JP2017146899A - Yield prediction device and yield prediction method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technique that models a cause-effect relationship between manufacturing time data and quality data differing in sampling frequency in a form verifiable with knowledge concerning a physical cause-effect relationship to make it easier to reconstruct.SOLUTION: Provided is a yield prediction device comprising: a representative value calculation unit for calculating a representative value using operating performance record data of a manufacturing device of a product that includes a plurality of variables; a nested network estimation unit for estimating a nested network that is a nested relation between object variables represented by a representative value for each representative value in accordance with the presence of a predetermined cause-effect relation and the hierarchical relation of variables; a prediction simulation unit for simulating prediction of quality data by a prescribed resolution using the nested network; and an output information generation unit for generating output information that shows the cause-effect relation between variables and quality data using the result of prediction simulation.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a yield prediction device and a yield prediction method.

  As a background art in this technical field, there is JP 2012-137813 A (Patent Document 1). This gazette states that, based on process operation data and quality data, an operation variable space having operation data as a basis vector is divided into a plurality of local regions, and a local relationship representing the relationship between the quality of each local region and the operation variables. An activity function for calculating the contribution ratio to the whole expression based on the operation data is calculated, and a first term representing the magnitude of the prediction error of the quality prediction value and the magnitude of the coefficient of the local relational expression are calculated. The coefficient of the local relational expression is determined by minimizing the weighted sum with the term 2. Further, a mathematical model expressing the relation between the operational variable and the quality as a superposition of the local relational expression and the local region having the activity function. If the error of the minimum error formula model is larger than a given convergence decision variable, the operation variable space is further subdivided and the above steps are performed. Repetitive Returns are described as display. "As an analysis result of a mathematical model of the convergence result.

JP 2012-13781 A

  The technique described in Patent Document 1 describes a technique for predicting quality with high accuracy from a large number of operating variables. However, when the sampling frequency of the data used for constructing the prediction model differs between the production data and quality data, the prediction model cannot be constructed. Moreover, since it is difficult to collate between the model and the physical causal relationship, it is difficult to reconstruct the model reflecting the consideration by the engineer and the result of the consideration.

  The object of the present invention is to model the causal relationship between production data and quality data with different sampling frequencies in a form that can be collated with knowledge about the physical causal relationship of the engineer, and reflect the collation result. The aim is to provide a technology that facilitates model reconstruction.

  The present application includes a plurality of means for solving at least a part of the above-described problems. Examples of such means are as follows. In order to solve the above problems, a yield prediction apparatus according to the present invention includes a representative value calculation unit that calculates a representative value using operation performance data of a product manufacturing apparatus including a plurality of variables, and a predetermined causal relationship. A nested network estimation unit that estimates a nested network that is a nested relationship between the variables of the target represented by the representative value for each representative value according to the presence / absence and the hierarchical relationship of the variables, and a predetermined value using the nested network A prediction simulation unit that performs a prediction simulation of the quality data at a resolution of, and an output information generation unit that generates output information that clearly indicates the causal relationship between the variable and the quality data using the result of the prediction simulation The yield prediction apparatus characterized by this is provided.

  According to the present invention, the cause-and-effect relationship between production data and quality data having different sampling frequencies is modeled in a form that can be collated with knowledge about the physical cause-and-effect relationship possessed by the engineer, and the collation result is reflected. Model can be rebuilt. Further, by using the reconstructed model to simulate the value of the target variable when the value of each variable is changed, it is possible to support the identification of the optimum variable value and improve the yield. Problems, configurations, and effects other than those described above will be clarified by the following description of embodiments.

It is a figure which shows the outline of the yield prediction system which concerns on 1st embodiment of this invention. It is a figure which shows the data structure stored in an apparatus operation performance memory | storage part. It is a figure which shows the data structure stored in a quality data storage part. It is a figure which shows the data structure stored in a representative value definition memory | storage part. It is a figure which shows the data structure stored in a synthetic variable definition memory | storage part. It is a figure which shows the data structure stored in a hierarchy definition memory | storage part. It is a figure which shows the data structure stored in a causal relationship information storage part. It is a figure which shows the data structure stored in a simulation granularity memory | storage part. It is a figure which shows the hardware constitutions of a yield prediction apparatus. It is a figure which shows the operation | movement flow of a yield prediction process. It is a figure which shows the operation | movement flow of the estimation process of a nested network. It is a figure which shows the example of a network structure. It is a figure which shows the example of a probability structure. It is a figure which shows the example of simulation result information. It is a figure which shows a simulation screen. It is a figure which shows the operation example of a simulation screen.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. Note that components having the same function are denoted by the same reference symbols throughout the drawings for describing the embodiment, and the repetitive description thereof will be omitted.

  Hereinafter, a yield prediction system 1 to which a first embodiment according to the present invention is applied will be described with reference to the drawings.

  FIG. 1 is a diagram showing an outline of a yield prediction system 1 according to the present invention. The yield prediction system 1 includes a yield prediction apparatus 100. The yield prediction apparatus 100 includes a control unit 120, a storage unit 130, an input / output interface control unit 150, and a bus 160 that connects them to each other. A user (such as a person in charge of quality control) uses the function of the yield prediction device 100 through the operation of the input / output device 170 connected to the input / output interface control unit 150. The yield prediction apparatus 100 can be configured by a general computer (PC or the like), and implements a characteristic processing function (each processing unit of the yield prediction apparatus 100) by software program processing, for example.

  The input / output device 170 is connected to the input / output interface control unit 150, and includes an input device that accepts user operations and an output device that outputs device allocation results and the like. For example, the input device includes a keyboard, a mouse, and a touch panel, and the output device includes a display, a printer, and the like. In this system, a graphical user interface (GUI) is configured on the output screen, and various types of information are displayed. Alternatively, the input / output device 170 may be a personal computer, a tablet, a smartphone, or the like connected via the Internet, a LAN (Local Area Network), a WAN (Wide Area Network), a VPN (Virtual Private Network), a mobile phone carrier network, or the like. It may be a terminal that can use the Web browser or the like.

  The storage unit 130 includes an apparatus operation result storage unit 131, a quality data storage unit 132, a representative value definition storage unit 133, a composite variable definition storage unit 134, a hierarchy definition storage unit 135, and a causal relationship information storage unit 136. And a simulation granularity storage unit 137.

  FIG. 2 is a diagram illustrating a data structure stored in the apparatus operation result storage unit 131. The apparatus operation result storage unit 131 stores information on the operation results of the apparatus used for production, and stores the time 131a, the construction product 131b, the temperature 131c, the pressure 131d, and the power 131e in association with each other. The The time 131a is information for identifying the time when the actual performance of the apparatus is acquired.

  The start product 131b is information for specifying a product to be produced by an apparatus used for the production. The temperature 131c, the pressure 131d, and the electric power 131e are predetermined measurement values indicating the operating state of the apparatus used for production at the time 131a, and are information for specifying the temperature, pressure, and electric power of the production apparatus, respectively.

  FIG. 3 is a diagram showing a data structure stored in the quality data storage unit 132. The quality data storage unit 132 stores the construction product 132a and the yield 132b in association with each other. The start product 132a is information for specifying a product to be produced by an apparatus used for the production. The yield 132b is information for specifying a yield as a result of producing the product specified by the start product 132a in the production apparatus.

  FIG. 4 is a diagram illustrating a data structure stored in the representative value definition storage unit 133. In the representative value definition storage unit 133, one or a plurality of characteristic values to be the representative value 133a are defined. Specifically, definitions of average values, standard deviations, and the like are included as characteristic values that should be representative values. This representative value defines a representative value for each of temperature, pressure, and electric power indicating the device operation performance. In addition, various statistical values such as a minimum value and a maximum value may be included as values of features to be representative values.

  FIG. 5 is a diagram illustrating a data structure stored in the composite variable definition storage unit 134. Here, the composite variable is a variable obtained by subjecting a representative value to a predetermined conversion. For example, a composite variable (principal component space variable) converted to the coordinate axis of the principal component space, a composite variable (Krylov subspace variable) converted to the coordinate axis of the Krylov subspace, or the like is combined with the representative value. Is a variable. Note that the synthetic variable is not limited to this, and may be another synthetic variable.

  FIG. 6 is a diagram illustrating a data structure stored in the hierarchy definition storage unit 135. In the hierarchy definition storage unit 135, a variable 135a and a hierarchy 135b are stored in association with each other. The variable 135a is information for specifying a variable for defining a hierarchy. The hierarchy 135b is information that defines the hierarchy of the variable specified by the variable 135a. Note that the variable hierarchy is information that can indicate the posterior relationship of variables between processes by setting the variables on the subsequent process side to a deep hierarchy and the variables on the preceding process side to a shallower hierarchy. is there.

  FIG. 7 is a diagram illustrating a data structure stored in the causal relationship information storage unit 136. The causal relationship information storage unit 136 stores the first variable 136a, the second variable 136b, and the causal relationship presence / absence 136c in association with each other. The 1st variable 136a and the 2nd variable 136b are information which shows the variable of the object which shows the presence or absence of the causal relationship between each other. The causal relationship presence / absence 136c is information specifying the presence / absence of a causal relationship between the first variable 136a and the second variable 136b. Note that the causal relationship of variables is assumed to be causal when there is a predetermined relationship in the change between the cause-side variable and the result-side variable. For example, in the relation of variables having a causal relationship, if the cause variable is changed, the result variable is affected.

  FIG. 8 is a diagram illustrating a data structure stored in the simulation granularity storage unit 137. In the simulation granularity storage unit 137, a variable 137a and a granularity 137b are stored in association with each other. The variable 137a is information for specifying a variable to be changed when the simulation process is performed. The particle size 137b is information for specifying the cardinality of the variable specified by the variable 137a in the simulation process. In the simulation process, the range of values that each variable can take is divided into an odd number, and the predicted value when each range is taken is simulated, so the value of the granularity 137b is used for this.

  Here, the storage unit 130 is connected via various networks such as a LAN (Local Area Network), a WAN (Wide Area Network), a cellular phone network, a wireless communication network, etc., or a network such as the Internet (not shown). Each function unit of the control unit 120 may be provided in the apparatus and access information stored in each storage unit via communication.

  The control unit 120 includes a representative value calculation unit 121, a variable synthesis unit 122, a hierarchy sorting unit 123, a nested network estimation unit 124, a prediction simulation unit 125, and an output information generation unit 126.

  The representative value calculation unit 121 calculates a representative value defined as a representative value of a variable for each construction product using the data on the operation of the apparatus. For example, the representative value calculation unit 121 refers to the acquired device operation result storage unit 131, and calculates the average value and the standard deviation according to the representative value definition for each temperature, pressure, and power for each construction product. Is calculated. In other words, the representative value calculation unit 121 calculates the representative value using the operation result data of the product manufacturing apparatus including a plurality of variables.

  The variable composition unit 122 calculates, with respect to the calculated representative value, a composite variable that is variable-converted to the coordinate axes of the principal component space according to the definition of the composite variable and a composite variable that is variable-converted to the coordinate axes of the Krylov subspace.

  The hierarchy sorting unit 123 arranges the variables of the operation result data of the apparatus based on the hierarchy definition set in advance. More specifically, the variables are arranged in a hierarchical relationship by using hierarchy definition information that defines the variables in advance according to the manufacturing process order of the product from which the variables are acquired.

  The nested network estimation unit 124 estimates a nested network. Specifically, the nested network estimation unit 124 estimates the network structure between variables and the probability structure. Here, the network structure is information regarding which variables and which variables have a parent-child relationship. The network structure is mainly expressed by Bayesian Network, but in this embodiment, the variable further includes a nested structure. The probability structure is information regarding the joint probability distribution of the parent variable and the child variable. In other words, the nested network estimation unit 124 estimates a nested network that is a nested relationship between target variables represented by representative values for each representative value according to the presence or absence of a predetermined causal relationship and the hierarchical relationship of the variables. It can be said that. More specifically, the nested network estimation unit 124 estimates a network between representative values for each variable, and estimates a network structure between variables in which the network between the representative values is regarded as one variable.

  The prediction simulation unit 125 divides the range of possible values of a variable into a predetermined number of radixes for each variable using a technique such as traditional sampling, and simulates the predicted yield of each range. To do. In other words, the prediction simulation unit 125 performs a prediction simulation of quality data with a predetermined resolution using a nested network. The prediction simulation unit 125 uses the nested network and the probability structure estimated by the nested network estimation unit 124 to calculate the predicted value of the quality data and the value of the variable when a predetermined value is set for the variable based on the resolution. The importance for each variable is calculated using the amount of change in the predicted value of the quality data due to the change.

  The output information generation unit 126 generates output information such as a screen indicating the result of the prediction simulation. In other words, the output information generation unit 126 generates output information that clearly indicates the causal relationship between the variable and the quality data using the result of the prediction simulation. Specifically, the output information generation unit 126 predicts quality data when a predetermined value is set for a variable based on resolution using information indicating the nested network in a predetermined diagram and the nested network and the probability structure. And information on the degree of importance for each variable calculated using the amount of change in the predicted value of the quality data caused by changing the value of the variable.

  FIG. 9 is a diagram illustrating a hardware configuration of the yield prediction apparatus 100. The yield prediction apparatus 100 includes an arithmetic device 204 such as a CPU (Central Processing Unit), a main storage device 205 such as a memory, an external storage device 203 such as a hard disk and an SSD (Solid State Drive), and an input such as a keyboard and a mouse. The apparatus 201 includes an output device 202 such as a display or a printer, a communication device 206 such as a NIC (Network Interface Card), and a bus 207 that connects these devices.

  The communication device 206 is a wired communication device that performs wired communication via a network cable, or a wireless communication device that performs wireless communication via an antenna. The communication device 206 communicates with other devices connected to the network.

  The main storage device 205 is a memory such as a RAM (Random Access Memory). The external storage device 203 is a non-volatile storage device such as a so-called hard disk, SSD, or flash memory that can store digital information.

  The input device 201 is a device that receives input information including a pointing device such as a keyboard and a mouse, a touch panel, or a microphone that is a voice input device.

  The output device 202 is a device that generates output information including a display, a printer, or a speaker that is an audio output device.

  The representative value calculation unit 121, the variable synthesis unit 122, the hierarchy sorting unit 123, the nested network estimation unit 124, the prediction simulation unit 125, and the output information generation unit 126 cause the arithmetic device 204 to perform processing. Realized by the program. This program is stored in the main storage device 205 or the external storage device 203, loaded onto the main storage device 205 for execution, and executed by the arithmetic device 204.

  In addition, the apparatus performance record storage unit 131, the quality data storage unit 132, the representative value definition storage unit 133, the composite variable definition storage unit 134, the hierarchy definition storage unit 135, the causal relationship information storage unit 136, and the simulation granularity The storage unit 137 is realized by the main storage device 205 and the external storage device 203.

  In addition, when the yield prediction apparatus 100 has a communication unit that is communicably connected to the Internet or the like (not shown), the communication unit is realized by the communication device 206. The input / output interface control unit 150 is realized by a device (not shown) that controls the output device 202 and the input device 201, or an arithmetic device 204.

  The above is the hardware configuration example of the yield prediction apparatus 100 of the yield prediction system 1 in the present embodiment. However, the configuration is not limited to this, and other hardware may be used. For example, a device that receives input / output via the Internet may be used.

  Although not shown, the yield prediction apparatus 100 has known elements such as an OS (Operating System), middleware, and applications, and has an existing processing function for displaying a GUI screen on an input / output device such as a display. Prepare.

  [Description of Operation] Next, the operation of the yield prediction system 1 in this embodiment will be described.

  FIG. 10 is a diagram illustrating a process flow of the yield prediction process. The yield prediction process performed by the yield prediction apparatus 100 according to the present embodiment is started when a process start instruction is received from a user (designer) while the yield prediction apparatus 100 is activated.

  First, the representative value calculation unit 121 reads operation result data and quality data (step S001). Specifically, the representative value calculation unit 121 includes, among information stored in the device operation result storage unit 131 and the quality data storage unit 132, operation result data and quality data in a time zone related to a predetermined range. And load.

  Then, the representative value calculation unit 121 calculates a representative value (step S002). Specifically, the representative value calculation unit 121 defines a representative value for each of the temperature 131c, the pressure 131d, and the power 131e for each construction product 131b of the apparatus operation result storage unit 131 read in step S001. Each of the representative values 133a stored in the storage unit 133 is calculated by a predetermined method. For example, the representative value calculation unit 121 adds the average values of the representative values of the temperature 131c, the pressure 131d, and the power 131e that are common to the construction products 131b, and divides the average value by the parameter. Find the value. Further, for example, the representative value calculation unit 121 uses a predetermined known algorithm for the standard deviation of the representative values for the device operation results of the temperature 131c, the pressure 131d, and the electric power 131e that are common to the construction product 131b. Ask for. The representative value calculated by the representative value calculation unit 121 is not limited to the average value and the standard deviation, and may be various known statistical values such as other median values, maximum values, and minimum values.

  Then, the variable synthesizer 122 synthesizes the variables (step S003). Specifically, the variable composition unit 122 calculates each of the composite variables 134 a stored in the composite variable definition storage unit 134 by using a predetermined method using each of the representative values calculated by the representative value calculation unit 121. For example, the variable composition unit 122 calculates a composite variable that is variable-converted to the coordinate axes of the principal component space and a composite variable that is variable-converted to the coordinate axes of the Krylov subspace. In addition, regarding the conversion process to each space when the variable synthesis unit 122 synthesizes the variable, the variable synthesis unit 122 uses a predetermined existing algorithm.

  Then, the hierarchy arranging unit 123 arranges the hierarchies (step S004). Specifically, the hierarchy sorting unit 123 refers to the hierarchy definition storage unit 135 and reads the information of the hierarchy 135b for each variable 135a. Note that the information on the level 135b indicates the level in which the variable is set. For this reason, the higher the value set as the hierarchy 135b, the higher the hierarchy alignment unit 123 arranges the variable to be arranged in a hierarchy close to the yield that is the target variable. In the present embodiment, it is assumed that variables are prepared for each manufacturing process, and the order of the processes is set as a hierarchy. For example, temperature is set as level 1, and both pressure and power are set as level 2. By doing this, it is possible to eliminate the need to specify the entire order of variables by simply setting the order of the processes, so that the resources for studying the setting for modeling can be simplified.

  Then, the nested network estimation unit 124 estimates a nested network (step S005). Specific processing contents will be described in the nested network estimation processing described later.

  And the prediction simulation part 125 performs prediction by simulation (step S006). Specifically, the prediction simulation unit 125 is a process of simulating a predicted value when each range is taken by dividing the range of possible values of each variable into radix. Such a simulation method of predicted values includes a method such as tradition sampling, and in this embodiment, simulation processing by tradition sampling is performed. However, the present invention is not limited to this, and other simulation methods may be used. There may be. The prediction simulation unit 125 stores the result of the simulation processing in an area (not shown) of the storage unit 130 in the format of the simulation result information example 500 illustrated in FIG. At that time, the prediction simulation unit 125 uses the radix defined by the particle size 137b of the simulation particle size storage unit 137 as the particle size used for the simulation.

  FIG. 14 is a diagram illustrating an example of simulation result information. In the data structure of the simulation result information example 500, a variable 500a, a value identifier 500b, and a yield 500c are stored in association with each other. In the variable 500a, information for specifying a variable corresponding to each of the temperature 131c, the pressure 131d, and the power 131e in the apparatus operation result storage unit 131 is stored. The value identifier 500b stores an identifier that specifies a range of values taken by the variable specified by the variable 500a. Note that the value taken by the variable specified by the variable 500a is calculated by the prediction simulation unit 125 according to the radix defined by the granularity 137b of the simulation granularity storage unit 137. In the yield 500c, a predicted value that is a result of a yield simulation in the value range specified by the identifier of the value identifier 500b is stored.

  Then, the output information generation unit 126 generates and outputs output information (step S007). Specifically, the output information generation unit 126 generates a simulation screen that is displayed so that the simulation result varies according to a predetermined operation, and outputs the simulation screen to the input / output device 170 as output information. Note that the display target of the output information is dynamically selected by controlling input reception, display change, redisplay, and the like by a browser or the like on the input / output device 170 side. Specifically, dynamic processing can be performed using dynamic HTML (Hyper Text Markup Language) or a predetermined animation display language.

  FIG. 15 is a diagram showing a simulation screen. The simulation screen 600 includes a hierarchical structure display area 610 and a variable importance display area 620. The hierarchical structure display area 610 shows a network structure in which variables are arranged in a predetermined figure according to the hierarchy. In the variable importance display area 620, the importance of each variable is displayed together with a predetermined diagram (for example, a bar graph having a higher height as the contribution is higher). The importance is a contribution to a result variable (a yield in the present embodiment) calculated by a predetermined algorithm. For example, the importance may be calculated by an average value of the fluctuation amount of the predicted value of the result variable.

  In the variable importance display area 620, when one of the displayed variables is selected and input, the variable in the hierarchical structure display area 610 is highlighted and displayed. In addition, a variable that is causally related to the variable and a link that indicates the relationship between the variable are also highlighted. In other words, output information is generated so as to highlight and display a variable selected and input among the variables, a variable that has a causal relationship with the variable, and a predetermined symbol that indicates the causal relationship.

  FIG. 16 is a diagram illustrating an operation example of the simulation screen. The simulation screen 650 is an example of a display that is displayed as a result of selection input for one of the variables displayed on the simulation screen 600. The simulation screen 650 includes a hierarchical structure display area 660 and a variable importance display area 670. The hierarchical structure display area 660 shows a network structure in which variables are arranged in a predetermined figure according to the hierarchy. In the variable importance display area 670, the importance of each variable is displayed together with a predetermined diagram (for example, a bar graph having a higher height as the contribution is higher).

  In the variable importance display area 670, when one of the displayed variables is selected, the variable in the hierarchical structure display area 660 is highlighted, for example, colored. In addition, a variable that is causally related to the variable and a link that indicates the relationship between the variable are also highlighted. In addition, by inputting one of the predetermined variables in the hierarchical structure display area 660 at a predetermined position in the variable, a range of values that can be taken by the variable according to the input position is specified. The amount of change in the result variable (in this embodiment, yield) is shown. In other words, for the variable selected and input among the variables, the adjustment bar accepts a change input of the variable setting value range based on the predetermined radix of the variable, and the quality data determined in advance according to the variable setting value range Output information is generated so as to reflect the increase or decrease of.

  By configuring the simulation screen 650 in this manner, the engineer can consider the validity of the model estimated by the present system. In addition, as a result of consideration, if there are concerns about validity, it can be reflected in the model by correcting the definition of the hierarchy and the causal relationship. In addition, the value of each variable to be set in order to improve the value of the objective variable such as the yield can be confirmed.

  FIG. 11 is a diagram illustrating an operation flow of the nested network estimation process. The nested network estimation process is started in step S005 of the yield prediction process.

  First, the nested network estimation unit 124 estimates a network between representative values (step S101). Specifically, the nested network estimation unit 124 estimates a network model between representative values calculated in the representative value calculation process for each variable. The network model estimation method includes, for example, the K2 algorithm, which is adopted, but is not limited thereto. Moreover, it can be said that the calculation amount can be reduced and the model can be simplified by estimating the network model between representative values instead of between variables.

  Then, the nested network estimation unit 124 selects a child variable candidate (step S102). Specifically, the nested network estimation unit 124 selects one unprocessed variable among the variables included in the range represented by the representative value, and sets it as a candidate for the child variable.

  Then, the nested network estimation unit 124 selects a parent variable candidate (step S103). Specifically, the nested network estimation unit 124 lists variables that can be parent variables of the variables selected as the candidate child variables in step S102.

  Then, the nested network estimation unit 124 calculates an index (Step S104). Specifically, the nested network estimation unit 124 calculates a probability that each parent variable candidate listed in step S103 is a parent variable and a probability that none of the parent variable candidates is a parent variable. As an example of the index value, K2 metric used in the K2 algorithm is used. However, it is not limited to this, and other index values may be used.

  Then, the nested network estimation unit 124 determines whether or not a parent variable is established (step S105). Specifically, the nested network estimation unit 124 selects a parent variable candidate having the highest probability of being a parent variable as a parent variable, and returns control to Step S103 (in the case of “Yes” in Step S105). However, when any parent variable candidate has the highest probability of not being a parent variable (in the case of “No” in step S105), the nested network estimation unit 124 does not select any variable as a parent variable, and proceeds to step S106. .

  If the parent variable has not been established (“No” in step S105), the nested network estimation unit 124 determines whether all the child variables have been selected (step S106). Specifically, the nested network estimation unit 124 determines whether all variables have been selected as child variable candidates in step S102. If all the child variables have not been selected (“No” in step S106), the nested network estimation unit 124 returns the control to step S102.

  Then, the nested network estimation unit 124 estimates a probability structure (step S107). Specifically, the nested network estimation unit 124 estimates a simultaneous probability distribution having two variables, the child variable selected in step S102 and the parent variable determined to be established in step S105.

  The above is the processing flow for estimating the nested network. According to the nested network estimation process, after estimating a network between representative values, a causal relationship between variables belonging to a range represented by the representative value is estimated. That is, by setting the representative value and modeling the variable in a nested structure, it is possible to avoid the brute force calculation by all the variables for estimating the causal relationship. Therefore, even when the number of types of variables increases, the amount of calculation can be suppressed, the responsiveness of the yield prediction apparatus 100 can be ensured, and the power consumption can be suppressed low.

  FIG. 12 is a diagram illustrating an example of a network structure. In the network structure example 300, variables that are child variables in the row direction (vertical direction) and variables that are parent variables in the column direction (horizontal direction) are enumerated to form a matrix. For example, the value of the column of pressure (standard deviation) in the row of temperature (standard deviation) is “1”. This means that the variable temperature (standard deviation) has the other variable pressure (standard deviation) in its parent, that is, is located on the causal result side. The relationship between the two variables when the value of the intersection of the two variables is “0” has a low causal relationship or does not affect the child variable from the parent variable. It shows that.

  FIG. 13 is a diagram illustrating an example of a probability structure. In the example 400 of the probability structure, the probabilities of variables as child variables in the row direction (vertical direction) and the probabilities of variables as parent variables in the column direction (horizontal direction) are respectively listed to form a matrix. There are as many matrices as there are child variables.

  Heretofore, the yield prediction system 1 according to the first embodiment has been described. According to the first embodiment, the causal relationship between production data and quality data with different sampling frequencies is modeled in a form that can be collated with the knowledge about the physical causal relationship that the engineer has, and the collation result is reflected. It is possible to perform yield prediction that makes it easy to reconstruct the model.

  The present invention is not limited to the first embodiment described above. The first embodiment described above can be variously modified within the scope of the technical idea of the present invention.

  For example, as a second embodiment, a method in which variable synthesis is not performed in step S003 can be used. In this case, since no synthetic variable is generated, the variable can be handled in the defined hierarchy, and the visibility of the model is increased. In addition, since the causal relationship specifying process is simplified and the load on the yield prediction apparatus 100 for calculation is reduced, it is possible to contribute to energy saving.

  In the above-described embodiment, the configuration has been described in detail for easy understanding of the present invention, and is not necessarily limited to the configuration having all the configurations described.

  Each of the above-described configurations, functions, processing units, and the like may be realized by hardware, for example, by designing a part or all of them with an integrated circuit. Further, the control lines and information lines are those that are considered necessary for the explanation, and not all the control lines and information lines on the product are necessarily shown. Actually, it may be considered that almost all the components are connected to each other.

  Each of the above-described configurations, functions, processing units, and the like may be realized in a distributed system by executing a part or all of them on another apparatus and performing integrated processing via a network, for example. In addition, the terminal used for the operation may be configured to be connectable via the Internet, so that the yield prediction may be performed as a cloud service.

  Further, the technical elements of the above-described embodiments may be applied independently, or may be applied by being divided into a plurality of parts such as program parts and hardware parts.

  In the above, this invention was demonstrated centering on embodiment.

DESCRIPTION OF SYMBOLS 1 ... Yield prediction system, 100 ... Yield prediction apparatus, 120 ... Control part, 121 ... Representative value calculation part, 122 ... Variable composition part, 123 ... Hierarchy alignment part, 124. ..Nested network estimation unit, 125 ... Prediction simulation unit, 126 ... Output information generation unit, 130 ... Storage unit, 131 ... Device operation result storage unit, 132 ... Quality data storage unit, 133 ... representative value definition storage unit, 134 ... composite variable definition storage unit, 135 ... hierarchy definition storage unit, 136 ... causal relationship information storage unit, 137 ... simulation granularity storage unit, 150 ..Input / output interface control unit, 160 ... bus, 170 ... input / output device

Claims (8)

A representative value calculation unit for calculating a representative value using operation result data of a manufacturing apparatus of a product including a plurality of variables and quality data indicating the quality of the product;
A nested network estimation unit that estimates a nested network that is a nested relationship between the variables represented by the representative value for each representative value according to the presence or absence of a predetermined causal relationship and the hierarchical relationship of the variables;
A prediction simulation unit that performs a prediction simulation of the quality data at a predetermined resolution using the nested network;
An output information generating unit that generates output information that clearly indicates a causal relationship between the variable and the quality data using the result of the prediction simulation;
A yield prediction apparatus comprising: The yield prediction device according to claim 1,
A hierarchical alignment unit that aligns the variables using hierarchical definition information that defines the variables in advance as hierarchical levels in accordance with a process order in which the variables are acquired;
A yield prediction apparatus comprising: The yield prediction device according to claim 1,
The nested network estimation unit estimates a network between the representative values for each variable, and estimates a network structure between variables in which the network between the representative values is regarded as one variable.
Yield prediction device characterized by that. The yield prediction device according to claim 1,
The nested network estimation unit estimates a probability structure between the variables by a predetermined algorithm,
The prediction simulation unit uses the nested network and the probability structure estimated by the nested network estimation unit, and predicts the quality data when a predetermined value is set for the variable based on the resolution, Calculating the importance for each of the variables using the amount of change in the predicted value of the quality data resulting from changing the value of the variable;
Yield prediction device characterized by that. The yield prediction device according to claim 1,
The nested network estimation unit estimates a probability structure between the variables by a predetermined algorithm,
The output information generation unit
Information indicating the nested network by a predetermined diagram;
The predicted value of the quality data when a predetermined value is set for the variable based on the resolution using the nested network and the probability structure, and the predicted value of the quality data by changing the value of the variable Information of importance for each of the variables calculated using a variation amount, and
Generate information to output,
Yield prediction device characterized by that. The yield prediction device according to claim 1,
The output information generation unit
The output information is generated so as to highlight a variable selected and input among the variables, the variable having a causal relationship with the variable, and a predetermined symbol indicating the causal relationship.
Yield prediction device characterized by that. The yield prediction device according to claim 1,
The output information generation unit
Highlighting the variable selected and input from among the variables, the variable having a causal relationship with the variable, and a predetermined symbol indicating the causal relationship,
For the variable selected and input from among the variables, the adjustment bar accepts a change input of a variable setting value range based on a predetermined radix of the variable, and the quality determined in advance according to the variable setting value range Generating the output information to reflect the increase or decrease in data,
Yield prediction device characterized by that. A yield prediction method that allows a computer to predict yield,
The computer
A representative value calculating step of calculating a representative value using operation result data of a manufacturing apparatus for a product including a plurality of variables;
Nested network estimation step for estimating a nested network that is a nested relationship between the variables represented by the representative value for each representative value according to the presence or absence of a predetermined causal relationship and the hierarchical relationship of the variables;
A prediction simulation step of performing a prediction simulation of the quality data at a predetermined resolution using the nested network;
An output information generation step of generating output information that clearly indicates a causal relationship between the variable and the quality data using a result of the prediction simulation.

Images