511128 A 7 B7 五、發明説明（i ) 發明背景 本發明係關於半導體處理裝置，特別是關於預測處理 結果，提升裝置之稼動率與信賴性之半導體處理裝置，以 及監視以及/或控制半導體製造裝置之方法。 近年來，半導體裝置之尺寸微細化，加工的尺寸精度 也嚴格至必須以10%以下之尺寸精度加工0.1 /2 m以下之閘極 電極。另一方面，在利用熱或電漿物理化學地加工半導體 晶圓之半導體製造裝置中，由於裝置內部之化學反應所產 生之反應產生物等附著於裝置之內壁而殘留，隨著時間使 晶圓之處理狀態變化。 因此，隨著多數片重複晶圓之處理，晶圓上之半導體 裝置的加工形狀逐漸變化，性能劣化。爲了處理此問題， 通常採取藉由電漿淸理處理室內壁之附著物，或提高處理 室壁之溫度，使得附著物不易附著等之對策。但是，在多 數之情形，這些之對策並不完全，結果爲：半導體裝置的 加工形狀逐漸變化。因此，在加工形狀變化成爲問題之前 ，必須進行製造裝置的構件之更換或洗淨。又，在堆積膜 以外，各樣之裝置狀態的變動與晶圓之加工形狀之變動有 關。因此，提案檢測半導體製造裝置內部之處理狀態的變 化，將檢測結果回饋於半導體製造處理裝置的輸入，一定 地保持處理狀態等之工夫。 監視電漿處理之變動的方法例如揭示在JP-A- 1 0- 1 25660 號。於此公報中，顯示利用電漿處理特性與裝置的電氣信 號之關係式，預測裝置性能、診斷電漿之狀態之方法。作 本紙張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） (請先閲讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製511128 A 7 B7 V. Description of the Invention (i) Background of the Invention The present invention relates to semiconductor processing devices, and in particular, to semiconductor processing devices that predict processing results, improve the productivity and reliability of the devices, and monitor and / or control semiconductor manufacturing devices. Method. In recent years, the size of semiconductor devices has been miniaturized, and the dimensional accuracy of processing has been strict enough to process gate electrodes of 0.1 / 2 m or less with a dimensional accuracy of 10% or less. On the other hand, in a semiconductor manufacturing apparatus that processes semiconductor wafers physically or chemically by heat or plasma, reaction products and the like generated by chemical reactions inside the apparatus adhere to the inner wall of the apparatus and remain. The processing status of the circle changes. Therefore, as many wafers are processed repeatedly, the processing shape of the semiconductor device on the wafer gradually changes and the performance deteriorates. In order to deal with this problem, countermeasures such as treating the attachments on the interior wall by plasma treatment or increasing the temperature of the treatment chamber wall are generally adopted. However, in most cases, these countermeasures are not complete, and as a result, the processing shape of the semiconductor device is gradually changed. Therefore, before a change in processing shape becomes a problem, it is necessary to replace or clean the components of the manufacturing apparatus. In addition to the deposited film, variations in the state of various devices are related to variations in the processing shape of the wafer. Therefore, it is proposed to detect changes in the processing state inside the semiconductor manufacturing apparatus, feed back the detection results to the input of the semiconductor manufacturing processing apparatus, and maintain the processing state to a certain extent. A method for monitoring changes in plasma processing is disclosed in, for example, JP-A-1 0-1 25660. In this publication, a method for predicting the performance of a device and diagnosing the state of a plasma using a relational expression between a plasma processing characteristic and an electrical signal of the device is shown. This paper size applies Chinese National Standard (CNS) A4 specification (210X297 mm) (Please read the precautions on the back before filling out this page) Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs
511128 A7 ____B7 五、發明説明（2 ) 爲其方法係揭示藉由多重回歸分析以求得表示3個之電氣 信號與裝置的電漿處理特性之關係的近似式之方法。又， 另一例係揭示在JP-A- 1 1 -87323號。於此公報中，顯示將安 裝既存之複數的檢測器之一般的檢測系統適用於半導體製 造裝置，由該檢測信號之相關信號監視裝置的狀態之方法 。作爲產生該相關信號之方法係揭示藉由6個之電氣信號 之比的計算式。又，另一粒係顯示在U.S.Patent 5658423。 於此公報中，顯示取入光或質量分析器之很多的信號，產 生相關信號，監視裝置之狀態的方法。作爲產生此相關信 號之方法係顯示利用主成分分析之方法。 發明摘要 經濟部智慧財產局員工消費合作社印製 但是，在JP-A- 10- 1 25660號之方法中，監視裝置之傳感 器資料有多種類時，與欲預測之處理性能無關之很多的信 號進入說明變數之故，藉由多重回歸分析之預測變得不太 順利。又，JP-A- 1 1 -87323號之方法係將取得由廣爲知悉的 複數的檢測手段來之複數的檢測信號之相關的信號使用於 診斷的一般之方法。取得所揭示之相關的方法也係取得幾 種之信號之比的習知的手法，適用於正確監視因應很多之 變動原因，取得多樣之狀態之半導體製造裝置的狀態之系 統有其困難。在U.S.Patent 5,658,423中，與以上之方法不 同，揭示：藉由主成分分析由裝置所監視之多量的資料， 捕捉裝置狀態之變動，以監視電漿之狀態之方法。但是， 在使用於實際的量產之半導體製造裝置中，單單如此適用 本紙張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） 511128 A 7 __B7_ 五、發明説明（3 ) 一般之統計處理之方法，並無法順利稼動。例如，幾乎都 是不知道主成分如何地一有變化，處理結果會變得如何之 情形之故。 本發明之目的在於提供：於處理各種種類之裝置的半 導體製造裝置中，藉由監視處理狀態，依據監視器輸出， 檢測異常處理，或預測處理結果，提升裝置的稼動率與信 賴性之半導體處理裝置及方法。 依據本發明之1種形態之半導體處理裝置爲了解決上 述課題，採用如下之手段。 經濟部智慧財產局員工消費合作社印製 處理裝置係包含：監視處理半導體晶圓之半導體處理 裝置的處理狀態之傳感器；以及輸入藉由前述半導體處理 裝置所處理之半導體晶圓的處理結果之量測値之處理結果 輸入手段；以及依據前述傳感器取得之傳感器資料以及前 述量測値，將前述傳感器資料當成說明變數，產生預測處 理結果之模型式之模型式產生部；以及依據前述模型式以 及前述傳感器資料，預測處理結果之處理結果預測部；以 及比較前述預測處理結果與預先設定之設定値，補正其之 偏差地控制前述半導體處理裝置之處理條件之處理條件控 制部。 藉由如此構成，如依據本發明，由半導體製造裝置藉 由傳感器取得監視器資料，在量測試料之處理結果前或不 用量測地，藉由模型式預測處理結果以提升半導體製造裝 置的稼動率與信賴性。 本紙張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） 511128 A7 B7 五、發明説明（4) 實施例之詳細說明 以下，依據圖面說明本發明之實施例。 圖1A係顯示本發明之第1實施例。同圖中，處理裝置 1具備有監視處理狀態之2個的處理狀態監視部2。處理狀 態監視部2也可以組裝在裝置1中，也可以設置在裝置1 之外側。又，也可以透過網路等設置於分離之場所。進而 ’如圖1B般地，機能之一部份透過網路等而被分離亦可。 經濟部智慧財產局員工消費合作社印製 處理狀態監視部2具有如下之構成。首先，具有監視 在處理裝置1之晶圓的處理狀態之傳感器3。傳感器通常 使用幾個種類之傳感器。例如，在電漿蝕刻裝置或電漿 CVD裝置之電漿處理中，傳感器利用分光器頻譜分解處理 中之電漿的發光，取得分解之各波長之發光強度以作爲傳 感器資料。例如，在使用具有1000頻道之CCD陣列之分光 器時，每1次之取樣可以取得1000個之傳感器資料。又， 裝置的壓力或溫度、氣體流量等也使用爲傳感器資料。又 ，可以將電流、電壓、阻抗握這些之高頻波成分等之電氣 量測結果當成傳感器資料使用。在晶圓之處理中，以適當 之時間間隔取得這些之傳感器資料，取得之傳感器資料保 存在傳感器資料保存部4。另一方面，處理完了之晶圓藉由 裝置1之外部或被組裝在裝置之處理結果量測器量測處理 結果。處理結果量測係藉由CDSEM之閘極寬幅之量測或藉 由剖面SEM之剖面形狀之加工形狀之量測，或加工之裝置 的電氣特性之量測。這些量測不需要於全部之晶圓進行， 通常抽出一部份之晶圓，量測處理結果即可。處理狀態監 本紙張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） 511128 A 7 B7 五、發明説明（5 ) 視部2爲了接受此處理結果之量測値，具有處理結果量測値 輸入手段5 。輸入手段5可以爲讀取被記錄在軟碟或 CDROM等之可搬動媒體之資訊之讀取器，也可以爲有線或 無線之網路接續裝置。由輸入手段5接受之處理結果之量 測値係被保存在處理結果量測値保存部6。於保存部6保 存各種裝置之處理結果的量測値。模型式產生部7係由傳 感器資料保存部4與處理結果量測値保存部6取出同種之 裝置，而且傳感器資料與處理結果量測値之兩方的資料被 保存之樣本。此樣本數例如在3個以上時，將傳感器資料 當成說明變數，製作預測處理結果量測値之模型式。此時 之傳感器資料之種類或數目爲多數，自動地選出使用於預 測之傳感器資料有困難。特別是在處理各式各樣之裝置的 情形等，個裝置在預測上有效之傳感器資料不同之故，預 先決定使用於預測之傳感器有其困難。 圖5A係說明藉由PLS法之模型式產生處理圖，如圖5A 所示般地，PLS法係由多數之傳感器資料自動產生與應預測 資料之變動其相關變得最強之說明變數。同時，由傳感器 資料也可以獲得計算說明變數用之函數。例如，以η片之 晶圓的處理結果量測値爲預測對象，以Yi表示第i號之晶 圓的處理結果量測値。由1個晶圓獲得m個之傳感器資料 時，以Sij表示第i號之晶圓的第j號之傳感器資料。此m 個之傳感器資料可以爲相同傳感器之不同時間的資料値， 也可以由不同傳感器來之資料。圖5B係說明傳感器資料Sij 之例圖。如同圖所示般地，裝置1之1片晶圓之處理被分 本紙張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） (請先閲讀背面之注意事項再填寫本頁) 訂 經濟部智慧財產局員工消費合作社印製 511128 A7 B7 五、發明説明（6 ) 成3步驟之處理條件，有傳感器A、B、C之3種類時，如 圖示般地，Sij係取得S11至Sn9即可。Sij也可以爲各步驟 處理中之傳感器資料之平均値’也有轉換爲平方或倒數等 之傳感器資料値比較好之情形。如使用PLS法，也可以將 傳感器資料Sij轉換爲以與Yi之變動的相關的強度之順序 排列之m個的說明變數Xik。由傳感器資料Sij轉換爲說明 變數Xk之函數Fk係如式（1 )所示。511128 A7 ____B7 V. Description of the Invention (2) Its method is a method for revealing an approximate expression of the relationship between three electrical signals and the plasma processing characteristics of the device by multiple regression analysis. Another example is disclosed in JP-A-1 1 -87323. In this bulletin, a method is shown in which a general detection system in which an existing plurality of detectors are installed is applied to a semiconductor manufacturing device, and the state of the device is monitored by a signal related to the detection signal. As a method for generating the correlation signal, a calculation formula for the ratio of six electrical signals is disclosed. Also, another germline is shown in U.S. Patent 5658423. In this publication, a method of displaying a large number of signals taken in by light or a mass analyzer, generating related signals, and monitoring the state of the device is shown. As a method for generating this correlation signal, a method using principal component analysis is shown. Summary of the Invention Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. However, in the method of JP-A-10-1 25660, when there are many types of sensor data of the monitoring device, many signals that have nothing to do with the processing performance to be predicted enter. Explaining the variables, the prediction by multiple regression analysis becomes less smooth. Further, the method of JP-A-1 1-87323 is a general method for diagnosing a signal by obtaining a plurality of detection signals from a plurality of well-known detection methods. Obtaining the disclosed related methods is also a conventional method of obtaining the ratio of several signals. It is applicable to a system for accurately monitoring the status of a semiconductor manufacturing device that has various states due to a large number of changes. In U.S. Patent 5,658,423, different from the above methods, it is revealed that the method of monitoring the state of the plasma by capturing the change of the state of the device by the principal component analysis of a large amount of data monitored by the device. However, in a semiconductor manufacturing device used in actual mass production, the paper size is only applicable to the Chinese national standard (CNS) A4 specification (210X297 mm) 511128 A 7 __B7_ V. Description of the invention (3) General statistical processing This method does not work smoothly. For example, it is almost impossible to know how the principal component changes and how the processing result will change. An object of the present invention is to provide a semiconductor process for improving the productivity and reliability of a device in a semiconductor manufacturing apparatus that processes various types of devices by monitoring the processing status, detecting abnormal processing, or predicting a processing result based on a monitor output. Device and method. In order to solve the above problems, a semiconductor processing apparatus according to one aspect of the present invention employs the following means. The printed processing device of the employee's consumer cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs includes: a sensor that monitors the processing status of a semiconductor processing device that processes semiconductor wafers; and a measurement that inputs the processing result of the semiconductor wafer processed by the aforementioned semiconductor processing device Means for inputting processing results of 以及; and a model generating section for generating a model formula for predicting processing results based on the sensor data obtained by the aforementioned sensors and the aforementioned measurement 値, and using the aforementioned sensor data as explanatory variables; and based on the aforementioned model formulas and the aforementioned sensors Data, a processing result prediction section that predicts the processing result; and a processing condition control section that compares the foregoing prediction processing result with a preset setting, and corrects the deviation thereof to control the processing conditions of the semiconductor processing device. With such a structure, as in accordance with the present invention, the semiconductor manufacturing device acquires monitor data through sensors and measures the processing result of the test material before or without measuring the amount, and predicts the processing result with a model to improve the performance of the semiconductor manufacturing device. Rate and reliability. This paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm) 511128 A7 B7 V. Description of the invention (4) Detailed description of the embodiments The following describes the embodiments of the present invention with reference to the drawings. Fig. 1A shows a first embodiment of the present invention. In the figure, the processing device 1 includes two processing state monitoring units 2 that monitor the processing state. The processing state monitoring unit 2 may be incorporated in the device 1 or may be provided outside the device 1. It can also be installed in a separate place via the Internet. Furthermore, as shown in FIG. 1B, a part of the functions may be separated through a network or the like. Printed by the Intellectual Property Office of the Ministry of Economic Affairs, Consumer Cooperatives. The processing status monitoring unit 2 has the following configuration. First, there is a sensor 3 that monitors the processing state of a wafer in the processing apparatus 1. Sensors usually use several types of sensors. For example, in the plasma processing of a plasma etching device or a plasma CVD device, the sensor uses the light emission of the plasma in the spectral decomposition processing of the spectroscope to obtain the light emission intensity of each wavelength as the sensor data. For example, when using a spectroscope with a CCD array of 1000 channels, 1000 samples of sensor data can be obtained for each sampling. In addition, device pressure, temperature, gas flow rate, etc. are also used as sensor data. In addition, the electrical measurement results of high-frequency components such as current, voltage, and impedance can be used as sensor data. During processing of the wafer, these sensor data are acquired at appropriate time intervals, and the acquired sensor data are stored in the sensor data storage unit 4. On the other hand, the processed wafer measures the processing result by a processing result measuring device outside the device 1 or assembled in the device. The measurement of the processing result is the measurement of the gate width of the CDSEM or the processing shape of the cross-sectional shape of the cross-sectional SEM, or the measurement of the electrical characteristics of the processed device. These measurements do not need to be performed on all the wafers. Usually, a part of the wafers are pulled out to measure the processing results. The paper size of the process is monitored according to the Chinese National Standard (CNS) A4 (210X297 mm) 511128 A 7 B7 V. Description of the invention (5) In order to accept the measurement result of the processing result, the visual inspection unit 2 has the measurement result measurement. Enter means 5. The input means 5 may be a reader that reads information recorded on a removable medium such as a floppy disk or CDROM, or a wired or wireless network connection device. The measurement results of the processing results received by the input means 5 are stored in the processing result measurement / storage storage unit 6. The storage unit 6 stores measurement results of processing results of various devices. The model generation unit 7 is a sample in which the same kind of device is taken out by the sensor data storage unit 4 and the processing result measurement / storage storage unit 6, and both the sensor data and the processing result measurement / storage data are stored. When the number of samples is, for example, three or more, the sensor data is used as an explanatory variable, and a model formula for measuring the prediction result is created. At this time, there are many types or numbers of sensor data, and it is difficult to automatically select sensor data to be used for prediction. Especially when dealing with a variety of devices, etc., the sensor data that is valid for prediction varies from device to device, so it is difficult to decide which sensor to use for prediction in advance. FIG. 5A is a diagram illustrating the generation of a processing diagram by using the model of the PLS method. As shown in FIG. 5A, the PLS method is an explanatory variable that automatically generates the strongest correlation between most sensor data and changes in the predicted data. At the same time, functions for calculating and explaining variables can also be obtained from the sensor data. For example, the measurement result of the processing result of the wafer of η wafer is used as the prediction object, and the processing result of the wafer of the i-th wafer is measured by Yi. When m pieces of sensor data are obtained from one wafer, Sij represents the j-th sensor data of the i-th wafer. The m sensor data may be data of the same sensor at different times, or data from different sensors. FIG. 5B is a diagram illustrating an example of the sensor data Sij. As shown in the figure, the processing of one wafer of device 1 is divided into paper sizes. The Chinese standard (CNS) A4 specification (210X297 mm) is applied (please read the precautions on the back before filling this page). Printed by the Consumer Cooperative of the Ministry of Intellectual Property Bureau 511128 A7 B7 V. Description of the invention (6) The processing conditions are divided into 3 steps. When there are 3 types of sensors A, B, and C, as shown in the figure, Sij obtains S11 to Sn9. Just fine. Sij can also be the average of the sensor data in each step processing. It is also better to convert to sensor data such as square or inverse. If the PLS method is used, the sensor data Sij can also be converted into m explanatory variables Xik arranged in order of the intensity related to the change in Yi. The function Fk converted from the sensor data Sij into a description variable Xk is shown in equation (1).
Xik = Fk(Si 1 ,Si2, ··· ,Sim) (1) 使用此說明變數Xik之中的幾個，預測處理結果量測 値。通常說明變數Xil與處理結果量測値Yi之相關最強之 故，將Xil、Xi2、Xi3等選擇爲說明變數。在PLS法中，如 式（2 )之預測式同時被.產生。但是也可以製作先前敘述之 Xij等之預測式（2)。Xik = Fk (Si 1, Si2, ···, Sim) (1) Using this explanation of several of the variables Xik, predict the processing result and measure 値. Usually, it is explained that the variable Xil has the strongest correlation with the measurement result 値 Yi. Therefore, Xil, Xi2, Xi3, etc. are selected as the explanatory variables. In the PLS method, a prediction formula such as (2) is generated simultaneously. However, it is also possible to make the prediction formula (2) of Xij et al.
Yi = (p(Xil，Xi2，Xi3) (2) 另一方面，在處理結果量測値之中，晶圓之處理狀態 變差，顯示異常之處理結果量測値之晶圓資料也包含在其 中。對於此種資料，以通常之多重回歸分析一進行預測， 如圖6所示般地，被異常資料影響之預測精度差的模型式被 產生。因此，於預測使用堅固回歸分析如使用堅固回歸分 析’如圖6所示之有異常的資料當成偏離者由預測對象被排 除之故，可以產生正確之預測模型式。 圖7係顯示說明模型式產生部7之模型式產生處理之流· 程圖。 於模型式產生部7中傳感器資料之種類在有很多時，進 本紙張尺度適用中國國家標準（CNS ) A4規格（210X297公產） 請 先 閱 讀 背 面 之 注 意 事 項 再 本 頁 經濟部智慧財產局員工消費合作社印製 511128 A7 B7 經濟部智慧財產局員工消費合作社印製 五、發明説明（7 ) 行傳感器資料之主成分解析（步驟701、702 )，利用獲得 之主成分，進行堅固回歸分析，預測處理結果（步驟505〜 706 )。此時，處理結果之預測所不必要之主成分也包含在 說明變數之故，如流程圖所示般地，去除回歸係數小之主 成分（步驟707 )，將別的主成分（第2主成分）追加於說 明變數（步驟704 )，再度進行多重回歸分析（步驟706 ) 之處理，重複此處理至預測誤差成爲設定値以下爲止（步 驟708 )。這些回歸分析可以爲線性，也可以利用由處理之 物理特性或經驗値所獲得之非線性之回歸分析。 以上述之方法所產生之模型式係被保存在圖1之模型式 保存部8。模型式係各種種之裝置地被製作之故，在模型式 保存部8只保存以處理裝置所處理之裝置的數目之模型式。 某裝置之晶圓被載入處理裝置1而處理時，對應此裝置之 模型式被載入預測部9。此裝置在處理中由傳感器3所獲得 之信號如果需要，利用藉由PLS法所獲得之式（1 )，被轉 換爲說明變數，或藉由主成分分析被轉換爲主成分，處理 結果之預測値藉由式（2 )之模型式被計算。計算之預測値 被傳遞於處理條件控制部10，控制部10補正處理結果之設 定値與預測値之偏差地變更處理條件。接著，說明在前述 之處理條件控制部10補正處理條件之具體的方法。此處再 度利用PLS法。在通常之半導體裝置的處理中，很多之加 工要求係要求幾種之相反之處理性能。例如，在閘極電極 之蝕刻加工等中，要求閘極電極之側壁的垂直性與底層氧 化膜與閘極多晶矽之蝕刻選擇性。及爲了達成側壁之垂直 - -1 〇- 請 先 閲 讀 背 之 注 意 事 項 再 f 本 頁Yi = (p (Xil, Xi2, Xi3) (2) On the other hand, during the processing result measurement, the processing state of the wafer becomes worse, and the wafer data of the abnormal processing result measurement is also included in the measurement result. Among them, for such data, a general multiple regression analysis is used to make a prediction. As shown in FIG. 6, a model formula of poor prediction accuracy affected by abnormal data is generated. Therefore, a robust regression analysis such as the use of robustness is used for prediction. Regression analysis. As shown in Figure 6, the abnormal data can be regarded as a deviation from the predicted object, so that a correct prediction model can be generated. Figure 7 shows the flow of the model generation process of the model generation section 7. When there are many types of sensor data in the model generation section 7, the paper size applies the Chinese National Standard (CNS) A4 specification (210X297). Please read the precautions on the back before the intellectual property of the Ministry of Economic Affairs on this page. Printed by the Consumer Cooperative of the Bureau of Staff of the State Council 511128 A7 B7 Printed by the Consumer Cooperative of the Bureau of Intellectual Property of the Ministry of Economic Affairs 5. Description of the invention (7) Analysis of the main components of the sensor data ( (Steps 701 and 702), use the obtained principal components to perform a robust regression analysis and predict the processing results (steps 505 to 706). At this time, the principal components unnecessary for the prediction of the processing results are also included in the explanation of the variables, such as the process As shown in the figure, the principal component with a small regression coefficient is removed (step 707), and another principal component (second principal component) is added to the explanatory variable (step 704), and the multiple regression analysis (step 706) is performed again. This process is repeated until the prediction error becomes less than the set value (step 708). These regression analysis can be linear, or non-linear regression analysis obtained from the physical characteristics or experience of the process can be used. The model formulas are stored in the model formula storage unit 8 of Fig. 1. Because the model formulas are produced in various devices, only the model formulas of the number of devices processed by the processing device are stored in the model formula storage unit 8. When the wafer of the device is loaded into the processing device 1 and processed, the model formula corresponding to the device is loaded into the prediction section 9. The signal obtained by the sensor 3 during the processing of this device If necessary, the formula (1) obtained by the PLS method is used to convert into explanatory variables or the main component is analyzed by principal component analysis. The prediction of the processing result is calculated by the model formula of the formula (2). The calculated prediction 値 is passed to the processing condition control section 10, and the control section 10 corrects the processing result setting 値 and the prediction 变更 to change the processing conditions. Next, the specific description of the processing condition control section 10 to correct the processing conditions Method. Here again, the PLS method is used. In ordinary semiconductor device processing, many processing requirements require several kinds of opposite processing performance. For example, in the gate electrode etching process, the side wall of the gate electrode is required Verticality and etching selectivity of underlying oxide film and gate polycrystalline silicon. And in order to achieve the verticality of the side wall--1 〇- Please read the notes on the back before f
訂 本紙張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） 511128 A7 _____B7 五、發明説明（8 ) 性，以使用堆積性低之鈾刻條件爲佳，爲了達成與底層氧 化膜之高選擇性，以使用堆積性高的蝕刻條件爲佳。如此 ，在有2個相反之要求的情形，處理條件之控制困難。 圖8〜10係說明滿足此種相反之要求之處理條件用之圖 。例如由於處理裝置1之時間經過變化，側壁之垂直性變 差時，一減少處理條件1 (此處，設爲氣體A之流量），垂 直性即使變好，但是同時，底層氧化膜之選擇性變差，作 爲處理條件並不理想。因此，例如，組合處理條件1與處理 條件2 (此處，設爲晶圓偏壓電力），必須一面改善側壁之 垂直性，而且，一面找出底層氧化膜之選擇性不會惡化之 條件。爲了解決此，如圖8般地，通常在進行處理之中心條 件之周圍設定改變數點至數十點之處理條件的實驗條件， 進行加工處理，量測處理結果。圖10中之點1〜4係相當於 圖8之實驗條件1〜4。此處，作爲處理結果量測値A，採用 側壁之垂直性之量測値，作爲處理結果量測値B，採用底層 氧化膜選擇比。 經濟部智慧財產局員工消費合作社印製The paper size of the edition is in accordance with Chinese National Standard (CNS) A4 (210X297 mm) 511128 A7 _____B7 V. Description of the invention (8) It is better to use uranium engraving conditions with low stackability. In order to achieve a high level with the underlying oxide film The selectivity is preferably based on the use of etching conditions with high buildability. Thus, in the case of two opposite requirements, it is difficult to control the processing conditions. Figures 8 to 10 are diagrams illustrating processing conditions that meet this opposite requirement. For example, when the time of the processing device 1 changes and the verticality of the sidewall deteriorates, as soon as the processing condition 1 is reduced (here, the flow rate of the gas A) is reduced, even if the verticality is improved, at the same time the selectivity of the underlying oxide film Deterioration is not ideal as a processing condition. Therefore, for example, in combination of processing condition 1 and processing condition 2 (here, set to wafer bias power), it is necessary to improve the verticality of the sidewall while finding the conditions under which the selectivity of the underlying oxide film does not deteriorate. In order to solve this problem, as shown in FIG. 8, the experimental conditions that change the processing conditions from a few points to a few tens of points are usually set around the central conditions of the processing, and the processing is performed to measure the processing results. Points 1 to 4 in FIG. 10 correspond to experimental conditions 1 to 4 in FIG. 8. Here, as the processing result measurement ， A, a measurement of the verticality of the sidewall is used, and as the processing result measurement 値 B, the bottom oxide film selection ratio is used. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs
如圖9所示般地，對於此實驗，適用PLS法，如使用二 種類之處理條件與二種類之處理結果量測値之相關，如圖 1 0所示般地，可以獲得與側壁垂直性相關強的條件之方向A 。又，同樣藉由PLS法所獲得之與底層氧化膜選擇比相關 之強的方向，可以計算與方向A正交，與底層氧化膜選擇 比相關強之方向B，將此條件方向A與條件方向B設定於 圖1A之處理條件控制部10。如此一來，依據模型式，在預 測部9預測側壁的垂直性惡化時，將處理條件往條件方向A 本紙張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） 511128 A7 _B7_ 五、發明説明（9 ) 修正，可以不犧牲底層氧化膜選擇比地改善側壁垂直性。 被計算之處理條件的控制方向係被保存在處理條件控制方 向保存手段14，使用於藉由模型式之處理結果預測値由設 定値偏離之情形，修正處理條件。 在此處所舉之例中，雖使二種類之處理條件變化，在 PLS法中可以使更多種類之處理條件變化，使愈多之處理條 件變化，可以獲得更好之結果。又，相反之處理結果預測 値也不單爲二種類，可以以更多數之處理結果爲對象。例 如，在側壁之垂直性與底層氧化膜選擇比之外，可以取用 光罩選擇比等。 圖2係顯示本發明之另一個之實施例。圖2之系統雖與 圖1 A之系統幾乎爲相同構成，但是代替處理條件控制部1 0 ，具有預測値顯示部11，於裝置之顯示器等顯示通知異常 之信息。此顯示部份也可以爲發出警報之蜂鳴器或電子郵 件之發送等。 圖3係顯示本發明之另一個之實施例。至此爲止之系統 係依據由模型式之處理結果之預測之故，無法監視不產生 模型式之種類的裝置之處理。在處理結果之量測很多需要 花非常多之時間的情形，有幾乎無法進行處理結果之量測 者，在那種裝置之情形，無法產生模型式。因此，如圖3般 地，在主成分抽出部12由多種之傳感器資料抽出主成分， 藉由以異常監視部1 3監視主成分之變動之偏差，檢測處理 之異常。在檢測異常之情形，停止裝置之下一晶圓之處理 動工即可。在異常之檢測上，例如可以使用稱爲As shown in FIG. 9, for this experiment, the PLS method is applied. If the correlation between the two types of processing conditions and the measurement results of the two types of processing is used, as shown in FIG. 10, the perpendicularity to the sidewall can be obtained. Correlation strong condition direction A. Also, the direction related to the selection ratio of the underlying oxide film obtained by the PLS method can also be calculated to be orthogonal to the direction A and the direction B related to the selection ratio of the underlying oxide film. This conditional direction A and conditional direction can be calculated. B is set in the processing condition control unit 10 in FIG. 1A. In this way, according to the model formula, when the prediction section 9 predicts the deterioration of the verticality of the side wall, the processing conditions are oriented toward the condition direction A. The paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm) 511128 A7 _B7_ V. Invention Note (9) The correction can improve the verticality of the sidewall without sacrificing the selection of the underlying oxide film. The calculated control direction of the processing condition is stored in the processing condition control direction storage means 14 and is used to correct the processing condition by predicting the model-based processing result (from the setting). In the example given here, although the two types of processing conditions are changed, more types of processing conditions can be changed in the PLS method, and more processing conditions can be changed to obtain better results. On the contrary, the prediction of the processing result 相反 is not limited to two types, and a larger number of processing results can be targeted. For example, in addition to the verticality of the sidewalls and the underlying oxide film selection ratio, a mask selection ratio may be used. Fig. 2 shows another embodiment of the present invention. Although the system of FIG. 2 has almost the same configuration as the system of FIG. 1A, it has a prediction display unit 11 instead of the processing condition control unit 10, and displays information notifying abnormality on the display of the device. This display part can also be a buzzer that sends out an alarm or the sending of an e-mail. Fig. 3 shows another embodiment of the present invention. The system so far is based on the prediction of the model-based processing result, and it is impossible to monitor the processing of a device that does not generate a model-type. In the case where the measurement of the processing result takes a lot of time, there are those who can hardly perform the measurement of the processing result, and in the case of that kind of device, the model formula cannot be produced. Therefore, as shown in FIG. 3, the main component extraction unit 12 extracts the main component from various kinds of sensor data, and monitors the deviation of the main component variation by the abnormality monitoring unit 13 to detect abnormality in the processing. In the case of abnormal detection, the processing of the next wafer under the device can be stopped. For abnormality detection, for example,
本紙張尺度適用中國國家標準（CNS ) A4規格（210X297公釐） H 注 意 項 經濟部智慧財產局員工消費合作社印製 511128 A7 _B7_ 五、發明説明（1〇) 請 先 閲 讀 背 面 之 注 意 事 項 SPC(Statistical Process Control)之偏差的管理方法。爲此， 記憶相符之裝置的處理中之主成分的平均値與方差，在被 量測之主成分由平均値分離方差之數倍以上時，將處理判 定爲異常。 圖4係說明合適於具備處理狀態監視部2以及副處理狀 態監視部2之雙方的處理裝置之處理流程圖。首先判定是否 產生保存對於欲處理之裝置之模型式（步驟501)。在模型 式有被保存之情形，藉由處理狀態監視部2實行監視控制（ 步驟502 )。在模型式未被保存之情形，藉由處理狀態監視 部V實行監視控制（步驟503 )。 不單使用圖1A、圖2之模型式之處理狀態監視部，如兼 具只使用如圖3之主成分之副處理狀態監視部之二個，如圖 4所示般地，即使爲模型式被產生保存之裝置，又即使爲未 被產生之裝置，也可以監視處理狀態。此種具有彈性之監 視系統適合於半導體製造裝置之監視。 圖面之簡單說明 經濟部智慧財產局員工消費合作社印製 圖1 A係顯示本發明之1實施例之半導體裝置的處理裝 置之if制系統的方塊圖。 示圖1A之控制系統的修正例之方塊圖。 圖2#^^圖1之處影狀態監視部之別的實施例之方塊 圖 圖3係顯示副處理狀態監視部之實施例之方塊圖。 圖4係顯示裝置之處理狀態的監視方法之例的流程圖 -13- 本紙張尺度適用中國國家標準（CNS〉A4規格（210X29*7公釐） ^Π128 A7 B7 五 、發明説明（ 11 表 圖5A係說明模型產生部之動作用之流程圖。 圖5B係例示η片之晶圓的處理結果量測値之預測法之 圖6係說明無異常資料之預測模型產生法之圖。 圖7係模型式至作用之流程圖。 圖8係說明處理條件控制部之補正動作用之表。 圖9係說明處理條件產生部之動作用之流程圖。 圖10係說明處理條件之控制方向用之圖。 請 先 閲 背 面 意 事 項 再 經濟部智慧財產局員工消費合作社印製 主要元件對照 1 處理裝置 2 處理狀態監視部 3 傳感器 4 傳感器資料保存部 5 處理結果量測値輸入手段 6 處理結果量測値保存部 7 模型式產生部 8 模型式保存部 9 預測部 10 處理條件控制部 11 預測値顯示部 本紙張尺度適用中國國家標準（CNS ) Α4規格（210Χ297公釐）This paper size applies to China National Standard (CNS) A4 (210X297 mm) H Note item Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 511128 A7 _B7_ V. Description of the invention (1) Please read the note on the back SPC ( Statistical Process Control). For this reason, the mean value and variance of the principal components in the processing of the matching device are memorized. When the measured principal component is separated from the mean value by several times the variance, the processing is determined to be abnormal. FIG. 4 is a flowchart illustrating a process suitable for a processing device including both a processing state monitoring unit 2 and a sub-processing state monitoring unit 2. FIG. First, it is determined whether or not a model expression for the device to be processed is generated (step 501). When the model expression is saved, the processing state monitoring unit 2 performs monitoring control (step 502). In the case where the model expression has not been saved, monitoring control is performed by the processing state monitoring unit V (step 503). Not only the model-type processing state monitoring unit of FIG. 1A or FIG. 2 is used, but if there are two sub-process state monitoring units that use only the main component of FIG. 3, as shown in FIG. Devices that have been saved can be monitored for processing status even if they are not. This flexible monitoring system is suitable for monitoring semiconductor manufacturing equipment. Brief description of the drawing Printed by the Consumer Cooperative of Intellectual Property Bureau of the Ministry of Economic Affairs Figure 1A is a block diagram of an if system showing a semiconductor device processing device according to an embodiment of the present invention. A block diagram showing a modified example of the control system of FIG. 1A is shown. Fig. 2 # ^^ Fig. 1 is a block diagram of another embodiment of the shadow state monitoring unit. Fig. 3 is a block diagram showing an embodiment of the sub-processing state monitoring unit. Fig. 4 is a flowchart of an example of a monitoring method for the processing status of a display device. 13- This paper size is applicable to the Chinese national standard (CNS> A4 specification (210X29 * 7 mm). ^ Π128 A7 B7 V. Description of the invention (11 Table) 5A is a flowchart for explaining the operation of the model generation unit. Fig. 5B is a diagram illustrating a prediction method for measuring the processing result of a wafer of n wafers. Fig. 6 is a diagram illustrating a prediction model generation method without abnormal data. Flow chart from model to action. Figure 8 is a table for explaining the corrective action of the processing condition control section. Figure 9 is a flowchart for explaining the action of the processing condition generating section. Figure 10 is a chart for explaining the control direction of the processing condition. Please read the notice on the back first, then compare the main components printed by the Intellectual Property Bureau of the Ministry of Economic Affairs and the Consumer Cooperatives. 1 Processing device 2 Processing status monitoring section 3 Sensor 4 Sensor data storage section 5 Processing result measurement (input means 6 Processing result measurement) Storage section 7 Model generation section 8 Model storage section 9 Prediction section 10 Processing condition control section 11 Prediction and display section The Chinese paper standard (CNS) ) Α4 specification (210 × 297 mm)