TWI227333B - In-situ detection of thin-metal interface using optical interference via a dynamically updated reference - Google Patents

Abstract

An invention for detecting an endpoint during a chemical mechanical polishing (CMP) process is provided. A reflected spectrum data sample is received that corresponds to a plurality of spectrums of light reflected from an illuminated portion of the surface of a wafer. The reflected spectrum data sample is normalized using a normalization reference comprising a first reflected spectrum data sample obtained earlier during the CMP process. In addition, the normalization reference is updated during the process using a second reflected spectrum data sample obtained earlier during the CMP process. The second reflected spectrum data sample is obtained after the first reflected spectrum data sample. In this manner, an endpoint is determined based on optical interference occurring in the reflected spectrum data.

Description

12273331227333

曰 修正 一、【發明所屬之技術領域】 、，t發明係關於一種在化學機械研磨程序中的終點偵 ’’則’更特別關於利用寬反射光譜之光干涉與連續更新基準 的終點偵測。 二、【先前技術】 I典型的半導體裝置製造中，積體電路裝置為多層結 士ft -V 丄 & ^ 在基板層，形成具有擴散區域的電晶體裝置。在 彳 將内接金屬線圖案化並電連接至電晶體裝置以 J立功此的裝置。如熟知的，圖案化的導電層藉由如 一氧化石夕之介質材料與其他導電層絕緣。當形成更多金屬 2與相連的介質層時，使介質材料平坦化的需求增加。若 =ί平坦化，由於表面地形更多變化，故附加金屬層的製 二只際3知更困難。在其他應用中，金屬線圖案用介質材 料形成及接著執行金屬化學機械研磨（CMP )作業以去 除過多的金屬。 在先 或刷拭站 亮晶圓的 研漿最常 並分佈在 預備的半 動、半導 生之摩擦 月’J的技術中 ，其中帶、 單面或雙面 用在移動預 預備表面及 導體晶圓之 體晶圓的移 之結合來完 ’ CMP糸統典型地完成帶、軌道的， 塾子，或刷子係用來擦洗、研磨及擦 。研漿係用來便利及增進CMP作業。 備表面上，例如帶、墊子、刷子等， 已研磨的、研磨的或其他由程序 表面。此分佈一般藉由預備表面的移 動及介於半導體晶圓與預備表面間I 成0 顯二^^製造程序之介質層丨〇2的橫剖面圖，Revision 1. [Technical field to which the invention belongs] The invention of t relates to the detection of an end point in a chemical mechanical polishing process, and more particularly to the detection of the end point by using light interference of a broad reflection spectrum and continuously updating a reference. 2. [Previous Technology] I In the manufacture of a typical semiconductor device, the integrated circuit device is a multilayer junction ft -V 丄 & ^ On the substrate layer, a transistor device having a diffusion region is formed. At 彳, the internal metal lines are patterned and electrically connected to the transistor device to make the device work. As is well known, the patterned conductive layer is insulated from other conductive layers by a dielectric material such as oxidized oxide. As more metal 2 is formed with the connected dielectric layer, the need to planarize the dielectric material increases. If = = flattening, the production of additional metal layers is more difficult because the surface topography changes more. In other applications, metal line pattern dielectric materials are formed and then a metal chemical mechanical polishing (CMP) operation is performed to remove excess metal. The polishing slurry of bright wafers in the first or brush station is most often distributed in the prepared semi-moving, semi-conductive friction month'J technology, in which strips, single-sided or double-sided are used to move the prepared surface and conductor The body of the wafer is combined with the movement of the wafer. The CMP system typically completes a belt, orbit, roller, or brush for scrubbing, grinding, and wiping. Grinding is used to facilitate and enhance CMP operations. Prepared surfaces, such as tapes, pads, brushes, etc., have been ground, ground or otherwise programmed. This distribution is generally based on the movement of the prepared surface and the cross section of the dielectric layer between the semiconductor wafer and the prepared surface.

第7頁 1227333 案號犯126205 五、發明說明（2) 二土建造鑲嵌與雙鑲嵌内接金屬.線。彳質層1〇2具有 =n質層丨02之蝕刻圖案表面之擴散阻障層1〇4。如孰知' 的，擴散阻障層典型為氮化鈦（TiN)、钽（τ〇、’氮化 ί/:”或(Μ)與纽（Τ〇的組合。擴散阻障 =4-旦沉積至期望厚度，銅層1〇6即以在介質層ι〇2中早 /蝕刻特被的方式形成在擴散阻障層上。一些過量的 J阻障及金屬材料亦無可避免地沉積在範圍區上。為了』 J覆蓋材料及為定義期望的内接金屬線與相關的 孔（未圖示），故執行化學機械平坦化（CMp)作業。 如上所述，CMP作業係設計自介質層1〇2上移去頂端金 f材料。例如，如圖1B所示，已移去銅層1〇6與擴散阻 層1〇4的覆蓋部分。如CMP作業中相同的，必持續cMp作 直到覆蓋金屬與擴散阻障材料104自介質層1〇2移去。然' 而，為了確保所有擴散阻障層104自介質層1〇2上移去了必 須有方法監控程序狀態及CMP進行期間的晶圓表面狀態。 此普遍稱為終點偵測。執行銅的終點偵測，因為銅不能& ，利用計時方式研磨。計時研磨無法用於銅，因為CMp^ 序的移除率對於銅層的計時研磨不夠穩定。CMp程 的移除率變化很大。因當到達終點時，需要監 定。在多步驟CMP作業中，需要確定多重終點：（1) 自擴散阻障層移去Cu，（ 2 )確定自介質層移去擴散阻 層。因此，利用終點偵測技術來確保已移去所有覆蓋材 料0 對於金屬CMP中的終點偵測已建議許多方法。先前 技術方^一般可分類為研磨之物理狀態的直接與間‘偵、Page 7 1227333 Case No. 126205 V. Description of the invention (2) Inlaid metal and double inlaid metal. The host layer 10 has a diffusion barrier layer 104 on the surface of the etched pattern of the n host layer 02. As known, the diffusion barrier layer is typically a combination of titanium nitride (TiN), tantalum (τ〇, 'nitride ί /:', or (M) and Ni (TO). Diffusion barrier = 4-denier Deposited to the desired thickness, the copper layer 10 is formed on the diffusion barrier layer in the manner of early / etching in the dielectric layer ι02. Some excess J barrier and metal materials are inevitably deposited on In the area. In order to cover the material and define the desired inscribed metal lines and related holes (not shown), a chemical mechanical planarization (CMp) operation is performed. As mentioned above, the CMP operation is designed from the dielectric layer The top gold f material is removed from 10 2. For example, as shown in FIG. 1B, the covering portion of the copper layer 10 6 and the diffusion resistance layer 10 4 has been removed. As in the CMP operation, the cMp operation must be continued until The covering metal and the diffusion barrier material 104 are removed from the dielectric layer 102. However, in order to ensure that all diffusion barrier layers 104 are removed from the dielectric layer 102, there must be a method to monitor the status of the program and the duration of the CMP. Wafer surface state. This is commonly referred to as end point detection. End point detection for copper is performed because copper cannot be & Method grinding. Timed grinding cannot be used for copper, because the removal rate of CMP ^ sequence is not stable enough for timed grinding of copper layer. The removal rate of Cmp process varies greatly. Because when the end point is reached, it needs to be monitored. In multiple steps In CMP operations, multiple endpoints need to be determined: (1) Cu is removed from the diffusion barrier layer, and (2) the diffusion barrier layer is removed from the dielectric layer. Therefore, endpoint detection technology is used to ensure that all covering materials have been removed. Many methods have been suggested for endpoint detection in metal CMP. The prior art methods generally can be classified as direct and indirect detection of the physical state of grinding,

第8頁 1227333Page 8 1227333

案號 92126205 五、發明說明（3) 測二直接的方法❹日月相外部信號源、或化學溶劑來探知 :晶圓狀態。s 一方面’間接的方 Π自然發生之物理或化學變化在機具内 = 5声間接：點偵測方法包含監控：研磨焊墊/晶圓表面。的 機具的震動、焊墊與研磨頭之間的摩擦力、研 2 Α研：學能，及聲音射出。溫度法利用放熱程序反應， 因為研磨研漿選擇性與研磨金屬臈反應。美國專 5美6043專^2為此方法的例子。美國專利第5,643，050號與 2ί: ’308, 438號揭示摩擦基準法，其中當研磨不 同金屬層時，監控馬達電流改變。 二終點偵測方法在歐洲專利申請案第肿0 739 W7 ’其將導因於研磨程序之聲音射出解調而產生 3 ί i ί貝訊。聲音射出監控-般用來偵測金屬終點。此 放詈I:研磨時發生之研磨行動。在距離晶圓預定位置處 ^風來感覺當材料移去之深度到達離介面特定之距 提二2 ί波’藉以產生輸出<貞測信號°所有這些方法 ，琴摆呈:彳癌的整體量測以及和程序參數裝置與消耗品的 ΐ產業的:然而，除了摩擦感測’沒有方法已達 /導雷直接：點偵測法利用聲波速度、光反射與干涉、阻抗 曰圓矣：利用特定化學溶劑產生之電化學能改變來監控 2曰^揭_。吴國專利第5,39 9, 234號與美國專利第5, 271， 控透7二利用聲波之金屬終點偵測法。這些專利說明監 二自二Ζ 1研聚傳送聲波速度以憤測金屬終點的方法。 —--二層通過至另一金屬層，聲波速度改變，且此已 1227333Case No. 92126205 V. Description of the invention (3) The direct method of measuring two: the external source of the moon and the moon phase, or a chemical solvent to detect: the state of the wafer. s On the one hand, the indirect method Π The physical or chemical changes that occur naturally in the machine = 5 sounds Indirect: The point detection method includes monitoring: polishing the pad / wafer surface. The vibration of the implement, the friction between the welding pad and the grinding head, research 2 Α research: academic ability, and sound emission. The temperature method uses an exothermic reaction because the grinding slurry selectively reacts with the ground metal rhenium. An example of this method is U.S.A. 5 US 6043. U.S. Patent Nos. 5,643,050 and 2 ':' 308, 438 disclose a friction reference method in which the motor current is monitored for changes as different metal layers are ground. The two-end detection method has been swollen in European Patent Application No. 0 739 W7 ′, which will demodulate the sound caused by the grinding process and generate 3 ί i ί bayon. Sound emission monitoring-generally used to detect metal end points. This release I: grinding action that occurs during grinding. At a predetermined position from the wafer, the wind feels when the depth of the material removal reaches a specific distance from the interface. Ti 2 2 'to generate an output < test signal. For all these methods, the pendulum presents: Measurement and program parameter devices and consumables for the ΐ industry: However, there is no method other than friction sensing '/ direct lead: point detection method uses sound wave speed, light reflection and interference, impedance, and circle: 利用Changes in the electrochemical energy produced by specific chemical solvents are used to monitor 2 ^^ _. Wu Guo Patent No. 5,39 9, 234 and U.S. Patent No. 5, 271, Controlled Transmission 72. The metal end point detection method using sound waves. These patents describe a method for monitoring the end point of a metal by transmitting sound velocities to a Z1 cluster. --- The second layer passes to another metal layer, the sound wave speed changes, and this has been 1227333

案號 9212620R 五、發明說明（4) 用來偵 感測器 終點偵 變，其 變一致 終點是 局部終 氣袖承 已 點。這 電磁光 信號。 偵測方 上及監 一金屬 測終點。 監控來自 測法。該 與當研磨 。不幸的 整體的， 點。此外 的線性研 有許多建 些建議可 譜之可見 美國專利 法，其中 控反射信 時，反射 此外，吳國專利第6，1 86,865號揭示利用 位於研磨帛墊下之流體軸承的济 1體壓力之 感測器係用來偵測研磨時流體壓力的改 自-材料層通過至另—材料層時剪力的改 ，此方法不耐用於製程改變。此外，偵測 此方法因而不能偵測晶圓表面上特定點的 ，專利第6’ 186, 86 5號的方法限於需要空 磨。 議利用來自晶圓表面的光反射來偵測終 分為兩類··利用雷射光源或利用涵蓋所有 範圍的寬帶光源，監控單一波長的反射光 第5, 433, 65 1號揭示利用單一波長的終點 來自雷射光源的光信號係衝擊在晶圓表面 號來偵測終點。當研磨自一金屬轉換至另 性的改變係用來偵測通過。 、 寬帶法依賴利用電磁光譜的多重波長之資訊。美國專 利第6，1 0 6，6 6 2號揭示利用分光計來獲得光學光譜之可見 範圍中反射光線的強度光譜。在光譜中選擇兩波胃長^，當 研磨自一金屬轉換至另一金屬時，提供對反射性改變的【 好敏感度。偵測信號接著藉由計算選擇之兩帶的平均強度 比率來定義。偵測信號中明顯的轉換表示自一金屬通過^ 另一金屬。 。 使用目前終點偵測技術的普遍問題為需要一些過度银 刻程度來確保所有導電材料（例如金屬材料或擴^阻$層 104 自介質層102上除去’來防止金屬線間不小心的電^ 1227333Case No. 9212620R V. Description of the invention (4) It is used to detect the end point change of the sensor. The end point is the local end point of the sleeve. This electromagnetic light signal. Detect and monitor a metal end point. Monitoring comes from the test method. The and when grinding. Unfortunately overall, point. In addition to the linear research, there are many suggestions that can be seen in the U.S. patent law, which controls the reflection of the letter, and the reflection. In addition, Wu Guo Patent No. 6,1,86,865 discloses the use of fluid bearings located under the abrasive pad. The sensor is used to detect the change of the fluid pressure during grinding from the change of the material layer to the change of the shear force when passing through the other material layer. This method is not resistant to process changes. In addition, this method cannot detect a specific point on the surface of the wafer. The method of Patent No. 6 '186, 86 5 is limited to the need for air grinding. It is recommended to use the light reflection from the wafer surface to detect the final classification into two types: · Use a laser light source or use a broadband light source covering all ranges to monitor the reflected light of a single wavelength No. 5, 433, 65 No. 1 reveals the use of a single wavelength The light signal from the laser light source hits the wafer surface number to detect the end point. When grinding changes from one metal to another, changes are used to detect passing. Broadband methods rely on the use of multiple wavelengths of information in the electromagnetic spectrum. U.S. Patent No. 6,106,62 discloses the use of a spectrometer to obtain an intensity spectrum of reflected light in the visible range of the optical spectrum. Select two wave lengths in the spectrum ^ to provide good sensitivity to changes in reflectance when grinding is switched from one metal to another. The detection signal is then defined by calculating the average intensity ratio of the two selected bands. A noticeable transition in the detection signal indicates passing from one metal to another. . A common problem with the current endpoint detection technology is that it requires some excessive silver engraving to ensure that all conductive materials (such as metallic materials or expansion layers 104 are removed from the dielectric layer 102 'to prevent accidental electricity between metal lines ^ 1227333

VJJU 月 五、發明說明（5) 接。不適當的終 a 質層1 02内之金屬·；广或過度研磨的副作用為在保持在介 質移去多於期望的\產生盤狀凹陷1 08。盤狀凹陷作用實 盤狀凹陷熟知為自=f材料並在金屬、線·留下盤狀特徵。 凹陷會導致期望的穑$;填裝金屬線之性能’❿太多盤狀 如U卜π f 體電路失去其原有的功能。 統及方法。另文善終咖準確性之終點谓測系 度。 糸、、先及方法應能準確地決定膜與層的厚 三、【發明内容】 长，=T t =田本發明藉由提供光終點偵測系統來滿足需 薄金屬區。為了處理周圍光線干 =本’二月之只施例利用連續更新動態正規化基準。在一 :=、#揭：了在化學機械研磨（cmp)程序期間偵測 、日:ΠΓ 射光譜資料樣本，其與自晶圓表面之 :射^反射之數條光譜一致。將該反射光譜資料樣本正規 化，其利用包含在該CMP程序早先得到之第一反射光譜資 料樣本的正規化基準。另外，更新該正規化基準，其利用 該CMP程序早先得到的第二反射光譜資料樣本。該第二反 射光譜資料樣本在該第一反射光譜資料樣本之後獲得。因 此’根據反射光譜資料中產生之光干涉決定終點。 在另一實施例中，揭示了向量中值濾波器方法，在化 學機械研磨（CMP)程序期間偵測終點。如上，接收反射 光譜資料樣本’其與自晶圓表面之照射區反射之數條光譜 第11頁 1227333VJJU Month 5. Description of Invention (5) Connect. Inappropriate end a metal in the stratum 102; the side effect of extensive or excessive grinding is to remove more than desired while maintaining the media, and to produce disc-shaped depressions. Disc-shaped depressions are well known as self-f materials and leave disc-like features on metals, wires, and wires. The depression will cause the desired performance of the filled metal wires to be too much disk-like. For example, a Ub π f body circuit loses its original function. System and methods. The end of the accuracy of Wenshan's final coffee is the measurement system. First, the first method should be able to accurately determine the thickness of the film and layer III. [Content of the invention] Long, = T t = This invention meets the need for thin metal areas by providing a light endpoint detection system. In order to deal with ambient light interference = this' February's example utilizes continuously updated dynamic normalized benchmarks. In one: =, # uncovering: The samples of the spectrum data detected during the chemical mechanical polishing (cmp) procedure are recorded, which are consistent with the spectra of the reflection from the surface of the wafer. The reflection spectrum data sample is normalized using a normalized reference of a first reflection spectrum data sample included earlier in the CMP procedure. In addition, the normalization benchmark is updated, using the second reflection spectrum data sample obtained earlier by the CMP program. The second reflection spectrum data sample is obtained after the first reflection spectrum data sample. Therefore, the end point is determined based on the light interference generated in the reflection spectrum data. In another embodiment, a vector median filter method is disclosed that detects an endpoint during a chemical mechanical polishing (CMP) procedure. As above, a sample of the reflection spectrum data is received ’and several spectra reflected from the illuminated area on the wafer surface. Page 11 1227333

一致。自該CMP程序早先得到之第一數個反射光譜資料樣 本選擇第-中值反射光譜資料樣本。將目前反射光譜資料 樣本正規化，其利用包含該中值反射光譜資料樣本的正規 化基準。另外，更新該正規化基準，其利用自該cMp程序 早先得到的第二數個反射光譜資料樣本選擇第二中值反射 光譜資料樣本。如上，根據反射光譜資料中產生之光干涉 決定終點。 本發明之另一實施例揭示了在化學機械研磨程序期間 偵測終點的另一方法。對晶圓的部分表面照射寬帶光，及 f收目前反射光譜資料樣本，其與自該晶圓之該表面的該 照射區反射之數條光譜一致。將目前反射光譜資料樣本正 ，化，其利用包含在該CMP程序早先得到之第一反射光譜 貧料樣本的正規化基準，及根據在該反射光譜資料中產生 之光干涉決定終點。另外，更新該正規化基準，其利用該 CMP程序早先得到的第二反射光譜資料樣本，其中該第二 反射，譜資料樣本在該第一反射光譜資料樣本之後獲得。 在一實施態樣中，該第一反射光譜資料樣本以預定之落後 基準延遲落後於該目前反射光譜資料樣本，其中該落後基 準延遲是代表介於目前反射光譜資料樣本與該第一反射光 譜資料樣本間反射資料樣本量的數目。另外，得到新的目 前反射光譜資料樣本，其與之後自該晶圓之該表面的該照 射區反射之數條光譜一致。在此例中，該第二反射光譜資 料樣本以該落後基準延遲落後於該新目前反射光譜資料樣 本〇 #矛"也’本發明實施例之動態正規化基準藉由處理光Consistent. The first sample of the reflectance spectrum data obtained earlier from the CMP program is the -median reflectance spectrum data sample. The current reflection spectrum data sample is normalized using a normalized benchmark that includes the median reflection spectrum data sample. In addition, the normalization benchmark is updated, which uses a second number of reflection spectrum data samples obtained earlier from the cMp program to select a second median reflection spectrum data sample. As above, the end point is determined based on the light interference generated in the reflection spectrum data. Another embodiment of the present invention discloses another method for detecting endpoints during a CMP process. A part of the surface of the wafer is irradiated with broadband light, and f is a sample of the current reflection spectrum data, which is consistent with several spectra reflected from the illuminated area of the surface of the wafer. The current reflection spectrum data sample is normalized, which uses the normalized reference of the first reflection spectrum lean sample included earlier in the CMP program, and determines the end point based on the light interference generated in the reflection spectrum data. In addition, the normalization benchmark is updated, which uses the second reflection spectrum data sample obtained earlier by the CMP program, wherein the second reflection, spectrum data sample is obtained after the first reflection spectrum data sample. In an implementation aspect, the first reflection spectrum data sample lags behind the current reflection spectrum data sample with a predetermined backward reference delay, wherein the backward reference delay represents a difference between the current reflection spectrum data sample and the first reflection spectrum data. The number of inter-sample reflection data samples. In addition, a new current reflection spectrum data sample is obtained, which is consistent with several spectra that are later reflected from the illuminated area of the surface of the wafer. In this example, the second reflection spectrum data sample lags behind the new current reflection spectrum data sample with the backward reference delay. # Spear " also '

第12頁 1227333 ---案號 92126205 五、發明說明（7)Page 12 1227333 --- Case number 92126205 V. Description of the invention (7)

路徑中外來周圍光線之改變特性來減少終點偵測錯誤。再 者’本發明之實施例使用光干涉來取代如習知終點偵測中 表面反射性的改變。因此，本發明之實施例有利地提供終 點偵測中增加的敏感度與耐用性。除了終點偵測，本發明 ^實施例可用來決定移去金屬覆蓋後晶圓中層的厚度。x以 前，需要度量衡工具來測量晶圓層的厚度。本發明之實施 例不需移走晶圓亦不需從單獨的機器測量即可測量晶^ ^ 的厚度。本發明之其他實施態樣與優點，自以下詳=說二 並參照依本發明原則之例子的附圖將可明瞭。 四、【實施方式】 一發明係揭示光終點偵測。本發明提供一種光終點伯 測系統，其利用動態更新基準來協助測量光干涉以決定、 CMP程序期間金屬層何時到達薄金屬區。特別地，^態更 新基準係用來減少偵測程序上周圍光線的影響。因此了終 點係根據發生在反射光譜資料中之光干涉決定，其為反射 自晶圓之不同層的光線中相差異之結果，且在上層金屬層 減為薄金屬區時發生。在下列說明中，為了完全了解本^ 明，將說明很多特定的細節。然而，對此技藝熟悉之人士 明瞭沒有這些特定的細節亦可操作本發明。在其他例子 中’為了不必要使本發明難懂，故不詳細說明熟知 步驟。 … 々π 圖2Α係顯示如本發明實施例的CMp系統，其中焊墊 250係設計來沿滾輪251轉動。滾筒2 54置於焊墊25〇之下以 & i、日日圓利用載子2 5 2將施加於其上之表面。如圖2 b所 1227333 i號 92m?nq 五、發明說明（8) ^ 、ς點偵/則係利用光偵測器2 6 0執行，豆中光線施$穿 透滾筒254、焊墊250，w ^ a ^ ^ τ 及轭加在研磨之晶圓20 0的表面。 :，到光終點偵測，焊塾溝槽25Qa係形成在焊塾25〇 實施例中，谭墊250可包含數個策略地放置在 ^ f & I 、不同位置的焊墊溝槽25〇a。焊墊溝槽250a係設 ;卜于；/二使研磨作業的衝擊最小。㊉了焊墊溝槽250a 一:亦定義在滾筒254中。滾筒溝槽254a係 X帶光束通過滾筒254、通過焊㈣0，及至在 研磨之晶圓200的期望表面上。 2利用光偵測器26〇 ’可確定自晶圓表面移走特定 伙、壬又此偵測技術係設計藉由檢驗由光偵測器2 6 0接 圖案來測量薄膜的厚度。因此，滾筒254係設計 加某種程度的後壓力至焊塾25G以便能自晶圓 200準確地移去層。 圖3係顯示如本發明實施例的在CMp程序間由寬帶光源 々射之晶圓300的-部份。曰曰曰圓3〇〇包含石夕基板3〇2、排列 在基板30 2上之氧化層304，與形成在氧化層3〇4上之銅層 306。銅層306表示在鑲㈣p程序間形成之覆蓋的銅。通 常，銅層30 6係放置在氧化層3〇4±，其在較早的步驟钱刻 j形成用以相接銅的渠溝。接著藉由研磨移去覆蓋銅以使 氧化層3 0 4暴露，因而僅留下溝渠内的傳導線。雙鑲嵌以 相似方法發生且允許同時形成金屬插塞與内連線。The changing characteristics of external ambient light in the path to reduce end-point detection errors. Furthermore, the embodiments of the present invention use light interference to replace changes in surface reflectivity as in conventional endpoint detection. Therefore, embodiments of the present invention advantageously provide increased sensitivity and durability in end point detection. In addition to endpoint detection, the embodiments of the present invention can be used to determine the thickness of the middle layer of the wafer after the metal cover is removed. Before x, metrology tools were needed to measure wafer thickness. The embodiment of the present invention can measure the thickness of the crystal without removing the wafer or measuring from a separate machine. Other implementation aspects and advantages of the present invention will be apparent from the following detailed description and reference to the accompanying drawings of examples according to the principles of the present invention. 4. [Embodiment] An invention discloses light end point detection. The present invention provides an optical endpoint measurement system that uses a dynamically updated reference to assist in measuring optical interference to determine when a metal layer reaches a thin metal region during a CMP process. In particular, the state update benchmark is used to reduce the effects of ambient light on the detection process. The end point is therefore determined based on the light interference that occurs in the reflection spectrum data, which is the result of phase differences in the light reflected from different layers of the wafer, and occurs when the upper metal layer is reduced to a thin metal region. In the following description, in order to fully understand this specification, many specific details will be explained. However, those skilled in the art will recognize that the invention may be practiced without these specific details. In other examples, to avoid obscuring the present invention, well-known steps are not described in detail. ... 々π Figure 2A shows a CMP system according to an embodiment of the present invention, in which the pads 250 are designed to rotate along the roller 251. The roller 2 54 is placed under the solder pad 25 and the surface to which it will be applied by the yen 2 using the carrier 2 5 2. As shown in Figure 2b, 1227333 i 92m? Nq V. Description of the invention (8) ^, ς spot detection / is implemented using the light detector 2 60, the light in the beans through the roller 254, solder pad 250, w ^ a ^ ^ τ and yoke are added to the surface of the polished wafer 200. : At the end of light detection, the solder pad groove 25Qa is formed in the solder pad 25. In the embodiment, the Tan pad 250 may include several pad pads 25 strategically placed at ^ f & I at different positions. a. The pad groove 250a is provided; and / / minimizes the impact of the grinding operation. The pad groove 250a is defined. It is also defined in the drum 254. The roller groove 254a is an X-band light beam that passes through the roller 254, through the welding pad 0, and onto the desired surface of the polished wafer 200. 2Using the photodetector 26o 'to determine the removal of a specific partner from the wafer surface. This detection technology is designed to measure the thickness of the film by inspecting the pattern of the photodetector 260. Therefore, the roller 254 is designed to apply a certain degree of back pressure to the solder pad 25G so that the layer can be accurately removed from the wafer 200. FIG. 3 shows a portion of a wafer 300 projected by a broadband light source between CMP processes according to an embodiment of the present invention. The circle 300 includes a Shi Xi substrate 302, an oxide layer 304 arranged on the substrate 302, and a copper layer 306 formed on the oxide layer 304. The copper layer 306 represents the copper that is formed between the damascene processes. Generally, the copper layer 306 is placed on the oxide layer 304 ±, which forms a trench for connecting copper in an earlier step. The cover copper is then removed by grinding to expose the oxide layer 304, leaving only the conductive lines in the trench. The dual damascene occurs in a similar way and allows the simultaneous formation of metal plugs and interconnects.

第14頁 在研磨程序間，本發明之實施例利用光干涉來決定銅 3 0 6何日$已移除。起初，如視線3 〇丨a所示，銅層μ 6相對地 j，約1 0, 0 0 0 A ，故因而不透光。因此，照射晶圓30〇表 1227333 曰 修正 案號 92126205 五、發明說明（9) 面的光線3 0 8係隨少數或無干涉反射回去。因此，當銅研 磨下來時，銅層306變成薄金屬，約3〇〇 —4〇〇 a 。此即為薄 金屬區。因此’如視線3 0 1 b所示，鋼層3 〇 6變透明及光線 可通過銅層3 0 6而照到下面的層。 當光線312開始穿透晶圓的各層時，產生光干涉。晶 圓的各層有反射指標’其為定義當光線312自一層通過至 另一層時，各層對光線3 1 2速度影響的特質。因此，光線 312的速度隨光線312自一物質通過至另一物質而改變。 在各層介面’光線3 1 2反射並返回至光偵測器。由於 在物質内部之速度已改變，故發生相變。因此，在自銅層 306表面反射之光線314與自氧化層3〇4表面反射之光線316 間有相差。同樣地，在自氧化層3 〇 4表面反射之光線3 1 6與 自基板3 0 2表面反射之光線3 1 8間有相差。當各種反射光線 314、316，與318互相影響時，產生光干涉。 因此，當銅層306是厚的時，由於光線3〇8無法穿透銅 層306，故不會發生相變，因而無干涉產生。然而，當銅 層30 6變很薄且透明時，因為在自晶圓30 0之各層反射光之 間發生相變，故發生干涉。此時，應停止研磨程序。 圖4係顯示如本發明實施例的在化學機械研磨程序期 間偵測終點方法4 0 0的流程圖。在作業4 〇 2中，獲得寬帶反 射資料及動態更新正規化基準。藉由照射寬帶光源至晶圓 的部分表面獲得寬帶反射資料。接著接收依照自晶圓表面 之照射部分反射光的光譜之反射光譜資料。Page 14 During the grinding process, an embodiment of the present invention uses optical interference to determine when copper 306 has been removed. At first, as shown by the line of sight 3a, the copper layer μ 6 is relatively j, about 10, 0 0 A, and is therefore opaque. Therefore, irradiating the wafer 30, table 1227333, amendment number 92126205 V. Description of the invention The light rays on the (9) plane 3 0 are reflected back with a few or no interference. Therefore, when the copper is ground down, the copper layer 306 becomes a thin metal, about 300-400a. This is the thin metal area. Therefore, as shown by the line of sight 3 0 1 b, the steel layer 3 06 becomes transparent and light can pass through the copper layer 3 6 to the lower layer. When light 312 starts to penetrate the layers of the wafer, light interference occurs. Each layer of the wafer has a reflection index, which is a characteristic that defines the effect of each layer on the speed of light 3 1 2 when light 312 passes from one layer to another. Therefore, the speed of the light 312 changes as the light 312 passes from one substance to another. At each layer interface, the light 3 1 2 is reflected and returned to the light detector. As the velocity inside the material has changed, a phase change occurs. Therefore, there is a difference between the light 314 reflected from the surface of the copper layer 306 and the light 316 reflected from the surface of the oxide layer 304. Similarly, there is a difference between the light 3 16 reflected on the surface of the oxide layer 3 04 and the light 3 1 8 reflected on the surface of the substrate 3 02. When various reflected rays 314, 316 and 318 interact with each other, light interference occurs. Therefore, when the copper layer 306 is thick, since the light 308 cannot penetrate the copper layer 306, no phase change occurs and no interference occurs. However, when the copper layer 306 becomes thin and transparent, interference occurs because a phase change occurs between the light reflected from each layer of the wafer 300. At this point, the grinding process should be stopped. FIG. 4 is a flowchart illustrating a method for detecting an end point during a chemical mechanical polishing process according to an embodiment of the present invention. In Assignment 202, obtain broadband reflection data and dynamically update the normalized benchmark. Broadband reflection data is obtained by irradiating a broadband light source onto a portion of the surface of the wafer. Then, reflection spectrum data according to the spectrum of the reflected light from the illuminated portion of the wafer surface is received.

圖5係顯示如本發明實施例的在CMP程序中來自不同點 之晶圓的寬帶反射光譜的光譜圖5 0 0。圖5 0 0繪製強度對應FIG. 5 is a spectrum diagram showing a broadband reflection spectrum of a wafer from different points in a CMP procedure according to an embodiment of the present invention. Figure 5 0 0 drawing intensity correspondence

第15頁 1227333 _ 案號 92126205__年月日___ 五、發明說明（10)Page 15 1227333 _ Case No. 92126205 __ Month and Day ___ V. Description of the invention (10)

於ι/又，其中；I為自由空間中的光波長。當發生光干涉 時，以；I為函數繪製強度提供無週期信號。因此，由於當 發生光干涉時，以1 /又為函數繪製強度提供週期信號，故 本發明之實施例以1/又為函數繪製強度。曲線5 02係顯示 當晶圓的銅層是厚的且因而不透光時之反射光譜。如先 前所述，當銅層是厚的時，由於光線無法穿透銅層且因而 不發生相變，故不發生干涉。此以曲線5 〇 2顯示，其不顯 示任何振盪。當銅層變更薄時，振盪開始在反射光譜中出 現，如曲線504a與504b所不，各表示當銅透明時之各點的 反射光譜。 更特定地’圖5 0 0係顯示當銅層厚度接近穿透深度 日守週期條紋或振盪開始沿1 /又或1 / n m軸之反射光譜中出 現，其中nm為10-9公尺。圖5中的各曲線為反射光譜R 〇/ 入）的例=，其中λ為3〇〇至7 0 0nm。反射波的電場大小與 基板上的單一介質層之入射波的比率之近似關係如 方程式（1 ): 之 化 基 1 ) R (1/ λ i2 π βYu //, where; I is the wavelength of light in free space. When light interference occurs, use I to provide an aperiodic signal for the function plotting intensity. Therefore, since the periodic signal is plotted as a function of the intensity when the optical interference occurs, the intensity is plotted as a function of the factor 1 /. Curve 5 02 shows the reflection spectrum when the copper layer of the wafer is thick and therefore opaque. As mentioned earlier, when the copper layer is thick, no interference occurs because light cannot penetrate the copper layer and therefore no phase change occurs. This is shown as curve 502, which does not show any oscillations. When the copper layer becomes thinner, oscillation starts to appear in the reflection spectrum, as shown by the curves 504a and 504b, each representing the reflection spectrum of each point when the copper is transparent. More specifically, FIG. 5 0 shows that when the thickness of the copper layer approaches the penetration depth, the sun-stroken periodic fringes or oscillations start to appear in the reflection spectrum along the 1 / or 1 / n m axis, where nm is 10-9 meters. Each curve in FIG. 5 is an example of a reflection spectrum R 0 /), where λ is 300 to 700 nm. The approximate relationship between the magnitude of the electric field of the reflected wave and the incident wave of a single dielectric layer on the substrate is as follows: Equation (1): Transforming basis 1) R (1 / λ i2 π β

一 r〇i 十 r12 e_ 其中，h與1^2為Fresnel係數。万為下列方程式（2 ) 相角：^2) /5=2n1d/A^ 其中d為介質層的厚度及〜為介質的反射係數。 。t ί中’因為外在因素可能發生反射資料中的小變 準使：ίΐ:點偵測程序上變化的影響，利用動態更新 曰n w : ^ t規化。更特別地，反射資料信號包含來 ---一·"、、且裝的，及光路徑中其他媒介如研磨程序中——翻 I1HA r〇i ten r12 e_ where h and 1 ^ 2 are Fresnel coefficients. The following equation (2) is the phase angle: ^ 2) / 5 = 2n1d / A ^ where d is the thickness of the dielectric layer and ~ is the reflection coefficient of the dielectric. . t ί 中 ’Because of external factors, small changes in reflection data may occur. Alignment: ίΐ: The effect of changes in point detection procedures, using dynamic updates. n w: ^ t normalization. More specifically, the reflection data signal contains -----, and installed, and other media in the light path, such as in the grinding process--turn I1H

第16頁 1227333 案號 92126205 五、發明說明（11) 本發明 光干涉 發明之 中其他 料樣本 減低周 分自光 自光組 間改變 之實施例藉由分析導因 的反射資料中的改變， 實施例將反射資料正規 媒介反射之周圍光線的 與較早之反射資料樣本 圍光線的影響。特別 組裝及光路徑中其他媒 裝及光路徑中其他媒介 使用之研漿的反射。然而， 於當接近薄金屬區時產生之 來執行終點偵測。因此，本 化以減少自光組裝及光路徑 影響。藉由創造目前反射資 間的比例，本發明的實施例 地，此比例有效地抑制大部 介反射之周圍光線。然而， 反射之周圍光線的影響隨時 圖6係圖表600，顯示與CMP程序之起始值一致之厚銅 層的反射強度之改變，其變化導因於光路徑特性隨時間的 改變。特別地，圖表6〇〇顯示不同波長，如波長⑽。、 602b，與6 0 2c之反射資料強度改變的百分比。如圖可見， ，變的百分比隨時間變化。另外，改變的百分比依波長而 定，各波長皆不同。例如，波長6 0 2a、6〇2b，與6〇2c的改 變互相皆不同。因此，在研磨程序期間，由於水的稀薄液 體媒介及光纖束與晶圓間注入之研漿，光路徑的傳送與反 射特徵改變。 此動態地改變水與研漿之成分可使在研磨程序開始時 枚集之基準光譜在隨後之研磨程序期間不適合於正規化。 為了處理此問題，本發明之實施例得到動態基準，其準確 反應任何時間反射光線之周圍狀況的狀態。 ~ 圖7係顯不如本發明貫施例的獲得寬帶反射資料及更 新正規化基準之方法4 0 2的流程圖。在最初的作業γ q 〇中， 執行前程序作業。前程序作業包含得到最初的正規化基 1227333 修正 案號 92126205 五、發明說明（12) f，例如第一反射貧料樣本，以及對此技藝熟悉的人士明 ’其他刖程序作業。一般，如以下更詳細之討論，將使 用敢初的正規化基準直至到達研磨程序之前程序延遲。 在作業702中，得到下個寬帶反射資料。如上所述， :圓表面部分係照有寬帶光源以得到與自晶圓表面的照射 邛分反射之光譜一致的反射光譜資料。如隨後更詳地的說 月此反射資料隨後根據光干涉為終點偵測分析。 =業704中，接著視終點偵測程序中之目前照射是否 刖f序延遲來做決定。本發明的實施例連續收集反射 貝；母移可彳于到一次反射資料樣本。各個反射資料樣本 :稱為「照射」。落後基準延遲為…照射」數的K ，數’正規化基準將落後目t「照射」。前程序延遲為將 弟一反射資料樣本用作正規化基準的「照 於落後基準延遲。 」数且相寺 圖8 A係顯示如本發明實施例的用以儲存反射資料樣 j之二維陣列80 0的圖。二維陣列8 0 0包含複數個陣列入口 Ma-802d，各陣列入口8〇2a —8〇2d包含在特定時間表示曰 圓的照射之反射資料的陣列。在圖8A的例子中，陣列入： =表一照射及陣列入口 802b為相等於前程序延遲的 …、射再者，陣列入口 802d為目前照射及陣列入口8 以洛後基準延遲落後於目前照射8〇2d的照射。如上 … 落後基準延遲為介於目前照射與將用來正規化之昭，L丄 照射數，及前程序延遲為將利用第一反射 ^ ;勺 化基準的照射數。 了十僳本為正規 1頭參考圖7，若終點偵測程序中的目前照射少於前 1227333 五、發明說明（13) — ίίίίτΟ^1^^ ° ^l，J ^ 射次^ 在作業706中，正規化基準為在照射〇之反 射：：娣* :其為第一反射資料樣本。跳至圖8A，第-反 、亲二；:' i刀以,列人σ 802a呈現。因此，直到研磨程序已 8 0 2 I I ^ Ϊ等於或大於前程序延遲的點時，陣列入口 8 0 2 a係用作為正規化基準。 目前KiKV在」乍業708中’當在終賴程序中的 特別-或大於則程序延遲時，正規化基準係更新。 準ίΐ㈡；；’利用最初陣列入口ma為正規化基 係更新。在：=於或大於前程序延遲時，正規化基準 口，其以等ϋΠ、隹正規化基準係更新至隨後之陣列入 杲之t招# 2:後=準延遲的量來落後於目前照射數。結 的落後吴準i ί接著用來使反射資料正規化至作業404中 曼土旱先，日，以下會更詳細討論。 中，：]:陸$圖8α中，落後基準延遲可定義為Η。在此例 目前8G2a係用作正規化基準直到照㈣。當 …、射數4於或大於前程序延遲，例 新至隨後之卩/列入在Ί射11期間，正規化基準係再次更 在此产、、兄下車j ，在此例中陣列入口相當於照射2。 之心傳送特：規化基準數係動態更新以更佳反射光路徑 第19頁 1227333 平 月 曰 案號921262防 五、發明說明（14) 為減少例如因照射期間之不成直線帶造成之偏遠劣的 照射之效果，纟發明之實施例可利用向量中值濾波器。圖 8B係顯示如本發明實施例的用以儲存反射資料樣本之二維 陣列800的圖，其中向量中值濾波器8〇4係用來產生正規化 基準。如上，二維陣列800包含複數個陣列入口 802a-802d，各陣列入口 802a_802d包含在特定時間表示晶 圓的照射之反射資料的陣列。陣列入口 8 〇 2 a _ 8 〇 2 c表示晶 圓表面的連續照射，而陣列入口 8〇 2d表示晶圓表面的目 前照射。 如上，以落後基準延遲落後於目前陣列入口 8〇Μ之陣 列入口 802b係用來決定正規化基準。然而，在圖8β所示之 實施例中，兩個其他陣列入口亦用來決定正規化基準。特 別地，三個連續陣列入口 802a —802c係提供給向量中值濾 波器804。在實施例中，陣列入口 8 〇2a先於陣列入口 〜 802b，及陣列入口 802c隨後之，而陣列入口8〇礼以落後基 準延遲落後於目前陣列入口8〇2d，以上皆提供至向量$ 濾波器804。 向量中值濾波器804藉由加總各像素值的平方來決定 各陣列入口 802a-802c的大小或無向量的距離量測。== 陣列入口 802a-802c之無向量量測接著以升冪順序分^， 及選擇中間的陣列入口為正規化基準。結果之正規化美 接著用來使反射資料正規化至作業4〇4中的落後基準光" 譜，以下會更詳細討論。在此情況下，向量中值濾波^ 8 04決定三個陣列入口的中位數值以用作正規化基“準，w 而防止偏遠值影響正規化程序。 ι ’因 I麵 第20頁 1227333 案號 92126205 曰 修正 五、發明說明（15) 回頭參考圖4，反射資料在作業404中正規化。使反射 資料正規化減少在資料中取樣變化的樣本。如先前所述， 當研磨帶中之終點窗移經終點偵測感測器時，晶圓表面照 射有寬帶光線及自晶圓表面反射之光線係記錄為反射資 料。因為由於外在因素可能發生資料的小變化，故反射資 料係正規化以減少終點偵測程序上變化之影響。 如上所述，反射資料信號包含來自晶圓、光組裝，與 光路徑中其他媒介的反射，如在研磨程序期間使用的研 漿。亦即，M = C + N，其中Μ為總量測信號、C相當於自 晶圓反射之光線的信號，及Ν為正規化信號。在正規化期 間’藉由透過基準光譜獲得之正規化信號分開各波長之測 量信號，及如下列之方程式（3 )所示，自此商數扣除1 以去除固定偏差：Page 16 1227333 Case No. 92126205 V. Description of the invention (11) The embodiment of the invention in which the sample of other materials in the optical interference invention of the present invention reduces the change in the interval between the light and the light is implemented by analyzing the changes in the reflection data of the cause. The example will reflect the effect of the surrounding light reflected by the regular medium of the reflected data and the surrounding light samples of the earlier reflected data. Special assembly and reflection of other media used in the light path. However, it is generated when the thin metal area is approached to perform end point detection. Therefore, it is localized to reduce the effect of self-assembly and light path. By creating the current reflection ratio, this embodiment effectively suppresses the ambient light reflected by most of the media. However, the effect of the reflected ambient light is at any time. Figure 6 is a graph 600 showing the change in the reflection intensity of a thick copper layer consistent with the initial value of the CMP procedure. The change is due to the change in the characteristics of the light path over time. In particular, the graph 600 shows different wavelengths, such as wavelength ⑽. , 602b, and the percentage change in the intensity of the reflection data at 60 2c. As can be seen in the figure, the percentage of change varies with time. In addition, the percentage change depends on the wavelength, and each wavelength is different. For example, the wavelengths 60 2a, 60 2b, and 60 2c are different from each other. Therefore, during the lapping process, the transmission and reflection characteristics of the optical path change due to the thin liquid medium of water and the slurry injected between the fiber bundle and the wafer. This dynamically changing composition of water and slurry can make the reference spectra of the set at the beginning of the milling process unsuitable for normalization during subsequent milling processes. In order to deal with this problem, an embodiment of the present invention obtains a dynamic reference which accurately reflects the state of the surrounding conditions of reflected light at any time. Fig. 7 is a flowchart of a method 402 for obtaining broadband reflection data and updating a normalized benchmark, which is inferior to the embodiment of the present invention. In the first job γ q 〇, a pre-program job is executed. The pre-program work includes obtaining the original normalized base 1227333 amendment No. 92126205 V. Description of the invention (12) f, such as the first reflection lean material sample, and persons familiar with the art will clarify other 'program work'. Generally, as discussed in more detail below, the initial normalization benchmark will be used until the process is delayed until the grinding process is reached. In operation 702, the next broadband reflection data is obtained. As described above, the round surface portion is illuminated with a broadband light source to obtain reflection spectrum data that is consistent with the spectrum of the radiation reflection from the wafer surface. As described in more detail later, this reflection data is then detected and analyzed based on light interference. In 704, a decision is then made based on whether the current exposure in the endpoint detection procedure is delayed. The embodiment of the present invention continuously collects reflection shells; the mother shift can be trapped in one reflection data sample. Each reflection data sample is called "irradiation". The delay behind the benchmark is K, which is the number of exposures, and the number 'normalized benchmark will be behind the exposure t. The pre-program delay is the "reflect to the backward reference delay." Using the first reflection data sample as the normalization reference. Figure 8A shows a two-dimensional array used to store the reflection data sample j according to the embodiment of the present invention. 80 0 figure. The two-dimensional array 800 includes a plurality of array entries Ma-802d, and each array entry 802a-802d contains an array of reflection data representing a circular illumination at a specific time. In the example of FIG. 8A, the array entry: = Table 1 irradiation and the array entry 802b are equal to the previous program delay ..., the second entry, the array entry 802d is the current irradiation and the array entry 8 lags behind the current irradiation by the reference delay 802d irradiation. As above ... The backward reference delay is between the current exposure and the number of L 丄 exposures that will be used for normalization, and the previous program delay is the number of exposures that will use the first reflection ^. A ten-headed copy is a regular one. Refer to Figure 7. If the current exposure in the endpoint detection program is less than the previous 1227333 V. Description of the invention (13) — ίίίτΟ ^ 1 ^^ ° ^ l, J ^ Shot times ^ at home 706 In the normalization criterion, the reflection at irradiation 0 :: 娣 *: it is the first reflection data sample. Skip to FIG. 8A, the first-anti, the second ;; 'i knife with, listed σ 802a. Therefore, until the grinding process has reached 80 2 I I ^ Ϊ equal to or greater than the point at which the previous process was delayed, the array entry 80 2 a is used as a normalization reference. At present, KiKV is in “Jiangye 708”. When the special-or-greater procedure is delayed, the normalization benchmark system is updated. Ϊ́㈡ίΐ㈡ ;; ’Use the initial array entry ma to update the normalized system. At: = at or greater than the previous program delay, the normalized reference port, which is updated by waiting for the 基准 Π, 隹 normalized reference system to the subsequent array entry # 2: after = the amount of quasi delay to lag behind the current irradiation number. The resulting backward Wu Zhun i is then used to normalize the reflectance data to operation 404. Manchurian drought first, Japan, will be discussed in more detail below. Middle,:]: Lu $ In Figure 8α, the backward reference delay can be defined as Η. In this example, the 8G2a system is currently used as a normalization benchmark until the photo is taken. When ..., the number of shots 4 is at or greater than the previous program delay, for example, new to the following / included in the period of shots 11, the normalized reference system is again produced in this, brother, get off j, in this example the array entry is equivalent于 illustrates2. Heart transmission feature: The standard reference number is dynamically updated to better reflect the light path. Page 19 1227333 Pingyue Case No. 921262 Prevention V. Description of the Invention (14) In order to reduce remote inferiority caused by non-straight bands during irradiation For the effect of irradiation, the embodiment of the invention can use a vector median filter. FIG. 8B is a diagram showing a two-dimensional array 800 for storing reflection data samples according to an embodiment of the present invention, in which a vector median filter 804 is used to generate a normalized reference. As described above, the two-dimensional array 800 includes a plurality of array entries 802a-802d, and each array entry 802a-802d contains an array of reflection data representing the irradiation of a crystal circle at a specific time. The array entrance 8 2 a _ 8 〇 2 c represents continuous irradiation on the wafer surface, and the array entrance 8 2 d represents the current irradiation on the wafer surface. As mentioned above, the array entry 802b, which is behind the current array entry 80M with the backward reference, is used to determine the normalization benchmark. However, in the embodiment shown in Fig. 8β, two other array entries are also used to determine the normalization benchmark. In particular, three consecutive array entries 802a-802c are provided to a vector median filter 804. In the embodiment, array entry 802a precedes array entry ~ 802b, and array entry 802c follows, and array entry 80 falls behind the current array entry 802d with a backward reference delay, all of which are provided to the vector $ filter器 804. The vector median filter 804 determines the size of each array entry 802a-802c or the vectorless distance measurement by summing the squares of the pixel values. == Array entry The 802a-802c vectorless measurement is then divided in ascending order ^, and the middle array entry is selected as the normalization benchmark. The normalized beauty of the results is then used to normalize the reflection data to the backward reference light " spectrum in job 404, which will be discussed in more detail below. In this case, vector median filtering ^ 8 04 determines the median value of the three array entries to use as the normalization basis "quasi, w" to prevent remote values from affecting the normalization process. Ι 'Case I, page 20, 1227333 No. 92126205, Revision V. Description of the Invention (15) Referring back to FIG. 4, the reflection data is normalized in operation 404. Normalizing the reflection data reduces sampling of changing samples in the data. As mentioned earlier, when the end point in the abrasive belt When the window moves past the end point detection sensor, the broadband surface is illuminated with broadband light and the light reflected from the wafer surface is recorded as reflection data. Because small changes in the data may occur due to external factors, the reflection data is normalized To reduce the impact of changes in the endpoint detection process. As mentioned above, the reflection data signal includes reflections from the wafer, light assembly, and other media in the light path, such as the slurry used during the grinding process. That is, M = C + N, where M is the total measurement signal, C is the signal of the light reflected from the wafer, and N is the normalized signal. During the normalization, it is obtained by passing through the reference spectrum The normalized signal separates the measurement signals of each wavelength and, as shown in the following equation (3), subtracts 1 from the quotient to remove the fixed deviation:

(3 ) R = (M/N ) -1= ( (C + N ) /N ) -1=C/N 在作 當銅 ，其 多項 曲線 成之 在作 高頻 利地 少高 衍生 ° 呈現 後， 資料 涉造 地， 可不 以減 業4 0 6中，利用多項式使正規化的反射資料反趨 層仍不透光時’反趨勢伸展反射曲線以減少振盪 可因不同於來自晶圓層之光干涉的因素造成。最 式適合於反射資料接著之後便減少。因此，反射 開始實質地平坦’接著準備由晶圓不同層的光千 較容易的振藍偵測。 業4 0 8中，移動平均濾波器沿丨/又軸應用。典型 率干擾的量呈現於反射資料曲線中。高頻率干擾 影響終點偵測程序。因此，濾波器係應用於曲線 頻率干擾。 變換接著應用於作業41 0中的反射資料。一般，(3) R = (M / N) -1 = ((C + N) / N) -1 = C / N is used as copper, and its multiple curves are formed after being presented as high frequency and low gain. When the data is related to the ground, it is not necessary to use the polynomial in 406 to reduce the vibration of the normalized reflective data when the counter-trend layer is still opaque. Caused by interference. The best formula is suitable for reflecting data and then reducing it. Therefore, the reflection starts to be substantially flat 'and then it is ready to be easily detected by the blue light of different layers of the wafer. In Industry 408, moving average filters are applied along the axis. The amount of typical rate interference is shown in the reflection profile. High frequency interference affects the endpoint detection process. Therefore, the filter system is applied to curve frequency interference. The transformation is then applied to the reflection data in job 410. general,

第21頁Page 21

12273331227333

號921?仍叩 五、發明說明（16) 持續的偏差或Dr S ϊ目Α ώ « 於反射光譜中的持圓表面收集的反射資料中。由 原點的大、、古疼丘κ 可以很大，故傅立葉變換可由在 峰。藉由將行：二i ΐ可控制及遮蔽在光譜較高區域的波 2糟m變換應用於反射資料，可降低 =中：表的形式中’反射資料曲線藉由除去固定偏差 曲線著應用至作業412中的反射資料。光譜窗使 之ii”不連續圓滑。光譜窗協助減少因反射光譜 ==續導致之傅立葉光譜中光侧，通常“ 射先禮包含循環或振盪的非整數時發生。 1+填塞（ZerT _Padding )接著應用至作業414中的反 、貝；、、。反射光譜資料的無填塞協助傅立葉變換調整至更 1的解析栅極。此程序基本在較細的柵極上***傅立葉變 才、此依序能增加波峰偵測中的準確度。如方法4〇〇稍後 之執行。在實施例中，藉由將反射光譜的像素數延伸至更 ί桃極纟執行無填I。未覆蓋有實際需求資料的延伸柵極 中之任何像素皆可填滿零值。 ^在作業41 6中，傅立葉變換應用於反射資料。傅立葉 變換將信號分為數個元件。因此，傅立葉變換可用來更佳 债測反射光譜中振盪圖案的發生。 —圖9係圖表9 0 0，顯示如本發明實施例的反射資料的傅 立葉、交換，其下方的介質層具有範圍6〇〇〇 — 1〇〇〇〇人之厚 度。傅立葉變換圖表90 0包含不透光銅反射曲線9〇2，其中 晶圓表面上銅層的厚度與穿透深度比較是非常大的，及包 έ薄金屬曲線9 0 4 ’其中銅層與穿透深度比輕异非堂撞No. 921? Still 叩 V. Description of the invention (16) Continued deviation or Dr S ϊ 目 Α «reflection data collected on a round surface in the reflection spectrum. Since the origin is large, the ancient pain mound κ can be large, so the Fourier transform can be made at the peak. By applying the line: 2i ΐ to control and shade the wave in the higher region of the spectrum, the 2m transformation can be applied to the reflection data, which can be reduced = Medium: In the form of a table, the reflection data curve is applied to Reflection data in homework 412. The spectral window makes it ii "discontinuous and smooth. The spectral window helps reduce the light side of the Fourier spectrum caused by the reflection spectrum == continued, which usually occurs when the first shot contains a non-integer number of cycles or oscillations. 1 + Padding (ZerT_Padding) is then applied to the inverse and shell in operation 414; ,,. The non-stuffing of the reflection spectrum data assists the Fourier transform adjustment to a more analytical grid. This procedure basically inserts a Fourier transformer on a thinner grid. This sequence can increase the accuracy in peak detection. This is performed as described later in Method 400. In the embodiment, the unfilled I is performed by extending the number of pixels of the reflection spectrum to a larger value. Any pixel in the extended gate that is not covered with the actual required data can be filled with zero values. ^ In Assignment 41.6, the Fourier transform is applied to the reflection data. The Fourier transform divides the signal into several components. Therefore, the Fourier transform can be used to better detect the occurrence of oscillation patterns in the reflection spectrum. —Figure 9 is a graph 900, which shows the Fourier, exchange of reflection data as in the embodiment of the present invention, and the dielectric layer below it has a thickness ranging from 600,000 to 100,000 people. The Fourier Transform Chart 90 0 contains the opaque copper reflection curve 902, where the thickness of the copper layer on the wafer surface and the depth of penetration are very large, and the thin metal curve 9 0 4 'where the copper layer and the through Penetration depth ratio

第22頁 1227333 修正 曰 Λ_Ά 五、發明說明（17) 的、。在以上之方程式（1 )與（2 )中，厚度d與波數I〆入 透過相表達疋相關的。因此，R ( 1 /又）的傅立葉變換缺 射至d空間： ' ' (4) RF(d) = F{R(1/X)} - -R(i/A) 圖9的傅立葉變換圖表9 0 0顯示CMP程序期間不同時間 的R (d)。自傅立葉變換圖表9〇〇可見，在銅厚度與穿遂 深度比較是非常大，即曲線9〇2時，在介質的厚度範圍中 傅立葉變換圖表900的大小，即60 0 0- 1 00 00 A是非常小 的。當研磨到達穿透深度時，如薄金屬曲線904所示，顯 著的波峰開始出現在介質厚度範圍内。如傅立葉變換圖表 900所示’薄金屬曲線904的波峰值出現在約8000A ，其在 此例中為銅層之下的介質屠厚度。 在其他實施例中，其中晶圓結構較複雜，傅立葉變換 的主要波峰呈現層結構的幾何配置。例如，在兩個厚度山 與4之層結構中，主要波峰會出現在屯與山+ d2。本發明之 貝&例利用此特性來偵測當金屬層到達薄金屬區時，CMP 程序期間的第一階段。若是銅，則穿透深度約為5 0 0 A ， 若是鎢，則穿透深度約為8 〇 〇 A 。 回頭參照圖4，在預定厚度範圍内的傅立葉變換光譜 中發現特定的波峰數。當知道下面的介質層之厚度時，窗 可集中在包含介質厚度之圖的區域。圖10係顯示在不同階 段時間之特定厚度限制的反射資料曲線之傅立葉變換的傅 立葉窗。在圖10的例子中，在銅層之下的介質層厚度範圍 在60 0 0- 1 0 0 0 0 A 。因此，傅立葉窗1 000係配置以顯示由 6 0 0 0 A上低厚度範圍（LTB )與1 0, 0 0 0 A之高厚度範圍 .............. ^一 _Page 22 1227333 Amendment Λ_Ά V. Description of the invention (17). In the above equations (1) and (2), the thickness d is related to the wave number I and the transmission phase expression 疋. Therefore, the Fourier transform of R (1 / again) is inferior to d space: '' (4) RF (d) = F {R (1 / X)}--R (i / A) Figure 4 Fourier transform chart 9 0 0 shows R (d) at different times during the CMP procedure. From the Fourier transform chart 900, it can be seen that the comparison between the copper thickness and the penetration depth is very large, that is, the curve 900, the size of the Fourier transform chart 900 in the thickness range of the medium, that is, 60 0 0-1 00 00 A Is very small. When the grinding reaches the penetration depth, as shown by the thin metal curve 904, significant peaks begin to appear within the thickness range of the medium. As shown in the Fourier transform chart 900, the peak value of the 'thin metal curve 904 appears at about 8000 A, which is the thickness of the dielectric layer below the copper layer in this example. In other embodiments, where the wafer structure is more complicated, the main peaks of the Fourier transform present the geometric configuration of the layer structure. For example, in a two-layer mountain structure with a thickness of 4, the main peaks will appear at Tun and Mountain + d2. The present invention ' s example uses this feature to detect the first stage during the CMP process when the metal layer reaches a thin metal region. In the case of copper, the penetration depth is about 500 A, and in the case of tungsten, the penetration depth is about 800 A. Referring back to Fig. 4, a specific number of peaks are found in a Fourier transform spectrum within a predetermined thickness range. When the thickness of the underlying dielectric layer is known, the window can focus on the area containing the map of the dielectric thickness. Figure 10 shows the Fourier transform of a Fourier transform of a reflection data curve with specific thickness limits at different stages of time. In the example of FIG. 10, the thickness of the dielectric layer under the copper layer ranges from 60 0 0 to 1 0 0 0 A. Therefore, the Fourier window 1000 series is configured to display the low thickness range (LTB) from 6 0 0 A and the high thickness range from 1 0, 0 0 A ......... ^ One _

12273331227333

(Η Ί B )建立之厚度内反身十咨♦丨1人 度範圍之間發現預定的波二間在繼與HTB定義之厚(Η Ί B) The thickness of the reflexes within the established thickness is found within the range of 1 person.

.乍業420中’加總在作業418中發現的波峰大 、；^業418中發現的波峰大小之總和稱為終點㈣（EPD )k唬。圖11係圖表1100 ’顯示以時間為函數之 期，現咖们虎，此時間係顯示為照射數。照射數 =在終點偵測程序的連續反覆期間得到之反射資料的順 1。如圖1100所示，在此例中，EPD信號11〇2在照射丨至約 84妆之CMP程序的較早階段期間維持低。接著，當在約昭 射90時銅接近薄金屬區，因為由於當銅層變薄且透明時之 光干涉，而在反射光譜資料中發生之振盪，故波峰大小曲 線1102陡升。 回頭參照圖4，在作業422中，依EpD信號是否大於預 先定義之臨限來做決定。通常選擇臨限來評估何時到達薄 金屬區。如圖11所示，當金屬層厚度與穿透深度比較為較 大時，選擇臨限11 04使得臨限11 04相對於EPD信號。再 者，選擇臨限11 04最好考慮落後基準延遲。 ° & 圖1 2係圖表1 2 0 0 ’顯不以時間為函數之作業41 8期間 發現的EPD信號。如上，在此例中，EPD信號1 2 0 2在照射1 至約84時之CMP程序的較早階段期間維持低。接著，當銅 接近薄金屬區時，EPD信號1202陡升，造成EPD信號1202 破壞臨限11 0 4。為保證適合的正規化，本發明之實施例在 E P D信號1 2 0 2破壞臨限11 〇 4之前通常利用大於所需時間的 落後基準延遲來橫越EPD信號1 202中的最後彎曲。此保證In Zhaye 420, the sum of the peaks found in job 418 is summed; the sum of the sizes of the peaks found in job 418 is called the end point (EPD). Fig. 11 is a graph 1100 'showing the period as a function of time, and the present time is displayed as the number of exposures. Number of exposures = 1 of the reflection data obtained during successive iterations of the endpoint detection process. As shown in Figure 1100, in this example, the EPD signal 1102 remained low during the earlier stages of the CMP procedure, which irradiated to about 84 makeup. Then, when the copper is near the thin metal region at about 90 °, the peak size curve 1102 rises sharply due to the oscillations in the reflection spectrum data due to the light interference when the copper layer becomes thin and transparent. Referring back to FIG. 4, in operation 422, a decision is made based on whether the EpD signal is greater than a predetermined threshold. Thresholds are usually chosen to assess when a thin metal area is reached. As shown in FIG. 11, when the thickness of the metal layer and the penetration depth are relatively large, the threshold 1 04 is selected so that the threshold 1 04 is relative to the EPD signal. Furthermore, it is better to consider the threshold delay behind the threshold of 11 04. ° & Fig. 12 is a graph of 1 2 0 0 ′ showing the EPD signal found during operation 41 8 as a function of time. As above, in this example, the EPD signal 1 2 0 2 remains low during the earlier stages of the CMP procedure at 1 to about 84 irradiance. Then, when the copper approaches the thin metal area, the EPD signal 1202 rises sharply, causing the EPD signal 1202 to break the threshold 1 104. To ensure proper normalization, embodiments of the present invention typically use a backward reference delay greater than the required time to traverse the final bend in the EPD signal 1 202 before the E P D signal 1 2 0 breaks the threshold 11 104. This guarantee

第24頁 1227333 修正 案號 9212fi2fl5 五、發明說明（19) 態更新程序以包含在到達薄金屬區之前僅得到的反射資 回頭參照圖4，若在作業418中發現之EpD信號少於 疋義之臨限，則繼續方法4〇〇以在作業4〇2中獲得下 帶反射資料。否則’方法例在作業424中完成。 因為此時已到達終點，故CMP程序在作業424中終止。 ^ f ^戶f不，本發明實施例之動態正規化基準藉由處理光 1 q ^岡1卜來周圍光線之改變特性來減少終點债測錯誤。圖 = = 1 300，顯示在作業418期間利用靜態正規化基準 ；；]n9態正規化基準發現之EPD信號。特別地，EPD信 ^。FPlT*利用包含最初反射資料照射之靜態正規化基準造 θ#仏唬1 304因利用包含如上述參照圖7、8人與^所述 落後…動態正規化基準造成。光路徑= 化臭準的、i之改變特性減少之後終點偵測程序中靜態正規 η。適用性。因此，在研磨程序中已到達真正終點 限1104 \ D信號1302可在圖上飄移更冑，最後通過臨 :。麸而* Ϊ ’當利用靜態正規化基準時可偵測出錯誤終 徑中實施例之動態正規化基準藉由處理光路 來減少終點偵測錯誤。 土早 設測i ί ί ΐ之其他實施例中’應用在epd信號的統計假 l丨^ i來決定何時到達薄金屬區。因為本發明之實施 2使=干涉來取代如習知終點债測中表面反射性的改 Ϊ盘發明之實施例有利地提供終點债測中增加的敏威 終點债測，本發明之實施例可用來tΐPage 24 1227333 Amendment No. 9212fi2fl5 V. Description of the invention (19) The state update procedure includes only the reflection data obtained before reaching the thin metal area. Referring back to FIG. 4, if the EpD signal found in operation 418 is less than the meaning of righteousness Limit, then proceed to method 400 to obtain lower band reflection data in operation 402. Otherwise, the method method is completed in operation 424. Because the end point has been reached at this time, the CMP procedure is terminated in job 424. ^ f ^ f f, the dynamic normalization benchmark of the embodiment of the present invention reduces the end-point debt measurement error by processing light 1 q ^ Gang 1 b to change the characteristics of the surrounding light. Figure = 1 300, showing the use of a static normalized reference during operation 418;] EPD signal found by the n9-state normalized reference. In particular, the EPD letter ^. FPlT * uses a static normalized reference that includes the initial reflection of the data. Θ # bluff1 304 is caused by using a dynamic normalized reference that includes backwards as described above with reference to Figures 7, 8 and ^. Light path = quasi-standard, static change in the end point detection procedure after i changes. applicability. Therefore, the true end point has been reached during the grinding process. The limit of 1104 \ D signal 1302 can drift more on the graph, and finally pass Pro:. Bran * Ϊ ’When using a static normalized reference, an erroneous terminal can be detected. The dynamic normalized reference of the embodiment reduces the end detection error by processing the optical path. As early as in the other embodiments of the invention, it is assumed that the statistical false of the epd signal is used to determine when to reach the thin metal area. Because the implementation of the present invention 2 makes the = interference instead of the modified surface reflectivity as in the conventional endpoint debt measurement. The embodiment of the invention advantageously provides an increase in the terminal debt measurement in the terminal test. The embodiment of the present invention can be used. Come tΐ

第25頁 1227333 案號 92126205 曰 修正 五、發明說明（20) 晶圓中層的厚度。以前，需要度量衡工具來測量晶圓層的 厚度。本發明之實施例不需移走晶1亦不需從單獨的機器 測量即可測量晶圓層的厚度。 雖然為了清楚的瞭解，已詳細說明前述的發明，但在 附加的申請專利範圍之範圍内可實行某些改變及修改。因 此，本實施例係認為是說明的且不受限制的，及本發明不 受限於本文之細節，但在附加之申請專利範圍的範圍與相 等内可修改。 iPage 25 1227333 Case No. 92126205 Revision V. Description of Invention (20) The thickness of the middle layer of the wafer. Previously, metrology tools were needed to measure wafer thickness. In the embodiment of the present invention, the thickness of the wafer layer can be measured without removing the crystal 1 and measuring from a separate machine. Although the foregoing invention has been described in detail for the sake of clarity, certain changes and modifications may be practiced within the scope of the attached patent application. Therefore, this embodiment is considered to be illustrative and not restrictive, and the present invention is not limited to the details herein, but may be modified within the scope and equivalent of the scope of additional patent applications. i

第26頁 1227333 修正 曰 案號9212防的 圖式簡單說明 五、【圖示簡單說明】 參照下列敘述並結合附圖可最了解其中有許多優點的 本發明： 圖1 A係顯示進行製造程序之介質層的橫剖面圖，其在 建造鑲嵌與雙鑲嵌内接金屬線； 圖1B係顯示已在CMP程序中移走之擴散阻障層及銅層 之覆蓋部分； 圖2 A係顯示如本發明實施例的CMp系統，其中焊墊係 設計來沿滚輪轉動； 圖2B係顯示如本發明實施例的終點偵測系統； 圖3係顯示如本發明實施例的在CMP程序間由多光譜光 照射之晶圓的一部份； 圖4係顯示如本發明實施例的在化學機械研磨程序期 間偵測終點方法的流程圖； 圖5係顯示如本發明實施例的在CMP程序中來自不同點 之晶圓的寬帶反射光譜的光譜圖； 圖6係顯示因光路徑角色的改變造成銅層的反射資料 隨時間改變之圖表； 圖7係顯示如本發明實施例的獲得寬帶反射資料及更 新正規化基準之方法的流程圖； 圖8A係顯示如本發明實施例的用以儲存反射資料樣本 之二維陣列的圖； 圖8B係顯示如本發明實施例的用以儲存反射資料樣本 之一維陣列的圖，其中向量中值濾波器係用來產生正規化Page 26 1227333 Brief description of the amendment to case number 9212 V. [Brief description of the illustrations] The invention with many advantages can be best understood by referring to the following description in conjunction with the drawings: Figure 1 A shows the manufacturing process A cross-sectional view of a dielectric layer, which is used to construct inlaid and double-inlaid metal wires. Fig. 1B shows the covering portion of the diffusion barrier layer and the copper layer that have been removed during the CMP process. Fig. 2 A shows the invention as shown in the present invention. The CMP system of the embodiment, wherein the pads are designed to rotate along the roller; FIG. 2B shows the end point detection system as in the embodiment of the present invention; FIG. 3 shows the multi-spectral light irradiation between CMP procedures according to the embodiment of the present invention A part of a wafer; FIG. 4 is a flowchart showing a method for detecting an end point during a CMP process according to an embodiment of the present invention; FIG. 5 is a flowchart showing a method from a different point in a CMP process according to an embodiment of the present invention; Spectral diagram of the wideband reflection spectrum of the wafer; Figure 6 is a graph showing the change of the reflection data of the copper layer over time due to the change of the role of the optical path; Figure 7 is a diagram showing the broadband reflection obtained according to the embodiment of the present invention A flowchart of a method for updating data and normalized benchmarks; FIG. 8A is a diagram showing a two-dimensional array for storing reflection data samples according to an embodiment of the present invention; FIG. 8B is a diagram for storing reflection data according to an embodiment of the present invention A graph of a one-dimensional array of samples, where a vector median filter is used to generate normalization

第27頁 1227333 案號 92126205 圖式簡單說明 基準 圖9係顯示如本發明實施例的反射資料的傅立葉變 換’其下方的介質層具有範圍6 0 0 0 - 1 0 0 0 0 A之厚度、；支 圖1 0係顯示如本發明實施例的在不同階段時^之特定 厚度限制的反射資料曲線之傅立葉變換的傅立葉窗； 圖11係顯示以時間為函數之作業期間發現的E p d信 圖1 2係顯示由於落後基準延遲以時間為函數之作業期 間發現的EPD信號；以及 … 圖13係顯示比較利用靜態正規化基準發現之EPD信號 人利用動態正規化基準發現之EPD信號的圖。 元件符號說明： 102〜介質層 1 〇 4〜擴散阻障層 106、306〜銅層 1 0 8〜盤狀凹陷 20 0、30 0〜晶圓 250〜焊墊 2 5 0 a〜焊塾溝槽 2 51〜滾輪 252〜載子 254〜滾筒 2 5 4 a〜滾筒溝槽 1227333 ___案號92126205_年月日 修正_ 圖式簡單說明 2 6 0〜光偵測器 3 0 1 a、3 0 1 b〜視線 3 0 2〜基板 3 0 4〜氧化層 308、312、314、316、318 〜光線 400、700〜方法Page 27 1227333 Case number 92126205 Schematic description of the reference Figure 9 shows the Fourier transform of the reflection data as in the embodiment of the present invention. The dielectric layer below it has a thickness of 6 0 0 0-1 0 0 0 0 A; Fig. 10 is a Fourier window showing a Fourier transform of a reflection data curve of a specific thickness limit at different stages as in the embodiment of the present invention; Fig. 11 is a graph showing Epd letters found during operation as a function of time. Series 2 shows that the EPD signal found during the operation as a function of time is delayed due to a backward reference; and ... FIG. 13 shows a graph comparing the EPD signal found using a static normalized reference and a person using the dynamic normalized reference. Description of component symbols: 102 ~ dielectric layer 1 〇4 ~ diffusion barrier layer 106, 306 ~ copper layer 1 0 8 ~ disc recess 20 0, 30 0 ~ wafer 250 ~ pad 2 5 0 a ~ solder groove 2 51 ~ Roller 252 ~ Carrier 254 ~ Roller 2 5 4 a ~ Roller groove 1227333 ___Case No. 92126205_Year Month and Day Revise_ Brief description of the drawing 2 6 0 ~ Light detector 3 0 1 a, 3 0 1 b ~ sight 3 0 2 ~ substrate 3 0 4 ~ oxide layer 308, 312, 314, 316, 318 ~ light 400, 700 ~ method

402 、 404 、 406 、 408 '410 、412 、414 、416 、418 、 420、422 > 424、70 0、702、704、70 6、708 〜作業 5 0 0〜光譜圖 502 、 504a 、 504b 、 902 、 904 、 1102 〜曲線 600 、 900 、 1100 、 1200 、 1300 〜圖表 602a、602b、602c 〜波長 8 0 0〜二維陣列 802a-802d〜陣列入口 804〜向量中值濾波器 1102、1 202、1 30 2、1 304 〜EPD 信號 11 0 4〜臨限 1 3 0 8〜終點402, 404, 406, 408, '410, 412, 414, 416, 418, 420, 422 > 424, 70 0, 702, 704, 70 6, 708 ~ Assignment 5 0 0 ~ Spectral graphs 502, 504a, 504b, 902, 904, 1102 ~ curve 600, 900, 1100, 1200, 1300 ~ graph 602a, 602b, 602c ~ wavelength 8 0 0 ~ two-dimensional array 802a-802d ~ array entry 804 ~ vector median filter 1102, 1 202, 1 30 2, 1 304 to EPD signal 11 0 4 to threshold 1 3 0 8 to end point

第29頁Page 29

Claims (1)

12273331227333 方法 •一種在化學機械研磨（CMP ) ，此方法包含以下作業： 接收目前反射光譜資料樣本， 射之數條光譜一致； 秋序期間偵測終點的 其與自晶圓表面之照射 區反 將該目前反射光譜資料樣本正規化，复 CMP程序早先得到之第一反 “ 匕含4 準，以及 $ &射光°曰貝枓樣本的正規化基 更新該正規化基準，其利用該CMp程 ^ 二反射光谁資料；1¾太 甘rb #斤 早先传到的第 久耵尤W貝枓樣本，其中該第二反射光詳Method • A kind of chemical mechanical polishing (CMP) method. The method includes the following operations: receiving a sample of the current reflection spectrum data, and the transmitted spectra are consistent; the end point detected during the autumn sequence and the irradiation area from the wafer surface are reversed. At present, the reflection spectrum data samples are normalized. The first inversion obtained by the complex CMP program earlier includes the 4 standard, and the normalization basis of the sample is updated to the normalization benchmark, which uses the CMp procedure ^ 2 Who is the reflected light? 1¾ 太 甘 rb # ^ The sample of the first Jiu Chiyou W Bei that was transmitted earlier, in which the second reflected light is detailed 第一反射光譜資料樣本之後獲得。 、〃，本在纪 2.如中請專利範圍幻項之在⑽程序期間偵測終點知 方法，更包含根據存在於該反射光譜資料令、+ ^ 定終點的作業。 卩中之先干涉來沒 期間偵測終點的 定之落後基準延 3.如申請專利範圍第1項之在CMP程序 方法，其中該第一反射光譜資料樣本以預 遲落後於該目前反射光譜資料樣本。 4·如申請專利範圍第3項之在CMP程序期間偵測終點的 方法，其中該落後基準延遲是代表介於該目前反射光譜資The first reflection spectrum data sample is obtained later. 2. The method of detecting the end point during the process of the patent scope magic item, as described in the above, further includes the operation of determining the end point based on the existence of the reflection spectrum data order, + ^. The first reference point in the interfering period is determined to lag behind the reference point. 3. For example, in the CMP method of the first scope of the patent application, the first reflection spectrum data sample lags behind the current reflection spectrum data sample in advance. . 4. The method for detecting the end point during the CMP procedure as in item 3 of the patent application, wherein the backward reference delay represents a difference between the current reflection spectrum information 料樣本與該第一反射光譜資料樣本間反射資料樣本量^數 目〇 5 ·如申請專利範圍第3項之在CMP程序期間偵測終點的 方法，更包含接收新的目前反射光譜資料樣本的作業，其 與之後自該晶圓之該表面的該照射區反射之數條光譜一 致。The number of samples of the reflection data between the sample of the material and the first reflection spectrum data sample ^ number 05. As in the method of detecting the end point during the CMP procedure of the third patent application, it also includes the operation of receiving a new sample of the current reflection spectrum data , Which is consistent with several spectra reflected from the illuminated area of the surface of the wafer later. 1227333 案號 92126205 六 、申請專利範圍 6. :範圍第5項之在CMP程序期間债測終點的 方法，，、中忒弟一反射光譜資料樣本以該落後基準延遲落 後於該新的目前反射光譜資料樣本。 7. : ρ奢專利範圍第2項之在CMP程序期間偵測終點的 方法，其中該光干涉是自該晶圓不同層反射之錢中相差 異的詰果。 、8·=請專利範圍第7項之在CMP程序期間谓測終點的 方法，/、中當上金屬層減為薄金屬區時，該光干涉產生。 9·如申請專利範圍第8項之在CMp程序期間偵測終點的 方法，更包含當振盪在根據該反射光譜資料之波數 生時決定之作業。 w τ I 1 〇·如申睛專利範圍第9項之在CMp程序期間偵測終點 的方法，其中該終點在當該波數圖中產生該振盪時產、生。 11· 一種在化學機械研磨（CMP)程序期間偵測終點的 方法，此方法包含以下作業： 接收目前反射光譜資料樣本，其與自 區反射之數條光譜一致； < …、町 亥c:p/序早先得到之第一數個反射光譜資料樣本 遠擇弟一中值反射光譜資料樣本； 將該目前反射光譜資料樣本正規化，其利用包 值反射光譜資料樣本的正規化基準，以及 ^新該正規化基準，其利用自該CMP程序早先得到的 數個反射光譜資料樣本選擇第二中值反射光譜資料樣1227333 Case No. 92126205 6. Application for Patent Scope 6 .: Method of Scope Item 5 for Determining the Debt During the CMP Procedure. The sample of the reflection spectrum data of Zhongdi is delayed behind the new current reflection spectrum by the backward reference Sample data. 7 .: The method of detecting the end point during the CMP procedure in item 2 of the luxury patent range, wherein the optical interference is a different consequence of money reflected from different layers of the wafer. 8 · = Please use the method of measuring the end point during the CMP procedure in item 7 of the patent scope. // When the upper metal layer is reduced to a thin metal area, this optical interference occurs. 9. The method for detecting an end point during a CMP procedure as described in item 8 of the scope of patent application, further including an operation determined when the oscillation is generated based on the wave number of the reflection spectrum data. w τ I 1 〇 The method of detecting an end point during a CMP procedure as described in item 9 of the patent scope, wherein the end point is generated and generated when the oscillation occurs in the wave number diagram. 11. A method for detecting an end point during a chemical mechanical polishing (CMP) process, the method includes the following operations: receiving a sample of current reflection spectrum data, which is consistent with several spectra of self-reflection; < The first number of reflection spectrum data samples obtained by p / order earlier is far from the first median reflection spectrum data sample; the current reflection spectrum data sample is normalized, which uses the normalized benchmark of the wrap-valued reflection spectrum data sample, and ^ The new normalization benchmark uses several reflection spectrum data samples obtained earlier from the CMP program to select a second median reflection spectrum data sample 第31頁 1227333 案號 92126205 曰 修正 六、申請專利範圍 1 2 ·如申請專利範圍第11項之在C Μ Ρ程序期間伯測終點 的方法，更包含根據在該反射光譜資料中產生之光干涉決 定終點的作業。 1 3 ·如申請專利範圍第11項之在C Μ Ρ程序期間彳貞測終點 的方法，其中該第一數個反射光譜資料樣本包含該CMP程 序早先得到之三個連續反射光譜資料樣本。 1 4 ·如申請專利範圍第1 3項之在C Μ P程序期間偵測終點 的方法，其中該第一數個反射光譜資料樣本之其一以預定 之落後基準延遲落後於該目前反射光譜資料樣^。 15·如申請專利範圍第14項之在CMP程序期間债測終點 的方法’其中該落後基準延遲是代表介於該目前反射光譜 資料樣本與該第一數個反射光譜資料樣本之其一間反射資 料樣本量的數目。 ' 16· —種在化學機械研磨（CMP)程序期間偵測終點的 方法，此方法包含以下作業： 對晶圓的部分表面照射寬帶光； )接收目前反射光譜資料樣本，其與自該晶圓之該表面 的δ亥知射區反射之數條光譜一致； ΓΜρΛ該目前反射光譜資料樣本正規化，用包含在該 CMP紅序早先得到之第一反射光譜資料樣本的正規化基 罕， 根據在該反射光譜資料中產生之光干涉決定終點，以 更新該正規化基準’其利用該CMp程序早先得到的第Page 31, 1227333, Case No. 92126205, Amendment VI. Application for Patent Scope 1 2 · If the method for the primary measurement of the end point during the CMP procedure is applied for item 11 of the patent application scope, it also includes interference based on the light generated in the reflection spectrum data Homework that determines the end point. 1 3 · The method of measuring the end point during the C MP procedure according to item 11 of the scope of the patent application, wherein the first plurality of reflection spectrum data samples include three consecutive reflection spectrum data samples obtained earlier by the CMP program. 1 4 · The method for detecting an end point during the C MP procedure according to item 13 of the patent application scope, wherein one of the first plurality of reflection spectrum data samples is delayed behind the current reflection spectrum data by a predetermined backward reference delay Like ^. 15. The method for measuring the end point of debt measurement during the CMP procedure according to item 14 of the scope of patent application, wherein the backward reference delay represents a reflection between the current reflection spectrum data sample and the first plurality of reflection spectrum data samples. The number of data samples. '16 · — A method for detecting the end point during a chemical mechanical polishing (CMP) process, the method includes the following operations: irradiating a part of the surface of the wafer with broadband light;) receiving a sample of the current reflection spectrum data, which is related to the wafer The spectra of the reflections in the delta region of the surface are consistent; ΓΜρΛ The current reflection spectrum data sample is normalized, and the normalization based on the first reflection spectrum data sample obtained earlier in the CMP red sequence is based on The light interference generated in the reflection spectrum data determines the end point to update the normalized reference 'which uses the 第32頁 1227333 92126205_年月 η_修正 六、申請專利範圍 二反射光譜資料樣本，其中該第二反射光譜資料樣本在該 第一反射光譜資料樣本之後獲得。 ^ 1 7·如申請專利範圍第1 6項之在CMP程序期間偵測終點 的方法，其中該第一反射光譜資料樣本以預定之落後基準 延遲落後於該目前反射光譜資料樣本。 ι 1 8·如申請專利範圍第1 7項之在CMP程序期間偵測終點 的方法’其中该落後基準延遲是代表介於該目前反射光譜 資料樣本與該第一反射光譜資料樣本間反射資料樣本量= 數目。 19·如申請專利範圍第18項之在CMP程序期間偵測終點 的方法，更包含接收新的目前反射光譜資料樣本的作業”’， 其與之後自该晶圓之該表面的該照射區反射之數條光j普一 致0 期間偵測終點 洛後基準延遲 20·如申請專利範圍第19項之在CMP程序 ，方法’其中該第二反射光譜資料樣本以該 落後於該新的目前反射光譜資料樣本。Page 32 1227333 92126205_year __ Amendment 6. Scope of patent application The second reflection spectrum data sample, wherein the second reflection spectrum data sample is obtained after the first reflection spectrum data sample. ^ 1 7. The method for detecting an end point during a CMP procedure according to item 16 of the patent application scope, wherein the first reflection spectrum data sample is delayed behind the current reflection spectrum data sample by a predetermined backward reference. ι 1 8 · The method for detecting the end point during the CMP procedure as described in item 17 of the patent application scope, wherein the backward reference delay represents a reflection data sample between the current reflection spectrum data sample and the first reflection spectrum data sample Amount = number. 19. The method for detecting the end point during the CMP procedure according to item 18 of the patent application scope, further including the operation of receiving a new current reflection spectrum data sample "', which is then reflected from the illuminated area of the surface of the wafer The number of light j is consistent with the zero detection period and the reference delay is 20. If the CMP procedure in item 19 of the patent application is applied, method 'where the second reflection spectrum data sample is behind the new current reflection spectrum. Sample data. 第33頁Page 33