TW531823B - Multi-variable monitoring method for semiconductor processing - Google Patents

Abstract

The present invention discloses a multi-variable monitoring method for semiconductor processing, which includes the following steps: selecting parameters in the computer for the processing steps to be monitored; next, selecting the statistical parameters from the selected parameters; after receiving the processing data by the computer from the processing machine, converting the processing data into the statistical data by the computer according to the selected statistical parameters; finally, forming the monitoring graph based on the statistical data.

Description

531823 五、發明說明（1) 詳細說明： 技術領域： 本發明係關於一種針對半導體製程之多變數監控的方 法，特別是關於一種提供機台異常偵測與診斷的二維和三 圍的監控圖，建立所謂機臺異常資料庫，幫助工程師偵測 機台異常，並指出發生異常的類型的方法。 發明背 半 片到可 製程又 可能南 良率， 常狀況 現 參數進 設定與 者，工 格。另 作法通 益。 半 不連續 體製程 景： 導體製 測試與 可以切 達四五 便必須 〇 行比較 行規格 範圍沒 程師常 外，因 常對半 程與設備是相當複雜與精密，一片晶片從裸 封裝，需經過約三至四百道製程， 割成好幾個步驟，而每一個步驟相 十個。若想要確保晶片的品質並提 針對每一個參數進行監控，避免機 先進半導體廠的作法是針對每個步 的設定，而每個參數因特性的不同 有一定的準則，造成此作法費時且 常因製程轉換或原件衰退而需重新 為半導體製程具有不連續性的關係 導體元件之故障分析與排除不會有 導體製程分析之困難與複雜，在於半導體 和批次的特性，兹以圖一說明。圖一為典 參數（例如薄膜沉積製程中的氣體流量）531823 V. Description of the invention (1) Detailed description: Technical field: The present invention relates to a method for multivariable monitoring of semiconductor manufacturing processes, and particularly to a two-dimensional and three-dimensional monitoring chart for providing machine abnormality detection and diagnosis. Establish a so-called machine anomaly database to help engineers detect machine anomalies and point out the types of abnormalities that occur. Invented the back half of the film to the process can be possible and the South yield rate, often the current situation is set by the parameters and workmanship. Another benefit. Semi-discontinuous system Cheng Jing: The conductor system test and can be cut to four or five years must be 0 line comparison line specifications range is not common, because often half-range and equipment is quite complicated and precise, a chip from bare packaging, need to pass about Three to four hundred processes are divided into several steps, and each step involves ten steps. If you want to ensure the quality of the chip and monitor each parameter, avoid the advanced semiconductor factory's method is to set for each step, and each parameter has certain guidelines due to different characteristics, resulting in this method is time-consuming and often Due to the process conversion or the decline of the original, it is necessary to re-construct the semiconductor process with discontinuity. The failure analysis and elimination of conductor components will not have the difficulty and complexity of conductor process analysis, which is due to the characteristics of semiconductors and batches. Figure 1 is a typical parameter (such as the gas flow rate in the thin film deposition process)

而每一道 關的參數 南製程的 台產生異 驟的每個 ，規格的 費力。再 更改規 ，習知的 太大的助 製程具有 型的半導 之曲線變And the parameters of each stage of the South process station are different each time, the specifications are laborious. Change the rules again. The too-large helper process has a semiconducting curve.

第5頁 531823 五、發明說明（2) 化，其橫轴代表時間，其縱轴則代表該製程參數之變化。 由圖一中可觀察得知，此製程可以分成四個步驟，由於每 步驟的設定值與變化要求都不一樣，因此在每一步驟切換 至下一步驟時，都會有不連續的現象發生。而傳統的單變 數統計控制只適用於產品品質變數的監控，大多屬於連續 的數據，對此具有非連續性的半導體製程無法有效應用， 致使增加問題分析之困難。Page 5 531823 V. Description of the invention (2) The horizontal axis represents time, and the vertical axis represents the change of the process parameters. It can be observed from Figure 1 that this process can be divided into four steps. Since the setting value and change requirements of each step are different, discontinuous phenomena will occur when each step is switched to the next step. The traditional single-variable statistical control is only applicable to the monitoring of product quality variables. Most of them belong to continuous data. The semiconductor process with discontinuities cannot be effectively applied, which makes it difficult to analyze the problem.

接下來請參考圖二，其為連續三片晶片之同製程參數 的曲線變化比較圖，其橫軸代表時間，其縱軸則代表該製 程參數之變化。值得注意的是這三片晶片的總反應時間不 同，分別為8 3秒、8 5秒和8 6秒，這是批次製程的特性。因 為半導體製程由許多步驟組成，其中許多步驟是以達到某 些反應條件為此步驟反應終點的判斷依據（例如沉積一層 薄膜至特定厚度），就因為每一片晶片的製程條件難以控 制到完全相同（例如晶片溫度可能會有正負攝氏0 . 5度的 誤差，因而造成薄膜沉積之速率不同），也因此造成相同 步驟會有不同反應時間的結果。Please refer to FIG. 2 for a comparison of the curve changes of the same process parameters for three consecutive wafers. The horizontal axis represents time and the vertical axis represents changes in the process parameters. It is worth noting that the total reaction time of the three wafers is different, which are 83 seconds, 85 seconds, and 86 seconds, respectively. This is a characteristic of the batch process. Because the semiconductor process consists of many steps, many of which are based on the determination of certain reaction conditions as the end point of the reaction of this step (such as depositing a thin film to a specific thickness), it is difficult to control the exact process conditions of each wafer ( For example, the wafer temperature may have an error of plus or minus 0.5 degrees Celsius, which results in different rates of thin film deposition), and also results in different reaction times for the same step.

以一個典型的八吋廠而言，需監控的參數可能多達上 萬個，並且參數間會彼此影響，使得元件之故障分析與排 除變得異常困難。假設此製程有五個步驟，而每個步驟有 1 0個參數，倘若依據習知的作法，需要5乘以1 0共5 0個監 控圖來控制此製程，通常此作法的缺點是：參數規格設定 太寬時，會發生機臺出了問題，卻沒有超出規格，失去監 控的目的；參數規格設定太嚴時，會發生參數常常超出規 531823 五、發明說明（3) 格，機臺卻沒有異常狀況，造成無謂的誤警與人員緊張。 另外，機台出現異常時，通常絕大多數的參數都會超出規 格，讓負責故障排除的工程師無從著手，不知道是以前曾 發生過的異常類型再度發生，還是產生新的異常類型，徒 使故障排除的時間增長。因此，傳統的單變數統計控制無 法提供系統性與有效的解決方法，不能協助工程師發現機 台異常，並指出發生異常的類型。 因此，發展出一種針對半導體製程之多變數監控的方 法，能監控每一步驟的每一參數，便成為半導體業界一 項十分重要的課題。 發明概述： 本發明 數監控的方 本發明 斷的二維和 助工程師偵 本發明係有 係藉由每個 的主要 法。 的次要 三圍的 測機台 關「半 目的為提供一種針對半導體製程之多變 目的 監控 異常 導體 參數按造一 用以代表此一製程。而 變數分析， 還提供機台 所謂機臺異 發生異常的 一指標的多 生。此方法 控圖，建立 常，並指出 為提供一種提供機台異常偵測與診 圖，建立所謂機臺異常資料庫，幫 ，並指出發生異常的類型的方法。 製程多變數監控之方法」，細言之 定的規則排列，組合成一組指標， 機台製程的異常監控，可以藉由此 判斷製程的正常操作與異常事件發 異常偵測與診斷的二維和三圍的監 常資料庫，幫助工程師偵測機台異 類型。For a typical 8-inch factory, there may be as many as tens of thousands of parameters to be monitored, and the parameters will affect each other, making component failure analysis and troubleshooting extremely difficult. Suppose there are five steps in this process, and each step has 10 parameters. If the conventional method is used, 5 times 10 and a total of 50 monitoring charts are needed to control the process. Generally, the disadvantages of this method are: parameters When the specification setting is too wide, there will be a problem with the machine, but it does not exceed the specification, and the purpose of monitoring will be lost. When the parameter specification is too strict, the parameter will often exceed the specification 531823 5. The invention description (3) grid, but the machine There is no abnormal situation, which causes unnecessary false alarms and personnel tension. In addition, when the machine is abnormal, usually most parameters will exceed the specifications, leaving the engineer responsible for troubleshooting unable to start, and it is not known whether the type of abnormality that has occurred in the past reoccurs, or whether a new type of abnormality is generated and the fault is caused. Excluding time increase. Therefore, the traditional single-variable statistical control cannot provide a systematic and effective solution, and cannot assist the engineer to discover the abnormality of the machine and point out the type of the abnormality. Therefore, the development of a method for multivariable monitoring of the semiconductor process, which can monitor every parameter of each step, has become a very important subject in the semiconductor industry. Summary of the invention: The method of the present invention for number monitoring The present invention is a two-dimensional and assisted engineer detection The present invention is related to each of the main methods. The secondary measurement of the measuring machine of the "Semi-purpose" is to provide a method for monitoring the abnormal conductor parameters of the semiconductor process to monitor the abnormal conductor parameters to create a representative of this process. The variable analysis also provides the machine so-called machine abnormality This method has multiple indicators. This method controls maps, establishes regularity, and points out a method for providing a machine abnormality detection and diagnosis map, establishing a so-called machine abnormality database, and pointing out the types of abnormalities that occur. The method of multi-variable monitoring ", detailed rules and arrangements, combined into a set of indicators, the abnormality monitoring of the machine process can be used to judge the normal operation of the process and the two-dimensional and three-dimensional measurement of the abnormal event detection and diagnosis The regular database helps engineers detect different types of machines.

第7頁 531823 五、發明說明（4) 本發明揭露一種針對 法，首先在電腦中針對欲 從所選取的參數中選取統 受製程數據之後，所述電 述製程數據轉換成統計數 監控圖。 所述統計參數係為有 值、平均值）、物理量（如 析值（如多項式分析）等等 工程師只需憑本發明 生過的異常類型再度發生 機臺回歸生產線的時間， 機臺之異常資料庫。此方 控圖，具有快速與正確的 半導體製程之多變數監控的方 監控之製程步驟選取參數，接著 計參數。所述電腦由製程機台接 腦依據所選取之統計參數，將所 據，最後依據所述統計數據形成 意義的統計值（如最大值、最小 斜率、反區點）、或數學代數分 〇 所形成之監控圖，就可以判定發 ，還是產生新的異常類型，縮短 藉此半導體廠可以建立所謂每一 法比起現行多達上萬個單變數監 優點。 發明的詳細說明： 本發明係有關「半導體製程多變數監控之方法」，細 言之係藉由每個參.數按造一定的規則排列，組合成一組指 標，用以代表此一製程。而機台製程的異常監控，可以藉 由此一指標的多變數分析，判斷製程的正常操作與異常事 件發生。此方法還提供機台異常偵測與診斷的二維和三圍 的監控圖，建立所謂機臺異常資料庫，幫助工程師偵測機 台異常，並指出發生異常的類型。 本發明係有關「半導體製程多變數監控之方法」，本Page 7 531823 V. Description of the invention (4) The present invention discloses a targeting method. After firstly selecting computer-controlled process data from the selected parameters in a computer, the electronic process data is converted into statistical monitoring charts. The statistical parameters are the value, average value, physical quantity (such as analytical value (such as polynomial analysis), etc.) The engineer only needs the time when the machine returns to the production line based on the type of abnormality generated by the present invention, and the abnormal data of the machine Library. This square control chart has the process steps of fast and accurate multivariable monitoring of the semiconductor process. The parameters are selected and then the parameters are counted. The computer is connected to the computer by the process machine according to the selected statistical parameters. Finally, according to the statistical data, a meaningful statistical value (such as the maximum value, the minimum slope, the inverse point), or a monitoring chart formed by mathematical algebra points 0 can be used to determine whether a new type of abnormality is generated or shortened. This semiconductor factory can establish the so-called advantages of each method over up to tens of thousands of single-variable monitors. Detailed description of the invention: The present invention is related to "methods of multi-variable monitoring of semiconductor processes". The numbers are arranged according to certain rules and combined into a set of indicators to represent this process. The abnormality monitoring of the machine process can be borrowed The multi-variable analysis of this indicator determines the normal operation of the process and the occurrence of abnormal events. This method also provides two-dimensional and three-dimensional monitoring maps of machine abnormality detection and diagnosis, and establishes a so-called machine abnormality database to help engineers detect The machine is abnormal, and the type of the abnormality is pointed out. The present invention relates to the "method of multivariable monitoring of semiconductor processes".

第8頁 531823 五、發明說明（5) 方法的技術手段在 程師經驗，而監控 由每個參數按照一 代表此一 標的多變 假設製程 習知的作 通常此作 出了問題 設定太嚴 狀況，造 台出現異 責故障排 異常類型 除的時間 參數選取 所有步驟 據。此方 相同的製 臨的問題 相同，使 程。 接下 製程之多 製程。而 數分析， 有五個步 法，需要 法的缺點 ’卻沒有 時，會發 成無謂的 常時，通 除的工程 再度發生 增長。本 其數據中 的每個參 法的突破 程，想要 是資料的 多變數分 來請參考 變數監控 於提出一個通用的模式，不需要所謂工 每一個步驟的每一個參數。細言之係藉 定的規則排列，組合成一組指標，用以 機台製程的異常監控，可以藉由此一指 判斷製程的正常操作與異常事件發生。 驟，而每個步驟有1 0個參數，倘若依據 5乘以1 0共5 0個監控圖來控制此製程， 是：參數規格設定太寬時，會發生機臺 超出規格，失去監控的目的；參數規格 生參數常常超出規格，機臺卻沒有異常 誤警、人員緊張與資源浪費。另外，機 常絕大多數的參數都會超出規格，讓負 師無從著手，不知道是以前曾發生過的 ，還是產生新的異常類型，徒使故障排 發明的技術手段在於將每一步驟的每個 具有統計意義的物理量，而形成一組由 數組成的數據，作為多變數監控的依 在於：多變數分析只能應用於數據長度 針對半導體製程進行多變數分析往往面 長度不一。此發明因為每組數據的長度 析可以應用在不連續且批次的半導體製 圖三，其為本發明所揭露之針對半導體 的方法的流程圖。首先在電腦中針對欲Page 8 531823 V. Description of the invention (5) The technical means of the method are based on the experience of the engineer, and the monitoring of each parameter according to a variable hypothesis process that represents this target is usually a problem. The problem setting is usually too strict. Select all the steps for the time parameter for the exception type and exception type. This side has the same problems and the same process. Too many processes. In numerical analysis, there are five steps. When the disadvantages of the required method are not present, it will become unnecessary and often, and the number of removal projects will increase again. For the breakthrough process of each parameter in the data, please refer to the multivariate points of the data. Please refer to the variable monitoring. To propose a general model, there is no need to call every parameter of every step of the process. In detail, the rules are arranged in accordance with a set of rules, which are combined into a set of indicators for abnormal monitoring of the machine process. One finger can be used to judge the normal operation of the process and the occurrence of abnormal events. Step, and each step has 10 parameters. If this process is controlled based on 5 times 10 times a total of 50 monitoring charts, it is: When the parameter specifications are set too wide, the machine will exceed the specifications and lose the purpose of monitoring. ; Parameter specifications The parameters often exceed the specifications, but the machine does not have abnormal false alarms, staff tension and waste of resources. In addition, most of the parameters of the machine often exceed the specifications, making it impossible for the negative teacher to start. It is not known whether it has occurred before or a new type of abnormality has occurred. The technical means of inventing the fault is to make every step of each step It is a statistically significant physical quantity that forms a set of data. As the basis for multi-variable monitoring, multi-variable analysis can only be applied to the length of the data. Multi-variable analysis for semiconductor processes often has different lengths. In this invention, the length analysis of each set of data can be applied to discrete and batch semiconductor mapping. This is a flowchart of the method for semiconductors disclosed in the present invention. First, target your desire in your computer.

第9頁 531823 五、發明說明（6) 監控之製程步驟選取參數（步驟1 0 1)。此步驟可選取所 有步驟的全部參數或依經驗選取部份步驟的部份參數。接 下來再從所選取的參數中選取統計參數（步驟1 0 2)，所 述統計參數係為有意義的統計值（如最大值、最小值、平 均值）、物理量（如斜率、反區點）、或數學代數分析值（如 多項式分析）等等。 舉例而言，圖四是快速熱製程（RTP， Rapid Thermal P r o c e s s )降溫步驟的A參數隨時間的變化圖，其中橫轴代 表時間，縱軸則代表A參數。其中曲線一表示第一批次之 晶片在此步驟反應了 2 5秒，曲線二表示第二批次之晶片在 此步驟反應了 2 6秒。面對製程數據資料點數不一的狀況， 處理這種接近直線的曲線，本發明會利用最大值、最小 值、斜率和平均值即（A_max、A_min、A_ave、A_slope)來 表示此步驟的統計參數，來取代原始參數。 接下來請參考圖五，其為快速熱製程（RTP， Rapid Thermal Process)降溫步驟的B參數隨時間的變化圖，其 中橫軸代表時間，縱軸則代表A參數。此製程B參數一共有 四個步驟，如圖五所示：則本發明會萃取[步驟一 (A — max、A — min、A — ave、A_s 1 ope ),步驟二（A — max、 A_min、A — ave、A — slope)，步驟三（A — max、A_min、 A__ave、A — slope)，步驟四（A — max、A_min、A — ave、 A _ s 1 o p e )]共十六個參數取代原先約1 7 5個不連續且批次的 原始資料。 請繼續參考圖三，在電腦由製程機台接受製程數據後Page 9 531823 V. Description of the invention (6) Select parameters for the process steps of monitoring (step 1 0 1). This step can select all parameters of all steps or select some parameters of some steps based on experience. Next, select statistical parameters from the selected parameters (step 102). The statistical parameters are meaningful statistical values (such as maximum value, minimum value, average value), and physical quantities (such as slope, inverse point). , Or mathematical algebraic analysis values (such as polynomial analysis), and so on. For example, Figure 4 is a graph of the A parameter change over time in the rapid thermal process (RTP, Rapid Thermal Process) step. The horizontal axis represents time, and the vertical axis represents A parameter. The curve one indicates that the wafers of the first batch reacted in this step for 25 seconds, and the curve two indicates that the wafers of the second batch reacted in this step for 26 seconds. Facing the situation that the number of process data points is different, when processing such a curve that is close to a straight line, the present invention uses the maximum value, minimum value, slope, and average value (A_max, A_min, A_ave, A_slope) to represent the statistics of this step. Parameter to replace the original parameter. Next, please refer to FIG. 5, which is a change diagram of B parameter with time of a rapid thermal process (RTP) cooling step, where the horizontal axis represents time and the vertical axis represents A parameter. There are four steps for the B parameter of this process, as shown in Figure 5. The present invention will extract [step 1 (A — max, A — min, A — ave, A_s 1 ope), step 2 (A — max, A_min , A — ave, A — slope), step three (A — max, A_min, A__ave, A — slope), step four (A — max, A_min, A — ave, A _ s 1 ope)] a total of sixteen The parameters replace about 175 discontinuous and batches of original data. Please continue to refer to Figure 3. After the computer receives the process data from the process machine,

第10頁 531823 五、發明說明（7) (步驟1 0 3)，電腦依據所選取之統計參數，將所述製程 數據轉換成統計數據（步驟1 0 4)，再依據所述統計數據 形成監控圖（步驟1 0 5)。其中不同批次製程之製程數據 之數據長度可能不相同，但經轉換成統計數據之後，不同 批次製程之統計數據之數據長度必然相同，使得多變數分 析可以應用在不連續且批次的半導體製程。 舉例而言，若某製程有兩步驟三個參數，其中步驟一 之參數為a、b、c，步驟二之參數為d、e、f，選取其最大 值（in a X )與平均值（a v e )為統計參數。假設晶片A在步驟一 中歷時1 0秒，在步驟二中歷時7 5秒，每秒取一製程數據， 共有8 5個製程數據。假設晶片B在步驟一中歷時8秒，在步 驟二中歷時7 2秒，每秒取一製程數據，共有8 0個製程數 據。此時製程數據之數據長度並不相同。本發明以最大值 (max)與平均值（ave)為其統計參數，使得晶片A與晶片B 各有如下1 2個統計數據，如此不同批次製程之統計數據之 數據長度必然相同。 晶片 A =[步驟一 (a_max, b_max, c — max，）， 步驟一（a — ave， b_ave, c — ave，）， 步驟二（d_max， e_max， f_max，）， 步驟二（d — ave， e — ave， f — ave，）] 晶片 B =[步.驟一 （a 一 max， b — max， c 一 max，）， 步驟一 （a_ave， b_ave， c_ave，）， 步驟二（d — max， e_max， f—max，），Page 10 531823 V. Description of the invention (7) (step 103), the computer converts the process data into statistical data according to the selected statistical parameters (step 104), and then forms a monitor based on the statistical data Figure (steps 105). The data length of the process data of different batch processes may be different, but after being converted into statistical data, the data length of the statistical data of different batch processes must be the same, so that multivariate analysis can be applied to discrete and batch semiconductors. Process. For example, if a process has two steps and three parameters, where the parameters of step one are a, b, and c, and the parameters of step two are d, e, and f, select the maximum value (in a X) and the average value ( ave) is a statistical parameter. It is assumed that wafer A took 10 seconds in step 1 and 75 seconds in step 2, and took one process data per second, and a total of 85 process data. Assume that the wafer B lasted 8 seconds in step 1, and took 72 seconds in step 2, and took one process data per second, and a total of 80 process data. At this time, the data length of the process data is not the same. The present invention uses the maximum value (max) and average value (ave) as its statistical parameters, so that each of wafer A and wafer B has the following 12 statistical data, so the data length of the statistical data of different batch processes must be the same. Chip A = [step one (a_max, b_max, c — max,), step one (a — ave, b_ave, c — ave,), step two (d_max, e_max, f_max,), step two (d — ave, e — ave, f — ave,)] chip B = [step. step one (a one max, b — max, c one max,), step one (a_ave, b_ave, c_ave,), step two (d — max , E_max, f—max,),

第11頁 531823 五、發明說明（8) 步驟二（d_ave， e_ave, f — ave，）] 接下來請參考圖六，其為本發明所形成之二維監控 圖。本發明的創新在於提供機台異常偵測與診斷的二維和 三維的監控圖，建立所謂機臺異常資料庫，幫助工程師偵 測機台異常，並指出發生異常的類型。其方法在於藉由多 變數分析得到的數據，選擇適當的三個或兩個數據當作三 維或二維空間的座標，藉由其在空間的分佈，可以定義機 臺的正常操作範圍，而遠離正常操作範圍的點，會依不同 的異常種類而有不同空間的分佈。茲以圖六之二維監控圖 作說明：圖六中最多點群聚的聚落，我們定義為正常操作 範圍，而當機臺出現異常時，會遠離此聚落。例如異常類 型1會分佈在此聚落的右上方，而異常類型2會分佈在此聚 落的正上方。工程師只需憑圖六，就可以判定發生過的異 常類型再度發生，還是產生新的異常類型，縮短機臺回歸 生產線的時間，藉此半導體廠可以建立所謂每一機臺之異 常資料庫。此方法比起現行多達上萬個單變數監控圖，具 有快速與正確的優點。 以上所述係利用較佳實施例詳細說明本發明，而非限 制本發明的範圍，而且熟知此技藝的人士亦能明瞭，適當 而作些微的改變與調整，仍將不失本發明之要義所在，亦 不脫離本發明之精神和範圍。Page 11 531823 V. Description of the invention (8) Step 2 (d_ave, e_ave, f — ave,)] Please refer to FIG. 6 for a two-dimensional monitoring chart formed by the present invention. The innovation of the present invention is to provide two-dimensional and three-dimensional monitoring maps of machine abnormality detection and diagnosis, establish a so-called machine abnormality database, help engineers detect machine abnormalities, and indicate the types of abnormalities that occur. The method is to use the data obtained by multivariate analysis, select the appropriate three or two data as the coordinates of the three-dimensional or two-dimensional space, and by its distribution in space, the normal operating range of the machine can be defined, and away from The points in the normal operating range will have different spatial distributions according to different types of anomalies. The two-dimensional monitoring chart in Figure 6 is used for illustration. The settlement with the most points in Figure 6 is defined as the normal operating range. When the machine is abnormal, it will be far away from the settlement. For example, anomaly type 1 will be distributed in the upper right of this settlement, and anomalous type 2 will be distributed directly above this settlement. The engineer only needs to rely on Figure 6 to determine whether the type of abnormality has occurred again, or whether a new type of abnormality has occurred. This shortens the time for the machine to return to the production line, so that the semiconductor factory can build a so-called abnormal database for each machine. Compared with the existing tens of thousands of single variable monitoring charts, this method has the advantages of quickness and correctness. The above description uses the preferred embodiments to explain the present invention in detail, but not to limit the scope of the present invention, and those skilled in the art will also understand that making small changes and adjustments appropriately will still lose the essence of the present invention. Without departing from the spirit and scope of the invention.

第12頁 531823 圖式簡單說明 圖式的簡要說明： 圖一為典型的半導體製程參數之曲線變化，其橫軸代 表時間，其縱軸則代表該製程參數隨時間之變化。 圖二為連續三片晶片之同製程參數的曲線變化比較 圖，其橫軸代表時間，其縱軸則代表該製程參數隨時間之 變化。 圖三為本發明所揭露之針對半導體製程之多變數監控 的方法的流程圖。 圖四疋快速熱製程（RTP，Rapid Thermal Pro cess)降 溫步，驟的A參數隨時間的變化圖，其中橫軸代表時間，縱 車由則代表A參數。 圖五為快速熱製程（RTP，Rapid Thermal Pro cess) P争 溫步，驟的B參數隨時間的變化圖，其中橫軸代表時間，縱 車由則代表A參數。 圖六為本發明所形成之二維監控圖。Page 12 531823 Brief description of the diagram Brief description of the diagram: Figure 1 shows the curve change of typical semiconductor process parameters. The horizontal axis represents time, and the vertical axis represents the change of the process parameter with time. Figure 2 is a comparison of the curve changes of the same process parameters for three consecutive wafers. The horizontal axis represents time, and the vertical axis represents the change of the process parameters with time. FIG. 3 is a flowchart of a method for monitoring multiple variables of a semiconductor process disclosed in the present invention. Figure 4: Rapid thermal process (RTP, Rapid Thermal Process) cooling step, the change of the A parameter over time, where the horizontal axis represents time, and the vertical axis represents the A parameter. Figure 5 shows the change of the B parameter over time in a rapid thermal process (RTP, Rapid Thermal Process). The horizontal axis represents time, and the vertical axis represents the A parameter. FIG. 6 is a two-dimensional monitoring chart formed by the present invention.

第13頁Page 13

Claims (1)

531823 六、申請專利範圍 申請專利範圍： 9 1. 一種針對半導體製程之多變數監控的方法，其包含： 在電腦中針對欲監控之製程步驟選取參數； · 從所選取的參數中選取統計參數； 所述電腦由製程機台接受製程數據； 所述電腦依據所選取之統計參數，將所述製程數據轉換 成統計數據； 依據所述統計數據形成監控圖。 2 .如申請專利範圍第1項所述之針對半導體製程之多變數 _ 監控的方法，其中所述統計參數係為統計數值。 3 .如申請專利範圍第2項所述之針對半導體製程之多變數 監控的方法，其中所述統計數值係為最大值。 4 .如申請專利範圍第2項所述之針對半導體製程之多變數 ^ 監控的方法，其中所述統計數值係為最小值。 5 .如申請專利範圍第2項所述之針對半導體製程之多變數 監控的方法，其中所述統計數值係為平均值。 6 .如申請專利範圍第1項所述之針對半導體製程之多變數 監控的方法，其中所述統計參數係為物理量。531823 6. Scope of patent application Patent scope: 9 1. A method for multivariable monitoring of semiconductor manufacturing processes, including: selecting parameters for the process steps to be monitored in a computer; selecting statistical parameters from the selected parameters; The computer receives process data from a process machine; the computer converts the process data into statistical data according to the selected statistical parameters; and forms a monitoring map according to the statistical data. 2. The method for monitoring multivariable variables of semiconductor manufacturing processes as described in item 1 of the scope of patent application, wherein the statistical parameters are statistical values. 3. The method for multivariable monitoring of a semiconductor process as described in item 2 of the scope of patent application, wherein the statistical value is a maximum value. 4. The method for monitoring multivariable variables of a semiconductor process as described in item 2 of the scope of patent application, wherein the statistical value is a minimum value. 5. The method for multivariable monitoring of semiconductor processes as described in item 2 of the scope of patent application, wherein the statistical value is an average value. 6. The method for multivariable monitoring of a semiconductor process according to item 1 of the scope of patent application, wherein the statistical parameter is a physical quantity. 第14頁 531823 六、申請專利範圍 7. 如申請專利範圍第6項所述之針對半導體製程之多變數 監控的方法，其中所述物理量為斜率。 8. 如申請專利範圍第6項所述之針對半導體製程之多變數 監控的方法，其中物理量為反區點。 9. 如申請專利範圍第1項所述之針對半導體製程之多變數 監控的方法，其中所述統計參數係為數學代數分析值。 1 〇 ·如申請專利範圍第9項所述之針對半導體製程之多變數 監控的方法，其中所述數學代數分析值係為多項式分 析0Page 14 531823 VI. Scope of Patent Application 7. The method for multivariable monitoring of semiconductor processes as described in item 6 of the scope of patent application, wherein the physical quantity is the slope. 8. The method for multivariable monitoring of semiconductor processes as described in item 6 of the scope of patent application, wherein the physical quantity is the anti-zone point. 9. The method for multivariable monitoring of semiconductor processes as described in item 1 of the scope of patent application, wherein the statistical parameters are mathematical algebraic analysis values. 1 〇 The method for multivariable monitoring of semiconductor processes as described in item 9 of the scope of patent application, wherein the mathematical algebraic analysis value is a polynomial analysis 第15頁Page 15