KR20090095694A - method for maintaining statistical process used semiconductor manufacture equipment control system - Google Patents

Abstract

A method for maintaining a statistical process used for a semiconductor manufacture equipment control system are provided to increase the reliability of a process management and maximize it by update a management limit time through QB statistical volume in real time. In a method for maintaining a statistical process used for a semiconductor manufacture equipment control system, measurement data corresponded to the property of a wafer from more than one measurement apparatus are collected(S10). The general statistics amount corresponding to a mean value and a distribution value of measurement data is calculated(S20). The distribution in a group is corresponded to the measurement data detected from a plurality of wafers. Measurement data are detected from a target wafer based on the several cassettes in which a plurality of wafers is mounted.

Description

Statistical process control method applied to the management system of semiconductor manufacturing equipment

The present invention relates to a statistical process control method applied to a management system of a semiconductor manufacturing facility. Specifically, the process control limit calculated from the QB statistics can be used to detect an abnormality in the semiconductor manufacturing process and analyze the cause. A statistical process control method applied to a semiconductor manufacturing facility management system.

Recently, in the semiconductor manufacturing industry, the minimum line width applied to the semiconductor integrated circuit process has been steadily decreasing to increase the operation speed of the semiconductor chip and increase the information storage capability per unit area. In addition, the size of semiconductor devices such as transistors integrated on semiconductor wafers has been reduced to sub-half microns or less.

The semiconductor device may be manufactured through a deposition process, a photo process, an etching process, and a diffusion process, and may be produced when these processes are repeated several times to several tens of times. In addition, there are a number of variables that can affect the results of each process.

Process variables are combined with each other and appear in each process and have a significant impact on results such as wafer surface inspection. The measurement results can be used to track individual process variables. Therefore, the measurement result is fed back in real time, and stabilization of each process is considerably performed.

In general, one of the well-known methods used to achieve process stabilization is online statistical process control. The most commonly used method of statistical process control is the control chart. The most important function of the online control chart is to detect as soon as possible the instability of the process, which is the cause of the process spread, so that the investigation of the cause of the change and the corresponding corrective action can be taken before a large amount of failure occurs. It is. Monitoring of process variation in control charts is accomplished by processing the process data measured in the process into some form of statistical model.

Hereinafter, with reference to the drawings will be described a statistical management method applied to the management system of the semiconductor manufacturing equipment according to the prior art.

1 is a diagram illustrating a statistical process control method applied to a management system of a semiconductor manufacturing facility according to the prior art.

As shown in Figure 1, the statistical process management method applied to the management system of the semiconductor manufacturing equipment according to the prior art calculates the control limits calculated by using the reference data collected in a stable state (S100), in real time during the process By comparing the measurement data collected by the control limit to detect the process variation (S200).

Here, the reference data is collected in a predetermined number or more in order to calculate a control limit by estimating a statistical process model as a stable process state as the process data (S110). Statistical model estimation may be made by removing an outlier generated in an ideally stable process state (S120). The reference data from which the outliers are removed may follow a normal distribution assumed by a general control chart, and statistical modeling may be performed (S130). For example, the result of a statistical model according to a traditional Schewhart-type control chart is the process mean and process variance. The control limit may be calculated using the process average and the process variance (S140). Therefore, the collection of the reference data and the calculation of the control limit can be regarded as the first state to present the criteria for monitoring the process in real time in the following.

On the other hand, the measurement data is data collected by measuring the state of the process made in the mass production of the semiconductor production line in real time (S210). The measurement data is converted into management statistics (S220) and it is determined whether or not to deviate from the management limit calculated in the first state (S230). If measurement data occurs outside the control limits, the engineer may be able to find and correct the cause of the process failure (S240). Thus, determining whether the measurement data is within control limits can be viewed as a second state of monitoring whether the process is normal.

The reference data is composed of various data corresponding to about 100 to about 150 wafers mounted on 4 to 6 cassettes on one basis of a cassette on which about 25 wafers are mounted. To calculate the control limit requires reference data collected from a plurality of wafers mounted in a certain number or more of cassettes. However, it is common for the production line of a multi-variate small quantity production method to lack the amount of reference data collected.

In addition, the measurement data may be expressed as management statistics of the control chart through in-group scattering corresponding to a plurality of wafers mounted in the cassette. However, it does not represent an intergroup distribution corresponding to the measurement data due to the association between cassettes. Therefore, the statistical management method of the prior art has a disadvantage in that the reliability of process management is lowered because only intra-group dispersion can be expressed but not inter-group dispersion.

An object of the present invention for solving the problems according to the prior art described above, the statistical applied to the management system of the semiconductor manufacturing equipment that can increase or maximize the reliability of the process management by using the statistics including intra-group and inter-group dispersion To provide a process control method.

The present invention relates to a statistical process control method applied to a semiconductor manufacturing facility management system suitable for a multi-product small quantity production method.

A statistical process management method applied to a semiconductor manufacturing facility management system according to an aspect of the present invention for achieving the above object comprises the steps of: collecting measurement data corresponding to characteristics of a wafer from at least one measuring device; Calculating general statistics corresponding to the mean and variance values of the measurement data; In-group dispersion corresponding to measurement data detected in the plurality of wafers mounted in one cassette from the general statistics and corresponding to the measurement data detected in the wafer based on the plurality of cassettes in which the plurality of wafers are mounted. Updating process control limits in real time using QB statistics that simultaneously include intergroup dispersion; And detecting whether a process variation occurs based on the process control limit.

Here, when the estimated value of the measurement data corresponding to the group spread is small, it is preferable to generate new measurement data using the bootstrap resampling method to use the QB statistics.

According to the present invention, there is an effect of increasing or maximizing the reliability of process management while updating control limits in real time using QB statistics including intergroup distribution and intragroup distribution.

In addition, if the measurement data produced by the short semiconductor manufacturing process is insufficient, new measurement data can be generated by the bootstrap resampling method and used in a semiconductor production line or a laboratory, which is suitable for a multi-product small quantity production method.

Hereinafter, a statistical process control method applied to a semiconductor manufacturing facility management system according to an exemplary embodiment of the present invention will be described with reference to the accompanying drawings. The embodiments of the present invention may be modified in various forms, and the scope of the present invention should not be construed as being limited by the embodiments described below. This embodiment is provided to more completely explain the present invention to those skilled in the art. Therefore, the shape of the elements in the drawings are exaggerated to emphasize a clearer description.

FIG. 2 is a diagram schematically showing a semiconductor manufacturing facility management system, wherein the semiconductor manufacturing facility management system of the present invention includes a plurality of manufacturing facilities 10 that perform a predetermined unit process of processing a wafer, and the unit process. A plurality of measurement facilities 20 for measuring the characteristics of the completed wafer, and a host computer 30 for overall control of the overall semiconductor manufacturing process while exchanging data with the measurement facility 20 and the manufacturing facility 10. It is configured to include.

Here, the plurality of manufacturing facilities 10 are unit processing apparatuses that individually perform a unit process for forming a semiconductor device pattern having a predetermined size on a wafer surface. For example, the unit process includes a deposition process, a photo process, an etching process, a polishing process, and a cleaning process. Although not shown, the plurality of manufacturing facilities 10 includes a plurality of first facility computers that are individually managed while automating the unit process.

The plurality of measurement facilities 20 are measurement devices that measure the characteristics of the wafer or the pattern by measuring the pattern formed on the wafer mechanically, optically, and electrically. For example, the metrology facility 20 includes an electron microscope such as an SEM, an optical microscope such as a TEM, and an X-ray inspection apparatus. Similarly, the plurality of metrology facilities 20 comprises a plurality of second facility computers (not shown) that automate the process, ie, the process.

The host computer 30 is formed so as to systematically manage almost all the manufacturing facilities 10 and the measurement facilities 20 throughout the semiconductor production line. In addition, all situations occurring in the manufacturing facility 10 and the measurement facility 20 can be grasped and recorded, and are set to be linked to subsequent processes. For example, the first facility computer, the second facility computer, and the host computer 30 exchange data with each other while communicating by means of TCP / IP (Transmission Control Protocol / Internet Protocol), which is a generally known communication protocol. The first facility computer, the second facility computer, and the host computer 30 use the Semi Equipment Communication Standard (SECS) protocol, which is a semiconductor equipment standard communication protocol that defines communication between each other so as to recognize and respond to the transmitted data. Are configured to communicate with each other and share data.

The statistical process control method applied to the management system of the semiconductor manufacturing equipment is as follows.

3 is a diagram illustrating a statistical process control method applied to a management system for semiconductor manufacturing equipment according to an embodiment of the present invention.

As shown in FIG. 3, the statistical process management method applied to the management system of the semiconductor manufacturing facility according to the embodiment of the present invention calculates the management limit while collecting measurement data in real time from the host computer 30, and Control limits are formed to detect process anomalies.

First, the host computer 30 sequentially collects measurement data and stores it in a database (S10). Here, the host computer 30 collects measurement data from the measurement facility 20 in real time. For example, the collection of measurement data presupposes that the process control state at the time of being applied to the control chart is stable.

Next, the general statistics of the measurement data are calculated (S20). For example, the general statistics correspond to the mean value and the variance value of the measurement data. At this time, the host computer 30 must determine the outlier of the measurement data before calculating the general statistics of the measurement data. Outliers are values that are unlikely to occur in probability, and include measurement errors and timing misses.

The determination of the outlier may be determined depending on whether or not the measurement data just before the present time point indicates the outlier. That is, it is possible to calculate the general statistics by determining that the measurement data currently collected is normal according to whether the measurement data exists in the previous average value or the variance value before the calculation of the general statistics. Accordingly, the host computer 30 calculates the general statistics corresponding to the mean value and the variance value using the measured data after determining the outliers of the measured data measured in real time.

)과 표본 평균(

)의 차이를 표준편차(

)으로 곱하고, 다시 분산추정치(S² _p,i)나누면서 표준정규분포화(w_i)되어 있다. 분산추정치(S² _p,i)는 분산값(S² _i)으로 이루어진다. 이때, 분산값(S² _i)은 군내 산포에 대응되는 값만을 포함하여 이루어진다. 결국, Q 통계량은 군내 산포를 포함하여 나타난다. 뿐만 아니라, Q 통계량은 평균의 차이(

-

)를 분산추정치(S² _p,i)나누면서 일반화(normalize)되면서 0을 기준으로 ±1 또는 ±숫자 범주의 관리 한계를 구하도록 할 수 있다.Next, the process control limits are updated in real time while converting the general statistics into Q statistics and QB statistics (S30). The QB statistic is a general form of the Q statistic and is shown in FIG. 4 compared with the Z statistic. Here, the Q statistic is expressed as the inverse of the normal distribution represented by the cumulative probability distribution function (G _{n1 + .. + ni-i} (w _i )) of the T distribution. n ₁ + .. + n _i -i is the degree of freedom of the T distribution. The cumulative probability distribution function (G _{n1 + .. + ni} (w _i )) of the T distribution is the mean (

) And sample mean (

) Is the standard deviation (

) And multiply by the variance estimate (S ² _{p, i} ) and standard normal distribution (w _i ). The variance estimate S ² _{p, i} consists of the variance value S ² _i . In this case, the dispersion value S ² _i includes only a value corresponding to the dispersion within the group. Eventually, the Q statistic includes intragroup distribution. In addition, the Q statistic is the difference in means (

-

) Can be normalized by dividing the variance estimate (S ² _{p, i} ) to obtain control limits of the ± 1 or ± numeric categories based on zero.

)과 표본 평균(

)의 차이를 루트(root) 군간 산포 분산추정치(σ² _b.i) 및 군내 산포 분산 추정치(σ² _w.i)로 나누면서 표준정규분포화(w_i)되어 있다. 군간 산포 분산추정치(σ² _b.i)는 군간 산포 분산값(Σ(

-

)/i-1)으로 나타나며, 군내 산포 분산 추정치(σ² _w.i)는 군내 산포 분산값(S² _i)으로 표현될 수 있다. QB 통계량은 군간 산포 및 군내 산포를 포함하는 분산값들을 이용하여 표준정규분포화할 수 있다. 때문에, QB 통계량을 이용하여 공정 관리 한계를 용이하게 산출할 수 있다.On the other hand, the QB statistic can be expressed as the inverse of the normal distribution represented by the cumulative probability distribution function G _{df (i)} (w _i ) of the T distribution. In this case, df (i) means the degree of freedom of the T distribution. The cumulative probability distribution function of the T distribution (G _{df (i)} (w _i )) is the mean (

) And sample mean (

) Is the difference between the root (root) groups scattered variance estimate (σ ² _bi) and gunnae dispensing nanumyeonseo a variance estimate (σ ² _wi) the standard normal distribution screen (w _i) of the. The scatter variance estimate between groups (σ ² _bi ) is the scatter variance value between groups (Σ (

-

) / i-1), the intra-group dispersion variance estimate (σ ² _wi ) may be expressed as the intra-group dispersion variance value (S ² _i ). The QB statistic can be standard normalized using variance values including intergroup distribution and intragroup distribution. Therefore, the QB statistic can be used to easily calculate the process control limits.

Therefore, the statistical process management method applied to the management system of the semiconductor manufacturing equipment according to the embodiment of the present invention increases the reliability of process management while updating control limits in real time using QB statistics including intergroup distribution and intragroup distribution. Or maximize.

On the other hand, in the case of dispersion within the group, it is rare to encounter a problem in the process of collecting the respective measurement data corresponding to about 25 wafers mounted in one cassette. However, in case of intergroup spread, measurement data collected in cassettes loaded with a plurality of wafers requiring a short process may be insufficient. In other words, process control limits can be calculated using QB statistics only when some measurement data corresponding to the distribution between groups exists.

Therefore, in the exemplary embodiment of the present invention, measurement data corresponding to intergroup spread may be arbitrarily generated by using a bootstrap resampling method. For example, the bootstrap resampling method is a method of creating new samples within the scope of existing sampled metrology data.

FIG. 5 is a diagram illustrating a bootstrap resample method. FIG. 5 illustrates generated data Snew1 corresponding to new measurement data in a category calculated according to a function f calculated based on existing measurement data S existing. , ... Snew10).

Here, the bootstrap resampling method again generates about n random numbers based on the existing n-person measurement data, selects sample data among the random numbers according to an arbitrary index, and generates new generated data based on the sample data. In this case, the arbitrary index may be a function f corresponding to a category calculated based on the existing measurement data. For example, the generated data can be generated to exist within the bell-shaped Gaussian distribution 40 calculated by the function f.

Accordingly, the statistical process management method applied to the management system of the semiconductor manufacturing equipment according to the embodiment of the present invention generates a new measurement data by using a bootstrap resampling method when the measurement data calculated due to the short semiconductor manufacturing process is insufficient. It can also be used in research institutes, so it is suitable for small quantity batch production.

Thereafter, by comparing and analyzing the value calculated by the QB statistics of each measurement data collected on the basis of the control limit to detect the change in the process (S40).

FIG. 6 is a control chart according to normalized QB statistics calculated using measurement data obtained from a semiconductor lab manufacturing line, and FIG. 7 is a control chart according to the non-normalized QB statistics of FIG. 6.

6 and 7, the control chart according to the generalized QB statistic is shown as a graph that converges to 0, but the control chart according to the non-generalized QB statistic is composed of actual measurement data having a unit of the corresponding variable. It is shown in the graph.

Here, each horizontal axis is a variable for each time zone or a corresponding wafer, and the vertical axis is different from each other. The vertical axis of the control chart according to the generalized QB statistics is symmetrical with ±. Although not shown, the vertical axis of the control chart according to the non-generalized QB statistic may be represented by a depth, area, volume, voltage, current, etc. of the measurement sample as a changing value of the unit of the measurement data.

The control chart according to the generalized QB statistic is ± 3, and has a constant control limit line in the linear direction, and the control chart according to the non-generalized QB statistic has a variable control limit line which varies according to the value of actual measurement data. Values beyond each control limit are indicated by red ellipses. The control chart of generalized QB statistic and the control chart of non-generalized QB statistic show the same contents with the difference of expression method.

On the other hand, control charts for non-generalized QB statistics are displayed to make it easier for engineers to understand process parameters, such as the characteristics and errors of the process. On the other hand, the control chart of the generalized QB statistic is displayed to help the public understand the process. Control charts for both generalized and non-generalized QB statistics can be displayed on a display, such as a monitor, to help non-experts and engineers.

For example, in the control chart of the generalized QB statistics, the probability of deviating from the statistical distance of the normal distribution, ± 3, is 0.0027. Therefore, if the control is out of this limit, the engineer can be stopped or alarmed to inform the engineer that a change in the process has been detected. At this time, if the QB statistic value is within the control limit, the host computer 30 determines that there is no process variation, returns to step S10 and collects the next measurement data.

In addition, the control chart of non-generalized QB statistics has the following advantages: The graph of the control chart of the non-generalized QB statistic in FIG. 6 shows that the red variable limit line is crossed one specification point first and then beyond the specification limit line indicated by a solid blue line. At this time, the standard limit line is a management limit line of the SPEC set in the laboratory manufacturing line. That is, it indicates that the moment of departure from the red variable limit line is one point earlier than the moment of departure from the blue standard limit line. For example, in a laboratory manufacturing line, one data point difference can correspond to a considerably long time, so detecting one data point first is very significant in terms of fast process anomaly detection. In fact, one data point difference in FIG. 5 corresponds to one week. This means that the cause of the process anomaly and corrective action can be advanced one week earlier.

Therefore, the statistical process management method applied to the management system of the semiconductor manufacturing equipment according to the embodiment of the present invention can perform the process management while calculating the control limit line in real time by using the QB statistics including the intra-group dispersion and the inter-group dispersion at the same time. Therefore, the reliability of process control can be improved.

Finally, if the data is found out of the control limit line, the engineer can find and solve the cause of the abnormal detection (S50).

In addition, the description of the above embodiment is merely given by way of example with reference to the drawings in order to provide a more thorough understanding of the present invention, it should not be construed as limiting the present invention. In addition, for those skilled in the art, various changes and modifications may be made without departing from the basic principles of the present invention.

1 is a diagram showing a statistical process control method applied to a management system of a semiconductor manufacturing facility according to the prior art.

2 is a schematic diagram of a semiconductor manufacturing equipment management system.

3 is a diagram illustrating a statistical process control method applied to a management system for semiconductor manufacturing equipment according to an embodiment of the present invention.

4 shows the calculation of each of the Q statistics and the QB statistics.

5 is a diagram illustrating a bootstrap resample method.

6 is a control chart according to generalized QB statistics calculated using measurement data obtained at a semiconductor lab manufacturing line.

7 is a control diagram according to the non-generalized QB statistics of FIG. 6;

* Description of the symbols for the main parts of the drawings *

10: manufacturing equipment 20: measuring equipment

30: host computer

Claims (6)

Collecting measurement data corresponding to characteristics of the wafer from at least one measurement device; Calculating general statistics corresponding to the mean and variance values of the measurement data; In-group dispersion corresponding to measurement data detected in the plurality of wafers mounted in one cassette from the general statistics and corresponding to the measurement data detected in the wafer based on the plurality of cassettes in which the plurality of wafers are mounted. Updating process control limits in real time using QB statistics that simultaneously include intergroup dispersion; And Statistical process management method applied to the management system of the semiconductor manufacturing equipment, characterized in that it comprises the step of detecting whether the process variation occurs based on the process control limit. The method of claim 1, If the estimate of the measurement data corresponding to the group spread is small, the statistical process control room applied to the management system of the semiconductor manufacturing equipment, characterized by generating new measurement data using the bootstrap resampling method to use the QB statistics. method. The method of claim 2, The bootstrap resampling method generates n random numbers based on the existing n individual measurement data, selects sample data among the random numbers according to an arbitrary index, and generates new measurement data based on the sample data. Statistical process control method applied to the management system of semiconductor manufacturing equipment. The method of claim 3, wherein The index is a statistical process control method applied to the management system of the semiconductor manufacturing equipment, characterized in that it comprises a function corresponding to the category calculated based on the existing measurement data. The method of claim 4, wherein And the measurement data is present in a bell-shaped Gaussian distribution calculated by the function. The method of claim 1, QB statistic is a statistical process control method applied to a management system of a semiconductor manufacturing facility, characterized by the inverse function of the normal distribution represented by the cumulative probability distribution function of the T distribution.

Images