JP2010267947A - Process parameter selection apparatus, process parameter selection method, process parameter selection program, program recording medium, and production process management apparatus containing the process parameter selection apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To select process parameters that can create a prediction model for predicting quality with high serviceability and high precision in a production process including multiple production apparatuses having the same function. <P>SOLUTION: The process parameters collected from the production process are set as explanatory variables and inspection data collected from an inspection process is set as target variables. PLS by production apparatus is performed and the process parameters having average process parameter VIP values equal to or larger than a reference value are selected as process parameters common to the production apparatuses and related to the quality. The process parameters that can create a prediction model with high precision are selected thus. Further, it is possible to reduce the number of process parameters to be managed, share the process parameters among all of the production apparatuses, reduce the number of steps and time for managing the process parameters, and obtain high serviceability. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a process parameter selection apparatus, a process parameter selection method, a process parameter selection program, and a program recording medium for selecting an effective process parameter from a large number of process parameters in a device manufacturing process.

  The adjustment of the manufacturing conditions in the device manufacturing apparatus affects the quality of the product, such as the determination result of good / bad and the characteristic value. Therefore, in the inspection process, check whether the quality of each product is within the target range, and if the quality is out of the target range, adjust the manufacturing conditions so that the quality is within the target range. There is a need to.

  However, in order to inspect all the quality, there are problems that many man-hours and time are required, and depending on the inspection process, the product cannot be inspected unless the product is destroyed, and the quality of all the products cannot be inspected.

  Therefore, there is a technique for predicting the quality of a product being manufactured from process parameters that can be obtained from the manufacturing process, for example, measured values in a state such as the temperature and pressure inside the device during the manufacturing, which are measured by various measuring instruments. is there. As a technique for predicting such quality, there is a method using PLS (Partial Least Squares). By creating a prediction model of quality expressed by the PLS in the line format of the process parameter, the quality is predicted from the process parameter using the prediction model. Therefore, it is possible to determine whether the quality is within the target range based only on the process parameters, and it is possible to adjust the manufacturing conditions without going through the inspection process.

  Here, when the quality is predicted using the process parameters of the manufacturing process, if the process parameters that have nothing to do with the quality to be predicted are used, the quality prediction accuracy decreases. In addition, managing many process parameters requires a lot of man-hours and time.

  For this reason, a method has been proposed in which process parameters are selected prior to creating the quality prediction model and a prediction model is created with limited process parameters (Japanese Patent Laid-Open No. 2004-335841 (Patent Document 1). )).

  In the “prediction apparatus for plasma processing apparatus” disclosed in Patent Document 1, operation data (process parameters) and process result data obtained for each wafer process are stored in an operation data storage unit and a process result data storage unit. Collect and collect multi-variate analysis on the collected operation data by the analysis processing unit, select the operation data to be used for prediction, obtain the correlation between the selected operation data and the processing result data, On the basis of this, the state of the plasma processing apparatus or the state of the object to be processed is predicted using the operation data when the wafer other than the wafer having the correlation is processed.

  Here, principal component analysis is used as the multivariate analysis. Then, in the selection of the operation data used for the prediction, a principal component analysis is performed on the collected operation data, and a loading that is a relative contribution of the original operation data with respect to each obtained principal component axis is obtained. Operation data corresponding to loading within a preset distance from the principal component axis is selected.

  However, the conventional plasma processing apparatus prediction apparatus disclosed in Patent Document 1 has the following problems.

  In order to manufacture a large number of products in a short time, there may be a case where a manufacturing process including a large number of manufacturing apparatuses having the same function is provided in the production process. In this case, since there is a difference in process parameter variation between manufacturing apparatuses having the same function, it is not efficient to perform exactly the same adjustment. Therefore, there is a problem that it is necessary to perform appropriate adjustment for each manufacturing apparatus.

  Specifically, Patent Document 1 only discloses that the state is predicted for a single manufacturing apparatus. Therefore, in the device manufacturing process having a plurality of manufacturing apparatuses having the same function, when the process parameters are selected for each manufacturing apparatus by the process parameter selection method used for the prediction disclosed in Patent Document 1, manufacturing is performed. Different process parameters may be selected between devices. Therefore, it is necessary to set and maintain process parameters to be acquired and managed in advance for each manufacturing apparatus, which increases man-hours. In addition, it is necessary to adjust the manufacturing conditions in consideration of which process parameter is managed for each manufacturing apparatus, which increases the burden on the adjuster.

  Furthermore, when there are a large number of manufacturing apparatuses, it is not practical to grasp the process parameters managed in all the manufacturing apparatuses.

  When there are a large number of manufacturing apparatuses, if there are process parameters that greatly affect the quality of only a specific manufacturing apparatus in a plurality of manufacturing apparatuses having the same function, the manufacturing apparatus may be abnormal. high. Therefore, rather than making predictions using the process parameters of the manufacturing equipment, the manufacturing equipment should be adjusted. However, such a point is not considered at all in Patent Document 1.

  Further, as another method, by applying the process parameter selection method used for the prediction collectively disclosed in Patent Document 1 to a plurality of manufacturing apparatuses constituting one manufacturing process, it is common to all the manufacturing apparatuses. Process parameters can be selected.

  FIG. 11 (a) shows a scatter diagram showing the relationship between the process parameter 1 and quality, and FIG. 11 (b) shows a scatter diagram showing the relationship between the process parameter 2 and quality. In FIG. 11 (a) and FIG. 11 (b), the three manufacturing apparatuses are indicated by ●, ▲, and ■, respectively. In this case, if the data of each manufacturing apparatus is substantially translated on the quality-process parameter plane shown in FIG. 11A or FIG. It is assumed that it substantially overlaps with the data distribution area related to the manufacturing apparatus.

  The process parameter 1 has a strong correlation with the quality of the three manufacturing apparatuses, but there is no correlation with the quality when viewed for each manufacturing apparatus. On the other hand, the process parameter 2 has no correlation with the quality when the three manufacturing apparatuses are combined, but has a strong correlation with the quality when viewed for each manufacturing apparatus. Therefore, when the process parameter selection method used for prediction disclosed in Patent Document 1 is applied to the above three manufacturing apparatuses, process parameter 1 having no correlation with quality is selected for each manufacturing apparatus. Therefore, when viewed by each manufacturing apparatus, the process parameter 2 having a strong correlation with the quality is not selected. Therefore, there is a problem that process parameters for predicting quality cannot be appropriately selected.

JP 2004-335841 A

  Accordingly, an object of the present invention is to select a process parameter that enables creation of a prediction model that predicts product quality with high maintainability and high accuracy in a manufacturing process having a plurality of manufacturing apparatuses having the same function. A process parameter selection device, a process parameter selection method, a process parameter selection program, and a program recording medium are provided.

In order to solve the above-described problem, a process parameter selection device according to a first aspect of the present invention provides:
In a manufacturing process in which process processing is performed on a material by a plurality of manufacturing apparatuses having the same function, a process parameter acquisition unit that acquires process parameters from each of the manufacturing apparatuses,
Multivariate analysis is performed for each manufacturing apparatus using the acquired process parameters, and a comparison result between a statistical value related to the process parameter in the multivariate analysis result for each manufacturing apparatus and a preset reference value is used. All manufacturing apparatuses by selecting the process parameters satisfying preset setting conditions or by selecting the process parameters having desirable statistical values by using the comparison result between the statistical values regarding the process parameters. And a process parameter selection unit for selecting a common process parameter.

  According to the above configuration, the multivariate analysis is performed for each manufacturing apparatus using the acquired process parameter, and the process parameter is selected based on the statistical value related to the process parameter in the multivariate analysis result for each manufacturing apparatus. ing. Therefore, it is possible to select process parameters common to all the manufacturing apparatuses in the manufacturing process. As a result, specific operations and knowledge for each manufacturing apparatus are not required, and any manufacturing apparatus can be managed in the same manner, and man-hours and time for process parameter management can be reduced to obtain high maintainability. be able to.

  Furthermore, by using the comparison result between the statistical value and a preset reference value, by selecting the process parameter that satisfies a preset setting condition, or using the comparison result between the statistical values, The process parameter is selected by selecting the process parameter having the desired statistical value. Therefore, only effective process parameters in which the statistical value of the multivariate analysis result exceeds the reference value or the statistical value is higher can be selected. As a result, by creating a prediction model using the selected effective process parameter, it is possible to increase the prediction accuracy of the prediction model.

Further, in the process parameter selection device of one embodiment,
In an inspection process in which an inspection apparatus performs an inspection on a product that has undergone a process process in the manufacturing process, the apparatus includes a quality data acquisition unit that acquires quality data from the inspection apparatus,
The process parameter selection unit uses the acquired process parameter as an explanatory variable, and uses the acquired quality data as a target variable, performs analysis by PLS for each manufacturing apparatus, and relates to the process parameter in the analysis value by the PLS. By using a comparison result between a statistical value and a preset reference value, by selecting the process parameter that satisfies a preset setting condition, or using a comparison result between statistical values related to the process parameter, By selecting a process parameter having a desired statistical value, a process parameter that is common to all the manufacturing apparatuses and is related to quality is selected.

  According to this embodiment, the acquired process parameter is used as an explanatory variable, the acquired quality data is used as a target variable, analysis is performed by PLS for each manufacturing apparatus, and the process parameter in the analysis value by the PLS is related to Process parameters are selected based on statistical values. Therefore, it is possible to select only process parameters that are common to all manufacturing apparatuses and that are related to quality. As a result, by creating a prediction model using the selected process parameters, it is possible to increase the quality prediction accuracy of the prediction model.

The process parameter selection method of the second invention is
In a manufacturing process in which process processing is performed on a material by a plurality of manufacturing apparatuses having the same function, a process parameter is acquired from each manufacturing apparatus by a process parameter acquisition unit,
The process parameter selection unit performs multivariate analysis for each manufacturing apparatus using the acquired process parameter, and a statistical value related to the process parameter in the multivariate analysis result for each manufacturing apparatus and a preset reference value The process parameter having a desired statistical value is selected by selecting the process parameter satisfying a preset setting condition using the comparison result of the above, or by using the comparison result between the statistical values regarding the process parameter. Thus, the process parameters common to all the manufacturing apparatuses are selected.

  According to the above configuration, the multivariate analysis is performed for each manufacturing apparatus using the acquired process parameter, and the process parameter is selected based on the statistical value related to the process parameter in the multivariate analysis result for each manufacturing apparatus. ing. Therefore, it is possible to select process parameters common to all the manufacturing apparatuses in the manufacturing process. As a result, specific operations and knowledge for each manufacturing apparatus are not required, and any manufacturing apparatus can be managed in the same manner, and man-hours and time for process parameter management can be reduced to obtain high maintainability. be able to.

  Furthermore, by using the comparison result between the statistical value and a preset reference value, by selecting the process parameter that satisfies a preset setting condition, or using the comparison result between the statistical values, The process parameter is selected by selecting the process parameter having the desired statistical value. Therefore, only effective process parameters in which the statistical value of the multivariate analysis result exceeds the reference value or the statistical value is higher can be selected. As a result, by creating a prediction model using the selected effective process parameter, it is possible to increase the prediction accuracy of the prediction model.

In the process parameter selection method of one embodiment,
In the inspection process in which the inspection apparatus is inspected for the product processed by the manufacturing process, the quality data acquisition unit acquires quality data from the inspection apparatus,
While the acquired process parameter is used as an explanatory variable by the process parameter selection unit, the acquired quality data is used as a target variable, and the PLS is analyzed for each manufacturing apparatus, and the process parameter in the analysis value by the PLS is analyzed. By using the comparison result between the statistical value regarding the process parameter and the preset reference value, selecting the process parameter that satisfies the preset setting condition, or using the comparison result between the statistical values regarding the process parameter By selecting a process parameter having a desired statistical value, a process parameter that is common to all the manufacturing apparatuses and is related to quality is selected.

  According to this embodiment, the acquired process parameter is used as an explanatory variable, the acquired quality data is used as an objective variable, analysis is performed by PLS for each manufacturing apparatus, and the statistics on the process parameter in the analysis value by the PLS are analyzed. A process parameter is selected based on the value. Therefore, it is possible to select only process parameters that are common to all manufacturing apparatuses and that are related to quality. As a result, by creating a prediction model using the selected process parameters, it is possible to increase the quality prediction accuracy of the prediction model.

In the process parameter selection method of one embodiment,
As the analysis value by the PLS, the relative contribution of the acquired process parameter with respect to each component axis obtained by the PLS from the acquired process parameter for each manufacturing apparatus, and the contribution to the quality of each component axis The VIP value calculated from the degree is used.

  According to this embodiment, a VIP value is used as an analysis value by the PLS. Therefore, process parameters important for predicting quality can be extracted.

In the process parameter selection method of one embodiment,
A model creation unit creates a prediction model for predicting product quality for each manufacturing apparatus using the selected process parameters.
The model storage unit stores the prediction model created for each manufacturing device by the model creation unit for each manufacturing device.

  According to this embodiment, since a prediction model with high prediction accuracy can be created for each manufacturing apparatus, it is possible to analogize quality with less error.

In the process parameter selection method of one embodiment,
A process parameter is acquired from the manufacturing apparatus by the process parameter acquisition unit, and a prediction model corresponding to the manufacturing apparatus related to the acquired process parameter is selected from the prediction models stored in the model storage unit. ,
Using the selected prediction model and the acquired process parameters, the product quality prediction value is calculated,
When the calculated predicted value deviates from a preset setting range, an abnormality warning is output by the abnormality warning unit.

  According to this embodiment, since quality anomalies can be detected early, early feedback can be performed. Further, since the process parameters are carefully selected by selection, there are few abnormality factor candidates, and the abnormality factor can be detected early. Furthermore, since the selected process parameter is common to all the manufacturing apparatuses, the number of process parameters handled by the coordinator of the manufacturing apparatus can be reduced and the burden can be reduced.

In the process parameter selection method of one embodiment,
A process parameter is acquired from the manufacturing apparatus by the process parameter acquisition unit, and a prediction model corresponding to the manufacturing apparatus related to the acquired process parameter is selected from the prediction models stored in the model storage unit. ,
Using the selected prediction model and the acquired process parameters, the product quality prediction value is calculated,
The predicted quality output unit outputs a predicted value of the calculated quality.

  According to this embodiment, it is possible to generate and output a quality prediction result based on the process parameters acquired from the manufacturing apparatus. Therefore, the quality can be estimated prior to the quality inspection. Furthermore, even if it is difficult to inspect the quality of all products due to the high cost of quality inspection, the predicted quality of all products can be obtained.

In the process parameter selection method of one embodiment,
As a statistical value regarding the process parameter in the analysis value by the PLS, a value obtained by averaging the VIP value calculated for each manufacturing apparatus for each process parameter is used.
The process parameter having an average value related to the process parameter of the calculated VIP value equal to or greater than the reference value is selected as a process parameter common to all manufacturing apparatuses and related to quality. It has become.

  According to this embodiment, the process parameter having an average value related to the process parameter of the VIP value calculated for each manufacturing apparatus is selected to be greater than or equal to the reference value. Therefore, it is possible to extract process parameters that can improve the overall prediction accuracy represented by the average prediction accuracy of all the manufacturing apparatuses.

In the process parameter selection method of one embodiment,
As a statistical value regarding the process parameter in the analysis value by the PLS, a value obtained by averaging the VIP value calculated for each manufacturing apparatus for each process parameter is used.
The process parameters corresponding to a preset number of the process parameters that are common to all the manufacturing apparatuses, in the descending order of the average value related to the process parameters of the VIP value obtained, are process parameters that are common to all manufacturing apparatuses and that are related to quality. It comes to choose as.

  According to this embodiment, a predetermined number of process parameters are selected in descending order of the average value regarding the process parameters of the VIP values calculated for each manufacturing apparatus. Therefore, it is possible to extract process parameters that can improve the overall prediction accuracy represented by the average prediction accuracy of all the manufacturing apparatuses. Furthermore, even when the number of variables to be used is limited, it is possible to extract process parameters that improve the prediction accuracy within the limit.

In the process parameter selection method of one embodiment,
While gradually increasing the reference value, the process parameters selected by the respective reference values are used as explanatory variables, while the obtained quality data is used as an objective variable, analysis is performed by PLS for each of the manufacturing apparatuses, and the quality When the prediction model is created sequentially, the process parameter that maximizes the prediction accuracy of the sequentially created prediction model is selected as a process parameter that is common to all the manufacturing equipment and is related to quality. To do.

  According to this embodiment, when the reference values are sequentially increased and quality prediction models are sequentially generated using the process parameters selected by the respective reference values, Process parameters that maximize the prediction accuracy are selected. Therefore, it is possible to select the smallest number of process parameters that can create a prediction model in which the quality prediction result is the best.

The process parameter selection method of the third invention is
In a manufacturing process having a plurality of partial manufacturing processes where a process is performed on a material by a plurality of manufacturing apparatuses having the same function,
A combination of a plurality of manufacturing apparatuses used in each of the partial manufacturing processes is handled as a manufacturing apparatus for the manufacturing process, and the process parameter selection method of the second invention is applied to the manufacturing apparatus for the manufacturing process. It is characterized by.

  According to the above configuration, it is possible to select a highly correlated parameter for each of the materials.

The process parameter selection program of the fourth invention is
The computer is caused to function as a process parameter acquisition unit and a process parameter selection unit in the process parameter selection device of the first invention.

  According to the said structure, the process parameter common to all the manufacturing apparatuses in the said manufacturing process can be selected similarly to the case of the process parameter selection apparatus of the said 1st invention. Therefore, specific operation and knowledge for each manufacturing apparatus are not required, and any manufacturing apparatus can be managed in the same way, and man-hours and time for process parameter management can be reduced to obtain high maintainability. Can do.

  Furthermore, only effective process parameters in which the statistical value of the multivariate analysis result exceeds the reference value or the statistical value is higher can be selected. Therefore, by creating a prediction model using the selected effective process parameters, it is possible to increase the prediction accuracy of the prediction model.

The computer-readable program recording medium of the fifth invention is
The process parameter selection program of the fourth invention is recorded.

  According to the above configuration, by reading and executing the program recording medium by a computer, the process parameters common to all the manufacturing apparatuses in the manufacturing process can be selected as in the case of the process parameter selecting apparatus of the first invention. can do. Therefore, specific operation and knowledge for each manufacturing apparatus are not required, and any manufacturing apparatus can be managed in the same way, and man-hours and time for process parameter management can be reduced to obtain high maintainability. Can do.

  Furthermore, only effective process parameters in which the statistical value of the multivariate analysis result exceeds the reference value or the statistical value is higher can be selected. Therefore, by creating a prediction model using the selected effective process parameters, it is possible to increase the prediction accuracy of the prediction model.

  As is clear from the above, according to the first, second, fourth, and fifth inventions, the multivariate analysis is performed for each manufacturing apparatus using the acquired process parameters, and the multivariate for each manufacturing apparatus is determined. A process parameter is selected based on a statistical value related to the process parameter in the analysis result. Therefore, it is possible to select process parameters common to all the manufacturing apparatuses in the manufacturing process. As a result, specific operations and knowledge for each manufacturing apparatus are not required, and any manufacturing apparatus can be managed in the same manner, and man-hours and time for process parameter management can be reduced to obtain high maintainability. be able to.

  Furthermore, a process parameter is selected based on a comparison result between the statistical value and a preset reference value or a comparison result between the statistical values. Therefore, only effective process parameters in which the statistical value of the multivariate analysis result exceeds the reference value or the statistical value is higher can be selected. As a result, by creating a prediction model using the selected effective process parameter, it is possible to increase the prediction accuracy of the prediction model.

  In addition, according to the third invention, in a manufacturing process having a plurality of partial manufacturing processes in which a process is performed on a material by a plurality of manufacturing apparatuses having the same function, the plurality of the above-described parts used in each partial manufacturing process Since a combination of manufacturing apparatuses is handled as the apparatus of the manufacturing process, an effective common process parameter can be selected from a large number of process parameters from the plurality of partial manufacturing processes. Therefore, it is not necessary to select process parameters for each of the enormous combinations of the manufacturing apparatuses due to an increase in manufacturing processes, and high maintainability can be obtained.

  In addition, when a prediction model is created using the effective process parameters selected by the first to fifth inventions, and the prediction result using the prediction model indicates an abnormality, it is necessary to adjust the process parameters. There is. At this time, since process parameters common to all the manufacturing apparatuses are selected, there are few process parameters to be adjusted, and the adjustment burden is reduced.

  Furthermore, even if 100% inspection of products is difficult or it is very expensive to inspect the quality of products, it is possible to predict the quality of all products with high prediction accuracy based on the above process parameters. it can.

It is a block diagram in the manufacturing process management apparatus carrying the process parameter selection apparatus of this invention. It is a block diagram of the production process containing the manufacturing process managed by the manufacturing process management apparatus shown in FIG. It is a flowchart of the prediction model creation process operation performed by the model creation part in FIG. It is a figure which shows an example of the format of a data set. It is a figure which shows the VIP value of each prediction model created from a process parameter, this process parameter, and the average value of this VIP value. It is a figure which shows the VIP value and average value of the process parameter following each prediction model, and each prediction model. It is the figure which arranged the process parameter and the average value of VIP value of each prediction model created from this process parameter in descending order from the above-mentioned highest average value. It is explanatory drawing of the method of putting together two partial manufacturing processes into one manufacturing process. It is explanatory drawing when the process position which processes a some lot simultaneously exists in two or more in one manufacturing apparatus. It is a block diagram in the manufacturing process management apparatus different from FIG. It is a scatter diagram which shows an example of the relationship between a process parameter and quality.

-1st Embodiment Hereinafter, this invention is demonstrated in detail by embodiment of illustration. FIG. 1 is a block diagram of a manufacturing process management apparatus equipped with a process parameter selection apparatus according to the present embodiment. FIG. 2 is a schematic configuration diagram of a production process including a manufacturing process managed by the manufacturing process management apparatus shown in FIG.

  As shown in FIG. 1, the manufacturing process management apparatus 1 includes a process parameter acquisition unit 2 that acquires various data (hereinafter referred to as process parameters) from the manufacturing apparatus of the manufacturing process 10 in the production process 9, and the production process 9. A quality data acquisition unit 3 that acquires an inspection result (hereinafter referred to as quality data) from an inspection apparatus in the inspection step 11, a process parameter storage unit 4 that stores the acquired process parameter, and the acquired quality data. A quality data storage unit 5 for storing, a model creation unit 6 for creating a prediction model for predicting product quality from the process parameters based on the process parameters and the quality data, and a model storage for storing the created prediction models Detecting that the quality predicted by the unit 7 and the prediction model is out of the predetermined target range And an abnormality warning unit 8 for outputting an abnormality warning.

As shown in FIG. 2, the production process 9 includes a manufacturing process 10 for manufacturing a product, and an inspection process 11 for inspecting the product after the manufacturing by the manufacturing process 10 is finished. The manufacturing process 10 includes A first manufacturing process 12 a, second manufacturing process 12 b,... A partial manufacturing process consisting of the A manufacturing process 12 c (partial manufacturing process 12 when represented by an arbitrary partial manufacturing process). Say). Further, each partial manufacturing process 12 includes a plurality of manufacturing apparatuses 13 that perform the same operation. For example, the Ath manufacturing process 12c includes L manufacturing apparatuses 13 including a first manufacturing apparatus 13 ₁ , a second manufacturing apparatus 13 ₂ ,..., An Lth manufacturing apparatus 13 _L. And in each partial manufacturing process 12, it processes by any one manufacturing apparatus 13, and progresses to the following test process 11. FIG.

  A product is managed with this lot number by assigning a number to each unit quantity (hereinafter referred to as a lot). The inspection process 11 includes B partial inspection processes consisting of a first inspection process 14a, a second inspection process 14b,..., A B manufacturing process 14c (referred to as a partial inspection process 14 when represented by an arbitrary partial inspection process). It consists of

  Based on the process parameters read from the process parameter storage unit 4 and the quality data read from the quality data storage unit 5, the model creation unit 6 of the manufacturing process management apparatus 1 having the above configuration performs quality by multivariate analysis. Create a prediction model that predicts. In that case, the manufacturing process 10 includes a plurality of partial manufacturing processes 12 including a plurality of manufacturing apparatuses 13 that perform the same operation. Therefore, the model creation unit 6 creates a prediction model for each manufacturing apparatus 13. That is, the prediction model in the present embodiment is a prediction formula for predicting quality from the process parameters.

  Therefore, the model storage unit 7 stores the prediction model created by the model creation unit 6 for each manufacturing apparatus 13 for each manufacturing apparatus 13.

  The abnormality warning unit 8 acquires a process parameter from the manufacturing process 10 and uses a prediction model corresponding to a manufacturing apparatus that is a target of quality prediction in the prediction model stored in the model storage unit 7, and uses a prediction model for quality. Is calculated. When the obtained quality prediction value is out of a predetermined target range, an abnormality warning is output.

  If it does so, the adjuster who received the warning of abnormality will adjust the manufacturing conditions regarding the corresponding manufacturing apparatus 13 in order to keep quality in the target range.

  Hereinafter, creation of a prediction model by the model creation unit 6 will be described in more detail. FIG. 3 shows a flowchart of the prediction model creation processing operation performed by the model creation unit 6. The model creation unit 6 selects and uses effective process parameters that improve the quality prediction accuracy, and creates a prediction model for each manufacturing apparatus 13. However, the process parameters used for creating the prediction model of each manufacturing apparatus 13 are common to one partial manufacturing process 12. Hereinafter, according to the flowchart shown in FIG. 3, a method for creating a prediction model that improves prediction accuracy under the condition that process parameters common to the respective manufacturing apparatuses 13 are used will be described.

  In step S1, an initial setting for creating a data set of process parameters and quality data for each manufacturing apparatus 13 is performed. In step S2, PLS using the process parameters as explanatory variables and the quality data as objective variables is executed for each manufacturing apparatus 13 for each data set created in step S1. In step S3, the average VIP (Variable importance in the projection) value is calculated for each process parameter of each data set.

  In step S4, the variable “V” representing the reference value of the VIP value set in the work area of the memory (not shown) and the variable “P0” representing the prediction accuracy of the previous prediction model are set to “0”. Is stored. In step S5, a process parameter whose average VIP value calculated in step S3 is V or more is selected. In step S6, for each data set created in step S1, PLS is executed for each manufacturing apparatus 13 using only the process parameters selected in step S5 as explanatory variables. In step S7, the prediction accuracy is calculated for each prediction model, that is, for each PLS executed in step S6 (for each manufacturing apparatus 13). Further, an average value of prediction accuracy for all prediction models (all manufacturing apparatuses 13) is calculated and stored in a variable “P1” representing the prediction accuracy of the current prediction model set in the work area.

  In step S8, the prediction accuracy “P0” of the previous prediction model is compared with the prediction accuracy “P1” of the current prediction model. If P0 is larger than P1, the process proceeds to step S10. Otherwise, the process proceeds to step S9. . In step S9, the value of the variable “P0” is updated with the value of the variable “P1”, and the value of the variable “V” is incremented by “0.1”. After that, the process parameters are selected by returning to step S5.

  In step S10, since the prediction accuracy “P1” of the current prediction model is lower than the prediction accuracy “P0” of the previous prediction model, the previous prediction model created by the PLS executed in step S6 is the previous prediction model. The manufacturing process management apparatus 1 outputs the prediction model used when predicting the quality of the product based on the process parameters from the manufacturing process 10. After that, the prediction model creation processing operation is terminated.

  That is, in the present embodiment, the process parameter selection unit in the above claims is configured by the above-described steps S1 to S9 of the prediction model creation processing operation by the model creation unit 6.

  Here, the “objective variable” is a variable to be predicted, and the “explanatory variable” is a variable that is a candidate factor of the objective variable.

  Hereinafter, the processing content in each step in FIG. 3 will be specifically described.

  (A) Step S1: A data set of process parameters and quality data is created for each manufacturing apparatus 13.

Hereinafter, the case where the manufacturing process for which the prediction model is created is the A-th manufacturing process 12c in FIG. 2 will be described as an example. In this case, the number of the manufacturing apparatus 13 constituting the second A manufacturing process 12c, the first manufacturing apparatus 13 _1, second manufacturing apparatus 13 _2, ... becomes an L stage of the L manufacturing apparatus 13 _L.

  In the following description, the unit quantity (lot) is 1, and a lot number is added to each product (material). Therefore, in the following description, individual products (materials) may be referred to as “lots”.

  Process parameters from the A manufacturing process 12c obtained when the manufacturing process is performed on the material in the A manufacturing process 12c, and the first inspection in the inspection process 11 for the product manufactured in the A manufacturing process 12c. L data sets including the quality data inspected in any of the partial inspection steps 14 in the step 14a, the second inspection step 14b,...

The number of materials processed in each of the first manufacturing apparatus 13 ₁ to the Lth manufacturing apparatus 13 _L is S _{1 to} S _L , the number of process parameters that can be acquired in the manufacturing process 10 is N, and the quality predicted in the inspection process 11 Let M be the number of. Also, the format of the first manufacturing apparatus 13 _first data set, as shown in FIG. 4, vertically lot number (1 to S _L), the process in the horizontal direction parameter 1 process parameters N and Quality 1 Quality M Is the ordered form.

  (B) Step S2: For each data set, PLS is executed with process parameters as explanatory variables and quality data as objective variables.

PLS, which is one of multivariate analysis, is performed on each of the L data sets created in step S1. PLS is a technique for creating a prediction model for predicting the objective variable Y from the explanatory variable X. For example, the in the case of the first manufacturing apparatus 13 ₁ in FIG. 4, "explanatory variables X" is a matrix of S ₁ × N representing the process parameters, "objective variable Y" S ₁ × M representing the quality Is a matrix.

（２） ａ＝１
（３） ｕ_a＝Ｙ_a-1の分散最大の列
（４） ｗ_a＝Ｘ_a-1 ^Tｕ_a/‖Ｘ_a-1 ^Tｕ_a-1‖
（５） ｔ_a＝Ｘ_a-1ｗ_a
（６） ｃ_a＝Ｙ_a-1 ^Tｔ_a/‖Ｙ_a-1 ^Tｔ_a‖
（７） ｕ_a＝Ｙ_a-1ｃ_a
（８） 直前のｕ_aと比較して収束したら(９)へ、そうでなければ上記(４)へ
（９） ｐ_a＝Ｘ_a-1 ^Tｔ_a/‖ｔ_a‖²
（１０）ｑ_a＝Ｙ_a-1 ^Tｔ_a/‖ｔ_a‖²
（１１）Ｘ_a＝Ｘ_a-1−ｔ_aｐ_a ^T
（１２）Ｙ_a＝Ｙ_a-1−ｔ_aｑ_a ^T
（１３）次のＰＬＳ成分が不要であれば終了、必要であれば「ａ＝ａ＋１］として上記(３)へ The PLS algorithm is shown below.

(2) a = 1
(3) Maximum variance of u _a = Y _a-1 (4) w _a = X _a-1 ^T u _a / ‖X _a-1 ^T u _a-1 ‖
(5) t _a = X _a-1 w _a
(6) c _a = Y _a-1 ^T t _a / ‖Y _a-1 ^T t _a
(7) u _a = Y _a-1 c _a
(8) Go to (9) if converged compared to the previous u _a , otherwise go to (4) (9) p _a = X _a-1 ^T t _a / ‖t _a ‖ ²
(10) q _a = Y _a-1 ^T t _a / ‖t _a ‖ ²
(11) X _a = X _a-1 −t _a p _a ^T
(12) Y _a = Y _a-1 −t _a q _a ^T
(13) If the next PLS component is not required, the process is terminated.

Here, “t” and “u” are latent variables of the explanatory variable X and the objective variable Y, “p” and “c” are loading vectors, “w” is a PLS weight vector, and “q” is t. Regression coefficient. The subscript “a” represents the a-th PLS component. Moreover, X ^T represents a transpose matrix of X.

ここで、Ｗ＝(ｗ₁,ｗ₂,…,ｗ_a)、Ｐ＝(ｐ₁,ｐ₂,…,ｐ_a)、Ｑ＝(ｑ₁,ｑ₂,…,ｑ_a)である。また、「Ｂ」は重回帰式における各プロセスパラメータに対する係数を表わし、「ｂ^T」は重回帰式における切片を表わす。 The expression of y ^ ^{T by} the multiple regression equation is as follows.

_{Here, W = (w 1, w} 2, ..., w a), P = (p 1, p 2, ..., p a), Q = (q 1, q 2, ..., q a) is. “B” represents a coefficient for each process parameter in the multiple regression equation, and “b ^T ” represents an intercept in the multiple regression equation.

  Prediction error is used to determine the number of components of PLS. Hereinafter, a method for calculating a prediction error using the cross-validation method will be described.

ここで、ｙ_obsはｙの実測値、ｙ^_predはバリデーションによる予測値である。上記PRESSが最小のＰＬＳ成分数を、最適な成分とする。 When there are n lots, except for one lot, a one-component PLS model is calculated for the remaining (n-1) lots. The estimated value y ^ _pred of one lot removed using the model is obtained. Change the lot to be removed and find y ^ _pred of n lots. Then, a prediction error PRESS (Prediction Residual Sum of Squares) is obtained using y ^ _pred as follows.

Here, y _obs is a measured value of y, and y ^ _pred is a predicted value by validation. The number of PLS components having the smallest PRESS is set as the optimum component.

  (C) Step S3: The average VIP value is calculated for each process parameter.

ここで、ＤはＰＬＳの成分数を、ｄはＰＬＳの各成分を、ｎは各プロセスパラメータを、ＳＳＹ_dはＹ_dの偏差平方和を、ｗ_dnはベクトルｗ_dのｎ番目の要素を表す。 The calculation formula of the VIP value is shown below.

Here, D is the number of PLS components, d is each component of PLS, n is each process parameter, SSY _d is the deviation sum of squares of Y _d , and w _dn is the nth element of the vector w _d. .

  According to the above calculation formula, VIP values of L prediction models are obtained for each process parameter. Then, an average is calculated as a statistical value of the VIP value for each process parameter.

  (D) Step S4: The initial value “0” is stored in the reference value “V” of the VIP value and the prediction accuracy “P0” of the previous prediction model.

  As will be described below, steps S5 to S9 are looped, and as the initial value of the loop, the value 0 is set to the reference value V of the VIP value, and the value 0 is set to the prediction accuracy P0 of the previous prediction model. .

  (E) Step S5: Select a process parameter whose average VIP value is V or more.

  (F) Step S6: For each data set, PLS is executed using only the selected process parameters as explanatory variables.

  For each of the L data sets created in step S1, only the process parameter selected in step S5 is used as an explanatory variable, and the same quality data as in step S2 is used as a target variable. I do. In this way, L PLS models are created.

  (G) Step S7: Prediction accuracy is calculated for each prediction model (that is, for each manufacturing device 13), and an average value of prediction accuracy for all prediction models (all manufacturing devices 13) is calculated.

The prediction accuracy is calculated by the cross-validation method as in the case of the prediction error calculated in (b) step S2. The calculation formula of the prediction accuracy Q ² of each prediction model is shown below.

  After calculating the prediction accuracy for each prediction model (for each manufacturing device 13) by the above formula, an average value is calculated as a statistical value of the prediction accuracy for all prediction models (all manufacturing devices 13), and the prediction accuracy of the current prediction model is calculated. Store in P1.

  (H) Step S8: The prediction accuracy “P0” of the previous prediction model is compared with the prediction accuracy “P1” of the current prediction model. If P0> P1, the process proceeds to step S10, and if not, the process proceeds to step S9. .

  The prediction accuracy “P0” of the previous prediction model is compared with the prediction accuracy “P1” of the current prediction model. If P0 is larger, reducing the process parameters further reduces the prediction accuracy. Determination is made and the process proceeds to step S10. Otherwise, it is determined that there is still a possibility that the process parameters can be further reduced to improve the prediction accuracy, and the process proceeds to step S9.

  (I) Step S9: The value of the variable “P0” is updated with the value of the variable “P1”, and the value of the variable “V” is incremented by “0.1”.

  Since the prediction accuracy “P1” of the current prediction model is higher than the prediction accuracy “P0” of the previous prediction model, the value “P0” is stored in “P1”. Then, 0.1 is added to the reference value V of the VIP value in order to tighten the process parameter selection conditions. After that, the process returns to step S5.

  (J) Step S10: The previous prediction model created in the previous step S6 is output.

  Since the prediction accuracy “P1” of the current prediction model is lower than the prediction accuracy “P0” of the previous prediction model, the manufacturing process management device 1 uses the manufacturing process 10 as the previous prediction model created in the previous step S6. Output as a prediction model to be used when predicting product quality based on the process parameters.

  Therefore, the process parameter selection unit in the present embodiment uses the process parameter selected in the previous step S5 as a process parameter that enables creation of a prediction model used by the manufacturing process management apparatus 1.

  Hereinafter, the process parameter selection method will be described by taking specific process parameters and quality data as examples.

The conditions of the partial manufacturing process 12 that are targets for creating the prediction model in this example are as follows.
The number L of manufacturing apparatuses in the partial manufacturing process 12 = 8 (first manufacturing apparatus 13 ₁ to eighth manufacturing apparatus 13 ₈ )
The number of process parameters N of the partial manufacturing process 12 = 41 (X1 to X41)
-Quality of inspection process 11 (number of partial inspection processes) M = 1 (Y1)
・ Number of lots for each production equipment 13
S ₁ = 70, S ₂ = 72, S ₃ = 91, S ₄ = 105,
S ₅ = 85, S ₆ = 73, S ₇ = 60, S ₈ = 48

First, a PLS model for each manufacturing apparatus 13 is created using all 41 process parameters. 5 and 6 show the VIP value of each prediction model created for each process parameter and the average value of the VIP values. In FIG. 5 and FIG. 6, 41 process parameters are arranged in the vertical direction, and the VIP value of each prediction model and the VIP value of the model average are arranged in the horizontal direction. Incidentally, it indicates that M1~M8 lateral is predictive model of the first manufacturing apparatus 13 ₁ through eighth manufacturing apparatus 13 _8, D of VIP [D] represents the number of PLS components. The blanks in FIG. 5 and FIG. 6 represent data that could not be obtained or data that did not fluctuate, are not treated as “0”, and are not used when obtaining an average value.

FIG. 7 relates to FIG. 5 and FIG. 6, for each VIP value, average the values of the first manufacturing device 13 ₁ to the eighth manufacturing device 13 ₈ , and in descending order from the highest average value of the VIP values. It is rearranged. In FIG. 7, the rank is shown in the first column, the process parameters corresponding to the ranks are shown in the second column, and the average value of the VIP values related to the process parameters is shown in the third column.

First, the reference value V of the VIP value is set to “0”, and in FIG. 7, a process parameter whose average value of VIP values is the reference value V = 0 or more is selected. Then, the prediction accuracy Q ² of the prediction model using the selected process parameters, that is, all the process parameters is obtained. Table 1 shows the prediction accuracy of each manufacturing apparatus 13 and the average prediction accuracy of all manufacturing apparatuses.

Subsequently, process parameter selection criteria are tightened. In step S9 in the flowchart shown in FIG. 3, the reference value V of the VIP value is incremented by "0.1". In this example, the reference value V of the VIP value is set to an appropriate value. . The same applies to the subsequent prediction models. In this case, assuming that the reference value V of the VIP value is 0.8, in FIG. 7, the top 20 process parameters whose average value of the VIP value is the reference value V = 0.8 or more are selected. Then, a prediction model is created using the selected process parameter, and the prediction accuracy Q ² is obtained.

Table 2 shows the prediction accuracy of each manufacturing apparatus 13 and the average prediction accuracy of all the manufacturing apparatuses using process parameters having an average VIP value of 0.8 or more. In the average prediction accuracy, the values in parentheses are expressed up to 4 digits after the decimal point, and the values outside the parentheses are values rounded off to the fourth decimal place.

The average prediction accuracy Q ² predictive models 2 "0.745" is greater than the average "0.663" prediction accuracy Q ² of the above-described prediction model 1. Therefore, the prediction model 2 has an average prediction accuracy higher than that of the prediction model 1.

Subsequently, assuming that the reference value V of the VIP value is 1.0, in FIG. 7, the top 12 process parameters whose average value of the VIP value is the reference value V = 1.0 or more are selected. Then, a prediction model is created using the selected process parameter, and the prediction accuracy Q ² is obtained.

Table 3 shows the prediction accuracy of each manufacturing apparatus 13 and the average prediction accuracy of all manufacturing apparatuses using process parameters having an average VIP value of 1.0 or more. In the average prediction accuracy, the values in parentheses are expressed up to 4 digits after the decimal point, and the values outside the parentheses are values rounded off to the fourth decimal place.

The average “0.7452” of the prediction accuracy Q ² of the prediction model 3 is larger than the average “0.7448” of the prediction accuracy Q ² of the prediction model 2 described above. Therefore, the prediction model 3 has an average prediction accuracy higher than that of the prediction model 2.

Subsequently, the reference value V of the VIP value is set to 1.2, and in FIG. 7, the top eight process parameters whose average value of VIP values is the reference value V = 1.2 or more are selected. Then, a prediction model is created using the selected process parameter, and the prediction accuracy Q ² is obtained.

Table 4 shows the prediction accuracy of each manufacturing apparatus 13 and the average prediction accuracy of all the manufacturing apparatuses using process parameters having an average VIP value of 1.2 or more.

The average “0.754” of the prediction accuracy Q ² of the prediction model 4 is larger than the average “0.745” of the prediction accuracy Q ² of the prediction model 3 described above. Therefore, the prediction model 4 has an average prediction accuracy higher than that of the prediction model 3.

Subsequently, the reference value V of the VIP value is set to 1.3, and in FIG. 7, the top four process parameters whose average value of the VIP value is the reference value V = 1.3 or more are selected. Then, a prediction model is created using the selected process parameter, and the prediction accuracy Q ² is obtained.

Table 5 shows the prediction accuracy of each manufacturing apparatus 13 and the average prediction accuracy of all the manufacturing apparatuses using process parameters having an average VIP value of 1.3 or more.

The average “0.597” of the prediction accuracy Q ² of the prediction model 5 is smaller than the average “0.754” of the prediction accuracy Q ² of the prediction model 4 described above. Therefore, the prediction model 5 has a smaller average of prediction accuracy than the prediction model 4.

  Therefore, when the manufacturing process management device 1 predicts the quality of the product based on the process parameters from the manufacturing process 10 based on the prediction model 4 created using the process parameters having the average value of the VIP value of 1.2 or more last time. It is adopted as a prediction model used for

Table 6 shows a summary of the above five analysis results. Table 1 column in 6 each prediction model and condition, the second column the number of process parameters used to create predictive models, the third column represents the average value of the prediction accuracy Q ² of models created.

In the prediction model 1 to the prediction model 4, the prediction accuracy increases as the reference value V of the VIP value is set higher. In contrast, in the case of the prediction model 5 also reduces the number of process parameters by increasing the reference value V of VIP value, prediction accuracy Q ² is is lower. This indicates that when the number of process parameters is too small, the prediction accuracy Q ² tends to be low. The selection criteria for selecting the process parameters, it is desirable to set such that the highest prediction accuracy Q ² becomes high. Therefore, in this example, as can be seen from Table 6, the best prediction model can be created when the VIP value is 1.2 or more.

  As described above, in the present embodiment, process parameters are acquired by the process parameter acquisition unit 2 from the partial manufacturing process 12 including a plurality of manufacturing apparatuses 13 having the same function. Further, quality data is acquired by the quality data acquisition unit 3 from the partial inspection step 14 for inspecting the product manufactured by the partial manufacturing step 12.

  Then, the model creation unit 6 creates a data set of process parameters and quality data for each manufacturing apparatus 13, and performs PLS with the process parameters as explanatory variables and the quality data as objective variables for each of the created data sets. Then, the average of the VIP values is calculated for each process parameter. Further, the reference value “V” of the VIP value and the prediction accuracy “P0” of the previous prediction model are set to the initial value 0.

  After that, a process parameter whose average value of the calculated VIP values is V or more is selected, and for each data set, PLS is performed using only the selected process parameter as an explanatory variable for each prediction model (manufacturing apparatus). The prediction accuracy is calculated every 13), and the average value of the prediction accuracy for all prediction models (all manufacturing apparatuses 13) is calculated to be “P1”. When the prediction accuracy “P0” of the previous prediction model is less than or equal to the prediction accuracy “P1” of the current prediction model, “P0” is updated with “P1” and “V” is only “0.1”. While incrementing, the selection of the process parameter and the comparison between “P0” and “P1” are repeated.

  When “P1” becomes lower than “P0”, the previous prediction model is output as the prediction model for the manufacturing process management apparatus 1.

  Accordingly, it is possible to select only effective process parameters by reducing process parameters not related to quality from among the process parameters acquired from the partial manufacturing process 12. As a result, by creating a prediction model using the selected process parameters, it is possible to increase the accuracy of quality prediction using the prediction model.

  In addition, the number of process parameters to be managed can be reduced, man-hours and time for process parameter management can be reduced, and high maintainability can be obtained. In addition, since the process parameters can be made common to all the manufacturing apparatuses 13 in the partial manufacturing process 12, specific operation and knowledge for each manufacturing apparatus are not required, and any manufacturing apparatus is managed in the same manner. It is possible to reduce the man-hours and time for process parameter management and to obtain high maintainability.

  In this way, by applying the process parameter selection method of the present embodiment, it is possible to select process parameters common to all the manufacturing apparatuses, and to reduce process parameter management work. Furthermore, prediction accuracy can be improved.

  As mentioned above, although preferred embodiment in this invention was described referring drawings, this invention is not limited to the said embodiment.

  For example, in step S3 in the flowchart shown in FIG. 3, a value obtained by performing PLS using each process parameter, such as a coefficient, can be used instead of the VIP value. In addition to the average used in the above embodiment, for example, a median or standard deviation can be used.

  In step S4, the initial value of the reference value V of the VIP value is set to “0”. However, a positive number such as “0.5” may be set from the beginning.

  Further, when the process parameters are selected in step S5, if the number of process parameters to be selected in advance is determined, the predetermined number is selected in order from the top of the average value of the VIP values. A prediction model may be created using the process parameters.

In the above-described step S7, it calculates the prediction accuracy Q ² as a value for comparison between the previous predictive model, and calculates an average value for all manufacturing equipment. However, the invention is not limited to prediction accuracy Q ^2, no problem even by using a value that can be calculated for each prediction model by performing PLS as the prediction error. Furthermore, in addition to the average value, for example, a median value or a standard deviation may be used.

  In step S8, the prediction accuracy of the previous prediction model is compared with the prediction accuracy of the current prediction model, but the present invention is not limited to this. For example, it may be compared whether the target prediction accuracy is close. Alternatively, it may be compared with the prediction accuracy of the previous prediction model.

  In step S9, the reference value V of the VIP value is incremented by “0.1” in order to narrow down the number of process parameters to be selected. However, the present invention is not limited to “0.1” and may be any positive number. Alternatively, the increase value of the reference value V may be varied so that the number of process parameter reductions is constant. Also, instead of increasing the reference value V of the VIP value, the number of process parameters is reduced by one, a prediction model is created for each process parameter, and the one with the best prediction accuracy is selected. It is also possible to select the best prediction model.

  In this embodiment, PLS is used as multivariate analysis, but it is also possible to use principal component analysis or a neural network, for example.

  When the principal component analysis is used, if there is no quality data in step S1 in the flowchart shown in FIG. 3, the data set can be created using only process parameters. In step S3, instead of the PLS VIP value, a value obtained by performing principal component analysis on each process parameter, such as loading on the principal component axis, can be used.

  When a neural network is used as the multivariate analysis, the neural network is applied to each process parameter, for example, as a coefficient, instead of the PLS VIP value in step S3 in the flowchart shown in FIG. The value obtained by doing so can be used.

  In this embodiment, PLS which is the same multivariate analysis is executed in step S2 and step S6 in the flowchart shown in FIG. However, the present invention is not limited to this, and different multivariate analyzes may be performed in step S2 and step S6. Similarly, instead of outputting the previous prediction model in step S10, a prediction model is generated and output by multivariate analysis different from that in step S6 using the process parameter selected last time. It doesn't matter.

-2nd Embodiment This Embodiment is related with the case where the manufacturing process used as the object of prediction model preparation is a some partial manufacturing process. Hereinafter, the case where the process parameter selection method of the present invention is applied to the two partial manufacturing steps 12 of the first manufacturing step 12a and the second manufacturing step 12b in FIG. 2 will be described as an example.

When the process parameter selection method of the present invention is applied to the two partial manufacturing steps 12 of the first manufacturing step 12a and the second manufacturing step 12b in FIG. 2, the first manufacturing step 12a is performed as shown in FIG. And the second manufacturing process 12b are combined into one partial manufacturing process. In addition, the 1st manufacturing process 12a is comprised with L ₁ manufacturing apparatuses, and the 2nd manufacturing process 12b is comprised with L ₂ manufacturing apparatuses. Then, the material processed by the manufacturing apparatus 1 of the first production process 12a is then treated with either of the manufacturing apparatus 1 production apparatus L ₂ of the second manufacturing step 12b. Also, material that has been processed in the manufacturing apparatus 2 of the first manufacturing step 12a are then treated with either of the manufacturing apparatus 1 production apparatus L ₂ of the second manufacturing step 12b. The same applies hereinafter.

And the 12th manufacturing process 12ab is produced combining the said 1st manufacturing process 12a and the 2nd manufacturing process 12b. As a result, the manufacturing apparatus 1 in the twelfth manufacturing process 12ab is a combination of the manufacturing apparatus 1 in the first manufacturing process 12a and the manufacturing apparatus 1 in the second manufacturing process 12b. The manufacturing apparatus 2 in the twelfth manufacturing process 12ab is a combination of the manufacturing apparatus 1 in the first manufacturing process 12a and the manufacturing apparatus 2 in the second manufacturing process 12b. In this way, each combination of the manufacturing apparatus of the first manufacturing process 12a and the manufacturing apparatus of the second manufacturing process 12b is regarded as one manufacturing apparatus, and the twelfth twelfth manufacturing apparatus configured by (L ₁ × L ₂ ) manufacturing apparatuses. It is handled as the manufacturing process 12ab. At this time, it is possible to create a prediction model capable of improving the quality prediction accuracy from all the process parameters obtained in the first manufacturing step 12a and all the process parameters obtained in the second manufacturing step 12b. Choose valid process parameters. The same applies to the case where the prediction model creation process is applied to three or more partial manufacturing steps 12.

  That is, the “manufacturing process” described in the claims corresponds to a manufacturing process in which a partial manufacturing process and a plurality of partial manufacturing processes in the present embodiment are combined.

  The preferred embodiments of the present invention have been described above with reference to the drawings, but the present invention is not limited to the above-described embodiments.

  For example, when the process parameter selection method of the present invention is applied to two partial manufacturing steps 12, the present invention can be applied to each partial manufacturing step 12 separately. In that case, first, the present invention is applied to each partial manufacturing process 12 separately, and effective process parameters are selected for each partial manufacturing process 12. Thereafter, the partial manufacturing steps 12 are combined into one partial manufacturing step, and the present invention is applied to all the process parameters selected for each partial manufacturing step 12 to select effective process parameters. In this way, the process parameters can be selected in two stages (or two or more stages).

  Further, the present invention can be applied by dividing the process parameters into several groups regardless of the partial manufacturing step 12. In that case, first, the present invention is applied to each process parameter group to select an effective process parameter, and then this process is again performed for all process parameters selected for each process parameter group. Apply the invention to select valid process parameters. In this way, the present invention is applied in two stages (or two or more stages). In that case, as a method of dividing the process parameter, it is possible to divide by a certain standard such as grouping highly correlated ones or randomly.

Third Embodiment This embodiment relates to a case where a processing lot, external requirements, manufacturing conditions, time conditions, and the like change in the manufacturing apparatus 13 constituting the partial manufacturing process 12 of the manufacturing process 10 described above.

For example, in the manufacturing apparatus 13 ₁ of the A manufacturing process 12c shown in FIG. 2, as shown in FIG. 9, in order to process a plurality of lots of the first lot 13 _1a to the fourth lot 13 _1d at the same time, the manufacturing apparatus 13 Assume a case where there are a plurality of positions for processing lots in _one . In Figure 9, there is a processing position for processing four lots in the production device 13 _1. In this case, consider the respective processing positions for each lot 13 _1a to 13 _1d and production equipment, handling the manufacturing apparatus 13 ₁ as a plurality of manufacturing apparatuses is to apply the present invention.

Thus, by treating the processing position, respectively in one manufacturing apparatus 13 ₁ as a manufacturing apparatus, the processing position is shown in FIG. 11 ●, ▲, even divided as the ■, in FIG. 11 (a) Instead of the process parameter 1 having no correlation for each processing position as shown, the process parameter 2 having a strong correlation for each processing position as shown in FIG. 11B can be selected.

  Further, in one manufacturing apparatus 13, the state in the manufacturing apparatus 13 may change due to external factors such as maintenance. In such a case, the present invention can also be applied by treating before and after maintenance as separate manufacturing apparatuses.

  Moreover, in one manufacturing apparatus 13, the state in the manufacturing apparatus 13 may change by changing manufacturing conditions. In such a case, the present invention can be applied by treating the manufacturing conditions before and after the manufacturing conditions as different manufacturing apparatuses.

  Further, in one manufacturing apparatus 13, there is a possibility that the state in the manufacturing apparatus 13 changes due to temporal changes. In such a case, the present invention can be applied by dividing each period as a separate manufacturing apparatus by dividing the period by a certain period or irregular period.

  In this way, when one manufacturing apparatus 13 is divided under some condition, each divided manufacturing apparatus can be handled as another manufacturing apparatus and the present invention can be applied.

  In addition, two or more manufacturing devices are handled as one manufacturing device, such as dividing a plurality of manufacturing devices 13 into several groups and integrating the manufacturing devices 13 belonging to each group as one manufacturing device. The invention can also be applied.

That was described in the above range claims as "production apparatus" is not limited to a single manufacturing apparatus, the manufacturing apparatus 13 _first processing position in this embodiment, before and after maintenance, before and after the manufacturing conditions change, When one manufacturing apparatus is divided under various conditions such as each period when divided by a certain period or irregular period, each divided part is regarded as a manufacturing apparatus, and a plurality of manufacturing apparatuses are combined into one manufacturing apparatus. It also corresponds to a summary.

-4th Embodiment This Embodiment is related with the manufacturing process management apparatus different from the manufacturing process management apparatus in the said 1st Embodiment.

  FIG. 10 is a block diagram of the manufacturing process management apparatus according to the present embodiment. The manufacturing process management apparatus in the present embodiment is obtained by replacing the abnormality warning unit 8 of the manufacturing process management apparatus shown in FIG. 1 in the first embodiment with a predicted quality output unit 15 as shown in FIG. .

  According to the manufacturing process management apparatus of the present embodiment, the predicted quality of all products predicted by the predicted model created by the model creating unit 6 is output from the process parameters obtained by the process parameter obtaining unit 2. Can do. By using the manufacturing process management apparatus having such a configuration, the quality of the product is predicted from the process parameters of a certain manufacturing process 10, and the following quality is set so that the final quality becomes the target quality based on the output value of the predicted quality. The manufacturing process can be adjusted. Moreover, even if it is difficult to inspect the quality of all products, the predicted quality of all products can be obtained. Therefore, the output value of the predicted quality can be used as a substitute for the quality inspection result.

  Further, the manufacturing process management apparatus of the present embodiment is provided with an abnormality warning unit similar to the abnormality warning unit 8 in the first embodiment, and includes the abnormality warning unit in addition to the predicted quality output unit 15. It doesn't matter. In that case, when the predicted quality is out of the target range and an abnormality is warned by the abnormality warning unit, the abnormality is suddenly abnormal by checking the predicted quality based on the output from the predicted quality output unit 15. It is possible to discriminate whether it is an abnormality that gradually moves away from the target range.

  The process parameter selection apparatus of the present invention is useful as a technique for increasing the prediction accuracy of a quality prediction model in a manufacturing process including a plurality of manufacturing apparatuses having the same function, and can be used for a semiconductor device manufacturing process or the like. In addition, by monitoring the predicted quality, it is possible to detect anomalies before passing through the inspection process. Even when anomalies are discovered, process parameters are carefully selected, so process parameter candidates that cause anomalies are detected. It is useful as a technique that can detect process parameters that cause few abnormalities at an early stage.

1 ... Manufacturing process management device,
2 ... Process parameter acquisition unit,
3 ... Quality data acquisition unit,
4 Process parameter storage unit,
5 ... Quality data storage unit,
6 ... Model creation department,
7 ... Model storage unit,
8 ... Abnormal warning part,
9 ... Production process,
10 ... Manufacturing process,
11 ... Inspection process,
12 ... Partial manufacturing process,
12a ... 1st manufacturing process,
12b ... 2nd manufacturing process,
12c ... Manufacturing process A
13 ... Manufacturing equipment,
13 ₁ ... 1st manufacturing apparatus,
13 ₂ ... second manufacturing device,
13 _L ... Lth manufacturing apparatus,
14 ... Partial inspection process,
14a ... 1st inspection process,
14b ... the second inspection step,
14c ... B manufacturing process,
15: Predictive quality output unit.

Claims (14)

In a manufacturing process in which process processing is performed on a material by a plurality of manufacturing apparatuses having the same function, a process parameter acquisition unit that acquires process parameters from each of the manufacturing apparatuses,
Multivariate analysis is performed for each manufacturing apparatus using the acquired process parameters, and a comparison result between a statistical value related to the process parameter in the multivariate analysis result for each manufacturing apparatus and a preset reference value is used. All manufacturing apparatuses by selecting the process parameters satisfying preset setting conditions or by selecting the process parameters having desirable statistical values by using the comparison result between the statistical values regarding the process parameters. And a process parameter selection unit for selecting a common process parameter. The process parameter selection device according to claim 1, wherein
In an inspection process in which an inspection apparatus performs an inspection on a product that has undergone a process process in the manufacturing process, the apparatus includes a quality data acquisition unit that acquires quality data from the inspection apparatus,
The process parameter selection unit uses the acquired process parameter as an explanatory variable, and uses the acquired quality data as a target variable to perform analysis by a partial least square method for each manufacturing apparatus, and performs analysis by the partial least square method. By selecting a process parameter that satisfies a preset setting condition using a comparison result between a statistical value related to the process parameter in the value and a preset reference value, or between statistical values related to the process parameter By selecting a process parameter having a desired statistical value using the comparison result, a process parameter that is common to all manufacturing apparatuses and that is related to quality is selected. Process parameter selection device. In a manufacturing process in which process processing is performed on a material by a plurality of manufacturing apparatuses having the same function, a process parameter is acquired from each manufacturing apparatus by a process parameter acquisition unit,
The process parameter selection unit performs multivariate analysis for each manufacturing apparatus using the acquired process parameter, and a statistical value related to the process parameter in the multivariate analysis result for each manufacturing apparatus and a preset reference value The process parameter having a desired statistical value is selected by selecting the process parameter satisfying a preset setting condition using the comparison result of the above, or by using the comparison result between the statistical values regarding the process parameter. A process parameter selection method characterized by selecting a process parameter common to all manufacturing apparatuses. The process parameter selection method according to claim 3, wherein
In the inspection process in which the inspection apparatus is inspected for the product processed by the manufacturing process, the quality data acquisition unit acquires quality data from the inspection apparatus,
The process parameter selection unit uses the acquired process parameter as an explanatory variable, and uses the acquired quality data as an objective variable to perform analysis by the partial least square method for each manufacturing apparatus. By selecting the process parameter satisfying a preset setting condition using a comparison result between a statistical value relating to the process parameter in the analysis value and a preset reference value, or between statistical values relating to the process parameter By selecting process parameters having desirable statistical values using the comparison results of the above, process parameters that are common to all manufacturing equipment and that are related to quality are selected. Process parameter selection method. The process parameter selection method according to claim 4, wherein
As an analysis value by the partial least square method, the relative contribution of the acquired process parameter regarding each component axis obtained by the partial least square method from the acquired process parameter for each manufacturing apparatus, A process parameter selection method characterized by using a VIP value calculated from the degree of contribution to the quality of the component axis. The process parameter selection method according to claim 4, wherein
A model creation unit creates a prediction model for predicting product quality for each manufacturing apparatus using the selected process parameters.
A process parameter selection method characterized in that the model storage unit stores the prediction model created for each manufacturing device by the model creation unit for each manufacturing device. The process parameter selection method according to claim 6, wherein
A process parameter is acquired from the manufacturing apparatus by the process parameter acquisition unit, and a prediction model corresponding to the manufacturing apparatus related to the acquired process parameter is selected from the prediction models stored in the model storage unit. ,
Using the selected prediction model and the acquired process parameters, the product quality prediction value is calculated,
A process parameter selection method characterized in that an abnormality warning is output by an abnormality warning unit when the calculated predicted value deviates from a preset setting range. The process parameter selection method according to claim 6, wherein
A process parameter is acquired from the manufacturing apparatus by the process parameter acquisition unit, and a prediction model corresponding to the manufacturing apparatus related to the acquired process parameter is selected from the prediction models stored in the model storage unit. ,
Using the selected prediction model and the acquired process parameters, the product quality prediction value is calculated,
A process parameter selection method, wherein the predicted quality output unit outputs a predicted value of the calculated quality. The process parameter selection method according to claim 5, wherein
As a statistical value related to the process parameter in the analysis value by the partial least squares method, a value obtained by averaging the VIP value calculated for each manufacturing apparatus for each process parameter is used.
The process parameter having an average value related to the process parameter of the calculated VIP value equal to or greater than the reference value is selected as a process parameter common to all manufacturing apparatuses and related to quality. A process parameter selection method characterized by comprising: The process parameter selection method according to claim 5, wherein
As a statistical value related to the process parameter in the analysis value by the partial least squares method, a value obtained by averaging the VIP value calculated for each manufacturing apparatus for each process parameter is used.
The process parameters corresponding to a preset number of the process parameters that are common to all the manufacturing apparatuses, in the descending order of the average value related to the process parameters of the VIP value obtained, are process parameters that are common to all manufacturing apparatuses and that are related to quality. A process parameter selection method characterized by being selected as: The process parameter selection method according to claim 6, wherein
While gradually increasing the reference value, the process parameters selected by the respective reference values are used as explanatory variables, and the obtained quality data is used as an objective variable to perform analysis by the partial least square method for each manufacturing apparatus. When the quality prediction models are created sequentially, the process parameters that maximize the prediction accuracy of the sequentially created prediction models are the process parameters that are common to all the manufacturing equipment and that are related to quality. A process parameter selection method characterized by selecting as being. In a manufacturing process having a plurality of partial manufacturing processes where a process is performed on a material by a plurality of manufacturing apparatuses having the same function,
A combination of a plurality of manufacturing apparatuses used in each of the partial manufacturing processes is treated as a manufacturing apparatus for the manufacturing process, and the process parameter selection method according to claim 3 is applied to the manufacturing apparatus for the manufacturing process. Process parameter selection method characterized by the above. A process parameter selection program that causes a computer to function as the process parameter acquisition unit and the process parameter selection unit according to claim 1.   14. A computer-readable program recording medium on which the process parameter selection program according to claim 13 is recorded.

Images