JP2003142361A - Method for analyzing data - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently and automatically extract a fault cause or the like, without depending on technologist's subjectivity, so as to analyze data to extract the fault cause or the like. SOLUTION: The method for analyzing the data comprises a step of selecting and extracting data to be an object to be analyzed from original data group, such as a yield value, various type measured values or the like (step S1). The method further comprises a step of extracting one or more data distribution features to the extracted data (step S2). The method also comprises steps of selecting a data distribution feature variables to be an object to be analyzed from them, and data mining a regression tree analysis or the like wit the feature variable as a target variable (step S3). The method also comprises the steps of finishing the regression tree analysis of the extracted data distribution feature (step S4), outputting the analyzed result and confirming it by the technologist (step S5), and making determination of his decision (step S6).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data analysis method for grasping relationships between data widely handled in industry and extracting meaningful results for producing industrially superior results. The present invention relates to a data analysis method for extracting knowledge or information that is difficult to discriminate by paying attention only to a target data value and its average value. The present invention also relates to a data analysis method and a data analysis device for evaluating the accuracy of analysis results.

[0002] For example, data used for grasping a yield variation condition based on a history of used devices, test results, design information or various measurement data acquired in a semiconductor manufacturing process, and thus extracting conditions advantageous for improving yield. Regarding analysis method. In particular, not only the original data stored in the computer system and its average value, but also the data distribution features obtained by editing these original data are automatically and quantitatively extracted and recognized, Based on low yield factors such as semiconductors,
The present invention relates to a data analysis method and a data analysis device for evaluation.

Further, in order to obtain a more efficient and reliable analysis result, a data analysis is made to deal with the case where a plurality of explanatory variables are entangled with each other (not independent) and it becomes difficult to extract a significant difference. The present invention relates to a data analysis method and a data analysis device for evaluating the accuracy of results and the like.

[0004]

2. Description of the Related Art A semiconductor data yield analysis will be taken as an example. In particular, when trying to obtain reference data for determining measures for quality and productivity improvement from the analysis results, such as process data analysis, the accuracy, reliability, etc. are important. The inventors have already filed an application (application number: Japanese Patent Application No. 2001-127534).
issue). In order to find the factors that reduce the yield as quickly as possible and implement countermeasures, equipment history, test results, design information,
Data analysis is performed to find factors that are effective in yield and other factors that are effective in that factor from various measurement data.

In the data analysis, what is to be analyzed, such as the yield value, is called the explanatory variable, the objective variable, the device history that causes the objective variable, the test result, the design information, and various measurement data. Various statistical methods are applied at that time, and by applying data mining as one of them, it is possible to extract valuable information and regularity that are difficult to discriminate from a large amount of various kinds of data.

In order to analyze the cause of failure of a semiconductor device, it is important to analyze the collected data in a multifaceted manner based on scientific grounds and extract more significant differences. Therefore, conventionally, the value of the original data accumulated in the computer system and its average value are often used. However, in some cases, it is difficult to extract a defect factor or the like from an original data group that is intricately entangled. In such a case, if there is a characteristic data distribution regarding various measurement results of wafer in-plane chips or wafers in a lot, yield, etc., analysis of defective data may proceed based on it.

[0007]

However, in the conventional computer system, although original data such as yield value and electric characteristic value is accumulated, a plurality of chips in a wafer surface and a plurality of wafers in a lot are accumulated. The characteristic data distribution over the whole is hardly accumulated. Therefore, the technician edits the original data,
It is necessary to acquire the data distribution status using various statistical analysis tools and chart creation tools. Then, it is necessary to compare the acquired data distribution status with the experience and know-how of the technician and to recognize the totalization and tendency of the data. Therefore, it is difficult to objectively grasp the feature amount related to the distribution of a large amount of original data. Further, there is a problem that an accurate result cannot be obtained even if the analysis is advanced based on the feature amount of the data distribution including the subjectivity of the engineer.

Further, conventionally, an engineer looks at the data distribution status obtained by various statistical analysis tools, chart creation tools, etc., and, for example, the distribution has a certain characteristic “whether” or “not present” or a certain characteristic. Whether the increase / decrease tendency is “increase” or “decrease”, whether a certain feature has “2” periodicity or “not”, or whether a certain feature has “3” periodicity ” , Etc., the feature quantity of the data distribution is represented by discrete values. For this reason, information indicating the extent to which a certain characteristic exists (or does not exist) or the degree to which a certain characteristic has an increasing tendency (or decreasing tendency) is missing. Further, for example, when a certain feature has a certain degree of periodicity of 2 and a certain degree of periodicity of 3 at the same time, there is a problem that only the stronger periodicity is recognized.

Further, considering various test results, measurement results, and combinations thereof, the combinations of assumed data distribution features become enormous, and it is extremely difficult to investigate all of them. In addition, there is a problem that the defect factor corresponding to the extracted data distribution feature is not always known, and much experience and know-how is required to determine an unknown defect factor.

[0010] Actually, even if data mining is applied to the yield analysis of semiconductor data, for example, it may not be successful. When applied to fields such as finance and distribution, there are millions of enormous amounts of data, and the number of explanatory variables is at most tens, so highly accurate analysis results were obtained. However, in the case of semiconductor process data analysis, the number of data is small, and the number of explanatory variables reaches several hundreds (apparatus history, in-process inspection value, etc.) despite the fact that the same type has about 200 lots at most. Multiple explanatory variables are no longer independent, and simple data mining may not give reliable results. This will be briefly described below by taking the yield analysis of semiconductor data as an example.

In process data analysis in which there are many explanatory variables (eg, LSI manufacturing process data) compared to the number of data (eg, lot number), a plurality of explanatory variables are entangled with each other (not independent), and statistically. In many cases, problems due to significant differences cannot be sufficiently narrowed down. Even when data mining (regression tree analysis, etc.) is applied, if this problem occurs, it is necessary to spend a lot of time to confirm the accuracy of the analysis result and the reliable range.

FIG. 40 shows the relationship between the lot flow and the abnormal manufacturing apparatus. A white circle “◯” indicates a normal device 101, and a black circle “●” indicates an abnormal device 102. Arrows indicate lot flow. The analysis of the difference between the devices in the LSI manufacturing data extracts, from the used device data for each process of each lot, which manufacturing device is used in which manufacturing process the yield is most affected.

FIG. 41 shows the yield distribution (box whisker diagram) by device in a certain process according to the prior art. The yield value of the lot is displayed in a box-and-whisker diagram for each device used in each manufacturing process, and each process is checked to identify the process with the most remarkable difference and the device.

However, this method requires a large number of steps at the present when the number of steps is several hundreds, and it is difficult to make a judgment when a difference is not apparent or when conditions are complicatedly entangled. A data mining method by regression tree analysis is effective for dealing with these, and the data is divided into a device group in which the value of the objective variable is high and a device group in which it is low.
As shown in FIG. 42, when the device used for each lot is fixed and the lot is flown, the abnormal device 102 indicated by a black circle "●"
May not be uniquely identified. That is, when the independence between the explanatory variables is low, the one in which the significant difference due to the two divisions of the set is large is not always “the true significant difference”.

The above is the confounding in the apparatus used in each step of semiconductor manufacturing, but the same applies to the confounding of the set divided into two as the result of the regression tree analysis. That is,
The same can be said for a set of a device group having a high yield and a device group having a low yield for each process. The same applies to the confounding of the two divided sets even when the explanatory variable is a continuous value.

The present invention has been made in view of the above problems, and edits original data to extract data distribution characteristic amounts such as various statistical values, and objectively recognizes and utilizes them. Therefore, it is an object of the present invention to provide a data analysis method for automatically extracting factors such as defects. Further, it can be included in the object of the present invention to provide a data analysis method and a data analysis device capable of clarifying the degree of confounding among a plurality of explanatory variables.

[0017]

In order to achieve the above object, the present invention is to automatically and quantitatively evaluate and extract various data distribution characteristics existing in an original data group accumulated in a computer system. Then, the extracted feature quantities are sequentially selected and analyzed to automatically and quantitatively evaluate and extract the factors causing the feature quantities. According to the present invention, a lot of information such as the tendency of data distribution, characteristic patterns, and relationships between data is extracted from the original data. Gender and significant differences are efficiently and quantitatively extracted based on scientific evidence.

Further, in order to clarify the degree of confounding among a plurality of explanatory variables, a step of preparing data results of explanatory variables and objective variables, and a degree of confounding between a plurality of explanatory variables based on the data result and / or Alternatively, there is provided a data analysis method including a step of calculating the degree of independence and a step of performing data mining using the degree of confounding and / or the degree of independence. By calculating the degree of confounding and / or the degree of independence between a plurality of explanatory variables, the degree of confounding of explanatory variables can be clearly understood. If regression tree analysis is performed based on this, it becomes possible to quantitatively evaluate the degree of confounding of explanatory variables based on the result of bisection of the set of regression tree analysis, and there is a large difference in the first branch in the regression tree. It is possible to clarify the explanatory variables that should be noted and are confounded with the explanatory variables that are

[0019]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiment 1 of the present invention will be described below.
2 will be described in detail with reference to the drawings.

(Embodiment 1) FIG. 1 is a diagram showing an example of a hardware configuration of a computer system used for implementing a data analysis method according to Embodiment 1 of the present invention.
As shown in FIG. 1, this computer system comprises an input device 1, a central processing unit 2, an output device 3 and a storage device 4.

FIG. 2 is a block diagram showing an example of the functional configuration of the data analysis apparatus realized by the computer system having the configuration shown in FIG. As shown in FIG. 2, this data analysis device has an original data group 42 including a database 41 containing a plurality of original data. The database 41 is constructed in the storage device 4 of the computer system shown in FIG.

Further, the data analysis device includes means 21 for quantitatively evaluating and extracting one or more data distribution features existing in the original data group 42, and analyzing from the extracted one or more data distribution feature quantities. A means 22 for selecting a target feature amount, data mining by a regression tree analysis method using the data distribution feature amount selected as an analysis target as a target variable, and a rule file 24 for features and regularities hidden in the data distribution is extracted. Means 23, and an analysis tool group 27 such as a statistical analysis component 25 for analyzing distribution characteristics of original data using the extracted rule file 24 and a chart creation component 26.

The above-mentioned means 21, 22, 23 and analysis tool group 27 are realized by the central processing unit 2 executing programs for performing the respective processing. The extracted rule file 24 is stored in the storage device 4 and also output device 3 such as a display device or a printing device.
Is output by. The decision 5 is made based on the analysis result by the analysis tool group 27.

The means 23 for extracting the above-mentioned rule file 24 is the original data in the original data group 42, the data distribution characteristics extracted by the means 21 for extracting the data distribution characteristics, or the analysis by the analysis tool group 27. Data mining is also performed on the results. The analysis tool group 27 also analyzes the original data in the original data group 42, the data distribution feature extracted by the means 21 for extracting the data distribution feature, or the output result of the analysis tool group 27. It has become. In addition, the analysis tool group 27
The analysis result by is fed back to the means 22 for selecting the data distribution characteristic amount to be analyzed and the original data group 42. The output of the means 21 for extracting the data distribution feature is fed back to the original data group 42.

FIG. 3 shows an example in which the data distribution characteristic amount extracted by the means 21 for extracting the data distribution characteristic shown in FIG. 2 is output in CSV format. Since each feature amount is calculated independently for each record, it is treated independently. For example, as shown in FIG. 3, since each feature amount is automatically output in the CSV format, it is possible to efficiently and significantly analyze each feature amount. Here, the data that is the feature amount is not only the original data value and its average value,
It may be the maximum value, the minimum value, the range, or the standard deviation value of the original data. Further, the periodicity of the data, the degree of similarity with the specific model, or the like may be used as the feature amount data.

Here, various features can be extracted according to the structure of the target data group, but the process of extracting what kind of feature amount according to the purpose is incorporated in the program in advance, or It is also possible to prepare a file that defines the feature quantity to be extracted and read the file. Each feature amount is not defined as a discrete value, but is defined as a continuous value indicating how strong the feature is. Therefore, since there is no loss of information due to the discrete value conversion as in the conventional case, a better analysis result is expected.

As an example of the data distribution feature amount, the feature of the intra-lot data distribution in the yield analysis of semiconductor data will be described. FIG. 4 is a list showing information focusing on the variation of the wafer attribute value. Here, the independent variable is the wafer number, and the dependent variable is the original data such as yield, category yield or various measured values.

Although not particularly limited, in the example shown in FIG.
(1) Center of overall data distribution, (2) Data variation, (3) Correlation of data with wafer number, (4)
Y-axis intercept at the time of first-order approximation, (5) slope of data with respect to wafer number, (6) strength of cycle 2 (sheets), (7) strength of cycle 3 (sheets), (8) most within lot Strong cycle, (9)
First half wafer-second half wafer average difference, (10) first half wafer-second half wafer variation difference, (11) first half wafer-second half wafer correlation difference, (12) first half wafer-second half wafer primary approximation y-axis Difference of intercepts, (1
3) Difference in inclination between first half wafer and second half wafer, (14)
Sixteen characteristic items are defined, namely, the strength of the second half lot cycle 2 (sheets), (15) the strength of the second half lot cycle 3 (sheets), and (16) the strongest cycle of the second half lot. The feature amount of each feature item is obtained in lot units.

The 16 characteristic items defined here will be briefly described. The feature amount of (1) is an average value of yields and various measured values of all wafers in the same lot.
The feature amount of (2) is a standard deviation value of yields of various wafers in the same lot and various measured values. The feature quantity of (3) is the correlation coefficient between the wafer number of wafers in the same lot and the yield and various measured values. The method of calculating this correlation coefficient is based on the target and purpose of analysis beforehand. It has been decided. In the feature quantity of (4), the wafer number of wafers in the same lot is x, and the yield and various measured values are y.
Is the value of the y-axis intercept when the relation between x and y is approximated to the linear expression y = b · x + a.

The characteristic amount of (5) is obtained by approximating the relation between x and y to a linear expression y = b · x + a, where x is the wafer number of wafers in the same lot, and y is the yield and various measured values. It is a mother regression coefficient. The feature amount of (6) is the distribution of yields and various measured values of all wafers in the same lot, and the wafer group of wafer numbers 1, 3, 5, ... Or the wafer numbers of 2, 4, 6 ,. .. is the ratio to the yield of wafers and the variance of various measured values. The feature amount of (7) is the variance of yields and various measured values of all wafers in the same lot,
The yield of wafer groups having wafer numbers 1, 4, 7, ..., Wafer groups having wafer numbers 2, 5, 8, ... Or the yield of wafer groups having wafer numbers 3, 6, 9 ,. It is the ratio to the variance of various measured values. The feature amount of (8) is the dispersion ratio of the cycles 2 (sheets) and 3 (sheets) obtained as described in (6) and (7) above for the wafers in the same lot, and the cycle 4 (which is similarly obtained. Of the dispersion ratios such as (sheet), 5 (sheet), ..., It is the value of the cycle in which the dispersion ratio is maximum.

The characteristic amount of (9) is to divide all wafers (for example, 50 wafers) in the same lot into the first half (for example, 25 wafers) and the latter half (for example, 25 wafers), and obtain the yield of the first half wafer group and various measured values. It is the difference between the average value and the average value of the yield of the latter half wafer group and various measured values. The reason why the device is divided into the first half and the second half is that the device history is different in the semiconductor manufacturing process. The feature amount of (10) is the difference between the yield of the first half wafer group and the standard deviation value of various measured values and the standard deviation value of the yield of the latter half wafer group and various measured values. The feature amount of (11) is the difference between the correlation coefficient of the first half wafer group and the correlation coefficient of the second half wafer group. The feature amount of (12) is the difference between the value of the first-order approximate y-axis intercept of the first half wafer group and the value of the first-order approximate y-axis intercept of the latter half wafer group.

The feature amount (13) is the difference between the mother regression coefficient of the first half wafer group and the mother regression coefficient of the second half wafer group.
The feature amount of (14) is the dispersion ratio for the period 2 similar to (6) above for the latter half lot group. The feature amount of (15) is the variance ratio for the period 3 for the latter half lot group as in (7) above. The feature quantity of (16) is the value of the cycle in which the variance ratio becomes maximum for the latter half lot group, as in (8) above. Regarding the first half lot group, the above (14)-
The strength of cycle 2 (sheets), the strength of cycle 3 (sheets), and the strongest cycle may be defined as in (16). Further, not only the characteristic items illustrated here, but various characteristic items are defined according to the analysis target and purpose.

When the characteristic items as described above are defined and analyzed, a significant difference depending on the apparatus used cannot be extracted only by analyzing the original data value such as the yield value or the average value as in the conventional case. However, it may be possible to extract a significant difference depending on the device used. For example, FIG. 5 shows an example in which variations in the yield value of wafers within a lot (corresponding to (2) above) are noted for a plurality of lots.

In the example shown in FIG.
Lot group 6 using Unit 2, Unit 24 or Unit 25
The average value of wafer yields and the overall distribution (variation between lots) of the lot group 7 (on the left side of the alternate long and short dashed line in FIG. 5) and the lot group 7 using the No. 28 machine (on the right side of the alternate long and short dashed line in FIG. 5) are almost the same. Therefore, no clear significant difference is observed even when the analysis is performed using the wafer yield value and the average value thereof. On the other hand, when attention is paid to the characteristic amount of the data distribution, that is, the variation of the wafer yield value within each lot, a clear significant difference is recognized between the two lot groups 6 and 7. The item of interest is not limited to the item (2), but may be the item (1), any of the items (3) to (16), or any other item.

As described above, since each data distribution characteristic amount exists as the attribute value of each lot, the means 22 for selecting the data distribution characteristic amount sequentially selects each data distribution characteristic amount as the objective variable. Then, the means 23 for performing the data mining and extracting the rule file 24 performs the regression tree analysis using the respective data distribution feature quantities as the objective variables in order. As a result, it is possible to determine the cause of the data distribution characteristic amount, and it is possible to extract more defect factors than the conventional analysis method. At that time, since the process of sequentially selecting the data distribution feature amount and the regression tree analysis process are automatically executed according to the program, the engineer does not have to consider which data distribution feature amount is selected as the objective variable. Analysis can be performed efficiently. This is especially effective when it is unknown what to analyze.

Further, even when a plurality of characteristic patterns are seen, as in the case where both the strength of cycle 2 (sheets) and the strength of cycle 3 (sheets) exist in the same record, both characteristics are evaluated. Since it is possible to do so, it is possible to obtain an analysis result that reflects the actual situation by eliminating the loss of information amount.

Next, the flow of the data analysis method according to the first embodiment of the present invention will be described. FIG. 6 is a flowchart showing the outline of an example of the data analysis method according to the present invention. As shown in FIG. 6, when this data analysis method is started, first, data to be analyzed, such as yield value and various measured values, is selected and extracted from the original data group 42 (step S1). ). Subsequently, a process of extracting one or more data distribution features is performed on the extracted data (step S2).

Then, the data distribution feature quantity to be analyzed is selected, and data mining such as regression tree analysis is performed using the selected data distribution feature quantity as an objective variable (step S3). When the regression tree analysis is completed for all the data distribution features extracted in step S2 (step S4), the analysis result is output, and the engineer confirms the result (step S4).
5). Then, the engineer makes a decision based on the analysis result (step S6).

Next, in order to further clarify the features of the present invention, a data analysis method using the data distribution feature will be described with a specific example. Generally, even in a group of wafers in the same lot, if the wafer numbers are different, the yield value and the electrical characteristic value of each wafer are different, and these values show various fluctuation patterns. The yield value and the electrical characteristic value are stored for each wafer. Therefore, the first embodiment
In this way, it is possible to analyze the variation pattern such as the yield value with respect to the wafer number as the data distribution characteristic over a plurality of lots. Here, the test substitute Nch transistor threshold voltage VT, which is an important electrical characteristic that greatly affects the product performance, is used.
An example of performing multifaceted analysis on _N2 (hereinafter, simply referred to as VT_N2) will be described. In addition, it is assumed that the history of used devices in each manufacturing process is effective for the yield.

FIG. 7 is a characteristic diagram showing the relationship between the yield and VT_N2. From the figure, the yield and VT_N2 seem to be irrelevant at first glance. Further, FIG. 8 is a histogram of VT_N2 data obtained from all wafers, and FIG. 9 is a box whiskers diagram showing all VT_N2 data for each wafer number. It is difficult to extract statistically significant differences from the results shown in these figures.

FIG. 10 is a diagram showing an example of the result of a regression tree analysis in which the objective variable is the average value of VT_N2 in each lot and the explanatory variable is the device name used in each step. It is a figure which shows the example of the statistical value list for evaluation showing the reliability information of this regression tree analysis. According to the result of this regression tree analysis, as shown in FIG. 10, the most significant value with respect to the variation of VT_N2 is No. 11 or No. 13 as the second wiring_device, or 1
It means whether Unit 2, Unit 14, Unit 17, or Unit 18 was used. All VT_N2 data on the second wiring_
FIG. 12 shows a box whiskers diagram displayed for each device name used, but no significant difference is seen in the diagram. The evaluation statistical value list is output together with the regression tree diagram, which will be described later.

FIG. 13 is a diagram showing an example of a result of performing a regression tree analysis with the objective variable being the value of VT_N2 of each wafer and the explanatory variable being the device name used in each step. FIG. It is a figure which shows the example of the statistical value list for evaluation with respect to a regression tree analysis. According to this regression tree analysis result,
As shown in FIG. 13, what is most significant with respect to the fluctuation of VT_N2 is whether the No. 11 machine, or No. 12 or No. 13 machine was used as the 2CON process_apparatus. FIG. 15 is a box whiskers diagram showing all VT_N2 data for each device name used in the 2CON process_device, but no significant difference is seen in this diagram either.

On the other hand, by extracting and analyzing the data distribution characteristics of VT_N2 as described below, the cause of failure can be clarified. FIG. 16 is a chart showing in CSV format an example of a file that defines each feature amount of VT_N2 for each lot. This file is output by the means 21 for extracting the data distribution characteristics of the device shown in FIG.

FIG. 17 is a histogram showing the characteristic quantities of various distributions within the lot of VT_N2 based on the CSV format data shown in FIG. Here, regarding VT_N2, of the 16 characteristic items (1) to (16) described with reference to FIG. 4, (1) average value (VT_N2_av
e), (2) standard deviation value (VT_N2_s), (3) correlation coefficient for wafer number (VT_N2_r),
(4) y-axis intercept (VT_N2_a) of the first-order approximation formula,
(5) mother regression coefficient (VT_N2_b), (6) periodicity of wafer number interval 2 (VT_N2_2), (7) periodicity of wafer number interval 3 (VT_N2_3), (9)
Difference between the average value of the first half wafer and the latter half wafer (VT_N2
_Ave_d), (10) difference in standard deviation between the first half wafer and the second half wafer (VT_N2_s_d), (11) difference in correlation coefficient between the first half wafer and second half wafer (VT_N2)
_R_d), (12) difference of the y-axis intercept of the first-order approximate expression of the first half wafer and the second half wafer (VT_N2_a_d), (1
3) Difference of the mother regression coefficient between the first half wafer and the second half wafer (V
T_N2_b_d) are extracted.

From FIG. 17, it can be seen that all the feature quantities are considerably varied. Therefore, if a regression tree analysis is performed using each feature amount as an objective variable, a factor causing a significant difference between the feature amounts, that is, a failure factor or the like can be analyzed.

In order to efficiently obtain the analysis result by using the data distribution feature as the analysis target, as the input data of the regression tree analysis, as shown in FIG. 18, the name of the device used in each process and the extracted data are extracted for each lot. A file is created in which the specified features are defined. This file is a rule file 24 that defines the device name used in each process and the lot yield as input data when analyzing the factor of yield variation by regression tree analysis.
(See FIG. 19) and the file shown in FIG. 16 are combined for the same lot number.

In FIG. 20, based on the file shown in FIG. 18, the above-mentioned (2) standard deviation value (VT_N2_s) is used as an objective variable, and the device name used in each step is used as an explanatory variable.
It is a regression tree diagram which shows the regression tree analysis result performed in order to extract the factor of the variation which has arisen in the lot of VT_N2. Further, FIG. 21 is a diagram showing an example of an evaluation statistical value list for this regression tree analysis.

According to the regression tree diagram shown in FIG. 20, VT_N
It is the Field_Ox process_apparatus that is most significant with respect to the fluctuation of the standard deviation value (VT_N2_s) of 2.
PM1 or PM3 was used, or P
It is whether or not M2 machine was used. This is because the S_ratio and t-value, etc. of the evaluation statistical value list are the respective values (S-ratio = 0.3767, t = 3.081) of the first Field_Ox process_apparatus that appears first and the second and subsequent values. In comparison with the respective values (S ratio> 0.43, t <2.2) of the second wiring_device and the DRY process_device appearing in Fig. 2, there is a clear significant difference. It is judged to be high.

In order to confirm this, FIG.
The distribution of the value of VT_N2 for each device used in the d_Ox step is shown in a box-and-whisker diagram. In FIG. 22, PM No. 1 or P
A clear significant difference is confirmed between the M3 machine and the PM2 machine. In other words, the effectiveness of the method of the present invention of performing analysis using the distribution characteristics of the original data was confirmed. The evaluation statistical value list, S ratio, and t value will be described later.

FIG. 23 is a histogram showing the distribution of VT_N2 of all wafers using PM No. 1 or PM No. 3 in the Field_Ox step, which is a problematic step, as a result of the regression tree analysis shown in FIGS. 21 and 22. Figure 24 ~
FIG. 26 is a histogram showing the distribution of VT_N2 of wafers for different one lot using PM1 machine or PM3 machine in the Field_Ox process. 27 is a histogram showing the distribution of VT_N2 of all wafers using PM2 machine in the Field_Ox process, and FIGS. 28 to 30 are respectively Field_O.
It is a histogram which shows distribution of VT_N2 of the wafer for each separate 1 lot which used PM2 machine in the x process.

As shown in FIGS. 23 and 27, PM1
VT_N of all wafers using No. 3 or PM No. 3
2 and the average value of VT_N2 (μ = 0.730) of all wafers using PM2.
It is almost the same as 2). Therefore, it is difficult to extract a significant difference even if an analysis is performed using an average value as in the conventional case.

However, the standard deviation value of VT_N2 of all wafers using PM No. 1 or PM No. 3 (σ = 0.
0835) and VT of all wafers using PM2
A significant difference is clearly seen when compared with the standard deviation value of _N2 (σ = 0.2351). Therefore, as in the first embodiment, by focusing on the data distribution characteristics such as the variation of the original data, a significant difference that could not be extracted by using only the original data as the analysis target is newly extracted as a failure factor. Is possible.

Based on the above analysis results, the PM2 is actually
As a result of a detailed investigation of Unit No. 1, it was found that the temperature distribution difference in the furnace was larger than that of Units PM1 and PM3. Furthermore, it was found that it was due to thermocouple deterioration, and the periodic inspection method was optimized. By the way, the result of performing the regression tree analysis with the lot yield as the objective variable and the device name used in each process as the explanatory variable did not reveal that PM2 was a factor of yield reduction. That is, the low-yield factor, which did not clearly appear in the yield value, was clarified by the method of the present invention in which the factor causing a significant difference in the standard deviation of the electrical characteristic values within the lot is analyzed. The first embodiment
Automatically edits the accumulated data, performs regression tree analysis, and quantitatively evaluates the results using a unique method.

FIG. 31 is based on the file shown in FIG. 18, and has the periodicity (V) of the interval 2 of (6) wafer number described above.
FIG. 9 is a regression tree diagram showing the result of performing a regression tree analysis with T_N2_2) as the objective variable and the device name used in each step as the explanatory variable. Further, FIG. 32 is a diagram showing an example of an evaluation statistical value list for this regression tree analysis. According to the regression tree diagram shown in FIG. 31, it is most significant for the intra-lot variation of VT_N2_2 to have a periodicity of 2 whether the F7 machine is used as the F diffusion process_apparatus, or F5. It means whether the No. 6, F6, F8 or F9 was used. It can be seen that the use of F7 machine shows a periodicity of 2 which is about 50% stronger.

To confirm this, in FIG. 33, the periodicity value of 2 (VT_N2_) is set for each device used in the F diffusion process.
The distribution of 2) is shown by a box-and-whisker diagram. In FIG. 33, a clear significant difference is confirmed between the F7 machine and the F5 machine, the F6 machine, the F8 machine, or the F9 machine. Note that all VT_N
It is not possible to see the periodicity of 2 from the box-and-whisker diagram of FIG. 9 in which 2 data is displayed for each wafer number. In this example as well, the effectiveness of the method of the present invention of performing analysis using the distribution characteristics of the original data was confirmed.

FIGS. 34 to 36 show results of the regression tree analysis shown in FIGS. 31 and 32, showing variation in VT_N2 within a lot for each one lot of wafers using the F7 machine in the F diffusion process which is a problematic process. It is a histogram showing. 37 to 39 are histograms showing the intra-lot variation of VT_N2 with respect to each wafer for one lot using F5 machine, F6 machine, F8 machine, or F9 machine in the F diffusion process. From the above analysis results, VT_N
The factor of the intra-lot variation was extracted, and the device of the F diffusion process, which is the device in which the wafers are used alternately, has been noticed, and it has been found that particles are often generated in one of the two chambers.

In the first embodiment, in the regression tree analysis, the feature variables extracted with the same explanatory variables are sequentially selected as the objective variables, and the regression tree analysis is automatically performed. Factors that influence each are extracted. In particular, when it is not clear what to analyze, all possible feature quantities are extracted and a regression tree analysis is performed using them as objective variables. As a result, since various analysis results are obtained as described above, the item considered to have the largest significant difference among them is taken as the candidate for the measure item for improving the yield. As described above, many significant differences that cannot be easily extracted by the conventional analysis method are efficiently extracted.

The regression tree analysis and evaluation statistical value list will be described below. First, the regression tree analysis will be briefly described. Regression tree analysis targets a set of records consisting of explanatory variables indicating a plurality of attributes and objective variables affected by the explanatory variables, and discriminates the attributes and attribute values that most affect the objective variables. From the means 23 (regression tree analysis engine) for performing the data mining and extracting the rule file 24, the rules indicating the characteristics and regularity of the data are output.

The process of regression tree analysis is performed by using each explanatory variable (attribute).
It is realized by repeatedly dividing the set into two based on the parameter value (attribute value) of. At the time of dividing the set, when the sum of squares of the objective variable before the division is S0 and the sum of squares of the respective objective variables of the two sets after the division is S1 and S2, ΔS shown in the equation (1) becomes maximum. In this way, the explanatory variable of the record to be divided and its parameter value are obtained.

[0060] ΔS = S0− (S1 + S2) (1)

The explanatory variables and their parameter values obtained here correspond to branch points in the regression tree. After that, similar processing is repeated for the divided sets, and the influence of the explanatory variable on the objective variable is examined. The above is the method of regression tree analysis that is generally well known, but in order to understand the clarity of the set partitioning in more detail, the following parameters (a) to (S d) is also used as a quantitative evaluation of the regression tree analysis results. These parameters are output as a list of statistical values for evaluation.

(A) S ratio: This is a reduction rate of the sum of squares due to the set division, and is a parameter indicating how much the sum of squares is reduced due to the set division. The smaller this value is, the greater the effect of the set partitioning is, and the set partitioning is clearly performed, so that the significant difference is large.

[0063] S ratio = ((S1 + S2) / 2) / S0 (2)

(B) t value: The regression tree analysis engine divides the set into two, and the average (/
X1, / X2) is a value for testing the difference. here,
"/" Indicates an overline. The statistical t-test serves as a standard for showing a significant difference in the average value of the objective variables in the divided sets.
If the degrees of freedom, that is, the number of data is the same, the larger t is, the more clearly the set is divided, and the significant difference is large.

At this time, when there is no significant difference in the variance of the divided sets, the t value is obtained by the following formula (3), and when there is a significant difference in the variance of the divided sets, the formula (4) is used. Find the t value. Here, N1 and N2 are the numbers of elements of the divided set 1 and set 2, respectively. Also, / X
1 and / X2 are the average of each set after division. S1 and S2 are sums of squares of the objective variables of each set after division.

[0066]

[Equation 1]

[0067]

[Equation 2]

(C) Difference in average values of objective variables of divided sets: The larger this value, the larger the significant difference.

(D) Number of data in each divided set: The smaller the difference between the two, the smaller the influence of the abnormal value (noise).

According to the above-described first embodiment, not only the original data and the average value thereof as in the prior art, but also various data distribution characteristics existing in the original data group such as variations of the original data and variation patterns within the lot. By extracting each feature quantity as an objective variable and performing an analysis, the factors that cause each feature quantity are automatically and quantitatively evaluated and extracted. Information can be extracted. Therefore, it is possible to objectively and efficiently and quantitatively extract the relation and the significant difference which are conventionally difficult to discriminate because they are buried in various data.

Further, in the first embodiment, since a series of procedures from the extraction of the characteristic amount to the extraction of the factor thereof are automatically performed, it is possible to automatically change the semiconductor manufacturing line etc. by setting a predetermined setting. It will be possible to constantly monitor the situation and its factors.

The present invention is not limited to the above-described first embodiment, but has a wide range of application. For example, when starting a new product, there are many causes of deterioration and lots with poor yield frequently occur.In addition to investigating the cause process using the original data and its average value, By investigating the cause process from the data distribution feature, it is possible to find the hidden cause and narrow down the cause.

(Second Embodiment) FIG. 56 is a diagram showing an example of a functional configuration of a data analysis device to which data mining according to the second embodiment of the present invention is introduced. The data mining unit 1703 creates a rule file 1704 by extracting features and regularity hidden in the data based on the individual original data extracted from each database 1702 in the original data group 1701. The analysis tool group 1705 has a statistical analysis component 1706, a chart creation component 1707, etc., and analyzes individual original data extracted from the database 1702 based on the rule file 1704.

The analysis result is fed back to the analysis tool group 1705 and the data mining unit 1703. The data mining unit 1703 includes the analysis tool group 17
Data mining is performed based on the analysis result of 05 and the original data group 1701. Analysis tool group 1705
Analyzes based on the rule file 1704, individual original data extracted from the database 1702, and its own analysis result. Decision making (part) 1708
Makes a decision based on the analysis result of the analysis tool group 1705.

When data mining is applied in the yield data analysis, a measure for improving the yield is determined based on the data mining result, whether or not the measure should be implemented is determined, and the effect of the measure is predicted. Will be. For that purpose, quantitative evaluation and accuracy of data mining results are required.

Among the discriminant tree analyzes that are one method of data mining, the regression tree analysis is particularly effective. One of the advantages of the regression tree analysis is that the result is output as an easy-to-understand rule, which is expressed in a general language or a database language such as SQL language. Therefore, it is possible to make effective use of the reliability and precision of these results, and to make effective decisions and take actions (that is, countermeasures) based on the results.

FIG. 43 shows a format of an example of data which is an input of the regression tree analysis. A record is a wafer number unit, and each record has a device 411 used in each manufacturing process, electrical characteristic data 412, and a wafer yield 413. The explanatory variable 401 is the used device 411 and the electrical characteristic data 4
It is 12 mag. The objective variable 402 is the yield 413. For example, the device used 41 has an effect on yield.
1 and electrical characteristic data 412. 44 and 45 show a regression tree diagram which is the result of the regression tree analysis based on this data and a list of statistical values for evaluation.

FIG. 44 is a regression tree diagram showing the result of the regression tree analysis. The root node n0 is divided into two nodes n1 and n2. The node n1 is divided into two nodes n3 and n4. The node n2 is divided into two nodes n5 and n6. The node n6 is divided into two nodes n7 and n8.

FIG. 45 shows evaluation statistical values of explanatory variables in the first two-division. For example, the average value Ave of the objective variables of all sets is 75, the standard deviation s is 12, and the number of data N is 1000. The lists 601 to 604 are, from the left, the ranks by significant difference, the S ratio, the t value, the difference in the average value of the objective variables of the divided sets, the number of data of each divided set, and the attribute name of the divided set ( Explanatory variables), the attribute value (parameter value) of the two divided sets, and the magnitude relation between the objective variables. The lists 601 to 604 are candidates for grouping according to the value of ΔS shown in the expression (1) of the attribute value (explanatory variable) to be divided, and are arranged in descending order of significant difference (ΔS). In FIG. 44, the node n0 is divided into nodes n1 and n2 based on the first candidate 601.

The set n0 of all wafers in FIG.
When the two sets n1 and n2 are divided on the basis of the evaluation value of ΔS, the yield is most affected by using either AM1 or AM2 in the process A, and the latter yields better. . This regression tree diagram can be obtained by repeating the same set division for the divided sets. For the wafer group using AM2 in the process A and CM2 in the process C, the state in which the electrical characteristic data RSP is 90 or less is most effective (high yield).

FIG. 46 is equivalent to FIG. 44 and shows the correlation between the yield of divided wafer sets, the equipment used in a specific process and the electrical characteristic data. The higher the explanatory variable that appears in the regression tree diagram of FIG. 44, the greater the influence on the objective variable.
The average yield of all wafers is 74.8%, but if we divide it into several sets in relation to the equipment used and electrical characteristic data, regression tree analysis automatically shows that there are such characteristics and regularity. To be used as a clue for yield analysis.

In the regression tree diagram of FIG. 44, since the upper two layers are all due to the difference in the used device, it is the difference in the used device that has a great influence on the yield in the analysis using all wafers even if the compound condition is included. . The electrical characteristics data appear to be less effective. However, it can be read from FIGS. 44 and 46 that RSP is most effective for the yield in the wafer group using AM2 in step A and CM2 in step C.

Next, an example of calculating the 2-division confounding degree and 2-division independence will be described. In the regression tree analysis, the degree of confounding of each set division state (confounding state, degree of independence) performed to find the most significant explanatory variable for the objective variable
Statistically grasping and clarifying the other explanatory variables confounding the explanatory variables with significant difference. With reference to FIG. 47, a method of calculating the 2-division confounding degree and 2-division independence will be described.

First, of the explanatory variables, the one for which the degree of confounding is to be evaluated is set as the reference explanatory variable 801.

Secondly, each record constitutes a table having a data value of "L" or "H" for each explanatory variable. Here, H belongs to a set in which the objective variable has a high value when the set is divided into two in the regression tree analysis, and L belongs to a set in which the target variable has a low value when the set is divided into two in the regression tree analysis. When the set is divided into two, L and H are determined for each explanatory variable of all records.

Thirdly, based on the reference explanatory variable 801, as the evaluation value of the degree of coincidence between L and H of each comparative explanatory variable 802, L,
Let Na be the number of records in which H matches and N be the total number of records, and define the 2-division confounding degree DEP as in Expression (5).
The range of the 2-division entanglement degree DEP is −1 to 1, and is 1 when completely entangled, 0 when completely entangled, and −1 when reverse entangled.

[0087] DEP = (2 × Na / N) −1 (5)

Further, the two-division independence degree IND is defined as in the equation (6) based on the two-division confounding degree DEP. The range of the 2-division independence degree IND is 0 to 1, and is 1 if completely independent, and 0 if not completely independent.

IND = 1- | DEP | (6)

Fourthly, the above-mentioned two-divided confounding degree DEP and two-divided independence degree IND are obtained between one reference explanatory variable 801 and the other explanatory variables 802 and used as an evaluation scale between the explanatory variables. Which explanatory variable is used as the reference explanatory variable is arbitrary, but it is assumed from the usefulness that there is a significant difference from the objective variable in the regression tree analysis, especially in the set partition at the highest hierarchy. Is effective.

Fifth, by obtaining the above-mentioned two-divided confounding degree DEP and two-divided independence degree IND, how much each comparative explanatory variable 802 belongs to each set of L and H and what is the reference explanatory variable 801? Whether there is a difference can be evaluated quantitatively.

By obtaining the degree of 2-part confounding and / or the degree of 2-part independence, it becomes possible to quantitatively evaluate the degree of confounding of explanatory variables based on the result of the set bipartition of the regression tree analysis, and in combination with the regression tree analysis. , It is possible to automatically extract another explanatory variable confounding with an explanatory variable having a large significant difference obtained by regression tree analysis.

The 2-division confounding degree can be evaluated with respect to any of the explanatory variables targeted in the regression tree analysis, but from the viewpoint of its effectiveness, the explanatory variable (= reference explanation) that is ranked higher than the first division candidate in FIG. Variables, listed in the statistical value list for evaluation)
And how much any other explanatory variables are confounded statistically, and the explanatory variables that are confounded with respect to the explanatory variables having a large significant difference are extracted. Standard explanatory variable 80
The explanatory variable for which the degree of confounding with 1 is to be analyzed is the comparative explanatory variable 802, and both are selected from the evaluation statistical value list of FIG. A calculation example of the 2-division confounding degree and 2-division independence degree will be described with reference to FIG. 47.

In FIG. 47, the horizontal axis represents the wafer number 803, the comparative explanatory variable 802, the reference explanatory variable 801, and the yield 804.
And the vertical axis represents the high yield group 81 of the reference explanatory variables.
1, a low-yield group 812 as a reference explanatory variable, a calculation formula 813 for 2-division confounding degree, a 2-division confounding degree 814, and 2-division independence degree 815 are shown.

From the upper candidate items (list of statistical values for evaluation) shown in FIG.
Decide as 1. In FIG. 47, ST3 is the reference explanatory variable 8
01. The other explanatory variables are referred to as comparative explanatory variables 802. In FIG. 47, ST1, ST2, and WET2 are comparative explanatory variables 802. Each comparative explanatory variable 802 is compared with the reference explanatory variable 801. Explanatory variables ST1, ST
2, ST3, WET2, low yield group "L"
Is indicated by a hatch, and “H” in the high yield group is indicated by no hatch.

ST3, which is the reference explanation variable 801, has a high yield group 811 of the reference explanation variables depending on its attribute value.
Can be divided into the low yield group 812 of the reference explanatory variables. High yield group 811 of the standard explanatory variable is 1
Low yield group 8 which is a set of 0 and has a standard explanatory variable
12 is also a set of 10.

Next, Na is counted by counting how much the lots of the high yield group and the low yield group, which are divided into two of the explanatory variables, match the same group of the reference explanatory variables. For example, ST which is the comparative explanatory variable 802
In No. 1, 10 high yield groups are included in the high yield group 811 of the reference explanatory variable, and 2 low yield groups are included in the low yield group of the reference explanatory variable 812.
It is an individual. That is, the number Na = 10 + 2 = 12 of the comparative explanatory variable ST1 and the standard explanatory variable ST3 that belong to the same group.

A formula 813 for substituting the above Na into the formula (5) is shown in the calculation formula 813 for the degree of entanglement. Here, the number of data N is 2
It is 0. The result of this calculation is shown in the degree of halving 814.
The value obtained by the equation (6) is shown as the 2-division independence degree 815. The two-division confounding degree 814 and the two-division independence degree 815 are
It is shown under each column in FIG. 47.

There are the following three basic methods of utilizing the two-division confounding degree and the two-division independence degree. Explanatory variables that were difficult to discriminate in the past can be obtained as quantitative information as follows.

(1) Confirming the range of significant explanatory variables: Grasping the candidates confounding with highly significant candidates and judging them as significant explanatory variables. There is no particular criterion for the degree of confounding, but it can be judged by comparing with the values of other explanatory variables. Also,
When a candidate that does not need to be considered as a technical target comes to the top, the candidate confounding this candidate can be clarified. Furthermore, meaningless candidates can be deleted and analyzed again for confirmation.

(2) Confirmation of highly independent candidate and its application: If all the candidates are confirmed to have independence with other candidates, and if there is a candidate with sufficiently high independence with other candidates, then this candidate is used. It becomes clear that the yield difference exists independently of other candidates. Furthermore, the same discriminant tree analysis is performed for each of the divided groups of the candidates, and the results are compared, and it is found that the reliability of the analysis result is high when the same analysis result is obtained. On the other hand, if the analysis results are different, there is an explanatory variable that influences the yield depending on the composite condition with the candidate considered to be independent, or it depends on peculiar data (due to the small number of data etc.) Conceivable.

(3) Discriminant Tree Analysis on Entangled Candidates: When a candidate considered to be important is entangled with the first branch candidate, it does not easily appear in the lower branch of the first branch. At that time, the data is divided by other division groups having a high degree of independence to perform the discriminant tree analysis, and the discriminant tree analysis results under this divided group are compared. If the result is similar, the important candidate cannot be distinguished from the first candidate, but the analysis itself is considered to have high reliability. On the contrary, if a candidate that seems to be important appears and the result is different, this result should be taken into consideration.
It is considered necessary to further analyze the data so that it can be analyzed separately from the candidates.

Next, a regression tree analysis is performed using the apparatus history, the electrical characteristic value as an explanatory variable, and the wafer yield as an objective variable, and the top 12 candidates of the first branch of the regression tree analysis result are divided into two divisions and the degree of convolution is divided into two. A case of obtaining the independence will be described.

The regression tree diagram and the evaluation statistical value list obtained in the second embodiment are shown in FIGS. 48 and 49. Figure 4
In 8, the node n900 is divided into nodes n901 to n914. FIG. 49 shows evaluation statistical values of the top 12 explanatory variables at the time of the first two-division. As a result, twelve candidates 1001 to 1012 of the set branch are listed.

FIG. 50 shows the top first candidate 10 of FIG.
ST1 listed as 01 is a reference explanatory variable 1101.
Then, when the other explanatory variables of the evaluation statistical value list are set to the comparative explanatory variable 1102, the degree of convolution of two divisions 1111 and 2
The division independence 1112 is shown.

FIG. 51 shows the third candidate 1 of the set branch of FIG.
ST3 listed as 003 is the reference explanatory variable 120.
2 shows the 2-division confounding degree 1211 and the 2-division independence degree 1212 when the other explanatory variable of the evaluation statistical value list is set to the comparative explanatory variable 1202.

The degree of 2-part entanglement shown in FIG. 50 exceeds 0.75 in ST2, ST4, ST5, ST6 and ST1.
0 and WET2, which do not appear in the regression tree diagram of FIG. 48, but may be a factor that greatly affects the yield. On the contrary, FIG. 51 shows that ST3 has a high degree of 2-division independence.

In FIG. 51, ST3, which has a high degree of 2-division independence in FIG. 50, is used as a reference explanatory variable, and 2-division confounding degree 1211 and 2-division independence 1212 with the other 11 explanatory variables.
Is shown. ST3 indicates that the degree of independence with any of the other explanatory variables is high.

FIG. 52 and FIG. 53 show the 2-part confounding degree, the 2-part independence degree, and their average value of the top 12 explanatory variables which are considered to have a large significant difference in the regression tree analysis of FIG.
The relationship between explanatory variables can be grasped at a glance. The bottom column of FIG. 52 shows the average value 1301 of the 2-division confounding degree, and the bottom column of FIG. 53 shows the average value 1401 of the 2-division independence degree.

Next, it was found that the difference in the equipment used in ST3 was effective for the yield independently of other explanatory variables. Therefore, the wafer group by the equipment group in ST3 (the defective wafer group : S3M2 and S3M3 are used) and the wafer group (the good wafer group, S3M1 and S3M4 are used) by the device group in ST3 that is good is separately subjected to the regression tree analysis. The resulting regression tree diagrams are shown in FIGS. 54 and 55.

FIG. 54 is a regression tree diagram showing the results of the regression tree analysis based on the defective wafer group.
506. FIG. 55 is a regression tree diagram showing the result of the regression tree analysis using the good wafer group.
0 to n1606.

The first branch of the defective wafer group in FIG. 54 is shown in FIG.
8 is the same as that of all the wafer groups, and the number of defective wafers in the top hierarchy of the regression tree diagram of FIG. 48 is small (n = 39). This is one of the factors that make analysis difficult. In the good wafer group of FIG. 55, ST3
The factors that were difficult to see due to defective equipment in the process were newly found.

According to the second embodiment, the degree of confounding the explanatory variables can be more clearly grasped by using the degree of convolution of two divisions and the degree of independence of two divisions. It is possible to clarify the explanatory variables to be noted that are confounded with the problem explanatory variables that have a large significant difference in the first branch.

Furthermore, the regression tree analysis is applied again by applying grouping of explanatory variables having high independence to improve the accuracy (reliability) and analysis efficiency of the regression tree analysis, and further detailed analysis is possible. Become.

The above-described embodiment can be realized by the computer executing the program. Also, means for supplying the program to the computer, for example, a CD-RO recording the program.
A recording medium such as M or a transmission medium such as the Internet that transmits such a program can also be applied as an embodiment of the present invention. The above program, recording medium, and transmission medium are included in the scope of the present invention.

The above-mentioned embodiments are merely examples of the implementation of the present invention, and the technical scope of the present invention should not be limitedly interpreted by these. It is a thing. That is, the present invention can be implemented in various forms without departing from its technical idea or its main features.

(Supplementary Note 1) A step of editing the original data values to quantitatively evaluate and extract one or more data distribution feature quantities existing in the original data group, and a step of extracting the extracted data distribution feature quantities. A data analysis method comprising: a step of performing an analysis by selecting an arbitrary data distribution feature amount from the inside, and a step of making a decision based on the obtained analysis result.

(Supplementary Note 2) A step of editing the original data values to quantitatively evaluate and extract two or more data distribution feature quantities existing in the original data group, and the extracted individual data distribution feature quantities. A data analysis method comprising: a step of sequentially selecting an amount for analysis, and a step of making a decision based on the obtained analysis result.

(Supplementary Note 3) The above-mentioned data distribution feature quantity is represented by a continuous value representing the degree of the feature, Supplementary Note 1
Or the data analysis method described in 2.

(Supplementary Note 4) Regarding each record, each data distribution feature quantity is independent from each other
3. The data analysis method described in any one of 3.

(Supplementary note 5) The data analysis method according to any one of supplementary notes 1 to 4, wherein analysis is performed by data mining using the data distribution feature quantity as an objective variable.

(Supplementary Note 6) The individual data distribution feature quantities are saved in a file for each record, and from the file,
6. The data analysis method according to appendix 5, wherein the same data distribution feature amount is sequentially selected for some or all of the records as an objective variable and a regression tree analysis is performed.

(Supplementary Note 7) Each of the above-mentioned steps is automatically performed by executing a software set up so as to be performed sequentially in a computer system. Data analysis method.

(Supplementary Note 8) One of the data distribution feature quantities is that the order of arrangement of original data is x, the original data value is y, and the relationship between x and y is a linear expression y = b · x.
The data analysis method according to any one of appendices 1 to 7, which is the value a of the y-axis intercept when approximated to + a.

(Supplementary note 9) One of the data distribution feature quantities is that the order of arrangement of the original data is x, the original data value is y, and the relationship between x and y is a linear expression y = b · x.
8. The data analysis method according to any one of appendices 1 to 7, wherein the value of the slope is b when approximated to + a.

(Supplementary note 10) One of the supplementary notes 1 to 7, wherein one of the data distribution feature quantities is the strength of the specific periodicity of the original data values with respect to the order of arrangement of the original data. Data analysis method described in.

(Supplementary Note 11) One of the supplementary notes 1 to 7 is characterized in that one of the data distribution feature quantities is a value indicating the strongest periodicity of the original data value with respect to the order of arrangement of the original data. Data analysis method described in section 3.

(Supplementary Note 12) (a) a step of preparing data results of explanatory variables and objective variables, and (b) a step of calculating the degree of confounding and / or independence between a plurality of explanatory variables based on the data results. And (c) a step of performing data mining using the degree of confounding and / or the degree of independence, the data analysis method.

(Supplementary note 13) The data analysis method according to supplementary note 12, wherein in the step (b), the degree of confounding and / or the degree of independence is calculated in units of a set divided into two by regression tree analysis.

(Supplementary Note 14) In the step (b), a plurality of explanatory variables that are factors of the division having a large significant difference are selected by regression tree analysis, and the degree of confounding and / or the degree of confounding among the plurality of explanatory variables are selected.
Alternatively, the data analysis method according to appendix 13, wherein the degree of independence is calculated.

(Supplementary Note 15) In the step (b), when the degree of confounding and / or the degree of independence between the reference explanatory variable and other explanatory variables are calculated, each set divided into two by regression tree analysis. 15. The data analysis method according to appendix 14, wherein the degree of confounding and / or the degree of independence is calculated based on the ratio of agreement and disagreement of data between the explanatory variables in the above.

(Supplementary note 16) The data analysis method according to supplementary note 15, wherein in step (c), data mining is performed by selecting explanatory variables based on the degree of confounding and / or the degree of independence. .

(Supplementary Note 17) Calculation means for calculating the degree of confounding and / or the degree of independence between a plurality of explanatory variables based on the data results of the explanatory variables and the objective variables, and data using the degree of confounding and / or the degree of independence. A data analysis device, comprising: a data mining means for performing mining.

(Supplementary note 18) The data analyzing apparatus according to supplementary note 17, wherein the calculating means calculates the degree of confounding and / or the degree of independence in units of sets divided into two by regression tree analysis.

(Supplementary Note 19) The calculating means selects a plurality of explanatory variables which are factors of division with a large significant difference by regression tree analysis, and calculates the degree of confounding and / or the degree of independence between the plurality of explanatory variables. 19. The data analysis device according to appendix 18, characterized in that.

(Supplementary Note 20) When calculating the degree of confounding and / or the degree of independence between a reference explanatory variable and other explanatory variables, the calculating means calculates the degree of confusion in each set divided into two by regression tree analysis. 20. The data analysis apparatus according to appendix 19, wherein the degree of confounding and / or the degree of independence is calculated based on the ratio of agreement and disagreement of data between explanatory variables.

(Supplementary note 21) The data analysis apparatus according to supplementary note 20, wherein the data mining means performs data mining by selecting the explanatory variables based on the degree of confounding and / or the degree of independence.

(Supplementary Note 22) (a) Procedure for preparing data results of explanatory variables and objective variables, and (b) Procedure for calculating confounding degree and / or independence degree between a plurality of explanatory variables based on the data results. And (c) a procedure for performing data mining using the degree of confounding and / or the degree of independence, a computer-readable recording medium recording a program for causing a computer to execute the procedure.

[0139]

According to the present invention, various data distribution features existing in the original data group accumulated in the computer system are extracted, and each feature amount is sequentially selected and analyzed to analyze each feature. Since the factor that causes the quantity is automatically and quantitatively evaluated and extracted, it is possible to extract a lot of information (trends, characteristic patterns, relationships between data, etc.) from a more multifaceted perspective. Therefore, it is possible to objectively and efficiently and quantitatively extract the relation and the significant difference which are conventionally difficult to discriminate because they are buried in various data.

[Brief description of drawings]

FIG. 1 is a diagram showing an example of a computer system used in a first embodiment of the present invention.

FIG. 2 is a block diagram showing an example of a functional configuration of a data analysis device implemented by the computer system having the configuration shown in FIG.

FIG. 3 is a chart showing each feature amount extracted by extraction of data distribution features in the first embodiment of the present invention in CSV format.

FIG. 4 is a chart showing information focusing on a variation of a wafer attribute value as a feature of a data distribution within a lot when performing a yield analysis of semiconductor data in the first embodiment of the present invention.

FIG. 5 is a diagram showing variations in yield values of wafers in a lot for a plurality of lots.

FIG. 6 is a flowchart showing an outline of an example of a data analysis method according to the first exemplary embodiment of the present invention.

FIG. 7 is a characteristic diagram showing a relationship between yield and VT_N2 as a specific example.

FIG. 8: V obtained from all wafers as a specific example
It is a figure which shows the histogram of T_N2 data.

FIG. 9 is a diagram showing a box whisker diagram in which all VT_N2 data are displayed for each wafer number as a specific example.

FIG. 10: As a specific example, the objective variable is V for each lot
It is a figure which shows the result of performing a regression tree analysis by setting the average value of T_N2 and using the explanatory variable as the device name used in each process.

11 is a diagram showing an example of an evaluation statistical value list for the regression tree analysis result shown in FIG.

FIG. 12 is a diagram showing a box whiskers diagram in which all VT_N2 data are displayed for each device name of the second wiring device used as a specific example.

FIG. 13 shows a specific example in which the objective variable is VT_ of each wafer.
It is a figure which shows the result of having performed the regression tree analysis by setting the value of N2 and using the explanatory variable as the apparatus name used in each process.

14 is a diagram showing an example of an evaluation statistical value list for the regression tree analysis result shown in FIG.

[FIG. 15] As a concrete example, all VT_N2 data is set to 2CON.
It is a figure which shows the box mustache figure displayed for every use apparatus name of process_apparatus.

FIG. 16 is a table showing a file defining each feature amount of VT_N2 for each lot as a specific example.

FIG. 17 is a diagram showing a histogram of each feature amount of the intra-lot distribution of VT_N2 as a specific example.

FIG. 18 is a table showing a file for performing a regression tree analysis on the feature amount of the intra-lot distribution of VT_N2 as a specific example.

FIG. 19 is a chart showing an input file for analyzing a yield variation factor by regression tree analysis as a specific example.

FIG. 20: As a specific example, the objective variable is V for each lot
It is a figure which shows the result of performing a regression tree analysis by setting the standard deviation value of T_N2 and using the explanatory variable as the device name used in each process.

FIG. 21 is a diagram showing an example of an evaluation statistical value list for the regression tree analysis result shown in FIG. 20.

[Fig. 22] As a specific example, all VT_N2 data is field
It is a figure which shows the box mustache figure displayed for every used device name of d_Ox process_device.

FIG. 23 shows PM1 in the Field_Ox process as a specific example.
VT_N of all wafers using No. 3 or PM No. 3
It is a figure which shows the histogram of 2.

FIG. 24 shows PM1 in the Field_Ox process as a specific example.
It is a figure which shows the histogram of VT_N2 of the wafer for 1 lot which used the No. 3 machine or PM3 machine.

FIG. 25 shows PM1 in the Field_Ox process as a specific example.
It is a figure which shows the histogram of VT_N2 of the wafer for 1 lot which used the No. 3 machine or PM3 machine.

FIG. 26 shows PM1 in the Field_Ox process as a specific example.
It is a figure which shows the histogram of VT_N2 of the wafer for 1 lot which used the No. 3 machine or PM3 machine.

FIG. 27 shows PM2 in the Field_Ox process as a specific example.
It is a figure which shows the histogram of VT_N2 of all the wafers which used the No. machine.

[FIG. 28] As a specific example, PM2 in the Field_Ox step
It is a figure which shows the histogram of VT_N2 of the wafer for 1 lot which used the No. machine.

FIG. 29 shows PM2 in the Field_Ox process as a specific example.
It is a figure which shows the histogram of VT_N2 of the wafer for 1 lot which used the No. machine.

FIG. 30 shows PM2 in the Field_Ox process as a specific example.
It is a figure which shows the histogram of VT_N2 of the wafer for 1 lot which used the No. machine.

FIG. 31 is a diagram showing a result of performing a regression tree analysis as a specific example, in which an objective variable is a periodicity value of a wafer number interval 2 and an explanatory variable is an apparatus name used in each step.

32 is a diagram showing an example of an evaluation statistical value list for the regression tree analysis result shown in FIG. 31.

FIG. 33 is a diagram showing a box whisker diagram in which the periodicity value of the wafer number interval 2 is displayed for each device name of the F diffusion process_device as a specific example.

FIG. 34 is a diagram showing intra-lot variation of VT_N2 for one lot of wafers using the F7 machine in the F diffusion process as a specific example.

FIG. 35 is a diagram showing intra-lot variation of VT_N2 for one lot of wafers using F7 machine in the F diffusion process as a specific example.

FIG. 36 is a diagram showing intra-lot variation of VT_N2 for one lot of wafers using F7 machine in the F diffusion process as a specific example.

FIG. 37 is a diagram showing intra-lot variation of VT_N2 for one lot of wafers using F5 machine, F6 machine, F8 machine, or F9 machine in the F diffusion process as a specific example.

FIG. 38 is a diagram showing intra-lot variation of VT_N2 for one lot of wafers using F5 machine, F6 machine, F8 machine, or F9 machine in the F diffusion process as a specific example.

FIG. 39 is a diagram showing intra-lot variation of VT_N2 for one lot of wafers using F5 machine, F6 machine, F8 machine, or F9 machine in the F diffusion process as a specific example.

FIG. 40 is a diagram showing a relationship between a lot flow and an abnormal manufacturing apparatus.

FIG. 41 is a diagram showing a yield distribution by device in a certain process according to a conventional technique.

FIG. 42 is a diagram showing the relationship between the lot flow and the confounding of the abnormal manufacturing apparatus.

FIG. 43 is a diagram showing an example of regression tree analysis input data.

FIG. 44 is a diagram showing an example of a regression tree.

FIG. 45 is a diagram showing an example of an evaluation statistical value list.

FIG. 46 is a diagram showing a relationship between a manufacturing apparatus used, electrical characteristic data, and a yield value.

[Fig. 47] Fig. 47 is a diagram illustrating a calculation example of a 2-division confounding degree and a 2-division independence degree.

FIG. 48 is a diagram showing an example of a regression tree.

FIG. 49 is a diagram showing an example of an evaluation statistical value list.

FIG. 50 is a diagram showing the degree of confounding and the degree of independence of each explanatory variable and the explanatory variable of the first candidate.

FIG. 51 is a diagram showing the degree of confounding and the degree of independence of each explanatory variable and the explanatory variable of the third candidate.

FIG. 52 is a diagram showing a degree of confounding among all candidates and an average thereof.

FIG. 53 is a diagram showing the independence of all candidates and their average.

FIG. 54 is a regression tree diagram showing a regression tree analysis result based on a defective wafer group.

FIG. 55 is a regression tree diagram showing the results of regression tree analysis using a group of good wafers.

FIG. 56 is a diagram showing an example of a functional configuration of a data analysis device.

[Explanation of symbols]

21 means for extracting data distribution features 22 Means for selecting feature quantity to be analyzed 23 Means for data mining 24 Rule file 25 Statistical analysis component 26 Chart Creation Component 27 Analysis tools 41 Database 42 Original data group 101 Normal device 102 Abnormal device 401 Explanatory variable 402 Objective variable 411 Equipment used 412 Electrical characteristics data 413 Yield 801 Standard explanatory variables 802 Comparison explanatory variables 803 Wafer number 804 Yield 811 High Yield Group of Standard Explanatory Variables 812 Low-yield group of reference explanatory variables 813 2-division entanglement calculation formula 814 2 division degree 815 2-division independence 1701 Original data group 1702 database 1703 Data Mining Department 1704 rule file 1705 Analysis tools 1706 Statistical Analysis Component 1707 Chart creation component 1708 decision-making department

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Hidehiro Shirai             3-2-1 Sakado, Takatsu-ku, Kawasaki City, Kanagawa Prefecture               Fujitsu LSI Technology Co., Ltd.             Within F term (reference) 5B056 BB61

Claims (7)

[Claims] 1. A step of quantitatively evaluating and extracting one or more data distribution feature quantities existing in the original data group by editing an original data value, and from the extracted data distribution feature quantities A data analysis method characterized by including a step of performing an analysis by selecting an arbitrary data distribution characteristic amount and a step of making a decision based on the obtained analysis result. 2. A step of editing an original data value to quantitatively evaluate and extract two or more data distribution feature quantities existing in the original data group, and a step of extracting each of the extracted data distribution feature quantities. A data analysis method comprising: a step of sequentially selecting and performing an analysis; and a step of making a decision based on the obtained analysis result. 3. The data analysis method according to claim 1, wherein the data distribution characteristic amount is represented by a continuous value indicating a degree of the characteristic. 4. The data analysis method according to any one of claims 1 to 3, wherein the respective data distribution feature quantities for each record are independent of each other. 5. The data analysis method according to claim 1, wherein analysis is performed by data mining using the data distribution feature quantity as an objective variable. 6. A regression tree in which the individual data distribution feature amounts are stored in a file for each record, and the same data distribution feature amount is sequentially selected for some or all of the records from the file as an objective variable. The data analysis method according to claim 5, wherein analysis is performed. 7. The process according to claim 1, wherein each of the steps is automatically performed by executing software that is configured to be sequentially performed on a computer system.
6. The data analysis method according to any one of 6.