JP2006318263A - Information analysis system, information analysis method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an information analysis system capable of easily and surely specifying an explanatory variable affecting a criterion variable about the criterion variable related to quality of a product and the explanatory variable related to production of the product. <P>SOLUTION: With respect to a production process of a semiconductor device, history information related to a processor used in each the process is stored in a process information collection part 3 as the explanatory variable of multi-valued information. A value of a yield that is the criterion variable is also stored in the process information collection part 3. When the yield that is the criterion variable is reduced as compared to a predetermined reference value, the history information of the process information collection part 3 is converted into a matrix of binary information by a cross totaling part 4. The matrix of the binary information is once stored in a data storage part 5, and an influence degree to the criterion variable is calculated by an analysis part 6. Contents of the influence degree that is a calculation result is displayed on a contribution description part 7 associatively to the explanatory variable. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an information analysis system, an information analysis method, and a program for performing quality control of a product, for example.

  For example, in quality control in the manufacture of semiconductor devices, history information representing processing devices used in the manufacturing process, manufacturing conditions, inspection results of completed characteristics, and quality control information such as yield are associated with products. Collecting. When the yield of a product decreases, for the purpose of investigating the cause, the relationship between the history information and the yield is analyzed using the yield as an objective variable and the history information as an explanatory variable. From this analysis result, a processing device causing a decrease in yield is identified, and the use of the identified processing device is stopped to recover the yield.

  In addition, if the product characteristics deteriorate, the inspection results are used as objective variables and the manufacturing conditions are used as explanatory variables.The relationship between the manufacturing conditions and the inspection results is analyzed, and the highly relevant manufacturing conditions are adjusted. It tries to restore the characteristics.

  As a method for performing this kind of quality control, there is a conventional method using regression tree analysis for a semiconductor manufacturing process (see Patent Document 1: Japanese Patent Laid-Open No. 2002-324206). In this quality control method, division into two parts is sequentially repeated for a set of quality control information based on parameter values of explanatory variables indicating processing devices at the time of manufacture. The parameter is the variance of the objective variable indicating the yield, and is divided so that the degree of separation between the two divided sets is maximized. By repeating the division into two, the most significant explanatory variable is extracted with respect to the objective variable, and the cause of the yield reduction in the semiconductor manufacturing process is specified.

  As another conventional quality control method, there is a method in which an explanatory variable is selected based on a predetermined criterion, and an objective variable is estimated from the selected explanatory variable using a regression model (for example, Patent Document 2: Japanese Patent Application Laid-Open No. Hei 11 (1999)). 7-325804).

  However, the conventional quality control method using regression tree analysis merely extracts information features and regularity based on the distribution of quality control information, and sequentially divides the information into groups that can be regarded as two relatives. . Therefore, since the factor governing the magnitude of the yield is not extracted, there is a problem that the cause of the decrease in the yield may not be immediately identified from the analysis result.

In addition, the conventional quality control method using a regression model needs to select explanatory variables according to predetermined criteria, but there are various selection criteria, and it is difficult to determine which explanatory variable is adopted. There is a problem that it is not easy to use in actual use.
JP 2002-324206 A JP 7-325804 A

  Therefore, an object of the present invention is to provide an information analysis system that can easily and reliably specify an explanatory variable that affects the objective variable, regarding the objective variable related to the quality of the product and the explanatory variable related to the production of the product. There is to do.

In order to solve the above problems, the information analysis system of the present invention includes an explanatory variable storage unit that stores explanatory variables related to production of a product as multi-value information,
An objective variable storage unit for storing objective variables related to the quality of the product;
A matrix creation unit that converts the explanatory variables of the multi-value information stored in the explanatory variable storage unit to create a matrix formed of binary information;
It is characterized by comprising an influence degree calculating section for calculating the influence degree indicating the degree of influence of the explanatory variable on the objective variable by calculating the matrix and the objective variable.

  According to the above configuration, the influence variable is obtained by calculating the matrix obtained by converting the explanatory variables and the objective variable by the influence calculation unit. The degree of the influence of the explanatory variable on the objective variable is represented by the magnitude of the influence degree. For example, a cause such as a decrease in yield as the objective variable is caused in the processing device corresponding to the explanatory variable having the large influence degree. It can be identified.

  Further, a matrix of binary information is created from the explanatory variables of the multi-value information stored in the explanatory variable storage unit by a matrix creation unit, for example, by cross tabulation. Since the explanatory variables of the multi-value information have a smaller data amount than the binary information matrix, the explanatory variable storage section that needs to store the explanatory variables sequentially with the production of the product is stored in a relatively small capacity. It can comprise using a part.

  In addition, since the influence of the explanatory variable on the objective variable is calculated by the calculation of the matrix derived from the explanatory variable and the objective variable, the operator's judgment like the quality control method using the conventional regression model. There is no need to select an explanatory variable as a reference. Therefore, it is possible to identify a factor that governs the change of the objective variable more easily than in the past.

  In one embodiment of the information analysis system, the matrix creation unit creates the matrix when the objective variable falls below or exceeds a predetermined reference value.

  According to the above-described embodiment, when the analysis of the influence level is necessary, since the explanatory variable of the multi-value information having a relatively small amount of data is converted into a matrix of binary information having a relatively large amount of data, The storage capacity required for the information analysis system can be minimized. For example, in the case where the objective variable is yield information, when the objective variable falls below the yield reference value, it is necessary to analyze the degree of influence. For example, when the objective variable is a defect rate, when the objective variable exceeds the reference value of the defect rate, an analysis of the influence level is required. As described above, the matrix may be created when the reference value is below or exceeds the reference value according to the contents of the objective variable. A matrix storage unit that temporarily stores the binary information matrix may be provided.

  In the information analysis system according to an embodiment, the explanatory variable is at least one of history information regarding processing performed on the product and design information of the product.

  According to the embodiment, for example, history information indicating a processing apparatus or the like in which the product is processed in a production process, and design information indicating a structure or the like in which the product is to be formed through the production process. The degree of influence on the objective variable is obtained for the explanatory variable that is at least one of the above.

  In one embodiment of the information analysis system, the objective variable is at least one of a yield, a defect rate, and a characteristic value of the product.

  According to the embodiment, the degree of influence of the explanatory variable is obtained on the objective variable that is at least one of the yield, defect rate, and characteristic value of the product.

  In the information analysis system of one embodiment, the influence degree calculation unit calculates the influence degree by a partial least square method.

  According to the above embodiment, even when the explanatory variables include elements having mutually dependent relationships, elements whose values are not changed, and the like, it is possible to calculate a highly reliable influence degree.

  An information analysis system according to an embodiment includes a first estimation unit that estimates an objective variable corresponding to a predetermined combination of the explanatory variables using the degree of influence obtained by the influence degree calculation unit.

  According to the embodiment, the first estimation unit estimates the objective variable for the production process of the combination represented by the predetermined combination of the explanatory variables, for example, in which production is not actually performed. For example, when the objective variable is a device characteristic and the explanatory variable is a processing device, it is possible to accurately estimate a device characteristic obtained by a combination of the processing device that is not actually performed.

An information analysis system according to an embodiment includes an interpolation influence degree calculation unit that calculates an interpolation influence degree by interpolation from the influence degree obtained by the influence degree calculation unit,
A second estimation unit that estimates an objective variable corresponding to an explanatory variable other than the explanatory variable used when the influence is calculated by using the interpolation influence;

  According to the embodiment, by using the interpolation influence degree, for example, when the objective variable is a device characteristic and the explanatory variable is a processing condition, a device characteristic obtained under a processing condition that is not actually performed is obtained. Can be estimated with high accuracy.

An information analysis system according to an embodiment includes an interpolation influence degree calculation unit that calculates an interpolation influence degree by interpolation from the influence degree obtained by the influence degree calculation unit,
A third estimation unit that estimates a plurality of objective variables corresponding to a plurality of explanatory variables having a predetermined probability distribution using the influence degree and the interpolation influence degree is provided.

  According to the above embodiment, the objective variable for the explanatory variable is estimated by the third estimator for the explanatory variable having a stochastic fluctuation in value, and as a result, the variation regarding the quality of the product can be estimated. it can.

The information analysis method of the present invention includes a step of storing, as multi-value information, history information relating to processing performed on a product, and an explanatory variable that is at least one of the design information of the product,
Storing an objective variable which is at least one of the yield, defect rate and characteristic value of the product;
When the objective variable falls below or exceeds a predetermined standard, converts the explanatory variable of the multi-value information to create a binary information matrix;
A step of calculating the matrix of binary information and the objective variable to obtain an influence indicating the degree of contribution of the explanatory variable to the objective variable.

  According to the above configuration, the degree of influence of the explanatory variable on the objective variable is represented by the magnitude of the degree of influence obtained by calculating the matrix derived from the explanatory variable and the objective variable. It can be specified that a cause such as a decrease in yield exists in the processing device corresponding to the explanatory variable having the large influence degree.

  Also, explanatory variables of multi-value information are stored, and when the objective variable falls below or exceeds a predetermined standard, the explanatory variables of the multi-value information are converted into a binary information matrix by, for example, cross tabulation Convert to Since the explanatory variables of the multi-value information have a smaller data amount than the binary information matrix, the explanatory variables can be stored with a small storage capacity. For example, when the objective variable is yield information, when the objective variable falls below the yield reference value, the binary information is converted into a matrix. On the other hand, for example, when the objective variable is a failure rate, when the objective variable exceeds a reference value of the failure rate, the binary information is converted into a matrix. As described above, the matrix may be created when the reference value is below or exceeds the reference value according to the contents of the objective variable.

  In addition, unlike the conventional information analysis method using a regression model, it is not necessary to select an explanatory variable as a reference based on the operator's judgment, so that the dominant factor for the objective variable can be identified more easily than in the past.

  The program of the present invention causes a computer to function as the information analysis system.

  According to the above configuration, the information analysis system of the present invention is obtained by executing the program by the computer.

  As described above, the information analysis system of the present invention converts the explanatory variables of the multi-value information stored in the explanatory variable storage unit to create a matrix of binary information, the matrix and the object And an influence degree calculation unit that calculates an influence degree indicating the degree of influence of the explanatory variable on the objective variable, so that the simple variable configuration can be achieved by using the explanatory variable storage part having a small storage capacity. The influence of explanatory variables on objective variables can be clarified.

  Hereinafter, the present invention will be described in detail with reference to the illustrated embodiments.

(First embodiment)
1st Embodiment demonstrates the information analysis system used for the manufacturing process of the semiconductor device as a product.

  The semiconductor device manufacturing process includes a number of processes from wafer reception to device completion, and each process is manufactured using a plurality of processing apparatuses. For example, for processing steps A, B,..., When there are three processing devices in processing step A and there are two processing devices in processing step B, the relationship between the processing steps and processing devices is shown in FIG. It can be expressed as Identification information called wafer ID is given to the wafer. The information analysis system records, in the database, a processing history having contents indicating which processing apparatus has processed the wafer in association with the wafer ID. FIG. 2 is a table in which a processing history is recorded for each wafer ID. The process is completed in the final process X, and the wafer is divided into products such as devices and panels. Products are inspected for operation, characteristics, etc., and products that satisfy certain standards are passed and shipped. As the inspection result, a yield, a defect rate, or the like is used. The trend of inspection results is monitored using, for example, a graph with the wafer ID as the horizontal axis and the yield as the vertical axis, and the cause is investigated when the yield decreases.

  The graph of FIG. 3 is a diagram showing an example of yield transition. With wafer IDs of 80 to 100, the yield is greatly reduced, and there may be some abnormality in the apparatus that processed these wafers. Therefore, the information analysis system according to the present embodiment uses the objective variable as the yield and the explanatory variable as the processing device, and analyzes the correlation between the yield and the processing device. First, a table as shown in FIG. 4 is created that associates numerical values for specifying processing devices with yields. Since the processing apparatus is randomly used on the wafer, there is no linear relationship between the numerical value representing the processing apparatus and the yield. Therefore, even if the correlation is analyzed using the yield as an objective variable and the processing device as an explanatory variable, it is difficult to extract a problematic device.

  The information analysis system according to the present embodiment performs cross-tabulation using the processing facility as a column header and the wafer ID as a row header, and converts the table into a table as shown in FIG. In this table, the processing devices arranged in the order of the manufacturing process and the processing device number are the explanatory variables X, and the yield is the objective variable Y.

  Next, the magnitude of the regression coefficient of each Xi with respect to Y is obtained, and the result is graphed. Here, Xi and the names of the processing devices are X1, X2, X3, X4, X5, X6, X7, X8,..., A1, A2, A3, B1, B2, C1, C2, C3, respectively. It corresponds to C4 ....

  FIG. 6 is a graph showing the above results. As shown in FIG. 6, among the explanatory variables, A1 and B1 have a large influence on the objective variable. Therefore, it becomes clear that the processing devices A1 and B1 have a high influence on the yield.

(Second Embodiment)
FIG. 7 is a schematic diagram showing the configuration of the information analysis system of this embodiment. In the present embodiment, three, two, and four processing apparatuses are used in the semiconductor device manufacturing processes A, B, and C, respectively. A-1 means the apparatus 1 of the process A. The wafer whose ID is indicated by AAA0001 has a processing history of A-1, B-1, and C-2, and wafer AAA0002 has a processing history of A-2, B-2, and C-3. The wafer is inspected for operation and performance in an inspection process downstream of the process C. The processing history information and the inspection information are collected and managed by the process information collection unit 3 as an explanatory variable storage unit and an objective variable storage unit.

  When the yield as inspection data falls below a predetermined reference value, the history information and inspection data information are cross-tabulated as a matrix creation unit from the process information collection unit in order to identify the cause process or processing device Sent to part 4. In the cross tabulation unit 4, history information composed of multi-value information is converted into a matrix composed of binary information. Specifically, a cross tabulation table composed of values of 0 or 1 is created using the explanatory variables developed with one history as one level as the row header and the wafer ID as the column header. Since the binary cross tabulation table requires a relatively large memory, it may be temporarily stored in the data storage unit 5. When the objective variable is the defect rate, when the defect rate exceeds a predetermined reference value, the inspection data that is the defect rate is sent from the process information collecting unit to the matrix creating unit together with the history information. To do.

  Next, the matrix 6 and the objective variable (yield) are calculated by the analysis unit 6 serving as the influence degree calculation part, and the influence degree on the objective variable is obtained, and this influence degree and the explanatory variable corresponding to this influence degree are obtained. Is displayed on the contribution display section 7. The cross tabulation unit 4, the data storage unit 5, the analysis unit 6 and the contribution drawing unit 7 may be realized by the same computer, or the cross tabulation unit 4 and the data storage unit 5 are provided in the server while the analysis unit 6 and the contribution drawing unit 7 may be realized by a client computer. The analysis unit 6 realized by the server and client computers is realized by executing a program stored in the computer.

  The process information collection unit 3 preferably accumulates information such as history information serving as explanatory variables as multi-value information. This is because the binary information obtained by cross-tabulation has a relatively large data scale, so if the historical information that is always collected during the manufacturing process is stored as binary data, a large storage capacity is required. is there. Therefore, when a problem such as a decrease in yield occurs, it is preferable to convert the multi-value history information into a binary cross tabulation table by performing cross tabulation when analyzing the cause. As a result, the scale of the storage device used for the process information collection unit 3 can be reduced, and an information analysis system that accurately extracts the cause of the problem can be realized.

(Third embodiment)
In the present embodiment, an example of extracting a process and a processing apparatus having a high defect rate for a semiconductor device manufactured from the wafer when about 1100 wafers each have 121 processing histories will be described.

  FIG. 8 is a graph in which the defect rate corresponding to each wafer is plotted on the coordinates, with the horizontal axis representing the wafer ID and the vertical axis representing the defect rate of the semiconductor device manufactured from the wafer with the ID. Considering that the value of the wafer ID increases with time in the production line, it can be seen that the defect rate changes randomly in time series as shown in FIG. Here, as in the table shown in FIG. 9, history information is cross tabulated for each wafer ID, and includes an explanatory variable that is a processing apparatus and a failure rate that is an objective variable, and is composed of 0 and 1 Convert to a matrix of value information.

  FIG. 10 is a network diagram showing the interrelationship between the semiconductor device manufacturing processes. Process B and process D have one facility, and the values of explanatory variables X3 and X9 are always 1. Further, since the number of facilities in process A is two, and either one of the facilities is used, explanatory variable X2 is 0 when explanatory variable X1 is 1, and explanatory variable X2 when explanatory variable X1 is 0. Is 1. Therefore, X1 and X2 have a strong correlation (correlation coefficient 1), X3 and X9 are invariant constants, and the independence data is included in the explanatory variables. Thus, when the correlation between explanatory variables is high and there is a dependency relationship, the regression coefficient of the explanatory variable in the dependency relationship becomes abnormally large, and the reliability of factor extraction may be impaired.

  In this case, the problem can be avoided by combining the explanatory variables into the principal components and using the partial least square method for performing regression analysis on the principal components. The regression coefficient of the principal component can be finally converted to the regression coefficient of the original explanatory variable, and the reliability of the regression coefficient obtained in this way is obtained when the principal component regression is not performed. Better than.

  In multiple regression analysis, a linear model for the objective variable is given to all explanatory variables, and a regression coefficient that minimizes the sum of squares of the estimation error due to the linear model is determined. A linear model is constructed between the potential principal components synthesized from the objective variables and the objective variable.

  Actually, with respect to the cross tabulation data of FIG. 9, when the defect rate that is an objective variable is set to Y and the history information that is an explanatory variable is set to X, using the result of the factor analysis, the partial least square method , 121 explanatory variables are aggregated into four potential principal components, as shown in FIG. 11B. FIG. 11A is a conceptual diagram showing the correspondence between explanatory variables and objective variables by multiple regression analysis. FIG. 12 shows the final analysis result. In FIG. 12, the contribution (influence) of the explanatory variable to the objective variable is shown in descending order. In multiple regression, the difference in contribution for each explanatory variable is large, but in the example using the partial least square method, the difference in contribution for each explanatory variable is small.

  FIG. 13 shows the distribution of failure rates corresponding to all data and the failure rate distribution when processing devices corresponding to the explanatory variables 4, 25, and 38 having a contribution of 5 or more are used in order to confirm the validity of the results. FIG. As can be seen from FIG. 13, when the processing device extracted by the factor analysis is used, the defect rate increases frequently, and it can be said that the result of the factor analysis is appropriate.

(Fourth embodiment)
In the present embodiment, design information is used as an explanatory variable, and device characteristics are used as an objective variable for a semiconductor device manufacturing process. The explanatory variable is composed of six pieces of design information from X ₁ to X ₆ , and one piece of design information can take the number of levels shown in Table 1 below.

Each design information is assigned to 0 and 1 for each level, and a total of 42 types of design information is used as column headings, and a cross tabulation table is created using 78 types of characteristic data as row headings. Next, calculation based on multiple regression analysis was performed using characteristic data as an objective variable, and the contribution degree (influence degree) of 42 pieces of design information was obtained. FIG. 14 is a diagram showing the calculation result at coordinates with the explanatory variable as the horizontal axis and the regression coefficient as the vertical axis. Here, the estimated value Y _est of the device characteristic Y is obtained by the following equation (1).

Here, in equation (1), C _oi is the degree of influence of the design variable X _{i on} the characteristic data Y.

FIG. 15 is a diagram showing the estimated value Y _est obtained from the above equation (1) and the measured data Y _mes superimposed on the same coordinates.

  As can be seen from FIG. 14, in each category of the design information A to F, the regression coefficient has a non-linear relationship with the level of the explanatory variable. Thus, even when the objective variable has a non-linear relationship with the explanatory variable, the linear regression model can be applied by decomposing the level that the design information can take and expressing it in a binary matrix. As shown in the above, it is possible to estimate the objective variable with good accuracy.

(Fifth embodiment)
In semiconductor device manufacturing processes, objective variables related to quality such as yield and characteristic values correspond to explanatory variables at discrete levels, so the regression coefficient is unknown for explanatory variables at levels other than the known level. And cannot be modeled. In such a case, modeling can be performed by interpolating a regression coefficient for a level other than a known level. For example, as shown in FIG. 16, for three explanatory variables A, B, and C, A has five levels (A = 10, 20, 30, 40, 50), and B has three C has three levels, and a regression coefficient corresponding to the level of each explanatory variable is obtained. The regression coefficient when A = 25 is 4 by interpolating 3 which is the regression coefficient of A = 20 and 5 which is the regression coefficient of A = 30. Similarly, for other design information, the regression coefficient for values other than the level can be estimated to estimate the value of the objective variable.

  The information analysis method of this embodiment can be executed by using an information analysis system in which the following configuration is added to the information analysis system shown in FIG. That is, the storage unit that stores the regression coefficient (influence value) calculated by the analysis unit 6, the input unit that inputs the level value to be estimated for the explanatory variable, and the value that is input to the input unit are stored as described above. An estimation unit that estimates the regression coefficient by interpolation based on the regression coefficient stored in the unit. By using the information analysis system having such a configuration, the value of the objective variable can be estimated for the new level of the explanatory variable.

(Sixth embodiment)
In the sixth embodiment, when the explanatory variable for which the objective variable is to be estimated does not take a deterministic value and has a stochastic fluctuation, the range of variation of the objective variable is estimated. For example, when the design information A has a fluctuation with a normal distribution with σ of 3 centered on 30, as shown schematically in FIG. 17, a distribution of regression coefficients corresponding to the design information of the normal distribution described above is obtained. By calculating the distribution of the objective variable from the distribution of the regression coefficients, the variation of the objective variable can be estimated.

For example, when the characteristic Y as the objective variable is represented by a regression coefficient (contribution) with the design information A, B, and C as explanatory variables as factors, the regression coefficients of A, B, and C are represented as _CoA , C _When expressed as _oB and C _oC , the characteristic Y is expressed as the following formula (2).
Y = C _oA + C _oB + C _oC (2)
Here, when only the design information A is fluctuating and B and C are constant values, the statistical fluctuation is given to the design information A to obtain the fluctuation of _CoA , and thereby the variation of the characteristic Y. Can be estimated. In the graph of the regression coefficients corresponding to the explanatory variables A, B, and C shown in FIG. 17, when 100 virtual prototypes were performed by giving a statistical fluctuation of σ = 5 around A2 = 30, The fluctuation of the characteristic Y is as shown in FIG. Further, the fluctuation of the characteristic Y when 100 virtual trial manufactures were performed with the statistical fluctuation of σ = 3 centered on A2 = 40 is as shown in FIG. FIG. 20 shows a comparison of fluctuations in the characteristic Y in FIGS. 18 and 19 using histograms. From FIG. 20, when the design condition of the design variable A is changed from 30 to 40 and the fluctuation (standard deviation) of A is reduced from 5 to 3, the characteristic value level of Y is improved and a characteristic with little variation can be obtained. It becomes clear.

  In the information analysis method of the present embodiment, the information analysis system in the fifth embodiment is provided with a second input unit to which the value of the standard deviation is input, and a plurality of values are calculated based on the standard deviation input to the second input unit. This can be executed by calculating the value of the explanatory variable and estimating a plurality of regression coefficients by the estimation unit corresponding to the value of the plurality of explanatory variables.

In the information analysis method of 1st Embodiment, it is the figure which showed the relationship between a process process and a processing apparatus. It is the table which recorded the processing history for every wafer ID. It is a figure which shows an example of transition of a yield. It is a table which matches the numerical value which specifies a processing apparatus, and a yield. It is a table in which explanatory variables and objective variables are converted by cross tabulation. It is a figure which shows the influence with respect to the objective variable of an explanatory variable. It is the schematic which shows the structure of an information analysis system. It is the graph which showed the defect rate corresponding to each wafer. It is a figure which shows the matrix of the binary information containing an explanatory variable and an objective variable. FIG. 5 is a network diagram showing the mutual relationship of semiconductor device manufacturing processes. It is a conceptual diagram which shows a response | compatibility with the explanatory variable and objective variable by multiple regression analysis about the matrix of FIG. It is a conceptual diagram which shows a response | compatibility with the explanatory variable by the partial least square method, and the objective variable about the matrix of FIG. It is the figure which showed the contribution degree of the explanatory variable with respect to the objective variable as a calculation result of the matrix of FIG. It is the figure which showed the failure rate distribution corresponding to all the data, and the failure rate distribution corresponding to the explanatory variable whose contribution is 5 or more in an overlapping manner. It is the figure which showed the contribution of the explanatory variable (design information) with respect to the objective variable (device characteristic) about the manufacturing process of a semiconductor device. It is the figure which piled up and showed the estimated value of the objective variable, and the actual measurement value. It is a figure which shows a mode that the contribution to the objective variable corresponding to the unknown level of an explanatory variable is estimated about the manufacturing process of a semiconductor device. It is a figure which shows a mode that the contribution to the objective variable (device characteristic) corresponding to the explanatory variable (design information) which has a predetermined probability distribution is estimated about the manufacturing process of a semiconductor device. It is a figure which shows the estimation result of the objective variable with respect to the explanatory variable which has a predetermined probability distribution. It is a figure which shows the estimation result of the objective variable about the explanatory variable which has a center value with a larger regression coefficient than in FIG. 18, and has predetermined | prescribed probability distribution. FIG. 20 is a diagram in which the estimation results of FIGS. 18 and 19 are superimposed.

Explanation of symbols

3 Process information collection unit 4 Cross tabulation unit 5 Data storage unit 6 Analysis unit 7 Contribution drawing unit

Claims (10)

An explanatory variable storage unit for storing explanatory variables related to production of the product as multi-value information;
An objective variable storage unit for storing objective variables related to the quality of the product;
A matrix creation unit that converts the explanatory variables of the multi-value information stored in the explanatory variable storage unit to create a matrix formed of binary information;
An information analysis system comprising: an influence degree calculation unit that calculates the degree of influence of the explanatory variable on the objective variable by calculating the matrix and the objective variable. The information analysis system according to claim 1,
The information analysis system, wherein the matrix creation unit creates the matrix when the objective variable falls below or exceeds a predetermined reference value. The information analysis system according to claim 1,
The information analysis system characterized in that the explanatory variable is at least one of history information relating to processing performed on the product and design information of the product. The information analysis system according to claim 1,
The information analysis system, wherein the objective variable is at least one of a yield, a defect rate, and a characteristic value of the product. The information analysis system according to claim 1,
The information analysis system, wherein the influence degree calculation unit calculates the influence degree by a partial least square method. The information analysis system according to claim 1,
An information analysis system comprising: a first estimation unit that estimates an objective variable corresponding to a predetermined combination of the explanatory variables using the degree of influence obtained by the degree of influence calculation unit. The information analysis system according to claim 1,
An interpolation influence calculation unit that calculates an interpolation influence degree by interpolation from the influence degree obtained by the influence degree calculation unit,
An information analysis system, comprising: a second estimation unit that estimates an objective variable corresponding to an explanatory variable other than the explanatory variable used when calculating the influence level using the interpolation influence level. The information analysis system according to claim 1,
An interpolation influence calculation unit that calculates an interpolation influence degree by interpolation from the influence degree obtained by the influence degree calculation unit,
An information analysis system comprising: a third estimation unit that estimates a plurality of objective variables corresponding to a plurality of explanatory variables having a predetermined probability distribution using the influence degree and the interpolation influence degree. A step of storing, as multi-valued information, explanatory variables that are at least one of history information relating to processing performed on the product and design information of the product;
Storing an objective variable which is at least one of the yield, defect rate and characteristic value of the product;
When the objective variable falls below or exceeds a predetermined reference value, converts the explanatory variable of the multi-value information to create a binary information matrix; and
An information analysis method comprising: calculating the degree of influence indicating the degree of influence of the explanatory variable on the objective variable by calculating the binary information matrix and the objective variable. A program that causes a computer to function as the information analysis system according to any one of claims 1 to 8.

Images