KR20090006437A - Process maintenance method of semiconductor manufacturing equipment - Google Patents

Abstract

The processing method of the semiconductor manufacturing facilities is provided to perform the process management irrespective of the size of the sample by improving the existing multivariate SPC. The first step(S100) is for managing the specification of the managed parameter about process samples extracted from the predetermined processing unit to the reference viewpoint. The second step(S200) is for determining the abnormality about extracted process samples after the reference viewpoint and then determining the cause and resetting the reference.

Description

PROCESS MAINTENANCE METHOD OF SEMICONDUCTOR MANUFACTURING EQUIPMENT}

The present invention relates to a process control method of a semiconductor manufacturing facility, and more particularly, to build a multivariate model and process sample size even when the process sample size is small so that a plurality of management variables can be efficiently managed through a single control chart. The present invention relates to a process control method of a semiconductor manufacturing facility that can enable process management from a small time.

Typically semiconductor devices are fabricated by repeatedly or selectively performing a number of processes such as diffusion, deposition, exposure, etching and cleaning on a wafer. In the plurality of processes, a semiconductor manufacturing facility in which each process is performed is introduced.

In a semiconductor manufacturing apparatus in which each of the above processes is performed, an abnormal state may occur during the process. In this case, the facility transmits the on-line data to the control unit to display the abnormal state or to generate an alarm to notify the abnormal state. Then, by identifying the abnormal state, the user takes measures to solve the abnormal state.

Here, the process control method adopted in the conventional semiconductor manufacturing equipment is as follows. In general, abnormalities detected through the process control are classified into fault groups among similar faults.

The conventional process management method includes a process of extracting a fault pattern due to the abnormality off-line, databaseting the extracted fault pattern, and recognizing the fault pattern online.

1 is a graph showing a method of identifying a conventional fault pattern.

Referring to FIG. 1, faults I to IV are abnormal pattern-specific areas extracted offline and located on the main component space. This fault area is formed based on the abnormal data that has been defended from the equipment that was operated in the past. Information provided with such an abnormal pattern-specific area is registered in a database.

And, the abnormal pattern recognition generated in the equipment operated online is as follows. As shown in FIG. 1, when data outside the normal management area is detected, the controller evaluates whether the detected data is similar to a plurality of abnormal pattern areas registered in the database. The evaluation of the similarity depends on the statistical distance between the sensed data and the abnormal pattern areas. Therefore, the larger the statistical distance is, the smaller the probability of matching the data with the abnormal pattern region becomes.

However, when the data is included in one boundary area of the abnormal pattern areas, it is determined that an abnormality corresponding thereto occurs. In addition, if the data is not included in any one of the abnormal pattern regions, it is determined that a new abnormality has occurred. In this case, the control unit goes offline and registers the pattern in which the new abnormality has occurred in the database.

In order to apply the multivariate SPC (Statistical Process Control) as described above, it is necessary to set a management statistic and to set a management limit of the statistic.

Typically, management statistics are calculated based on a multivariate model. At this time, more than a certain number of process samples are required to perform modeling.

For example, in order to estimate the covariance matrix of the control variables, sufficient samples of 5 to 10 times the number of control variables are usually obtained under stable process conditions.

Therefore, in the conventional multivariate SPC method, modeling is not performed until sufficient samples are secured as described above.

Accordingly, conventionally, in the case of mass production initial process, there is a problem that the start point of management becomes longer than a certain period.

In addition, conventionally, even in a multi-product small quantity production process, a situation in which multivariate modeling is difficult due to the small number of samples obtained for each product.

In addition, when frequent changes occur in the unit process for the same reason, the process conditions are frequently changed according to the conventional multivariate SPC has a problem that is difficult to apply in this case.

The present invention has been made to solve the above problems, the first object of the present invention is to efficiently manage a plurality of management variables through a control chart in the unit process through a semiconductor manufacturing facility Even if the process sample size is small, a multivariate model is established and the process management method of the semiconductor manufacturing equipment that enables process management from the time when the process sample size is small by sequentially updating the model each time a process sample is obtained. In providing.

It is a second object of the present invention to provide a process control method of a semiconductor manufacturing facility that can apply a multivariate SPC technique to an initial production process of a product, a small quantity production process, and a process in which frequent condition changes occur.

Process management method of the semiconductor manufacturing equipment of the present invention for achieving the above object is a first step of managing the specification of the management variable for the process samples extracted in a series of processes up to a reference time point, and extracted after the reference time point And a second step of determining whether there is an abnormality for the process samples, and resetting a criterion for determining the cause of the abnormality.

The second step may include calculating management statistics of process samples extracted after the reference time point, and comparing the calculated first management statistics with a first reference control limit to determine whether there is an abnormality of the process samples. Preferably, the method includes one or more detection steps, and a first or more identification step for determining the cause of the abnormality by calculating the contribution of the management variable of the process sample in which the abnormality is detected.

In addition, it is preferable to recalculate the first management statistics and the first reference control limits for the normal process samples determined in the first abnormality detecting step.

The reference time point is preferably a time point until the size of the process sample is smaller than the number of the management variable plus two.

On the other hand, in the second step, when the number of the extracted process sample is more than a certain multiple of the number of the management variable, the third step of determining whether there is an abnormality for the process samples extracted after the predetermined multiple or more It is preferred to further comprise a step.

Here, the third step is to calculate the management statistics for the process samples after the number of the extracted process samples is more than a predetermined multiple of the number of the management variable, the calculated second management statistics and the second reference A second abnormality detecting step of determining whether or not abnormalities of the process samples are compared with a control limit, and a second abnormality identifying step of identifying a cause of the abnormality by calculating a contribution of a management variable of the process sample in which the abnormality is detected. It is desirable to.

The present invention can improve the existing multivariate SPC that can efficiently manage a plurality of management variables through a single control chart to build a multivariate model even when the sample size is small, and to update the model whenever a process sample is obtained. Thus, when the size of the process sample is small, the process management can be enabled.

In addition, the present invention can overcome the difficulty of securing a large number of process samples over a certain number for the management of the semiconductor manufacturing process, suggesting a way to overcome these difficulties, the initial production process, multi-quantity small quantity production process and frequent conditions It has an effect that the multivariate SPC technique can be easily applied to a process in which a change occurs.

Hereinafter, a process control method of a semiconductor manufacturing apparatus of the present invention will be described with reference to the accompanying drawings.

1 is a graph showing a method of confirming a conventional abnormal pattern. 2 is a flowchart illustrating a process control method of a semiconductor manufacturing facility of the present invention. 3 is a flowchart showing a first step and a second step according to the present invention. 4 is a flowchart showing a second step according to the present invention. FIG. 5 is a graph illustrating whether abnormality is determined according to control statistics of process samples as the second step according to the present invention is applied to a semiconductor manufacturing unit process. 6 is a flowchart showing a third step according to the present invention. 7 is a graph showing that abnormality is determined and abnormal causes are identified as the third step according to the present invention is applied to a semiconductor manufacturing unit process.

2 and 3, a process management method of a semiconductor manufacturing apparatus of the present invention first goes through a first step of managing a specification of a management variable for process samples extracted in a predetermined unit process up to a reference time point ( S100).

Although not shown in the drawings, the facility is provided with a plurality of measuring instruments for detecting and extracting process samples when the unit process is in progress.

These meters, when the process starts in the facility, sequentially extract the process samples until the size of the process samples is smaller than the number N of the control variable plus two. The section at this time is called 'A' section.

As shown in FIG. 2, a value obtained by adding 2 to a management variable is referred to as 'N'.

The first step according to the present invention manages the specification in the case where there is a specification of the management variables since the sample size for implementing multivariate modeling typically does not reach a certain size in the section A.

Subsequently, the present invention goes through a second step of determining whether there is an abnormality for the process samples extracted after the reference time point and resetting a criterion for determining the cause of the abnormality (S200).

2 to 4, in the second step, multivariate statistical process management is performed from the time when the size of the extracted process sample is equal to 'N'.

Such a section is referred to as a 'B' section as shown in FIG.

4 and 5, the second step first goes through a first abnormality detection step.

The first abnormality detecting step calculates first measurement statistics measured in real time from a measuring instrument of the facility, that is, first management statistics of extracted process samples, preferably, process samples extracted from the reference time point (①, S210).

The calculated first management statistics may be T2 statistics and Square Prediction Error (SPE) statistics.

Subsequently, it is determined whether the process samples are abnormal by comparing the calculated first management statistics with the first reference control limits (2, S220).

The second step is followed by a first abnormal identification step after the first abnormal detection step as described above (③, S230).

That is, the first abnormality identifying step may calculate the contribution of the management variable of the process sample in which the abnormality is detected to identify the cause of the abnormality.

For example, (a) and (b) shown in FIG. 10 are graphs showing the percentages of contributions of different management variables for different process samples.

Therefore, in the case of the process sample of (a), the cause of the abnormality is more likely due to the management variable of 'CD5', and in the case of the process sample of (b), the cause of the abnormality due to the management variables of 'CD11' and 'CD5'. It can be seen that it is likely to be.

Meanwhile, when the process sample extracted in the first abnormality detecting step of the second step is a normal process sample or when the first reference control limit is less than or equal to k multiples, a first management statistic and a first criterion for the normal process sample Recalculate or update control limits.

Here, k may be a value selected in an arbitrary range.

In addition, when k is '1', the process sample determined as abnormal is not used for updating the mean and the covariance matrix.

When k is smaller than '1', the update of the first management statistics is robust, and when k is larger than '1', the frequency of updating the first management statistics can be increased.

That is, the mean vector and the covariance matrix for the normal process sample can be updated (④).

Here, the updating of the matrix may use the method proposed by "Sullivan" and Jones (2002).

Next, a reduced space is found based on the updated average and covariance matrix. This is accomplished by multivariate projection techniques such as principal component analysis. The principal component model is thus set (⑤).

In this case, the dimension of variance may be defined by the number of management variables, and the dimension of the reduced space may be equal to the number of principal components calculated through the principal component analysis.

This updated multivariate model is then used to calculate the first control statistic when the process sample is measured.

Subsequently, a first reference management limit of the first management statistic may be set based on the updated multivariate model (⑥).

On the other hand, in the second step, when the number of the extracted process sample is a certain multiple, preferably more than 10 times the number of the management variable, the predetermined multiple or more as shown in Figs. Afterwards, the process may be subjected to a third step of determining whether there is an abnormality in the extracted process samples (S310 to S330).

Here, the criterion of 10 times or more is a value that is selectively selected, and if the value is not selected, the third step may be omitted.

If the number of the extracted process samples is more than 10 times the number of the management variables, the three steps corresponding to the 'C' section are performed.

In step 3, process samples of a predetermined size or more exist while passing through section 'A' and section 'B'.

Therefore, referring to FIG. 7, the multivariate model as in step 2 is set through process samples having a predetermined size (which may be historical data) (7).

Subsequently, the second reference control limit of the second management statistic for the process samples after the number of process samples to be extracted (which may be a measured value measured from a measuring device of the facility) is 10 or more times the number of the control variables. (8, 9, S310).

Then, the second management statistics are calculated. The second management statistics may be T2 statistics and SPE statistics.

In addition, a second abnormality detection step of determining whether there is an abnormality of the process samples is compared with the calculated second management statistics and the second reference control limit (S320).

In addition, as mentioned in the second step, the second abnormality identification step of determining the cause of the abnormality by calculating the contribution of the management variable of the process sample is detected (이상, S330).

Next, with reference to FIGS. 8 to 10, an example in which the process management method of the semiconductor manufacturing equipment of the present invention is applied will be described in more detail.

Referring to FIG. 8, it can be seen that a total of six management variables are measured for the extracted process samples when the unit process is in progress.

8 shows the management results through a second step in accordance with the present invention for 417 process samples.

Here, the "standardized T2 statistic" is a value obtained by dividing the T2 statistic by the corresponding first reference management limit.

Therefore, the management limit of the standardized T2 statistic is set to '1'. In addition, in FIG. 5, black origins represent process samples determined to be normal, and white origins represent process samples determined to be abnormal.

In the case shown in FIG. 7, the transition to the management through the third step according to the present invention is not made, and the management of the second step is performed on the entire process samples.

However, it may be possible to switch to the management through the third step according to the present invention at the time when a process sample of about 10 times the management variable is obtained (in this case, the time when the number of 60 samples is obtained) according to the operator's selection.

FIG. 9 is an enlarged view of a portion A of FIG. 7.

As shown in Fig. 9, management through the first step is started from the eighth sample (six management variables + 2).

FIG. 10 is a bar chart representing contribution ratios of management variables of the 144th and 246th process samples among the process samples determined as abnormal. Since these charts have been described above, they will be omitted below.

Through this, the present invention can automatically present a cluster of management variables that are determined to be the cause in the abnormally detected process sample, and can further provide data for further cause analysis of the worker.

1 is a graph showing a method of confirming a conventional abnormal pattern.

2 is a flowchart illustrating a process control method of a semiconductor manufacturing facility of the present invention.

3 is a flowchart showing a first step and a second step according to the present invention.

4 is a flowchart showing a second step according to the present invention.

FIG. 5 is a graph illustrating whether abnormality is determined according to control statistics of process samples as the second step according to the present invention is applied to a semiconductor manufacturing unit process.

6 is a flowchart showing a third step according to the present invention.

7 is a graph showing that abnormality is determined and abnormal causes are identified as the third step according to the present invention is applied to a semiconductor manufacturing unit process.

8 is a graph illustrating whether abnormality is determined according to control statistics of process samples as the second step according to the present invention is applied to a semiconductor manufacturing unit process.

FIG. 9 is an enlarged graph of part A of FIG. 5.

10 is a graph showing the contribution rate of management variables according to the present invention.

Claims (6)

A first step of managing a specification of management variables for process samples extracted in a series of processes to a reference time point; And And a second step of determining whether there is an abnormality for the process samples extracted after the reference time point and resetting a criterion for identifying the cause of the abnormality. The method of claim 1, The second step may include calculating a first management statistic of process samples extracted after the reference time point, and comparing the calculated first management statistic with a first reference control limit to determine whether there is an abnormality of the process samples. With one or more detection steps, And a first abnormality identifying step of determining the cause of the abnormality by calculating the contribution of the management variable of the process sample in which the abnormality is detected. The method of claim 2, And resetting the standard by recalculating the first management statistics and the first standard control limits for the normal process samples determined in the first abnormality detecting step. The method of claim 1, And said reference time point is a time point until the size of said process sample is smaller than the number of said management variables plus two. The method of claim 4, wherein In the second step, if the number of the extracted process sample is more than a predetermined multiple of the number of the management variable, And a third step of determining whether there is an abnormality in the process samples extracted after the predetermined multiples or more. The method of claim 5, The third step is to calculate the second management statistics and the second reference control limits for the process samples after the number of the extracted process samples is more than a predetermined multiple of the number of the management variable, the calculated second A second abnormality detecting step of determining whether there is an abnormality of the process samples by comparing the control statistics with the second standard control limit; And a second abnormality identifying step of calculating the contribution of the management variable of the process sample in which the abnormality is detected to identify the cause of the abnormality.

Images