12949711294971
V 九、發明說明： ' 【發明所屬之技術領域】 ' 本發明係為一種系統模式維度鑑識與重要變數篩選 之方法及系統，尤指~種用於模式建構時之系統模式維度 鑣識與重要變數篩選之方法及系統。 • 【先前技術】 • 在變數分析與系統模式建構時，有雨個因素非常的重 要：系統自由度(degree of freedom)的推測及系統獨立自變 數，亦即重要變數的篩選。在真實系統中，變數的個數往 往十分繁雜眾多，加上時間序列資料項的拓展，將增加建 立糸統板式的困難度’以經驗权式為例，過多的輸入變數 將導致對資料數據量的要求上升、增加建立時間、提高模 式複雜度及浪費計算的資源等等問題，並進而影響到整體 权式預測的表現。然而事實上，在系統模式中的變數，彼 • 此間都存在某種程度的線性/非線性之相依關係，真正獨立 的自變數僅存在於其中的-小部分而已，所以在進行系統 板式建立或錯誤分析之i，若能先針對資料進行變數分析 與判定，則將可以有效減少因輪人過多變數所造成之問 . 題。在進行變數分析與判定的步驟上，首先，必須先推求 • 系、統的自由度(或者可以稱之其獨立維度(independent dim—〇ns))以了解自變數的個數數量，接著再從眾多的變 數中挑選不少於此數目的合適者，在不過度失真的前提 1294971 下=二^維度，藉以簡化其所要處理的_。 在業界上有兩種方式可賴得上述 推測及系統獨立自變數的筛選，第一個 ' 自由度的 業經驗的工程師或摔作人α 疋和具有專 义知作人貝合作，透過工程師或 多年累積的經驗所形成的、、 訊斑建議，來辑;?曰多〃 1厍、，、口予糸統各種相關的資 I、魏认㈣、統自由度及系統獨立 部份專業人員所具有的4。$ θ ^丄 羔而大 不-定-韵入“有識驗而獲得的推測，並 肥句王面、咪入，也未必有充足的計算 所得到的結果常常是因人而異。 刀析故 第二個方式，則是依賴一些基本的統計技巧，如主值 二析法(PCA) ’來對系統自由度做推測，並經由相關係數 等比Ϊ方式來 1選取合適的變數，這樣做的好處是可以較廣 泛且冰入地^系統’分析變數彼此之間存在的各種關 係’但問題在於主值分析法本身並不提供任何明確的系統 自由度及重要交數資訊，使用者仍需由主值分析法的結果 自行U出#]疋—般而言，由主值分析法的結果來判定系 統自由度的職準則係透過由使用者自訂依系統特性及自 身經驗自，了-特徵值及系統變異解釋程度之最小門根值， 滿足此門檻值的最少主值個數即為推測之系統自由度，通 ，準則來做選擇變數結論，然而此通用準則中門檻值的設 疋兵系，本身的特性有十分密切的關係，不同的系統有不 同的汉疋’除了依賴自身經驗或透過試誤法(trial and error) 外，並/又有其它比較明確有效的做法，除此之外主值分析 7V IX. Description of invention: 'The technical field to which the invention belongs>> The invention is a method and system for system pattern dimension identification and important variable selection, especially the system mode dimension knowledge and importance when the model is constructed. Method and system for variable screening. • [Prior Art] • In the analysis of variables and the construction of system models, the rain factor is very important: the degree of freedom of the system and the independent independent variables of the system, that is, the screening of important variables. In the real system, the number of variables is often very complicated, and the expansion of time series data items will increase the difficulty of establishing SiS board. Taking the empirical weight as an example, too many input variables will lead to the amount of data. The requirements for rising, increasing setup time, increasing model complexity, and wasting computing resources, and thus affecting the performance of overall weighted forecasts. However, in fact, there are some linear/nonlinear dependencies in the variables in the system mode. The truly independent self-variables only exist in the small part of them, so in the system board establishment or If the error analysis is i, if the variable analysis and judgment can be performed on the data first, it will effectively reduce the problem caused by the excessive number of variables. In the steps of performing variable analysis and determination, first, we must first derive the degree of freedom of the system (or its independent dimension (independent dim-〇ns)) to understand the number of independent variables, and then Among the many variables, choose the appropriate number of the number, in the absence of excessive distortion of the premise 1294972 = two ^ dimension, in order to simplify the _ to be processed. There are two ways in the industry that can rely on the above speculation and the screening of independent independent variables of the system. The first engineer with the degree of freedom of the industry or the singer α 疋 and the professional knowledge of the person, through the engineer Or the experience accumulated over many years, the recommendations of the spot, the series; 曰 曰 〃 , , , , , , , 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种 各种Have a 4. $ θ ^ 丄 而 大 大 大 大 - 韵 - rhyme into "the speculation obtained by the test, and the fat sentence of the king face, immigrants, there may not be sufficient calculations to get the results often vary from person to person. The second way is to rely on some basic statistical techniques, such as the main value analysis method (PCA) to speculate on the degree of freedom of the system, and to select the appropriate variable through the correlation coefficient and other methods. The advantage is that the system can be used to analyze the various relationships between variables. The problem is that the main value analysis itself does not provide any clear system degrees of freedom and important intersection information. The user still needs to The results of the main value analysis method are self-extracting.] In general, the criteria for determining the degree of freedom of the system by the results of the main value analysis method are self-adjusted according to the characteristics of the system and its own experience. The minimum gate root value of the value and the degree of system variation interpretation. The minimum number of principal values that satisfy this threshold is the presumed system degree of freedom. The criterion is used to select the variable. However, the threshold of the general criterion is system, There are very close relationships between their own characteristics. Different systems have different habits. In addition to relying on their own experience or through trial and error, and / other clear and effective practices, in addition to the main Value analysis 7
12^4971 法針對重要變數之_選亦 輯，因此董卜般使用者而:φ確提供任何有效之計算邏 便，报難單·賴其㈣分析妓剌上並不方 析法的結果大多是_、^的結論’換言之，主值分 只會得到-組特徵值與翻有關，主值分析法最後 由度與決定變數對象軸；：:旦如何從中決定系統自 使用時必須同時兼且兩方^賴於經驗與系統知識，因此 果亦可能會因人而显。事^知，才能做出判斷，且其結 據應用的領域與對象，結:相對2 = ί狀況下=根 析人員建翻方絲進行分析^專業人貝，針統計分 鐘許多專利前案揭露利用主值分析法來執行各 、^、應用，如中華民國專利公告號第123〇263號「基 於二工具開發及控制目的之用以量化均一性類型並納入專家 ^識的方法」，利用主值分析法收集、標定數量資料以描緣 斗寸级及刀析半導體晶圓上不均勻性及提供回饋及控制以引 導半導體製造過程。 又’如中華民國專利公告號第1235311號「供資料分 析用之專家知識方法與系統」，利用主值分析法以產生新數 據組之一模型及多變數統計數，以分析在晶圓處理設備進 行的晶圓處理操作之良窳的方法。 再，如中華民國專利公告號第200515112號「利用自 適應多變項分析法來診斷一處理系統的方法與系統」，利用 主值分析法來診斷一處理系統的方法，尤其關於更新的 8 1294971 PCA之運用。請參考第一圖，係為半導體製造過程期間用 以處理基板的處理系統之監測方法流程圖，該方法包括自 處理系統獲得多個觀察的數據(S100)，運用靜態集中及比 例係數建立主值分析模型（S102)，自處理系統得到其他數 據(S104)’利用自適應集中及比例係數(sl〇6)，由其他的數 據及主值分析模型決定至少一個統計量（S108)，設定一控 制界限（Sll〇)，將至少一個統計量與控制界限作比較 (S112) ’備測到處理系統之失誤（§114)，及通知操作者 (S116) 〇 由上述該些專利文獻中，不難發現主要是利用主值分 析所後知之變數以製作出模型，但該些專利文獻皆沒有一 個月t的準則來做較具決定性的結論，無法明確決定系統 自由度與決定變數對象，是故由上述該些專利文獻所製作 出的模型無法提供準確的模型。 【發明内容】 於以上的問題，本發明的目的在於提供-種系統名 =又鑑識與重要變數篩選之方法及系統，本發明之方$ 、出精確的系統自由度及篩選出系統所要之變數，利月 逮：土由度及變數即可精確的建立出系統模式，再使用戶 器設用於-電腦，可對外部祕^ 、’、板式建構之變數功能之系統。 為了達成上述之目的，本發明係提出n统模式乡 1294971 度鑑識與重要變數篩選之方法，係包括讀取資料；利用使 用者所定義之相關條件將所讀取之資料分為輸入變數及輸 出又數’利用主值分析法以求得輸入變數之特徵向量及特 1值；由特徵值所計算之停止點以決定系統自由度；獲得 特，向量之第-個變數；訂定輸出變數與輸入變數之估計 目標規則；依據估計目標規則、第―個變數及祕自由度 為基^崎彳福餘的賴糾魏數量滿^為止；獲得初 =建礒之變數數量；及建立初始建議之變數數量之系統模The 12^4971 method is for the selection of important variables, so the user of Dongbo: φ does provide any effective calculation logic, report the difficulty list, and rely on it. (4) Analysis is not the result of the analysis method. Conclusion of ^, ^ In other words, the main value will only get - the group eigenvalue is related to the flip, the main value analysis method finally determines the variable object axis;:: How to decide from the system must be used at the same time ^Rely on experience and system knowledge, so the fruit may also be obvious to people. If you know what you know, you can make a judgment, and the field and object of the application are related. Conclusion: Relative 2 = ί under the situation = the root of the analysis of the staff to analyze the square wire for the analysis ^ Professional people, the needle statistics minute many patents before the case disclosure Use the main value analysis method to implement the various applications, such as the Republic of China Patent Bulletin No. 123〇263 “Methods for quantifying the type of homogeneity and incorporating expert knowledge based on the development and control of the two tools”, using the main The value analysis method collects and calibrates the quantity data to describe the level of non-uniformity on the semiconductor wafer and provides feedback and control to guide the semiconductor manufacturing process. 'For example, the Republic of China Patent Bulletin No. 1235311 "Expert knowledge methods and systems for data analysis", using the main value analysis method to generate a new data set model and multivariate statistics to analyze the wafer processing equipment A good method of performing wafer processing operations. Further, as in the Republic of China Patent Publication No. 200515112 "Method and System for Diagnosing a Processing System Using Adaptive Multivariate Analysis Method", a method for diagnosing a processing system using a principal value analysis method, especially regarding the updated 8 1294971 The use of PCA. Please refer to the first figure, which is a flow chart of a monitoring method for a processing system for processing a substrate during a semiconductor manufacturing process, the method comprising obtaining a plurality of observed data from a processing system (S100), and establishing a main value by using static concentration and a proportional coefficient. Analyze the model (S102), obtain other data from the processing system (S104) 'Using the adaptive concentration and the scale factor (sl〇6), and determining at least one statistic from other data and the main value analysis model (S108), setting a control The limit (S11〇), comparing at least one statistic with the control limit (S112) 'preparing the error of the processing system (§114), and notifying the operator (S116) 不 from the above patent documents, it is not difficult The discovery is mainly to use the main value analysis to know the variables to make the model, but these patent documents do not have a one-month t criterion to make more decisive conclusions, can not clearly determine the system degrees of freedom and determine the variable object, it is the reason The models produced by the above patent documents do not provide an accurate model. SUMMARY OF THE INVENTION In view of the above problems, an object of the present invention is to provide a method and system for screening a system name=and forensic and important variables, the method of the present invention, the precise system degree of freedom, and the variables required to screen out the system. , Li Yue arrest: soil and degree can accurately establish the system mode, and then the user device is set to - computer, can be used for external secret ^, ', plate construction variable function system. In order to achieve the above object, the present invention proposes a method for screening 1294971 degree identification and important variable screening in the n-mode mode, which includes reading data; and dividing the read data into input variables and outputs by using relevant conditions defined by the user. In addition, the main value analysis method is used to obtain the eigenvector and the special value of the input variable; the stopping point calculated by the eigenvalue determines the degree of system freedom; the special variable is obtained, and the first variable is determined; the output variable is determined The estimated target rule of the input variable; based on the estimated target rule, the first variable and the secret degree of freedom, the number of the 纠 魏 魏 ^ ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; Variable number of system modules
Wn雜式維度賴與重要變 數=之糸統以達成本發明之目的，本發明之系統應用於 數:功式建構之變 、功此，该系統包括輸入資料模組，其可由資料來 處取得資料；變數數量模組 、:μ 组，用w —連結於雜入資料模 量；變數八^ 模式建構所需之一最大變數數 將該，貝料依該使用者所定義之相關條 I二 入變數及至少—輸出變數；主值 、為至广輸 組，其可由特徵值中之第二次大幅 ’糸、、先自由度模 及獲得特徵值中之第二次大幅下降^做為系統自由度 數數量；變數個數計算模組，係電^所有特被值的變 模組，其可由系統自由度模組捉取：=系統自由度 ㈢加、交數；選擇變數 1294971 :::數刪數個數計算—，其可由增加變 獲得之選:估:目標單元為基礎所 所獲得之選擇變將選擇變數模組 ^本發明可解決習知之主值 在使用者無須得狀結果進行進-步的分析， 必要介入的情況下，自動系統相關知識且非 獲得建議之系& σ斤亚技今罘—次大幅下降點以 建，=:本度方 自動篩選出決定性之重要=疋： -=:::=_式_ 與‘ 【實施方式】 請同時參考第 圖 、第三圖及第四圖，第二圖係為去 勒明弟-貫關之纟频式維度麵與重錢數篩選ς 法流程圖，第三圖係為本發明第-實施例之增加變數數! 之方法流程圖’第四圖係為本發明第—實施例之選擇合= 變數之方法流程圖。於第二圖中，其方法流程係包括讀^ 11 1294971 複數個資料（S2〇〇) ’其中該些資料係可由線上連結之資料 庫、一資料庫或任何形式的檔案來取得。使用者定義該模 式建構所需之一最大變數數量（S2〇2)，其中該變數分類包 括但不限於一流量、一壓力、一溫度或一體積。利用使用 者所疋義之相關條件將所讀取之該些資料進行複數個輸出 變數及複數個輸入變數之變數分類（S2〇4)，利用主值分析 法以求得該些輸入變數之至少一特徵向量及至少一特徵值 (S206)，設定一相關群組，將經由該主值分析後之該些輸 入變數進行群組分類（S208)，以及由步驟S2〇6後可獲得該 些特徵向量之第一個變數（S210)，於步驟S2〇8中，該相關 群組係為計算後之，但不限於，一流量、一壓力、一溫度 或一體積，而於步驟S210中，該第一個變數係為最大特徵 值所對應的特徵向量，在各輸入變數方向上具有最大投影 量者。 除了上述步驟S210所獲得之第一個變數之外，其餘 的系統變數須接著經由步驟S206後，進行決定該些特徵值 所计异之一停止點為一糸統自由度（S212)，其中該停止點 係為該些特徵值之第二次大幅下降點，該系統自由度係為 該些特徵值之第二次大幅下降點之前所有特徵值的變數數 量。 於輸出端變數處，訂定該些輸出變數與該些輸入變婁文 之一估計目標規則（S214)，其中該估計目標規則之計算公 式如下： 12 1294971 /(X) =|α * Coc(x5Output)/b^CoC{X,Selected __Inputs) | · * *(1) 或 f{X) ~\\^^CoC[X^Output) \~\b^CoC{X,Selected_Inputs) || ee#(2) 數 上述之公式（1)或（2)中，該a及該b係為可調整變 該CoC係為相關係數函數，該X係為所選擇之變數。 上述估计目標規則之計算公式僅為實施例而已，不應 該被解釋為本發明之限制，然而只要符合“所獲得之系統 、艾數與已選擇之系統變數之間關係愈小愈好（亦即愈獨立 愈好），而與輸出變數之間關係愈大愈好，，條件，使用者可 依此條件而定義出多種的估計目標規則。 、依據該估計目標規則、該第一個變數及該系統自由度 為基礎以獲得其餘的變數直到該變數數量滿足為止 ^S216^)’其中步驟S216之輸入來源係包括上述獲得該些特 + 2畺之第一個變數（兑1〇)步驟、執行增加變數數量（S224〕 =或訂定該些㈣變數與該些輸人變數之—估計目標規 括S2^4)步驟’步驟S216之詳細流程請參考第四圖，係包 赖待增加魏(S234)，檢測職增加變數是否符合 -人2標細（S236)，若符合，麟待增加變數即成為 及數(S238)’料符合’則持續取得該待增加變數， 則择^否已達到該變數數量（湖）1判斷結果為是， 為^传—减建議之變數數量（S218)，反之，若判斷結果 、、^則回到取得該待增力口變數（S234)之步驟。 後得-初始建議之變數數量（S218)，其中該初始建議 1294971 之變數數量之輸入來源可為該相關群組之該些輸入變數， 接著建立该初始建議之變數數量之一系統模式（S22〇)，檢 測邊糸統模式所產生之一誤差是否為可接受（S222)，其中 該檢測該系統模式之步驟，若檢測結果為是，則結束該模 式建構之變數選擇流程，反之，若檢測結果為否，更包括 執行增加變數數量（S224)之步驟。 關於增加變數數量（S224)之步驟，請參考第三圖所 • 不，該增加變數數量之輸入來源包括決定系統自由度（S212) 以及由上述第二圖之S222步驟完成後進入S334之步驟所 組成者，當決定系統自由度或者增加系統變數個數之後， 則計算解釋程度（S228)，比對使用者設定的條件（s23〇)， 其中该預設條件係為一使用者所設定之一限定條件庫中所 取得者，該比對之步驟中，若比對結果為是，則輸出經比 對f合後之該待增加變數（S232)，如不符合所需解釋程度 之最小要求，即增加變數個數建議值（S334)。 丨於上述核測该糸統模式所產生之一誤差是否為可接 文（S222)步驟中，當檢測結果為否，則進入增加系統變數 (^334)及執行增加變數數量（S224)之步驟，其増加變數數 里之方法流程如第三圖所示。 請麥考第五圖，係為本發明第二實施例之系統模式維 度鑑識與重要變數篩選之方法詳細流程圖，係包括由線上 連結之貧料庫、一資料庫或任何形式的檔案中讀取系統所 要處理的資料⑽〇)，使用者定義該模式建構所需之一最 14 1294971 大變數數量（S302)，再將這些資料利用使用者所定義之相 關條件將所讀取之該些資料進行變數分類的動作⑽4)， ^述之變數包括但祕於流量、壓力或温料，再者，將 這些變數分類後的變數分成輸人變數⑽6)及輸出變數 (S308)兩種，接著，以主值分析⑽）法將以變數進行分 斤（S312)，並且將分析後的變數根據依存性將輪入變數進 2同群組的分類（S31G)，該不_群組分賴如：溫度、 或流…因執行主值分析的動作，所以輸入 =數、、坐由主值分析後將會產生至少—特徵向量⑻⑷及至 (::碰：綱，系統由該些一 由最大㈣上奴域分析所完成的該些特徵值的排列是 值’輯技術是依據所獲得之特徵 解睪百刀比疋否滿足使用者所定之^丨欠 :法;1會峨這些特徵值中第二次二== 所幅下_是從钟雜錢此之間比例關係 而本發明之方法所提及之第二次大幅下降點之 计-Τ方式在此舉一例子來說明，若 八個特徵值，iL中4丨至分析法求得出 方式 \ ^ 的值’特徵值為遞減的排列 來的，大幅：:二算方勸^ 一 為5%’而第二組 次大幅如此可得之第三'组係為本發明所謂之第二 15 1294971The Wn heterogeneous dimension depends on the important variable = the system of the present invention. The system of the present invention is applied to the number: the construction of the power structure, and the system includes an input data module, which can be obtained by data. Data; variable quantity module, : μ group, with w - linked to the miscellaneous data modulus; variable eight ^ mode construction required one of the maximum number of variables, the bead material according to the user defined correlation I The second-in variable and at least the output variable; the main value is the maximum-transmission group, which can be determined by the second large 糸 of the eigenvalues, the first degree of freedom, and the second significant drop in the eigenvalues. The number of system degrees of freedom; the variable number calculation module, which is a variable module of all special values, which can be captured by the system degree of freedom module: = system degree of freedom (3) plus, intersection number; selection variable 1294972 ::: Counting the number of calculations - which can be obtained by adding variables: Estimating: the selection of the target unit is based on the selection of the variable module. The invention can solve the conventional main value without the user having to obtain the result. Step-by-step analysis, necessary intervention In the case, the knowledge of the automatic system and the non-recommended system & σ 亚 亚 罘 罘 罘 罘 罘 罘 罘 罘 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅 大幅_ and ' [Embodiment] Please refer to the figure, the third picture and the fourth picture at the same time. The second picture is the flow chart of the 维度 式 弟 贯 贯 贯 , , , , , , The third figure is a flowchart of the method for increasing the number of variables of the first embodiment of the present invention! The fourth figure is a flow chart of the method for selecting the combination = variable of the first embodiment of the present invention. In the second figure, the method flow includes reading a plurality of data (S2〇〇) ’, wherein the data can be obtained by an online linked database, a database or any form of file. The user defines one of the maximum number of variables (S2 〇 2) required for the construction of the pattern, wherein the variable classification includes, but is not limited to, a flow rate, a pressure, a temperature, or a volume. Using the relevant conditions defined by the user, the read data is classified into a plurality of output variables and a plurality of input variables (S2〇4), and the main value analysis method is used to obtain at least one of the input variables. a feature vector and at least one feature value (S206), setting a correlation group, grouping the input variables analyzed by the primary value into groups (S208), and obtaining the feature vectors after step S2〇6 The first variable (S210), in step S2〇8, the relevant group is calculated, but is not limited to, a flow rate, a pressure, a temperature or a volume, and in step S210, the first One variable is the eigenvector corresponding to the largest eigenvalue, and has the largest projection amount in each input variable direction. In addition to the first variable obtained in the above step S210, the remaining system variables must be determined through step S206 to determine that the stop points of the feature values are one degree of freedom (S212), wherein the stop The point is the second significant drop point of the feature values, and the system degrees of freedom are the number of variables of all the feature values before the second significant drop point of the feature values. At the output variable, the output variables are determined and the target rule is estimated (S214), wherein the estimated target rule is calculated as follows: 12 1294971 /(X) =|α * Coc(x5Output) /b^CoC{X,Selected __Inputs) | · * *(1) or f{X) ~\\^^CoC[X^Output) \~\b^CoC{X,Selected_Inputs) || ee#(2 In the above formula (1) or (2), the a and the b are adjustable, and the CoC system is a correlation coefficient function, and the X system is the selected variable. The calculation formula of the above estimation target rule is only an embodiment and should not be construed as a limitation of the invention, but as long as the relationship between the obtained system, the number of the selected system and the selected system variable is as small as possible (ie, The more independent, the better, and the greater the relationship with the output variable, the better, condition, the user can define a variety of estimated target rules according to the condition. According to the estimated target rule, the first variable and the Based on the system degree of freedom to obtain the remaining variables until the number of variables is satisfied, the input source of step S216 includes the step of obtaining the first variable (for the first parameter) of the special + 2畺, and executing Increase the number of variables (S224) = or set the (four) variables and the input variables - the estimated target includes S2^4) Steps of the detailed process of step S216, please refer to the fourth figure, the package depends on the increase of Wei ( S234), whether the test increase variable meets the -person 2 standard (S236), if it is met, the lining to increase the variable becomes the sum (S238) 'material meets' and continues to obtain the variable to be added, then the choice has been reached The change The number (lake) 1 judges that the result is yes, and the recommended number of variables is reduced (S218). On the other hand, if the result is judged, the step returns to the step of obtaining the variable to be boosted (S234). - the number of variables of the initial recommendation (S218), wherein the input source of the variable number of the initial recommendation 1294972 may be the input variables of the relevant group, and then establish one of the system variables of the initial suggested variable number (S22〇), Detecting whether one of the errors generated by the edge mode is acceptable (S222), wherein the step of detecting the system mode, if the detection result is yes, ending the variable selection process of the mode construction, and vice versa, if the detection result is no And the step of performing the step of increasing the number of variables (S224). For the step of increasing the number of variables (S224), refer to the third figure. • No, the input source of the increased number of variables includes determining the degree of system freedom (S212) and After the step S222 of the second figure is completed, the process proceeds to the step S334. After determining the degree of system freedom or increasing the number of system variables, the degree of interpretation is calculated (S228). Comparing the condition set by the user (s23〇), wherein the preset condition is obtained by one of the user-defined conditional libraries, and in the step of comparing, if the comparison result is yes, the output is output. After the comparison of the f-addition variable (S232), if the minimum requirement of the required degree of interpretation is not met, the variable number recommendation value is increased (S334). 丨 One of the above-mentioned verification results is generated. Whether the error is in the splicable text (S222) step, when the detection result is no, the steps of increasing the system variable (^334) and executing the increasing variable number (S224) are entered, and the method flow in the variable number is as shown in the third figure. The fifth diagram of the McCaw test is a detailed flow chart of the method for identifying the system mode dimension and the screening of the important variables according to the second embodiment of the present invention, which includes a poorly connected library, a database or any form of The file in the file is read by the system (10) ,), the user defines one of the most needed variables for the construction of the mode (S302), and then the data is read using the relevant conditions defined by the user. Some of these The action of classifying the variables (10) 4), the variables described include but the flow, pressure or warmth. Furthermore, the variables classified by these variables are divided into two types: input variable (10) 6) and output variable (S308). According to the main value analysis (10) method, the variable will be divided into variables (S312), and the analyzed variables will be rounded into variables according to the dependency into the classification of the same group (S31G), and the non-group is divided into : Temperature, or flow...Because of the action of performing the main value analysis, the input = number, and the seat is analyzed by the main value will produce at least - feature vector (8) (4) and to (:: touch: class, the system from the one by the maximum (four) The arrangement of the eigenvalues completed by the slave domain analysis is the value of the technique. The technique is based on the obtained feature, and the knives are determined by the user to determine the deficiencies: 1; The second two == the bottom of the frame is the second significant drop point mentioned in the method of the present invention from the proportional relationship between the clock and the miscellaneous money - the method is illustrated in this example, if eight Eigenvalue, i丨4丨 to analysis method to get the way \ ^ value 'feature value To the arrangement, two substantially :: operator side of a 5% advised ^ 'and a second set of available views sharply so the third' group of the so-called second system of the present invention 151,294,971
丄％少哪MU4之進行變數分類， ，曰如2之該些特徵向量進行比對_)，經由比對後^ 传取大的魏’此變數即為第—個變數(S324)，系統將4 一個-個地從步驟S32G中所決定的系統變數數量依序 ，變數解釋PCA變異度(S326)，於步驟⑽之解釋程度 疋由系統所定義者，在此使用者先於系統喊立限定條件 庫(S328)，上述之限定條件庫包含定義麟擇之變數對整 置匯整在一起，舉例來說， 系所需之變數總量為20，由 ’，該些輸入變數(S306)與步 個系統之解釋程度值。 將限疋條件與步驟S326之變數作一比對動作 (S330)，如果比對結果為符合時，就將所選擇之變數成為 建議選擇變數個數(S332)，假定由此步驟(S322)獲得建議變 數數量為4個，因為由步驟S322已獲得第一個變數，因 此目别還需求得之系統變數為3個，反之，若比對結果不 符合時，系統將會從步驟S320再增加系統變數個數 (S334)，系統為了求得剩餘的變數，系統將會在步驟S328 之後定義一個估計目標規則以依序選擇出符合估計目標規 則之剩餘的變數(S336)，上述之估計目標規則係為所選擇 的系統變數與輸出變數關係最大，而與輸入變數關係最小 者，其計算之公式已揭露於上，在此不再重述。 16 1294971 依據估計目標規則從步驟S332之選取最合適的變數 (S338) ’接著判斷系統變數個數是否滿足(S34〇)，如果判斷 結果為是，财獲㈣統她_變數輝342)，其中該 初始建議之變數數量之輸人來源可為步驟S3ig談相關群 組之該些以變數，如此⑽可由相騎組中找出相近的 變數來使用，反之，若S34G步驟之判斷結果為否，則回 到估計目標值(讀)之步驟，計算目標值之後，回到丄% less MU4 is subject to variable classification, for example, the eigenvectors of 2 are compared _), and the larger Wei's variable is passed through the comparison ^ is the first variable (S324), the system will 4 one by one, the number of system variables determined in step S32G is sequentially, the variable interprets the PCA variability (S326), and the degree of interpretation in step (10) is defined by the system, where the user is prior to the system The conditional library (S328), wherein the qualifier library includes a variable defining a lining to integrate the whole set, for example, the total number of variables required by the system is 20, by ', the input variables (S306) and The degree of interpretation of the system. The limit condition is compared with the variable of step S326 (S330). If the result of the comparison is met, the selected variable is changed to the number of recommended selection variables (S332), and it is assumed that the step (S322) is obtained. It is recommended that the number of variables be four, because the first variable has been obtained by step S322, so the system variable required for the target is three. Conversely, if the comparison result does not match, the system will increase the system from step S320. The number of variables (S334), in order to obtain the remaining variables, the system will define an estimated target rule after step S328 to sequentially select the remaining variables that meet the estimated target rule (S336), the estimated target rule For the selected system variable and the output variable has the largest relationship, and the relationship with the input variable is the smallest, the formula of the calculation has been disclosed above, and will not be repeated here. 16 1294971 According to the estimation target rule, the most suitable variable is selected from step S332 (S338) 'Next, it is judged whether the number of system variables is satisfied (S34〇), and if the judgment result is yes, the financial acquisition (four) unified her _ variable 342), wherein The input source of the variable number of the initial suggestion may be the variable of the relevant group for step S3ig, such that (10) may be used to find similar variables in the phase riding group, and if the judgment result of the S34G step is negative, Then return to the step of estimating the target value (read), after calculating the target value, return to
Hi 路核糊網路或依其他方式所建立的模式 模式的準確性，本發明之方法將會對所建Μ 糸統 模式誤差的動作⑻46)，當系統發 U統核式作 生誤差時，會判斷糸絲M 、， 建立的糸統模式產 若誤差為可是否為系統可接受範圍， 可接受時，夺咣二二:個流程，反之，若誤差為不 數個數。糸、、、先就會回到步驟S334以進—步增加系統變 要鐵圖，係為本發明之系統模式維声”盘舌 要又數師奴系統示意圖，本發明 績Μ與重 腦，且對外部連社之一電哭… 糸、'先1可應用於一電 :處取得複數個資料’該資料來源處係由線資料來 料庫或任何形式的檔案。—變數數連結之資料 〃連結於該輪人資料模110 m I ‘二 1294971 建構所需之一最大變數數量。一主值分析模組16，其可將 該些輸入變數進行主值分析以獲得至少一特徵向量及至少 一特徵值。一變數分類模組14，係電性連結於該變數數量 模組12 ,其可將該些資料依該使用者所定義之相關條件分 類為至少一輸入變數及至少一輸出變數。 一相關群組模組18，將經由該主值分析模組分析後之 該些輸入變數進行群組分類。一第一變數模組20，其可由 該些特徵值中最大特徵值所對應該些特徵向量之最大特徵 1 向量。一系統自由度模組22,其可由該些特徵值中之第二 次大幅下降點做為一系統自由度及獲得該些特徵值中之第 二次大幅下降點之前所有特徵值的變數數量。 一變數個數計算模組24,係電性連結於該系統自由度 模組22，其可由該系統自由度模組22捉取至少一待增加 變數，該變數個數計算模組24包括一限定條件庫240，其 預先儲存有複數個限定條件，用以限定所要選擇之變數， 一解釋程度單元242，其選取該系統自由度模組之該些變 數數量之一變數及解釋所選取之該變數的主值分析變異 度，一比較單元244，其將該些限定條件與所選取之該變 數作比較，及一輸出變數單元246，其將符合該比較單元 比較後之變數進行輸出。 一選擇變數模組26,係電性連結於該變數個數計算模 組24，其可由該變數個數計算模組24、該第一變數模組 20及一估計目標單元260為基礎所獲得之複數個選擇變 18 1294971 數’該選擇變數模組26包括—估計目標單 =義=些輪出變數關係最大且與該些選關== 之-估计目標值，及一選擇變數單元262 =係取小 目標值、該第—變數模組及該系統自由度_ Ί亥估計 數計算模組中選擇出至少一變數。又、、、’该變數個 一判斷變數模組28，係電性連纟士 18，其用明_選擇魏模組2 =關_組模組 讀量是料㈣魏數量麻變數 里，一建立模式模組3〇 ,其可將兮垂又η亥、交數數 該些選擇變數建立模式，一模式:差模：模:所獲得之 =自由度’接著若該模式誤差模組二;:1:調 =連結至該變數個數計算模組24以增一需之=變 由上述之實施例可知，本發日狀方法係 =據資料的輸人，透過上述之本發明之方法流可: =析決定系統自由度與筛選重要變數的結果。：= 外部連m設備提2 押ttr ’可精確地依照所獲得之系統模式以 設備，本發明之方法可應祕各種電器 s +導體設備絲庭電H設備以增進“準確度。 於羽本^確能藉上述所揭露之技術，提供—種迴然不同 =α者的設計’堪能提高整體之使用價值，又其申請 見於刊物或公開使用’誠已符合發明專利之要件，麦依 19 1294971 法提出發明專利申請。 惟，上述所揭露之圖式、說明，僅為本發明之實施例 而已，凡精于此項技藝者當可依據上述之說明作其他種種 之改良，而這些改變仍屬於本發明之發明精神及以下界定 之專利範圍中。 【圖式簡單說明】 第一圖係為習知之半導體製造過程期間用以處理基板的處 > 理系統之監測方法流程圖； 第二圖係為本發明第一實施例之系統模式維度鑑識與重要 變數篩選之方法流程圖； 第三圖係為本發明第一實施例之增加變數數量之方法流程 圖， 第四圖係為本發明第一實施例之獲得其餘的變數之方法流 程圖； 第五圖係為本發明第二實施例之系統模式維度鑑識與重要 變數篩選之方法詳細流程圖；及 第六圖係為本發明之系統模式維度鑑識與重要變數篩選之 系統示意圖。 【主要元件符號說明】 模式建構之變數選擇系統1 電器設備 2 20 1294971 輸入貢料模組 10 變數數量模組 12 變數分類模組 14 主值分析模組 16 相關群組模組 18 第一變數模組 20 系統自由度模組 22 變數個數計算模組 24 限定條件庫 240 解釋程度單元 242 比較單元 244 輸出變數單元 246 選擇變數模組 26 估計目標單元 260 選擇變數單元 262 判斷變數模組 28 建立模式模組 30 模式誤差模組 32 21The accuracy of the mode of the Hi-channel nucleus network or other established mode, the method of the present invention will act on the error of the built-in mode (8) 46), when the system sends a U-core nucleus error, Will judge the silk M, the established system mode error is whether the system is acceptable range, when acceptable, win two or two: a process, and vice versa, if the error is not a few.糸, ,, and then will return to step S334 to further increase the system to change the iron map, which is a schematic diagram of the system mode of the invention. And one of the external company is crying... 糸, 'First 1 can be applied to one electricity: to obtain multiple data'. The source of the data is from the data source library or any form of file. 〃 is connected to the wheel data module 110 m I '21294971 to construct one of the required maximum number of variables. A main value analysis module 16 can perform main value analysis on the input variables to obtain at least one feature vector and at least A eigenvalue. A variable classification module 14 is electrically coupled to the variable quantity module 12, and the data can be classified into at least one input variable and at least one output variable according to relevant conditions defined by the user. A related group module 18 performs group classification on the input variables analyzed by the main value analysis module. A first variable module 20 can be corresponding to the largest feature value among the feature values. Maximum feature vector 1 vector. A system degree of freedom module 22, which can be used as a system degree of freedom from the second significant drop point of the feature values and obtain all the feature values before the second significant drop point of the feature values. The variable number calculation module 24 is electrically coupled to the system degree of freedom module 22, and the system degree of freedom module 22 can capture at least one variable to be added, the variable number calculation module 24 A qualifier library 240 is included, which prestores a plurality of qualification conditions for defining a variable to be selected, and an interpretation degree unit 242 that selects one of the variables of the system degree of freedom module and the interpretation is selected. The main value analysis variability of the variable, a comparison unit 244, compares the qualification conditions with the selected variable, and an output variable unit 246, which outputs the variables that match the comparison unit. A variable selection module 26 is electrically coupled to the variable number calculation module 24, which can be based on the variable number calculation module 24, the first variable module 20, and an estimation target unit 260. The plurality of selections obtained are 18 1294971. The selection variable module 26 includes - an estimated target list = meaning = the round-off variable relationship is the largest and the selection target == - the estimated target value, and a selection variable unit 262=Selecting a small target value, the first variable module, and the system degree of freedom _ Ί 估计 估计 估计 算 算 选择 选择 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计 估计Electrician Gentleman 18, its use _ select Wei module 2 = off _ group module reading is material (four) Wei number of hemp variables, a set of mode module 3 〇, which can be 兮 又 η 、, The number of intersections is selected to establish a mode of change, a mode: differential mode: mode: obtained = degree of freedom 'following the mode error module 2; 1: 1 = tune to the variable number calculation module 24 It is known from the above embodiments that the present method is based on the input of the data, and the flow of the method of the present invention described above can be: = the result of determining the degree of system freedom and screening important variables. := Externally connected to the device, 2 ttr ' can be precisely according to the system mode obtained by the device, the method of the invention can be applied to various electrical appliances s + conductor equipment silk court electric H equipment to improve "accuracy. Yu Yuben ^ It is indeed possible to use the above-disclosed technology to provide a design with a different kind of = a person's design can improve the overall use value, and its application can be found in the publication or public use. "I have already met the requirements of the invention patent, Mai Yi 19 1294971 The invention claims the patent application. However, the drawings and descriptions disclosed above are only examples of the present invention, and those skilled in the art can make other improvements according to the above description, and the changes still belong to The inventive concept of the present invention and the patent scope defined below. [Simplified description of the drawings] The first figure is a flow chart of a monitoring method for a substrate for processing a substrate during a conventional semiconductor manufacturing process; A flowchart of a method for identifying a system mode dimension and an important variable for filtering according to a first embodiment of the present invention; the third figure is a variable number of the first embodiment of the present invention Method flow chart, the fourth figure is a flow chart of the method for obtaining the remaining variables according to the first embodiment of the present invention; the fifth figure is a detailed flow chart of the method for system mode dimension identification and important variable screening according to the second embodiment of the present invention. And the sixth figure is a system diagram of the system mode dimension identification and important variable screening of the invention. [Main component symbol description] Model construction variable selection system 1 Electrical equipment 2 20 1294971 Input tribute module 10 Variable quantity module 12 variable classification module 14 main value analysis module 16 related group module 18 first variable module 20 system degree of freedom module 22 variable number calculation module 24 qualification library 240 interpretation degree unit 242 comparison unit 244 output variable Unit 246 selects variable module 26 estimates target unit 260 selects variable unit 262 determines variable module 28 builds mode module 30 mode error module 32 21