發明的實施例可實施成一或 更夕電腦程式產品，即實體收錄於機器可讀取儲存媒體 的一或更多電腦程式，以供資料處理設備（如可程式處 理器、電腦、或多個處理器或電腦）執行或控制運作。 電腦程式（亦稱為程式、軟體、軟體應用或編碼）可以 任何包括編譯或解譯語言的程式語言編寫，電腦程式並 可部署成任何形式’包括獨立程式或模組、部件、副程 式.‘或其他適合運算環境的單元。電腦程式不一定要對 應一個標案。程式可存儲在含有其他程式或資料的部分 播案、提問程式專用的單-標案、或多重座標槽案（如 模組、副程式或部分編碼的標案)。電腦程 式伐置供早一電腦或多個電腦執行，電腦位於一網點 或分散遍及多個網點且由通信網路相連。 ‘、 說明書所述的製程和邏輯流程可由_ 更多電腦程式的可程式處理器進行，以 執仃―， 料及產生輸出而發揮功能。t "术作輸入資 用途的邏輯電路進行，且讯 '、σ由特定 且-備也可實施做為特定用途的 36 201223702 邏輯電路’特定用途的邏輯電路例如為現場可程式閘陣 列（FPGA)或特定功能積體電路（ASIC)。 上述研磨設備和方法可應用到各種研磨系統。無論是 研磨墊承载頭或二者都可移動而提供研磨表面與基板 間的相對移動。例如，平臺可不自轉、而是進行公轉。 研磨墊可為固定於平臺的圓形墊（或其他形狀)。如當研 磨f為線性移動的連續式或捲盤式研磨帶時，某些終點 偵測系統態樣可應用到線性研磨系統。研磨層可為桿準 研磨剛如含有或不含填料的聚胺酯)、軟材料或固定 研磨材料。在此採用相對位置的敘述方式；應理解研磨 表面_基板可保持朝垂直位向或其他位向。 /本發明的特定實施例已揭露如上。其他實施例亦落在 後附申請專利範圍所界定的保護範圍内。 【圖式簡單說明】 第1A及1B圖為研磨前後的基板截面圖。 第2圖圖示為一研磨設備實例的截面圖。 第3圖圖示為具有多個區域的基板的上視圖。 第4圖圖示為研磨墊的上視圖，圖顯示基板上進行 位測量的位置。 原 第5圖圖示出自原位光學監測系統的測量光譜 第6圖圖示參照光譜圖庫。 第7圖圖示指數軌跡。 37 201223702 第8圖圖示具線性函數的指數執跡，此線性函數配適 在偵測到清除上層後收集的指數值。 第9 g a製造基板及债測研磨終點的示例製程流程 圖。 第1〇圖圖示複數個指數軌跡。 苐11圖圖不依據參照區域的指數執跡達目標指數的 寺間计算複數個調整區域的複數個預定斜率。 驾 ® ®不依·據參照、區域的指數軌跡達目標指數的 時間，計算終點。 I’ 13圖為不例製程流程圖’用以調整複數個基板的複 數個區域的研磨速率’使複數個區域於目標時間有近乎 相同的厚度。 =14圖顯示偵測上層清除的流程圖。 身，15A圖顯示單_掃掠期間，於研磨之初收集的光错 圖c & 15B圖顯示單一掃掠期間，於幾乎清除阻障層時收 集的光譜圖。 第16圖顯示隨研磨時間變化的光譜標準偏差圖。 17圖為顯示不同♦宁县几邮 , +丨·）决疋最佳匹配參照光譜技術的比 較圃。 各圖中相同的元件符號和稱號代表相似的元件。 【主要元件符號說明】 38 201223702 10 基板 12、 14' 16 層 18 導電材料 100 研磨設備 108 ' 118 窗口 110 研磨墊 112 研磨層 114 背層 120 平臺 121 ' > 154 馬達 124、 152 傳動軸 125 、155 軸 128 凹槽 129 聯結器 no 埠口 132 研磨液 140 承載頭 142 定位環 144 彈性膜 146a-146c 腔室 148a -148c 區域 150 支撐結構 160 光學監測系統 162 光源 164 偵測器 166 電路 168 光學頭 170 光纖 172 主幹 174 、176 分支 190 控制器 201 位置 201a-201k 點 202a-204a 光譜 202b-204b 光譜 204 箭頭 210 、220、230 軌跡/序列 212 、222 ' 232 214 、224、234 直線 300 測量光譜 310 圖庫 312 指數值 314 函數 320 參照光譜 330 指數值 340 記錄 350 資料庫 900 方法 指數值 39 201223702 902、 903 ' 904 ' 906 、 908 ' 908a 、908b ' 910 914、916、918 步驟 1300 流程圖 1302 、1304、1306、1308、 1310、 1312、1314、 1318、 1330 步驟 1400 方法 1402 ' 1404 > 1406 ' 1408 步驟 1500a ' 1500b 光譜 1600 標準差圖 1602 繪製點 1610 時期 1702 、1704 、 1706 、 1708 軌跡 、912、 1316、 40Di = ["(u λ1))·[Σλ=λ" xb [ΙΚλ) _[Average (λ)]2]]"2 or 〇丨= [l/(Ia - λΐ»).[Σλ ^3_ u |1; (person)_ I average (nine)|]], where a to 乜 is the summed wavelength range. Once the difference in each spectrum in the spectral group is calculated, The difference calculates the value of the group's scatter parameters. Various scatter parameters are possible, such as standard deviation, quartile range, magnitude of change (maximum minus minimum), mean difference, median absolute difference, and mean absolute difference The sequence dispersion value can be analyzed and used to detect whether the upper layer is removed (step 1408). Figure 16 shows the spectral standard deviation as a function of the polishing time. Figure 16 (where the standard deviation is the difference from the spectral group) Calculated by calculation.) Therefore, each plotted point 1602 in the figure is the standard deviation of the spectral group difference that is collected when the optical monitoring system performs a specific sweep. As shown in the figure, the standard deviation of the first period is still Quite small. However, after the period of 161, the standard deviation is getting bigger and more divergent. Without being limited to any particular theory, the thick barrier layer has a tendency to control the reflectance spectrum, resulting in a sub-degree difference between the barrier layer itself and any underlying layer that is not readily detectable. As the grinding progresses, the barrier layer is removed more or more, and the reflection spectrum becomes more susceptible to the thickness variation of the lower layer. Therefore, the spectral dispersion will increase as the barrier layer is removed. Various algorithms can be used to detect changes in the behavior of the dispersion values when the upper layer is cleared. For example, comparing the sequence dispersion value and the smell value; if the dispersion value exceeds the smell value, a signal is generated indicating that the upper layer has been cleared. For another example, calculating the slope number of the partial sequence dispersion value in the mobile window 34 201223702 indicates that the upper layer has been cleared. If the slope exceeds the threshold, the signal is generated as a partial algorithm for detecting the increase in dispersion. The sequence dispersion value can be processed by a filter (such as a low-pass filter or a band-pass filter) to remove high-frequency noise. Examples of low pass filters include moving averages and Butterworth filters. Although the above description focuses on detecting whether to remove the barrier layer, this technique can also be used to detect whether other layers are removed, for example, in another semiconductor process, whether to remove the upper layer of the dielectric layer stack structure (such as between layers) Dielectric WILD)), or detect whether to remove the thin metal layer on the dielectric layer. In addition to the above-mentioned techniques for triggering the initialization feature structure, this technique for deciding whether to clear the upper layer may also be used for other purposes of the material grinding operation, for example, as the end point signal itself to trigger the timer 'to expose the lower layer' to grind the lower layer. - The predetermined time of the segment, or the modification of the grinding parameters: to change the carrier head pressure or the slurry composition immediately after the lower layer is exposed. In addition, although the above description uses a rotating platform with an optical end point monitor mounted to the platform, the system can also be applied to other types of relative motion between the monitoring system and the substrate. For example, in some embodiments, such as orbital motion, the light source traverses different locations on the substrate, but does not cross the edge of the substrate. In this case, the collected spectra can still be grouped, for example, the spectra collected at a specific frequency are collected as a group of 4 knives. The collection time should be long enough for each group to have 5 to spectra. The term "substrate" as used in this specification may include, for example, a product substrate (e.g., 35 201223702 substrate comprising a plurality of memory or processor dies), a test substrate, a bare substrate, and a latch substrate. The substrate can be in various stages of integrated circuit fabrication, such as the substrate can be a bare wafer, or the substrate can include one or more deposited layers and/or patterned layers. The term "substrate" can include discs and rectangular sheets. Embodiments and descriptions of the invention are described in digital electronic circuits, or computer software, firmware or hardware, including structural devices referred to in the specification, structural equivalents of the structural devices described above, or combinations of the above structural devices. Embodiments of the present invention may be implemented as one or more computer program products, that is, one or more computer programs embodied in a machine readable storage medium for data processing devices (such as a programmable processor, a computer, or a plurality of The processor or computer) performs or controls the operation. Computer programs (also known as programs, software, software applications or encodings) can be written in any programming language including compiled or interpreted languages. Computer programs can be deployed in any form 'including stand-alone programs or modules, components, sub-programs.' Or other units suitable for the computing environment. The computer program does not have to correspond to a standard. The program can be stored in a part of a program containing other programs or data, a single-marker for a question-solving program, or a multi-coordinate slot (such as a module, a subprogram, or a partially coded standard). Computer-based logging is performed by a computer or multiple computers, and the computer is located at a network or distributed across multiple sites and connected by a communication network. ‘The process and logic flow described in the manual can be performed by a programmable processor of more computer programs, functioning and generating output. t " The logic circuit for the input application, and the ', σ is specific and can be implemented as a specific use. 36 201223702 Logic circuit 'Special-purpose logic circuit such as field programmable gate array (FPGA ) or a specific function integrated circuit (ASIC). The above grinding apparatus and method can be applied to various grinding systems. Either the abrasive pad carrier or both can be moved to provide relative movement between the abrasive surface and the substrate. For example, the platform may not rotate, but instead conduct a revolution. The polishing pad can be a circular pad (or other shape) that is fixed to the platform. For example, when the grinding f is a linear moving continuous or reeling belt, some end point detection system aspects can be applied to the linear grinding system. The abrasive layer can be a quenched polyurethane such as a polyurethane with or without a filler, a soft material or a fixed abrasive material. Here, the relative position is described; it should be understood that the abrasive surface_substrate can be oriented in a vertical orientation or in other orientations. / Specific embodiments of the invention have been disclosed above. Other embodiments are also within the scope of protection defined by the scope of the appended claims. BRIEF DESCRIPTION OF THE DRAWINGS FIGS. 1A and 1B are cross-sectional views of a substrate before and after polishing. Figure 2 is a cross-sectional view showing an example of a grinding apparatus. Figure 3 illustrates a top view of a substrate having multiple regions. Figure 4 is a top view of the polishing pad showing the position of the position measurement on the substrate. The original Figure 5 shows the measured spectrum of the in-situ optical monitoring system. Figure 6 shows the reference spectrum library. Figure 7 illustrates the exponential trajectory. 37 201223702 Figure 8 illustrates an exponential trace with a linear function that fits the index value collected after the upper layer is detected. An example process flow diagram for the fabrication of the substrate and the end point of the bond measurement at 9 g. The first diagram illustrates a plurality of exponential trajectories. The 苐11 map calculates a plurality of predetermined slopes of the plurality of adjustment regions between the temples based on the index of the reference region and the target index. The driving end is calculated based on the time when the index trajectory of the area reaches the target index. The I'13 figure is a non-exemplary process flow diagram 'for adjusting the polishing rate of a plurality of regions of a plurality of substrates' such that the plurality of regions have nearly the same thickness at the target time. The =14 figure shows the flow chart for detecting upper layer clearing. Body, Figure 15A shows the optical error collected at the beginning of the grinding during the single sweep. Figure c & 15B shows the spectrum collected during the almost sweep of the barrier layer. Figure 16 shows the spectral standard deviation plot as a function of grinding time. 17 is a comparison of the best matching reference spectroscopy techniques for different numbers of ♦ Ning County, +丨·). The same component symbols and symbols in the various drawings represent similar components. [Major component symbol description] 38 201223702 10 Substrate 12, 14' 16 layer 18 Conductive material 100 Grinding device 108 '118 Window 110 Abrasive pad 112 Abrasive layer 114 Back layer 120 Platform 121 ' > 154 Motor 124, 152 Drive shaft 125, 155 Axis 128 Groove 129 Coupling no Nozzle 132 Grinding fluid 140 Carrier head 142 Positioning ring 144 Elastic film 146a-146c Chamber 148a - 148c Area 150 Support structure 160 Optical monitoring system 162 Light source 164 Detector 166 Circuit 168 Optical head 170 fiber 172 trunk 174, 176 branch 190 controller 201 position 201a-201k point 202a-204a spectrum 202b-204b spectrum 204 arrow 210, 220, 230 track/sequence 212, 222 ' 232 214, 224, 234 line 300 measurement spectrum 310 Gallery 312 Index Value 314 Function 320 Reference Spectrum 330 Index Value 340 Record 350 Library 900 Method Index Value 39 201223702 902, 903 ' 904 ' 906 , 908 ' 908a , 908b ' 910 914 , 916 , 918 Step 1300 Flowchart 1302 , 1304 , 1306, 1308, 1310, 1312, 1314, 1318, 1330 Step 1400 Method 1402 ' 1404 & Gt 1406 ' 1408 Step 1500a ' 1500b Spectrum 1600 Standard Deviation 1602 Draw Point 1610 Period 1702 , 1704 , 1706 , 1708 Trajectory , 912 , 1316 , 40