KR100446926B1 - Method of monitoring and/or controlling a semiconductor manufacturing apparatus and a system therefor - Google Patents

Abstract

It is an object of the present invention to provide a method and system for controlling or monitoring a semiconductor manufacturing apparatus while predicting the processing result thereof.

The system includes a sensor for monitoring a processing state, a sensor data storage unit for storing sensor data transmitted from the sensor, an input device for inputting measurement values of processing results of semiconductor devices processed by the manufacturing apparatus, and input processing result measurement values. A preservation unit, a model equation generation unit for generating model equations from the stored sensor data and processing result measurements, a model equation preservation unit for storing the generated model equations, and a prediction result of the processing results from the stored model equations and sensor data. And a processing condition control unit for controlling the processing conditions of the manufacturing apparatus from the predicted processing result.

Description

METHOD OF MONITORING AND / OR CONTROLLING A SEMICONDUCTOR MANUFACTURING APPARATUS AND A SYSTEM THEREFOR}

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a semiconductor processing apparatus, and more particularly, to a method for monitoring and / or controlling a semiconductor processing apparatus and a semiconductor manufacturing apparatus for predicting a processing result to improve the operation rate and reliability of the apparatus.

In recent years, the size of semiconductor devices has become finer, and the dimensional accuracy of processing has become so severe that a gate electrode of 0.1 µm or less must be processed with a dimensional accuracy of 10% or less. On the other hand, in the semiconductor manufacturing apparatus that processes heat and plasma physically and chemically, the semiconductor wafer is formed by the reaction product generated by the chemical reaction inside the device, which remains on the inner wall of the device and changes the processing state of the wafer with time. Let's do it. For this reason, as the number of wafer processes is overlapped, the processing shape of the semiconductor device on the wafer gradually changes, and performance deteriorates. In order to cope with this problem, measures are usually taken such that the deposit on the inner wall of the chamber is cleaned by plasma, or the temperature of the chamber wall is raised to make the deposit difficult to attach. However, in most cases, these countermeasures are not perfect, and eventually the processing shape of the semiconductor device gradually changes. For this reason, as the machining shape becomes a problem, the parts of the manufacturing apparatus must be replaced or cleaned before the change. In addition to the deposition film, variations in the state of various apparatuses are involved in variations in wafer processing shape. For this reason, studies have been devised to detect a change in the processing state inside the semiconductor gymnastics apparatus and to keep the processing state constant by feeding back the detection result to the input of the semiconductor manufacturing processing apparatus.

A method of monitoring variation in plasma processing is disclosed in, for example, Japanese Patent Laid-Open No. 10-125660. This publication shows a method for predicting device performance or diagnosing the state of plasma using the relationship between plasma processing characteristics and electrical signals of the device. As a method, an approximation equation showing the relationship between the three electrical signals and the plasma processing characteristics of the apparatus is shown by means of multiple regression analysis. Another example is shown in Japanese Patent Laid-Open No. 11-87323. This publication discloses a method of applying a conventional detection system provided with a plurality of existing detectors to a semiconductor manufacturing apparatus and monitoring the state of the device from the correlation signal of the detection signal. As a method of generating the correlation signal, a calculation formula based on a ratio of six electrical signals is disclosed. Another example is shown in US Pat. No. 5,565,223. This publication shows how to introduce a number of signals from a light or mass spectrometer to generate a correlation signal to monitor the condition of the device. As a method of generating this correlation signal, a method using principal component analysis has been shown.

However, in the method of Japanese Patent Application Laid-Open No. 10-125660, when there are many kinds of sensor data for monitoring a device, many signals irrelevant to the processing performance to be predicted are included in the explanatory variable, so the prediction by the multiple regression analysis becomes difficult. . In addition, the method of Japanese Patent Laid-Open No. 11-87323 is a general method in which a signal obtained by correlating a plurality of detection signals from a plurality of well-known detection means is used for diagnosis. The method of taking the disclosed correlation is also a conventional method of taking a ratio of several signals, and it is difficult to apply it to a system that accurately monitors the state of a semiconductor manufacturing apparatus taking various states depending on many causes of variation. US Pat. No. 5,565,23 discloses a method of monitoring the state of plasma by principal component analysis of a large amount of data monitored by the device, unlike the method described above, to promote the variation of the state of the device. However, the semiconductor manufacturing apparatus used for actual mass production does not operate well by simply applying the method of general statistical processing. This is because, for example, it is often impossible to know how the treatment result will change if the main component changes.

SUMMARY OF THE INVENTION An object of the present invention is to provide a semiconductor processing apparatus for processing various types of devices, wherein the semiconductor processing apparatus improves the operation rate and reliability of the apparatus by monitoring the processing state and detecting abnormal processing or predicting the processing result based on the monitor output; To provide a way.

1A is a block diagram of a control system of a processing apparatus for a semiconductor device, showing an embodiment of the present invention;

1B is a block diagram showing a modification of the control system of FIG. 1A;

2 is a block diagram illustrating another embodiment of the processing state monitoring unit of FIG. 1;

3 is a block diagram showing an embodiment of a subprocess state monitoring unit;

4 is a flowchart showing an example of a method for monitoring a processing state of a device;

5A is a flowchart for explaining an operation of a model generator;

5B is a table for illustrating a method of predicting a processing result determination value of n wafers;

6 is a view for explaining a prediction model generation method of the abnormal data free;

7 is a flow chart for creating a model formula;

8 is a table for explaining a correction operation of the processing condition controller;

9 is a flowchart for explaining an operation of a processing condition generating unit;

10 is a diagram for explaining a method of controlling processing conditions.

In order to solve the above problems, the following method was adopted.

According to one aspect of the present invention, a semiconductor processing apparatus includes a sensor for monitoring a processing state of a semiconductor processing apparatus for processing a semiconductor wafer, and a processing result input for inputting measurement values of processing results of the semiconductor wafer processed by the semiconductor processing apparatus. Means, a model formula generator for generating a model formula for predicting a process result as the explanatory variable based on the sensor data acquired by the sensor and the measured value, and a process result based on the model formula and the sensor data And a processing condition control unit for controlling the processing conditions of the semiconductor processing apparatus to correct the deviation by comparing the predicted processing result with a preset setting value.

With this configuration, according to the present invention, the operation rate of the semiconductor manufacturing apparatus is obtained by acquiring the monitor data from the semiconductor manufacturing apparatus by the sensor and predicting the processing result by a model equation before or without measuring the processing result of the sample. And improve the reliability.

Embodiments of the present invention will be described below with reference to the drawings.

1A shows a first embodiment of the present invention. In the figure, the processing apparatus 1 is provided with two processing state monitoring units 2 for monitoring the processing state. The processing state monitoring unit 2 may be assembled in the apparatus 1 or may be provided outside the apparatus 1. It may also be installed in a place away from the network or the like. In addition, as shown in FIG. 1B, some of the functions may be separated through a network or the like. The processing state monitoring unit 2 has the following configuration. First, it has the sensor 3 which monitors the processing state of the wafer in the processing apparatus 1. The sensor usually uses several kinds of sensors. For example, in a plasma processing in a plasma etching apparatus, a plasma CVD apparatus, or the like, the sensor 3 spectroscopically decomposes the emitted light of the plasma being processed by using a spectrometer, and acquires the light emission intensity for each decomposed wavelength as sensor data. For example, when a spectrometer having a 1000-channel CCD array is used, 1000 sensor data can be acquired for each sampling. The pressure, temperature and gas flow rate of the apparatus are also used as sensor data. In addition, electrical measurement results such as current, voltage, impedance, and their harmonic components can be used as sensor data. During the processing of the wafer, these sensor data are acquired at appropriate time intervals, and the obtained sensor data is stored in the sensor data storage section 4. On the other hand, the processed wafer is measured by a processing result measuring instrument which is assembled to the outside of the apparatus 1 or the apparatus. The measurement of the processing result is the measurement of the gate width by CDSEM, the measurement of the processed shape such as the cross-sectional shape by SEM, or the electrical characteristics of the processed device. These measurements do not have to be performed on all wafers. Usually, some of the wafers are extracted and the processing results are measured. The processing state monitoring unit 2 has processing result measurement value input means 5 to receive the measurement value of this processing result. The input means 5 may be a reader for reading information recorded on a portable medium such as a flexible disk or a CDROM, or may be a wired or wireless network connection device (input means 56). The measurement value of the received processing result is stored in the processing result measurement value storage section 6. The storage unit 6 stores the measured values of the processing results in various devices and the like. The model generator 7 is a device of the same type from the sensor data storage section 4 and the processing result measurement value storage section 6, and draws out samples in which both data of the sensor data and the processing result measurement values are stored. . For example, when the number of samples is three or more, a model equation is generated that predicts the measurement result of the treatment using sensor data as an explanatory variable. Usually, there are many types and numbers of sensor data at this time, and it is difficult to select the sensor data used for prediction automatically. In particular, in the case where various devices are processed, it is difficult to determine the sensor to be used for the prediction in advance because the sensor data valid for the prediction is different for each device.

It is a figure explaining the model formulating process by the PLS method. As shown in FIG. 5A, the PLS method automatically generates explanatory variables from a plurality of sensor data so as to have the strongest correlation with the fluctuation of data to be predicted. At the same time, a function for calculating explanatory variables from sensor data is also obtained. For example, it is assumed that the measurement result of the n wafers is a prediction target, and the measurement result of the i-th wafer is represented by Yi. When m pieces of sensor data are obtained from one wafer, Sij represents the jth sensor data of the i-th wafer. The m sensor data may be data at different times of the same sensor, or may be data from different sensors. 5B is a diagram for explaining an example of sensor data Sij. As shown in this figure, the processing of one wafer of the apparatus 1 is divided into three stages of processing conditions. When the sensors are A, B, or C three types, Sij may be taken as S11 to Sn9. Sij may be an average value of the sensor data during each step processing, and may be a value obtained by converting sensor data such as square and reciprocal. Using the PLS method, the sensor data Sij can be converted into m explanatory variables Xik, which are arranged in the order of the intensity of correlation with the variation in the measurement result Yi. The function Fk for converting the sensor data Sij into the explanatory variable Xk is expressed by Equation 1 below.

Some of these explanatory variables (Xik) are used to predict treatment outcome measurements. Usually, since the explanatory variable Xi1 has the strongest correlation with the processing result measured value Yi, Xi1, Xi2, Xi3 and the like are selected as the explanatory variable. In the PLS method, it is sometimes preferable to prepare a prediction equation as shown in Equation 2 using explanatory variables such as Xi1 described as simultaneously generating a prediction equation as in Equation 2.

On the other hand, the measured value of the processing result also includes the data of the wafer which indicates that the processing state of the wafer is bad and that the abnormal processing result is measured. When such data is predicted by a normal regression analysis, as shown in Fig. 6, a model equation with poor prediction accuracy influenced by abnormal data is generated. For this reason, robust regression analysis is used for prediction. Using robust regression analysis, an accurate prediction model equation can be generated because abnormal data as shown in FIG. 6 is out of the prediction object as an outlier.

7 is a flowchart for describing a model formulating process of the model formula generating unit 7.

When there are many kinds of sensor data in the model generator 7, principal component analysis of the sensor data is performed (steps 701 and 702), and robust regression analysis is performed using the obtained principal components (step 705). , 706). At this time, since the principal component that is not necessary for the prediction of the processing result is included in the explanatory variable, as shown in the flowchart, the principal component having a small regression coefficient is removed (step 707) to add another principal component (second principal component) to the explanatory variable ( In step 704, the process of performing the regression analysis again (step 706) is repeated until the prediction error becomes less than or equal to the set value (step 708). These regression analyzes may be linear, or nonlinear regression analysis derived from the physical characteristics of the treatment or the experience value may be used.

The model formula generated by the method as described above is stored in the model formula storage section 8 of FIG. Since the model formulas are created for each device of the same type, the model formula storage unit 8 stores as many model formulas as the number of devices processed by the processing apparatus.

When a predetermined device wafer is loaded and processed in the processing apparatus 1, a model equation corresponding to the device is loaded in the predicting unit 9. During the processing of the device, the signal obtained from the sensor 3 is converted into explanatory variables using Equation 1 obtained by the PLS method, or converted into principal components by principal component analysis if necessary, and processed according to the model equation of Equation 2. The prediction of is calculated. The calculated predicted value is passed to the processing condition control unit 10, and the control unit 10 changes the processing condition so as to correct the deviation between the set value of the processing result and the predicted value.

Next, a specific method of correcting the processing condition in the processing condition controller 10 will be described. Here we use the PLS method again. In the processing of a normal semiconductor device, some opposing processing performances are often required as processing requirements. For example, in etching the gate electrode, the verticality of the sidewalls of the gate electrode and the etching selectivity of the underlying oxide film and the gate polysilicon are required. That is, it is better to use etching conditions having a lower deposition property in order to achieve verticality of the sidewall, and to use high etching conditions attaining high selectivity with an underlying oxide film. If there are two conflicting demands as described above, it is difficult to control the processing conditions. 8 to 10 are diagrams for explaining processing conditions satisfying such an upper half requirement. For example, if the treatment condition 1 (here, referred to as the flow rate of gas A) is reduced, the verticality may be improved, and at the same time, the selectivity of the underlying oxide film is deteriorated, which is not preferable as the treatment condition.

For this reason, for example, a combination of processing condition 1 and processing condition 2 (herein, referred to as wafer device power) must be found to improve the verticality of the sidewalls and also to deteriorate the selectivity of the underlying oxide film.

In order to solve this problem, as shown in Fig. 8, experimental conditions in which several to several tens of processing conditions are changed around a central condition for processing are usually set, and processing is performed to measure processing results. Points 1 to 4 in FIG. 10 correspond to experimental conditions 1 to 4 of FIG. 8. Here, the measurement of the perpendicularity of the sidewall is taken as the processing result measured value A, and the underlying oxide film selectivity is taken as the processing result measured value B. As shown in FIG.

As shown in Fig. 9, if the correlation between two types of processing conditions and two types of processing result measurements is applied to the experiment by applying the PLS method, the direction A of a condition having a strong correlation with the sidewall perpendicularity as shown in Fig. 10 is shown. Is obtained. Similarly, the direction B having a strong correlation to the underlying oxide film selectivity can be calculated from the direction strongly correlated with the underlying oxide film selectivity obtained by the PLS method. This conditional direction A and the conditional direction B are set in the process condition control part 10 of FIG. 1A. In this way, when it is predicted that the verticality of the sidewall is deteriorated by the predictor 9 based on the model equation, if the processing conditions are modified in the conditional direction A, the sidewall perpendicularity is improved without sacrificing the underlying oxide selectivity. can do. The control direction of the calculated processing condition is stored in the processing condition control direction preservation means 14, and is used to correct the processing condition when the predicted value of the processing result by the model equation deviates from the set value.

In the examples here, two kinds of processing conditions were changed, but in the PLS method, many kinds of processing conditions could be changed again, and the more the processing conditions were changed, the more preferable the result was obtained. In addition, there are not only two types of measurements of contradictory treatment results but also a larger number of treatment results. For example, in addition to the perpendicularity of the sidewalls and the underlying oxide film selectivity, a mask selectivity and the like can be taken.

2 shows another embodiment of the present invention. The system of FIG. 2 has a configuration substantially the same as that of the system of FIG. 1A, but has a predictive value display section 11 instead of the processing condition control section 10 to display a message informing of an abnormality in the display of the apparatus or the like. This display portion may be an alarm buzzer or an e-mail transmission.

3 shows another embodiment of the present invention. The system up to now is based on the prediction of the processing result by the model expression, and thus cannot monitor the processing of the device of the type which does not generate the model expression. In many cases, the measurement of the processing result is very time consuming, and in most cases, the measurement of the processing result is not performed, and such a device cannot generate a model equation. For this reason, the main component extraction part 12 extracts a main component from many kinds of sensor data as shown in FIG. 3, and the abnormality monitoring part 13 monitors the nonuniformity of the fluctuation of a main component, and detects the abnormality of a process. When the abnormality is detected, the processing start of the next wafer of the apparatus may be stopped. For the above detection, a nonuniform management method called, for example, SPC (Statistical Process Control) may be used. To this end, the average value and variance of the main components in the processing of the corresponding device are stored, and the processing is judged abnormal when the measured main component has fallen several times or more from the average value.

4 is a diagram showing a processing flow suitable for a processing apparatus including both the processing state monitoring unit 2 and the subprocessing state monitoring unit 2. First, it is determined whether or not a model equation for a device to be processed is generated and stored (step 501). If the model expression is stored, the monitoring state control unit 2 executes monitoring control (step 502). If the model expression is not stored, the monitoring control is executed by the processing state monitoring unit 2 '(step 503).

In addition to the processing state monitoring unit using the model formulas shown in Figs. 1A and 2 as well as two subprocessing state monitoring units using only the main components as shown in Fig. 3, the device having the model formula shown in Fig. 4 is generated and stored. Even if the device has not been created yet, the processing status can be monitored. Such a flexible monitoring system is suitable for monitoring semiconductor manufacturing devices.

Claims (17)

In the semiconductor plasma processing apparatus for processing a semiconductor wafer, A sensor for monitoring a processing state of the semiconductor plasma processing apparatus for processing a semiconductor wafer; Processing result input means for inputting an external measurement value of the processing result of the semiconductor wafer processed by the semiconductor processing apparatus in a preliminary test; A model equation generation unit for generating a model equation for predicting a processing result using the sensor data acquired by the sensor as an explanatory variable based on the sensor data acquired by the sensor and the external measurement value; A processing result predicting unit predicting a processing result based on the model equation and the sensor data; And a processing condition controller for controlling processing conditions of the semiconductor plasma processing apparatus to correct the deviation between the predicted processing result and the preset value by comparing the predicted processing result with a preset value. The method of claim 1, The model generating unit is a semiconductor plasma processing apparatus for generating a model using a partial least square method (PLS). In the semiconductor plasma processing apparatus for processing a semiconductor wafer, A sensor for monitoring a processing state of the semiconductor plasma processing apparatus for processing a semiconductor wafer; Processing result input means for inputting an external measurement value of the processing result of the semiconductor wafer processed by the semiconductor processing apparatus in a preliminary test; A model equation generation unit for generating a model equation for predicting a processing result by using the sensor data acquired by the sensor as an explanatory variable based on the sensor data acquired by the sensor and the external measurement value; A processing result predicting unit predicting a processing result based on the model equation and the sensor data; And a processing condition controller which controls the processing conditions of the semiconductor plasma processing apparatus to correct the deviation between the predicted processing result and the preset value by comparing the predicted processing result with a preset value. The model expression generator generates a model equation by using robust regression (Robust Regression). In the semiconductor plasma processing apparatus for processing a semiconductor wafer, A sensor for monitoring a processing state of the semiconductor plasma processing apparatus for processing a semiconductor wafer; Processing result input means for inputting an external measurement value of the processing result of the semiconductor wafer processed by the semiconductor processing apparatus in a preliminary test; A model equation generation unit for generating a model equation for predicting a processing result by using the sensor data acquired by the sensor as an explanatory variable based on the sensor data acquired by the sensor and the external measurement value; A processing result predicting unit predicting a processing result based on the model equation and the sensor data; And a processing condition controller which controls the processing conditions of the semiconductor plasma processing apparatus to correct the deviation between the predicted processing result and the preset value by comparing the predicted processing result with a preset value. The model expression generator generates a model equation using a principal component robust regression method (Principal Component Robust Regression). In the semiconductor plasma processing apparatus for processing a semiconductor wafer, A sensor for monitoring a processing state of the semiconductor plasma processing apparatus for processing a semiconductor wafer; Processing result input means for inputting an external measurement value of the processing result of the semiconductor wafer processed by the semiconductor processing apparatus in a preliminary test; A model equation generation unit for generating a model equation for predicting a processing result by using the sensor data acquired by the sensor as an explanatory variable based on the sensor data acquired by the sensor and the external measurement value; A processing result predicting unit predicting a processing result based on the model equation and the sensor data; And a display unit for displaying a deviation of the predicted value from the predicted value or the preset value. In the semiconductor processing apparatus for processing a semiconductor wafer, A sensor for monitoring a processing state of the semiconductor processing apparatus for processing the semiconductor wafer; Processing result input means for inputting a measurement value of the processing result of the semiconductor wafer processed by the semiconductor processing apparatus; A model formula generator for generating a model formula for predicting a processing result based on the sensor data acquired by the sensor and the measured value; A processing result predicting unit predicting a processing result based on the model expression and the sensor data; A processing condition controller which controls the processing conditions of the semiconductor processing apparatus to correct the deviation by comparing the predicted processing result with a preset set value; A main component extracting unit extracting a main component based on the sensor data acquired by the sensor; The extraction unit includes an abnormality detection unit for detecting the abnormality of the process based on the nonuniformity of the variation of the main component extracted, If no model expression is generated in the model generation unit, the process stops when the abnormality detection unit detects an abnormality, and the model expression generation unit generates a model expression using the partial least square method (PLS). A semiconductor processing apparatus characterized by the above-mentioned. The method of claim 1, A main component extracting unit extracting a main component based on the sensor data acquired by the sensor; The extraction unit further includes an abnormality detection unit for detecting an abnormality of the process based on the nonuniformity of the variation of the main component extracted, And when the abnormality detection unit detects an abnormality, when a model expression is not generated in the model expression generating unit, the semiconductor plasma processing apparatus. The method of claim 1, A sensor data storage unit for storing the sensor data, and And a processing result measurement value storage section for storing the processing result inputted by the processing result input means, wherein the model expression generating unit generates the model expression based on the sensor data and the measurement values stored in the respective preservation sections, and generates the result. A semiconductor plasma processing apparatus for storing the model formula in the model formula storage section. A method of controlling the processing of a processing target in a semiconductor processing apparatus by using a sensor for monitoring a processing state of a semiconductor processing apparatus for processing a processing target, Inputting a measurement value of a processing result of the semiconductor object processed by the semiconductor processing apparatus; Generating a model equation for predicting a processing result using the sensor data as an explanatory variable based on the sensor data acquired by the sensor and the measured value; Predicting a processing result of the semiconductor object based on the model equation and the sensor data; Controlling the processing conditions of the semiconductor processing apparatus so as to compare the predicted processing result with a preset set value and correct the deviation based on the comparison result. How to control. The method of claim 9, The model generation step is a method for controlling the processing of the processing target to generate a model equation using the partial least square method (PLS). The method of claim 9, The model generation step is a method of controlling the processing of the processing target to generate a model equation using the robust regression (Robust Regression). The method of claim 9, The method of generating a model equation is a method of controlling the processing of the processing target to generate the model equation using the principal component robust regression (Principal Component Robust Regression). A method for monitoring the processing of a processing object in a semiconductor device by using a sensor for monitoring a processing state of a semiconductor processing device for processing a processing object, Inputting a measurement value of a processing result of the semiconductor object processed by the semiconductor processing apparatus; Generating a model equation for predicting a processing result using the sensor data as an explanatory variable based on the sensor data acquired by the sensor and the measured value; Predicting a processing result based on the model equation and the sensor data; And displaying the deviation of the predicted predicted value or the predicted value and a predetermined set value. In the semiconductor processing monitoring method for monitoring the processing of a semiconductor wafer in a semiconductor processing apparatus using a plurality of sensors for monitoring the processing state of the semiconductor processing apparatus for processing a semiconductor wafer, Extracting a main component based on a plurality of sensor data acquired by the plurality of sensors; And detecting an abnormality of the processing on the basis of the unevenness of the variation of the extracted main component. The method of claim 9, Extracting a main component based on a plurality of sensor data acquired by the plurality of sensors; Detecting an abnormality of the process based on the nonuniformity of the variation of the extracted main component; And stopping the processing when the abnormality detecting step detects an abnormality, when no model formula is generated in the modeling step. The method of claim 9, A sensor data preservation step for preserving the sensor data and a process result preservation value preservation process for preserving the processing result input in the processing result input step are further provided. A method of controlling a process to be processed in a semiconductor process that generates the model formula based on the method. In the semiconductor plasma processing apparatus for processing a semiconductor wafer, A sensor for monitoring a processing state of the semiconductor plasma processing apparatus for processing a semiconductor wafer; Receive and store a model equation for predicting a processing result using the sensor data acquired by the sensor, which is generated based on the sensor data acquired by the sensor and the external measurement value, as an explanatory variable, and the measurement value is stored in the preliminary test. A model type preservation section which is a processing result of the semiconductor wafer processed by the semiconductor plasma processing apparatus; A model result storage unit for predicting a processing result based on the model expression and the sensor data and a processing result predictor connected to a sensor; And a processing condition controller for controlling the processing conditions of the semiconductor plasma processing apparatus to correct the deviation between the predicted processing result and the preset value by comparing the predicted processing result with a preset value. Processing unit.