KR20120128251A - Fault detection method - Google Patents
Fault detection method Download PDFInfo
- Publication number
- KR20120128251A KR20120128251A KR1020110046083A KR20110046083A KR20120128251A KR 20120128251 A KR20120128251 A KR 20120128251A KR 1020110046083 A KR1020110046083 A KR 1020110046083A KR 20110046083 A KR20110046083 A KR 20110046083A KR 20120128251 A KR20120128251 A KR 20120128251A
- Authority
- KR
- South Korea
- Prior art keywords
- result
- value
- estimate
- deriving
- result estimate
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- General Factory Administration (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Description
The present invention relates to a problem detection method, and more particularly to a problem detection method capable of accurate analysis.
Products produced in various industries such as semiconductors, displays, chemicals and steel are manufactured through a number of processes. At this time, it is repeatedly inspected whether the requirements of the product is satisfied in the middle of the process, and finally, after completion of the process, the yield and quality measurement are examined.
In particular, the yield is a very important factor because it indicates whether the process is completed or not and the quality. Therefore, it is very important to maximize the yield and keep the quality in top condition.
Statistical process control (STC) is performed to improve quality and identify defects early, and state-of-the-art precision measuring devices and statistical techniques are applied.
However, it is difficult to determine how all variables associated with an individual process affect yields, which can lead to significant losses if the problem is not detected and properly addressed during the process. Since there are many processes, devices, parameters, etc. in a single production line, the problem must be solved by accurately finding the correct process, device, and parameter.
Data mining and the like have been applied to detect problematic processes, devices and parameters. Data mining is the process of finding useful correlations hidden among many data, extracting actionable information in the future, and using it in decision making. Such data mining includes an analysis of variance (ANOVA) method and a sequential pattern method.
However, according to the variance analysis method and the sequential pattern method, it may be difficult to obtain accurate analysis results in a complicated environment having various processes and devices. This is because the results of the variance analysis method and the sequential pattern method itself are difficult to adequately represent the variation of the results generated by the individual processes and devices. In addition, when the analysis using the result value, the average of the result values for each process of the samples are the same, it is impossible to search for the problem process, and if the number of devices in each process is one problem problem may be all the same.
That is, according to the conventional method, it may be difficult to obtain accurate analysis results.
The present invention seeks to provide a problem detection method that can statistically and accurately detect problematic processes, devices, and parameters.
Problem detection method according to an embodiment of the present invention, the monitoring data of the parameters (or representative value of the monitoring data) for the samples processed by the plurality of devices in a plurality of processes and the results of the process Collecting result values for confirmation; Deriving a first result estimate for each sample from each device using the monitoring data as an independent variable and the result value as a dependent variable; Deriving a second result estimate for each sample using the first result estimate as an independent variable as a representative value of each process and the result value as a dependent variable; And detecting a problem process by calculating a contribution of each process based on the second result estimated value.
The method may further include rearranging the first result estimate value and the result value for each process, based on the samples, between the first result estimate value and the second result estimate value.
In each of deriving the first result estimate and deriving the second result estimate, a relationship between the independent variable and the dependent variable may be derived using regression modeling.
The regression modeling may be at least one of ordinary least square (OLS), principal component regression (PCR), and partial least square (PLS).
In the detecting of the problem process, the samples may be divided into a defective group and a good group based on the second result estimation value, and the contribution of each process may be derived by the following equation.
≪ Equation &
Contribution = {(average of bad group)-(average of good group)} * regression coefficient
(Where, the regression coefficient is a coefficient derived by the regression modeling for deriving the second result estimate.)
When there are a plurality of devices for the problem process, after detecting the problem process, the method may further include calculating a total loss of each device for the problem process to detect the problem device.
In the deriving of the second result estimate, a relationship between the independent variable and the dependent variable may be derived using regression modeling. In the detecting of the problem device, the samples are divided into a bad group and a good group based on the second result estimate value, and the total loss of each device may be derived by the following equation.
≪ Equation &
Total loss = {(average of bad group)-(average of good group)} * number of samples in bad group * regression coefficient
(Where, the regression coefficient is a coefficient derived by the regression modeling for deriving the second result estimate.)
Problem detection method according to an embodiment of the present invention, the monitoring data (or representative value of the monitoring data) of the parameters (parameters) for the samples processed in the plurality of devices for a plurality of processes, and the results of the process Collecting result values for confirmation to derive a first result estimate for each sample at each device; At least one of a problem process, a problem apparatus, and a problem parameter is derived by deriving a second result estimate for each sample using the first result estimate as an independent variable as the representative value of each process and the result value as a dependent variable. To derive.
According to the present embodiment, the result value is estimated two times, so that the problem process, the problem apparatus, and / or the problem parameter can be statistically accurately detected even in a complicated environment having a plurality of processes using a plurality of devices. In addition, it is possible to compensate for the problem that the result value estimation varies according to the number of samples of the device by two result value estimation.
1 is a flowchart of a problem detection method according to an embodiment of the present invention.
Hereinafter, a problem detection method according to an embodiment of the present invention will be described in detail.
The problem detection method according to the present embodiment is a process, apparatus and / or parameter when a problem such as a decrease in yield occurs in a production line that is performed by various processes using various apparatuses according to a work recipe. This is to detect whether the problem is caused by a parameter.
Such a problem detection method may be performed by a problem detection system including a monitoring unit, a data collection unit, an analysis unit through regression modeling, and the like. Various problems can be used as such a problem detection system.
In this embodiment, the monitoring data (or representative values of the monitoring data) of the parameters for the samples processed by the various apparatuses in the various processes, and the result value for confirming the results of each process, are used twice. The resulting value can be estimated over time to accurately detect the problem process and / or problem device. This will be described in more detail with reference to FIG. 1 as follows.
1 is a flowchart of a problem detection method according to an embodiment of the present invention.
Referring to FIG. 1, in the problem detecting method according to the present exemplary embodiment, the method may include collecting monitoring data and result values (ST10), deriving a first result estimate value (ST20), and deriving a second result estimate value ( ST30), detecting the problem process (ST40) and detecting the problem device (ST50). The method may further include rearranging data between the step ST20 of deriving the first result estimate and the step ST30 of deriving the second result estimate. These steps ST10, ST20, ST30, ST40, ST50 will be described with specific reference examples.
First, in the step ST10 of collecting monitoring data and result values, the monitoring data of the parameter (or the representative value of the monitoring data, the same below) in all the devices used in all the processes of one production line, and the process Collect the result (Y) to confirm the result of.
Here, the monitoring data of the parameter may be monitoring data for a fault detection and classification (FDC). The monitoring data may be actual monitoring data or may use a representative value thereof. For example, as the representative value, a mean value, a median value, a mode value, a minimum value min, a maximum value max, a standard deviation, or the like may be used. However, the present invention is not limited thereto, and various types of monitoring data may be used.
In addition, the result value Y for confirming the result of the process may be a value measured to confirm the result in each process or a yield measured after completion of the final process. However, the present invention is not limited thereto, and various types of result values Y may be used.
For a more detailed description, the case where the total number of processes is four and the measurement is made after the processes are completed will be described with reference to the example.
Devices 1, 2 and 3 are used in step 1, devices 4 and 5 are used in step 2, devices 6, 7 and 8 are used in step 3, and devices 9 and 10 are used in step 4. The monitoring data of the parameters and the resultant values Y are collected in the devices 1 to 10. In this example, the number of samples is limited to 30 and it is assumed that the number is statistically sufficient. The results collected in each of Steps 1 to 4 are shown in Tables 1 to 4, respectively.
Subsequently, in the step ST20 of deriving the first result estimate value, the relationship between the independent variable and the dependent variable in each device is identified by using the collected data as the independent variable and the collected result value Y as the dependent variable. Accordingly, each device derives a first result estimate Y1 ^ for each sample.
In this case, general regression modeling may be used as a method of identifying the relationship between the independent variable and the dependent variable. For example, regression modeling may include ordinary least square (OLS), principal component regression (PCR), and partial least square (PLS). However, the present invention is not limited thereto, and regression modeling may be performed in various ways.
Such a calculation can be easily obtained using an existing statistical tool. In addition, since regression modeling according to other methods or statistical tools for obtaining regression coefficients, compensation values, and the like are widely known, a detailed description thereof will be omitted.
In the above-mentioned reference example, the first result estimate Y1 ^ obtained by modeling the results of Tables 1 to 4 by the partial least square method is shown in Tables 5 to 8, respectively. In this case, since the apparatuses 1, 2 and 3 are used in the process 1, the apparatuses 1, 2 and 3 are modeled, and the processes 2 to 4 are similarly modeled for the apparatuses 4 to 10, respectively. According to modeling, a first result estimate Y1 ^ of each sample is obtained for each device.
Subsequently, the first result estimate Y1 ^ and the result value Y may be rearranged based on a sample so that the second result estimate Y ^ may be smoothly derived. That is, using the first result estimate Y1 ^ as a representative value of the process, the first result estimate Y1 ^ and the result value Y of each process are rearranged based on the samples.
In the above-described reference example, the first result estimate Y1 ^ is used as the representative value of the process, and the first result estimate Y1 ^ and the result value Y of each process are rearranged on a sample basis. Table 9 shows.
Subsequently, in the step ST30 of deriving the second result estimate value, the first result estimate value Y1 ^ for the processes based on each sample is used as the independent variable and the result value Y is the dependent variable in each sample. The relationship between the independent variable and the dependent variable is identified, and accordingly, a second result estimate Y ^ for each sample is derived. In this case, as a method of identifying the relationship between the independent variable and the dependent variable, a method using the regression coefficient and the compensation value obtained by the general regression modeling described above may be used.
Referring to the above-described reference example, the regression coefficients and the compensation values obtained through regression modeling through partial least squares from the rearranged data in Table 9 are shown in Table 10 below.
The equation for obtaining the second result estimate Y ^ from the first result estimate Y1 ^ using the regression coefficient and the compensation value is shown in Equation 1 below.
&Quot; (1) "
Second result estimate (Y ^) = 0.46900 * (first result estimate of step 1) + 0.22800 * (first result estimate of step 2) + 0.23700 * (first result estimate of step 3) + 0.55600 * (step 4 First estimate of result)-60.60300
The second result estimate Y ^ for each sample calculated according to this equation is shown in Table 11 below.
Next, in the step ST40 of detecting a problem process, the problem process is detected based on the second result estimated value Y ^.
More specifically, according to a predetermined criterion, the second result estimated value Y ^ is determined to be good if it is within a desired level, and to be bad if it is outside the level. The contribution is calculated to detect the process with the largest contribution value as the problem process. In this case, the contribution may be calculated as in Equation 2 below.
&Quot; (2) "
Contribution = {(average of bad group)-(average of good group)} * regression coefficient
That is, after subtracting the average of the good group consisting of the sample determined to be good from the average of the bad group consisting of the sample determined to be defective, multiplying the regression coefficients to obtain the contribution that contributed to the result value (Y) in each process.
In the above-described reference example, assuming that the second result estimated value Y ^ is less than 120, it is determined to be good, and if it is 120 or more, it is determined to be bad, and samples 1 to 12 are good groups and samples 13 to 30 are bad groups. to be. In each process, multiplying the difference between the mean of the defective group (ie, the average of samples 13 to 30) and the mean of the good group (ie, samples 1 to 12) by the regression coefficient, the contribution of each process is calculated. This is shown in Table 12.
In Table 12, it can be seen that the contribution of Step 4 is 8.12446, which is the largest among Steps 1-4. Therefore, step 4 is detected as a problem step.
Next, in the step ST50 of detecting the problem device, the problem device is derived from the second result estimate Y ^ for the device and the sample. The specific method of detecting the problem apparatus is similar to the process of deriving a problem process.
That is, the total loss is calculated to detect the device having the largest total loss value as the problem device. In this case, the total loss may be calculated as in Equation 3 below.
&Quot; (3) "
Total loss = {(average of bad group)-(average of good group)} * number of samples in bad group * regression coefficient
That is, after subtracting the average of the good group consisting of the sample determined to be good from the average of the bad group consisting of the sample determined to be defective, multiplying the regression coefficients to obtain the contribution that contributed to the result value (Y) in each process. At this time, the average of the good group is calculated as the average of the samples determined to be a good group in the samples processed in all devices, the average of the bad group is calculated as the average of the samples determined by the bad group among the samples processed in the device. This is because the difference between the equipment can be determined.
Referring to the above-described reference example, as shown in Table 13, samples 1 to 12 of the device 9 are good groups and samples 13 to 30 of the device 10 are bad groups. Calculating the total loss as shown in Table 14, the total loss of device 9 is zero and the total loss of device 10 is 149.56705. Thus, device 10 is detected as a problem device.
As described above, in the present embodiment, after passing through a plurality of processes using a plurality of devices, the resultant value Y is measured, and from the resultant value Y, the first result estimated value Y1 ^ and the second result estimated value Y ^) This makes it possible to accurately detect the problem step and the device even in a complicated environment having a plurality of steps using a plurality of devices.
At this time, a second result estimate Y ^ for each sample is derived by using the first result estimate Y1 ^ obtained through modeling in each apparatus as a representative value of each process. Since the first result estimate Y1 ^ through modeling in each apparatus is used as a representative value of each process, it is possible to easily detect a problem process, and the first result estimate Y1 ^ varies depending on the number of samples of the apparatus. The problem may be complemented by a second result estimate Y ^.
In addition, after detecting the problem process, the problem device may be derived based on the devices corresponding to the problem process and the second result estimate Y ^. Although not described separately, after the problem apparatus is derived, the problem parameter may be detected based on the parameters in the problem process and the problem apparatus and the second result estimated value Y ^. The detection method of the problem parameter is very similar to the detection method of the problem process and / or the problem apparatus and will not be described separately.
Features, structures, effects and the like according to the above-described embodiments are included in at least one embodiment of the present invention, and the present invention is not limited to only one embodiment. Further, the features, structures, effects, and the like illustrated in the embodiments may be combined or modified in other embodiments by those skilled in the art to which the embodiments belong. Therefore, it should be understood that the present invention is not limited to these combinations and modifications.
Claims (8)
Deriving a first result estimate for each sample from each device using the monitoring data as an independent variable and the result value as a dependent variable;
Deriving a second result estimate for each sample using the first result estimate as an independent variable as a representative value of each process and the result value as a dependent variable; And
Detecting a problem process by calculating a contribution of each process based on the second result estimated value
Including a problem detection method.
Between deriving the first result estimate and the second result estimate,
And rearranging the first result estimate value and the result value for each process based on the respective samples.
Deriving a relationship between the independent variable and the dependent variable using regression modeling, respectively, in deriving the first result estimate and deriving the second result estimate.
The regression modeling is at least one of ordinary least square (OLS), principal component regression (PCR), and partial least square (PLS).
In the detecting of the problem process, the samples are divided into a defective group and a good group based on the second result estimate value, and the contribution of each process is derived by the following equation.
≪ Equation &
Contribution = {(average of bad group)-(average of good group)} * regression coefficient
(Where, the regression coefficient is a coefficient derived by the regression modeling for deriving the second result estimate.)
When there are a plurality of devices for the problem process,
After detecting the problem process, further comprising calculating a total loss of each device for the problem process to detect the problem device.
In the deriving of the second result estimate, a relationship between the independent variable and the dependent variable is derived using regression modeling.
In the detecting of the problem device, the samples are divided into a bad group and a good group based on the second result estimate value, and the total loss of each device is derived by the following equation.
≪ Equation &
Total loss = {(average of bad group)-(average of good group)} * number of samples in bad group * regression coefficient
(Where, the regression coefficient is a coefficient derived by the regression modeling for deriving the second result estimate.)
At least one of a problem process, a problem apparatus, and a problem parameter is derived by deriving a second result estimate for each sample using the first result estimate as an independent variable as the representative value of each process and the result value as a dependent variable. Derivation, problem detection method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR20110046083A KR101482758B1 (en) | 2011-05-17 | 2011-05-17 | Fault detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR20110046083A KR101482758B1 (en) | 2011-05-17 | 2011-05-17 | Fault detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
KR20120128251A true KR20120128251A (en) | 2012-11-27 |
KR101482758B1 KR101482758B1 (en) | 2015-01-16 |
Family
ID=47513029
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
KR20110046083A KR101482758B1 (en) | 2011-05-17 | 2011-05-17 | Fault detection method |
Country Status (1)
Country | Link |
---|---|
KR (1) | KR101482758B1 (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101466798B1 (en) * | 2014-05-20 | 2014-12-01 | 삼성전자주식회사 | Method and apparatus for discovering the equipment causing product faults in manufacturing process |
KR20180011048A (en) | 2015-05-25 | 2018-01-31 | 닛산 가가쿠 고교 가부시키 가이샤 | Composition for applying resist pattern |
KR20190059893A (en) | 2016-10-04 | 2019-05-31 | 닛산 가가쿠 가부시키가이샤 | Method for preparing composition for applying resist pattern using solvent substitution method |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20090077601A (en) * | 2008-01-11 | 2009-07-15 | (주)아이세미콘 | The method of multi-variate parameter control limit by yield management |
-
2011
- 2011-05-17 KR KR20110046083A patent/KR101482758B1/en active IP Right Grant
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101466798B1 (en) * | 2014-05-20 | 2014-12-01 | 삼성전자주식회사 | Method and apparatus for discovering the equipment causing product faults in manufacturing process |
KR20180011048A (en) | 2015-05-25 | 2018-01-31 | 닛산 가가쿠 고교 가부시키 가이샤 | Composition for applying resist pattern |
KR20190059893A (en) | 2016-10-04 | 2019-05-31 | 닛산 가가쿠 가부시키가이샤 | Method for preparing composition for applying resist pattern using solvent substitution method |
Also Published As
Publication number | Publication date |
---|---|
KR101482758B1 (en) | 2015-01-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Lin et al. | The worst ill-conditioned silicon wafer slicing machine detected by using grey relational analysis | |
CN111275307B (en) | Quality control method for high-frequency continuous observation data of automatic online station for water quality | |
WO2015115432A1 (en) | Quality management device and method for controlling quality management device | |
WO2008076601B1 (en) | Determining physical property of substrate | |
CN107063099A (en) | A kind of machinery manufacturing industry online quality monitoring method of view-based access control model system | |
TW201401231A (en) | Detection of defects embedded in noise for inspection in semiconductor manufacturing | |
CN103646899B (en) | Wafer defect detection method | |
JP4568786B2 (en) | Factor analysis apparatus and factor analysis method | |
US20130138415A1 (en) | Method and model for monitoring pretreatment process of low-k block layer | |
KR101482758B1 (en) | Fault detection method | |
JP7354421B2 (en) | Error factor estimation device and estimation method | |
US7337033B1 (en) | Data mining to detect performance quality of tools used repetitively in manufacturing | |
CN114840932A (en) | Method for improving TC4 titanium alloy surface roughness prediction precision through multi-factor coupling | |
US11422542B2 (en) | Workpiece surface quality issues detection | |
CN109614758B (en) | Method for monitoring circular shape error with spatial correlation | |
KR20090077601A (en) | The method of multi-variate parameter control limit by yield management | |
CN110021027B (en) | Edge cutting point calculation method based on binocular vision | |
TWI461871B (en) | Monitor method for multi tools | |
CN110057330B (en) | Line width measuring method and line width measuring system | |
CN104535033B (en) | Accurately measure the method and system of different flow product resist thicknesses | |
US20100014745A1 (en) | Inspecting method and inspecting equipment | |
US7137085B1 (en) | Wafer level global bitmap characterization in integrated circuit technology development | |
JP3981911B2 (en) | Manufacturing process monitoring method | |
CN110287610B (en) | Offline mass production product technological parameter adjusting method and adjusting system thereof | |
KR101098044B1 (en) | A prediction system and method on static inverter quality abnormality |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
A201 | Request for examination | ||
E902 | Notification of reason for refusal | ||
E902 | Notification of reason for refusal | ||
E601 | Decision to refuse application | ||
J201 | Request for trial against refusal decision | ||
J301 | Trial decision |
Free format text: TRIAL DECISION FOR APPEAL AGAINST DECISION TO DECLINE REFUSAL REQUESTED 20140124 Effective date: 20141031 |
|
S901 | Examination by remand of revocation | ||
GRNO | Decision to grant (after opposition) | ||
GRNT | Written decision to grant | ||
FPAY | Annual fee payment |
Payment date: 20180205 Year of fee payment: 4 |
|
FPAY | Annual fee payment |
Payment date: 20181023 Year of fee payment: 5 |