Manufacturing condition optimization device, regression model generation device, program, and manufacturing condition optimization method

Abstract

A production conditions optimization device (100) that comprises: a yield estimation unit (117) that estimates the yield of product of a quality that would pass inspection by inspection equipment if conditions for production of a product by production equipment were to be varied; and an optimization processing unit (113) that calculates how much to vary the production conditions to achieve a maximum yield.

Description

[Technical field]

[0001] The present invention relates to a manufacturing condition optimization device, a program, and a manufacturing condition optimization method for optimizing manufacturing conditions in a manufacturing facility. [Background technology]

[0002] When manufacturing products and shipping those that have passed quality inspection, the product yield (inspection pass rate) is important and serves as a manufacturing indicator. The plant operating condition optimization system described in Patent Document 1 includes an operating state data acquisition means for acquiring operating state data indicating the operating state of the plant measured by a plurality of sensors, an operating index data acquisition means for acquiring operating index data for evaluating the operation of the plant, the operating state data being measured by sensors provided in the plant or calculated based on the operating state data acquired by the operating state data acquisition means, a measurement data recording means for associating the operating state data acquired by the operating state data acquisition means with the operating index data calculated by the operating index data acquisition means based on a predetermined item, and for recording the measurement data in a data recording unit, and a measurement data recording means for recording the measurement data in a data recording unit based on the plurality of sets of measurement data recorded in the data recording unit, the measurement data recording means recording the operation index data in a data recording unit, and the operation index data in a data recording unit. A regression model creation means performs a predetermined multivariate analysis using a driving index variable representing the data side as a response variable to create a regression model, the regression model having a component conversion means for converting explanatory variables into components that are uncorrelated with each other and have a smaller number than the original explanatory variables, a prediction means for predicting the response variable from the components converted by the component conversion means, and an inverse conversion means for estimating the explanatory variable from the components by a method corresponding to the component conversion means, and a driving index variable optimization means for determining a driving state variable that optimizes the driving index variable based on the regression model created by the regression model creation means, characterized in that the explanatory variable value is determined when optimizing an evaluation function related to the response variable predicted by the prediction means while satisfying a constraint condition related to the explanatory variable estimated by the inverse conversion means, and the explanatory variable value is set as the optimal driving condition. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2012-074007 A Summary of the Invention [Problem to be solved by the invention]

[0004] The quality of a product depends on the manufacturing conditions of the product. Even if appropriate manufacturing conditions are set once, the quality may change due to disturbances that are difficult to completely control, such as temperature and material viscosity, and the product yield may decrease. The same applies to disturbances that are not measured or monitored in the manufacturing environment. For this reason, it is required to adjust the manufacturing conditions in real time during the manufacturing of the product. However, the operating condition optimization system described in Patent Document 1 does not assume that the manufacturing conditions will be adjusted in real time in anticipation of changes due to disturbances. For this reason, optimization corresponding to changes in uncontrollable disturbances cannot be performed.

[0005] The present invention has been made in consideration of the above background, and has an objective of providing a manufacturing condition optimization device, program, and manufacturing condition optimization method that enable the improvement of manufacturing indexes by adjusting the manufacturing conditions of a product even in the presence of disturbances. [Means for solving the problem]

[0006] The above object of the present invention can be achieved by the following means.

[0007] (1) The manufacturing conditions of the product by the manufacturing equipment At a given time When you change Based on the quality of the product at the predetermined time, a yield estimating unit that estimates a yield of the product having a quality that passes inspection by an inspection facility; In order to achieve this, the predetermined timing The above manufacturing conditions from and an optimization processing unit that calculates an amount of change in the manufacturing condition.

[0008] (2) A manufacturing condition optimization device as described in (1), further comprising a quality change estimation unit that estimates a change in quality of the product manufactured by the manufacturing equipment from a change in the manufacturing conditions, wherein the yield estimation unit estimates a probability that the product will pass inspection by the inspection equipment from the change in quality of the product estimated by the quality change estimation unit, and estimates a yield of the product having a quality that passes inspection by the inspection equipment from the probability of passing the inspection.

[0009] (3) A manufacturing condition optimization device as described in (2), further comprising a regression model generation unit, which generates a regression model that estimates a change in quality of the product from a change in the manufacturing conditions based on correspondence data between past changes in the manufacturing conditions and changes in the quality of the product at the changes in the manufacturing conditions, and the quality change estimation unit estimates a change in quality of the product from the change in the manufacturing conditions using the regression model.

[0010] (4) The manufacturing condition optimization device according to (3), wherein the regression model is a generalized linear model, a Gaussian process regression model, or a neural network model.

[0011] (5) The manufacturing condition optimization device described in (1), wherein the optimization processing unit calculates the amount of change in the manufacturing conditions that maximizes the yield using at least one of an optimization algorithm that does not use derivatives, a local search method, a simulated annealing method, a tabu search method, and a genetic algorithm.

[0012] (6) The manufacturing condition optimization device described in (2), wherein the yield estimation unit, when estimating the probability that the product will pass the inspection by the inspection equipment from the amount of change in quality, estimates the probability that the product will pass the inspection by the inspection equipment based on the quality of a predetermined number of products most recently manufactured.

[0013] (7) The manufacturing condition optimization device described in (2), wherein the yield estimation unit calculates a probability distribution of the quality of the product from the amount of change in the quality of the product, and estimates the probability that the product will pass inspection by the inspection equipment based on the probability distribution.

[0014] (8) The manufacturing condition optimization device according to (7), wherein the probability distribution is a normal distribution, a binomial distribution, or a Poisson distribution. (9) Based on the corresponding data of the change in the manufacturing conditions in the past and the change in the quality of the product due to the change in the manufacturing conditions, At a given timing The above manufacturing conditions Manufacturing conditions changed from Change in is used as an explanatory variable, and the change in the quality of the product at the specified time is A regression model generating unit that generates a regression model for estimating the amount of change in the quality of the product. Regression model generation Device.

[0015] (10) Computer (1)~( 8 ) The manufacturing condition optimization apparatus according to any one of the above items or (9). A program to function as a

[0016] (11) A manufacturing condition optimization method for a manufacturing condition optimization device, comprising: At a given time When you change Based on the quality of the product at the predetermined time, estimating a yield of the product having a quality that passes inspection by an inspection facility; In order to achieve this, the predetermined timing The above manufacturing conditions from A manufacturing condition optimization method that executes the steps of: (12) Regression model generation Equipment Regression model generation A method for manufacturing a product according to the present invention, comprising: At a given timing The above manufacturing conditions Manufacturing conditions changed from Change in is used as an explanatory variable, and the change in the quality of the product at the specified time is generating a regression model for estimating the amount of change in quality of the product; Regression model generation method. Effect of the Invention

[0017] According to the present invention, it is possible to provide a manufacturing condition optimization device, a program, and a manufacturing condition optimization method that make it possible to improve a manufacturing index by adjusting the manufacturing conditions of a product even when a disturbance occurs. [Brief description of the drawings]

[0018] [Figure 1] 1 is an overall configuration diagram of a manufacturing system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram for explaining an outline of processing performed by the manufacturing condition optimization device according to the present embodiment. [Diagram 3] FIG. 1 is a functional block diagram of a manufacturing condition optimization device according to an embodiment of the present invention. [Figure 4] FIG. 2 is a data configuration diagram of learning data according to the embodiment. [Diagram 5] FIG. 2 is a data configuration diagram of manufacturing management data according to the embodiment. [Figure 6] 11 is a flowchart of a regression model generation process according to the present embodiment. [Figure 7] 11 is a flowchart of an objective function process executed by an objective function processing unit according to the present embodiment. [Figure 8] 11 is a flowchart of an optimization process according to the present embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0019] <Overall configuration of the manufacturing system> A manufacturing condition optimization device in an embodiment for carrying out the present invention will be described below. 1 is an overall configuration diagram of a manufacturing system 10 according to this embodiment. The manufacturing system 10 is configured to include a manufacturing facility 410, an inspection facility 420, a manufacturing management device 430, and a manufacturing condition optimization device 100. The manufacturing facility 410, the inspection facility 420, the manufacturing management device 430, and the manufacturing condition optimization device 100 are capable of communicating with each other via a network 499.

[0020] The manufacturing facility 410 manufactures the product 460 under the manufacturing conditions set by the manufacturing management device 430. The manufacturing facility 410 transmits the individual manufacturing conditions of the product 460 to the manufacturing management device 430. The product 460 is transported to the inspection equipment 420 and undergoes inspection. The product 470 that passes the inspection is shipped as a non-defective product. The product 480 that fails the inspection is a defective product and is discarded.

[0021] The inspection equipment 420 measures one or more qualities (quality items, inspection items) of each product 460, and if all the measured values ​​are within the respective standard ranges (between upper and lower limits), the product is deemed to have passed. The inspection equipment 420 transmits the measured quality values ​​(also simply referred to as quality or quality value) of each product 460 to the manufacturing management device 430. The manufacturing management device 430 stores manufacturing management data 440 (see FIG. 5 described later) that stores the individual manufacturing conditions and quality of the product 460. The manufacturing condition optimization device 100 obtains manufacturing conditions that maximize the yield of the product 460 based on the manufacturing management data 440. This makes it possible to improve the manufacturing index.

[0022] <Outline of manufacturing condition optimization device> Fig. 2 is a diagram for explaining an outline of the processing of the manufacturing condition optimization device 100 according to this embodiment. In the regression model generation processing, a regression model 150 is generated from learning data 140. The learning data 140 is data that is a set of a change amount of the manufacturing conditions in the manufacturing equipment 410 (see Fig. 1) and a change amount of the quality of the product 460 in the inspection equipment 420 when the manufacturing conditions are changed by the change amount. The regression model 150 is a model that estimates a change amount of the quality (quality change amount) of the product 460 from the change amount of the manufacturing conditions (manufacturing condition change amount), and is, for example, a linear regression model.

[0023] In the optimization process, the manufacturing condition optimization device 100 uses the manufacturing condition variation as an input (variable) and a yield estimation function that estimates the yield of the product 460 as an objective function to solve an optimization problem that maximizes the yield. The yield estimation function uses the regression model 150 to calculate an estimate of the quality variation from the manufacturing condition variation that is the input, and estimates the yield by obtaining an estimate of the quality from the quality variation. The solution to this optimization problem becomes the manufacturing condition variation that maximizes the yield. The manufacturing condition optimization device 100 transmits this manufacturing condition variation to the manufacturing management device 430. The manufacturing management device 430 can maximize the yield of the product 460 by setting the manufacturing conditions of the manufacturing equipment 410 according to the manufacturing condition variation.

[0024] <Configuration of manufacturing condition optimization device> 3 is a functional block diagram of the manufacturing condition optimization apparatus 100 according to this embodiment. The manufacturing condition optimization apparatus 100 includes a control unit 110, a storage unit 130, a communication unit 170, and an input / output unit 180. The communication unit 170 transmits and receives communication data to and from other devices including a manufacturing management apparatus 430. The input / output unit 180 is connected to user interface devices such as a display, a keyboard, and a mouse.

[0025] The storage unit 130 is composed of a ROM (Read Only Memory), a RAM (Random Access Memory), an SSD (Solid State Drive), etc. The storage unit 130 stores a program 131, learning data 140 (see FIG. 4 described later), and a regression model 150. The program 131 describes the procedures for a regression model generation process (see FIG. 6 described later), an optimization process, and an objective function (yield estimation function) process (see FIG. 7 described later).

[0026] <<Configuration of manufacturing condition optimization device: learning data>> 4 is a data configuration diagram of the learning data 140 according to this embodiment. The learning data 140 is, for example, data in a table format, and one row (record) includes columns (attributes) of manufacturing condition variation 141 and quality variation 142. The manufacturing condition change amount 141 indicates the amount of change when the manufacturing conditions are changed in the manufacturing equipment 410. The manufacturing conditions include one or more items, and the manufacturing condition change amount 141 indicates the amount of change in the one or more items. The quality change amount 142 indicates the amount of change in quality of the product 460 in the inspection equipment 420 when a change in the manufacturing conditions indicated in the manufacturing condition change amount 141 is made. The quality change amount 142 includes the amount of change in one or more qualities (inspection measurement values, quality values) related to one or more products 460.

[0027] <Configuration of manufacturing condition optimization device: Control unit> Returning to FIG. 3, the control unit 110 is configured to include a CPU (Central Processing Unit), and includes a learning data generation unit 111, a regression model generation unit 112, an optimization processing unit 113, an objective function processing unit 114, a quality change estimation unit 115, a quality estimation unit 116, and a yield estimation unit 117. The learning data generating unit 111 acquires manufacturing management data 440 (see FIG. 5, which will be described later) stored in the manufacturing management device 430, and generates learning data 140 (see FIG. 4).

[0028] 5 is a data configuration diagram of manufacturing management data 440 according to this embodiment. The manufacturing management data 440 is, for example, data in a table format, and one row (record) includes columns (attributes) of product identification information 441 (referred to as product ID (identifier) ​​in FIG. 5), manufacturing date and time 442, manufacturing conditions 443, quality 444, and inspection results 445. Product identification information 441 is information for identifying each individual product 460, such as a serial number. Manufacturing date and time 442 is the date and time when the product 460 was manufactured. Manufacturing conditions 443 are the manufacturing conditions in the manufacturing equipment 410 when the product 460 was manufactured. Quality 444 is a measurement value of the inspection of the product 460 by the inspection equipment 420. Inspection result 445 indicates the pass / fail (OK / NG) that is the inspection result of the product.

[0029] Returning to FIG. 3, the learning data generation unit 111 acquires the manufacturing management data 440 and divides the data into groups of products that are manufactured consecutively and have the same manufacturing conditions 443. Next, the learning data generation unit 111 obtains the amount of change in manufacturing conditions 443 for each group compared to the previous group, and sets this as the amount of change in manufacturing conditions. Next, the learning data generation unit 111 obtains the difference between the quality 444 of each product in the group and the average value of the quality 444 in the previous group, and sets this as the amount of change in quality. The learning data generation unit 111 adds a record to the learning data 140 (see FIG. 4), and sets the amount of change in manufacturing conditions 141 and the amount of change in quality 142 as the obtained amount of change in manufacturing conditions and the amount of change in quality.

[0030] The regression model generating unit 112 generates a regression model 150 based on learning data 140. The learning data 140 is data indicating the amount of change in product quality (amount of quality change 142) when the manufacturing conditions are changed by the amount of manufacturing condition change 141. The regression model 150 is a model that is referenced when estimating (calculating) the amount of quality change from the amount of manufacturing condition change, and is, for example, a linear regression model. The regression model 150 is calculated as β shown in the following formula (1).

[0031]

number

[0032] X is a matrix indicating the manufacturing conditions referred to when generating the learning data 140. ΔX is a matrix indicating the amount of change in the manufacturing conditions included in the learning data 140. ΔX T denotes the transpose matrix of ΔX. Δy is a vector indicating one quality item (measured value of inspection) of a product included in the learning data 140. There are as many Δy as there are quality items (N, which will be described later). β is an estimate of the partial regression coefficient, and there are as many as quality items.

[0033] The optimization processor 113 solves an optimization problem that maximizes the yield score shown in equation (2) below, using a yield estimation function described later as an objective function, and calculates the amount of change in the manufacturing conditions that maximizes the yield. Methods for solving the optimization problem include an optimization algorithm that does not use derivatives, a local search method, a simulated annealing method, a tabu search method, and a genetic algorithm.

[0034]

number

[0035] N is the number of qualities (quality items). i is the index of the quality items, from the first quality item to the Nth quality item. M is the number of products for which the yield is to be estimated, for example, the number of products manufactured in a given period in the past. The optimization problem in this embodiment is to find the amount of change in manufacturing conditions that maximizes the number of products that pass inspection among the M products to be manufactured. j is the index of the M products to be manufactured, going from the first product to the Mth product. l L,i is the lower limit of the i-th quality item, and is the lower limit of the acceptance range of the i-th quality item. l U,i is the upper limit of the i-th quality item, and is the upper limit of the acceptance range of the i-th quality item. y i,j is the predicted value of the i-th quality item in the j-th product. y i,j follows the probability distribution shown in equation (3) below.

[0036]

number

[0037] N(μ,σ 2 ) is the mean μ, variance σ 2 shows a normal distribution. Δy iis an estimate of the average change in the i-th quality item and is calculated using equation (4) described later. y 0,i,j is the actual measured value of the i-th quality item in the j-th product among the M products most recently manufactured. SE(Δy i ) is the standard error of the estimated value of the change in the i-th quality item and is calculated using equation (5) described later. s i is the standard error of the regression for the i-th quality item, and is calculated using equation (6) described below.

[0038]

number

[0039] Here, Δx in formula (4) is a vector indicating the amount of change in the manufacturing conditions. β is an estimate of the partial regression coefficient of the trained linear model and is calculated using the above-mentioned equation (1).

[0040]

number

[0041] M0 is the number of products when β is calculated. X is a matrix indicating the manufacturing conditions referred to when generating the learning data 140. X T denotes the transpose of X. Δx T denotes the transpose vector of Δx.

[0042]

number

[0043] K is the number of manufacturing conditions. RSS is the residual sum of squares.

[0044] The objective function processor 114 (yield rate estimation function processor) estimates (calculates) the yield rate from the amount of change in manufacturing conditions as a yield rate estimation function, which is an objective function. In detail, the objective function processor 114 obtains a probability distribution of the quality of the product 460 shown in formula (3) from the amount of change in the quality of the product 460, and estimates the probability that the product 460 will pass the inspection by the inspection equipment 420 based on this probability distribution. The objective function processor 114 estimates the yield rate using a quality change amount estimation unit 115, a quality estimation unit 116, and a yield rate estimation unit 117, which will be described later.

[0045] The quality variation estimation unit 115 calculates the quality variation of the product 460 manufactured by the manufacturing equipment 410 from the manufacturing condition variation of the manufacturing equipment 410 using the regression model 150. In detail, the quality variation estimation unit 115 calculates the estimated value Δy of the quality variation from the manufacturing condition variation Δx using the formula (4). i In addition, β in formula (4) is the β corresponding to the i-th quality item among the N βs.

[0046] The quality estimation unit 116 estimates y i,j In detail, the distribution of the estimated quality change Δy i and the measured value y of the i-th quality item of the j-th product among the M products most recently manufactured. 0,i,j The sum of these is calculated as the mean value of the distribution. The variance of the distribution is the estimated value of the quality change Δy i The standard error of SE(Δy i ) (see equation (5)) and the standard error of the regression s i (See formula (6)). The normal distribution of the mean and variance is calculated as y i,j (see equation (3)).

[0047] The yield estimation unit 117 calculates y i,j From the distribution of y i,j The probability that P(l L,i ≦y i,j ≦l U,i ) and calculate the yield score (see formula (2)). As described above, the objective function processing unit 114 estimates the yield using the quality variation estimation unit 115, the quality estimation unit 116, and the yield estimation unit 117.

[0048] <Regression model generation process> 6 is a flowchart of the regression model generation process according to this embodiment. The regression model generation process is executed at a predetermined timing, for example, after a predetermined cycle or after a manufacturing condition is changed and a predetermined number of products are inspected. In step S11, the learning data generating unit 111 acquires the manufacturing management data 440 (see FIG. 5) from the manufacturing management device 430 (see FIG. 1). In step S12, the learning data generating unit 111 generates learning data 140 (see FIG. 4) from the acquired manufacturing management data 440. In step S13, the regression model generating unit 112 generates a regression model 150 (see equation (1)) from the training data 140.

[0049] <Objective function processing> 7 is a flowchart of the objective function processing executed by the objective function processing unit 114 according to this embodiment. The objective function processing is called and executed at a necessary timing when the optimization processing unit 113 solves the optimization problem (see step S31 in FIG. 8 described later). In step S21, the objective function processing unit 114 starts the process of repeating steps S22 to S26 for each of the M products whose yields are to be estimated. In step S22, the objective function processing unit 114 starts the process of repeating steps S23 to S25 for each of the N qualities (quality items).

[0050] In step S23, the quality variation estimation unit 115 calculates the quality variation from the manufacturing condition variation using the regression model 150. In detail, the quality variation estimation unit 115 calculates the estimated value Δy of the quality variation from the manufacturing condition variation Δx using the formula (4). i Calculate. In step S24, the quality estimation unit 116 estimates y i,jCalculate the distribution (see formula (3)).

[0051] In step S25, the yield estimation unit 117 calculates y i,j From the distribution of y i,j The probability that P(l L,i ≦y i,j ≦l U,i ) is required. In step S26, the yield estimation unit 117 estimates the probability Π i=1,···,N P(l L,i ≦y i,j ≦l U,i In this embodiment, Π i=1,···,N P(l L,i ≦y i,j ≦l U,i ) instead of Σ i=1,···,N log(P(l L,i ≦y i,j ≦l U,i )) is required. In step S27, the yield estimation unit 117 calculates the yield score (see formula (2)).

[0052] <Optimization process> 8 is a flowchart of the optimization process according to this embodiment. The optimization process is executed at a predetermined timing, for example, at a predetermined cycle or after a predetermined number of products have been manufactured. In step S31, the optimization processor 113 solves an optimization problem that maximizes the yield score (see formula (2)) using the yield estimation function as an objective function. In step S32, optimization processing unit 113 transmits the manufacturing condition change amount, which is the optimal solution, to manufacturing management device 430. Upon receiving the optimal solution, manufacturing management device 430 instructs manufacturing equipment 410 to change the manufacturing conditions by the manufacturing condition change amount.

[0053] <Features of the manufacturing condition optimization device> The manufacturing condition optimization device 100 calculates the actual measured value (y 0,i,j), the amount of change in the manufacturing conditions (amount of change in manufacturing conditions) that maximizes the yield is calculated. More specifically, the manufacturing condition optimization device 100 calculates the amount of change in the manufacturing conditions that results in the amount of change in quality that maximizes the yield. Since the manufacturing conditions are changed based on the quality of the most recently manufactured product for quality caused by unmeasured or uncontrollable disturbances, it becomes possible to set the manufacturing conditions so that the quality is improved according to the state of the current disturbance, which is considered to be the same as or almost the same as the most recent disturbance. By adjusting the manufacturing conditions according to the state of the uncontrollable disturbance, it becomes possible to maximize the product yield and improve the manufacturing index.

[0054] <<Variation: Regression Model>> The regression model 150 in the above embodiment is a linear regression model, but may be another model. For example, a Gaussian process regression model may be used instead of a (generalized) linear model. Alternatively, a machine learning model such as a neural network model may be used. In detail, a machine learning model that predicts the amount of quality change from the amount of manufacturing condition change may be generated using the learning data 140 (see FIG. 4) as teacher data and used in the yield estimation function (objective function). By using a Gaussian process regression model or a neural network model, it is possible to improve the prediction accuracy when there is a high-order interaction between explanatory variables (see manufacturing condition change 141 in FIG. 4). Furthermore, in addition to the manufacturing condition variation 141, the manufacturing condition before being converted into the variation (see manufacturing condition 443 in FIG. 5) may be added as an explanatory variable in the learning data 140 used to generate the regression model 150. In this case, the regression model 150 becomes a model that estimates the quality variation from the manufacturing condition and the variation from the manufacturing condition.

[0055] <<Variation: Actual quality measurements>> In the above embodiment, the manufacturing condition optimization device 100 calculates the quality (y 0,i,j Based on the distribution of product quality (N(Δy i +y 0,i,j ,SE(Δy i ) 2 +s i2 The yield is estimated by calculating the average product quality (y 0,i ) and calculate the product-independent (one product) quality distribution (N(Δy i +y 0,i ,SE(Δy i ) 2 +s i 2 )) may be calculated to estimate the yield. Since the repeated process (j=1, ,M) for each product is eliminated (regarding M=1), the manufacturing condition optimization device 100 can perform the optimization process at high speed and more frequently.

[0056] <<Variation: Quality Inspection Results>> In the above embodiment, the manufacturing condition optimization device 100 acquires a quality measurement value as the inspection result (see step S11 in FIG. 6). Instead of a quality (quality item, inspection item) measurement value, an inspection result expressed as a binary value of pass or fail, such as an external inspection, or the number of defects in one product may be acquired. In the case of an inspection result expressed as a binary value, the regression model 150 is a logistic regression model, and in the case of the number of defects, the regression model 150 is a Poisson regression model. Both the logistic regression model and the Poisson regression model are included in the generalized linear model.

[0057] Other variations Although some embodiments and modifications of the present invention have been described above, these embodiments are merely illustrative and do not limit the technical scope of the present invention. For example, in the above embodiment, the objective function processing unit 114, the quality variation estimation unit 115, the quality estimation unit 116, and the yield estimation unit 117 are divided for convenience of explanation, but the yield estimation function may be one functional unit, for example, the yield estimation unit. In other words, the yield estimation unit may use a regression model to estimate the yield of the product 470 having a quality that passes the inspection by the inspection equipment 420 when the manufacturing conditions of the product 460 by the manufacturing equipment 410 are changed. In the above embodiment, the predicted value of quality, y i,jThe distribution of (see equation (3)) is a normal distribution, but may also be a binomial distribution or a Poisson distribution.

[0058] The present invention can take various other embodiments, and furthermore, various modifications such as omissions and substitutions can be made without departing from the gist of the present invention. These embodiments and modifications are included in the scope and gist of the invention described in this specification, etc., and are included in the scope of the invention described in the claims and their equivalents. [Explanation of symbols]

[0059] 100 Manufacturing condition optimization device 110 Control section 111 Learning Data Generation Unit 112 Regression model generation unit 113 Optimization Processing Unit 114 Objective function processing unit (yield estimation unit) 115 Quality change estimation unit (yield estimation unit) 116 Quality Estimation Department (Yield Estimation Department) 117 Yield Estimation Section 140 Training Data 150 Regression Models. 410 Manufacturing equipment 420 Inspection Equipment 430 Manufacturing control equipment 440 Manufacturing Management Data 460 products

Claims

1. a yield estimation unit that estimates a yield of the product having a quality that passes an inspection by an inspection facility based on a quality of the product at a predetermined timing when a manufacturing condition of the product by the manufacturing facility is changed at the predetermined timing; an optimization processing unit that calculates an amount of change from the manufacturing conditions at the predetermined timing so as to maximize the yield; Manufacturing condition optimization device.

2. a quality change amount estimation unit that estimates a change amount of quality of the product manufactured by the manufacturing equipment from the change amount of the manufacturing condition, The yield estimation unit includes: estimating a probability that the product will pass an inspection by the inspection facility based on the amount of change in quality of the product estimated by the quality change amount estimation unit; A yield of the product having a quality that passes the inspection by the inspection equipment is estimated from the probability of passing the inspection.

2. The manufacturing condition optimization device according to claim 1 .

3. A regression model generating unit is further provided, The regression model generation unit generating a regression model for estimating a change in quality of the product from a change in the manufacturing condition based on corresponding data between a change in the manufacturing condition in the past and a change in quality of the product due to the change in the manufacturing condition; The quality change amount estimation unit The amount of change in the quality of the product is estimated from the amount of change in the manufacturing conditions using the regression model.

3. The manufacturing condition optimization device according to claim 2.

4. The regression model is A generalized linear model, a Gaussian process regression model, or a neural network model.

4. The manufacturing condition optimization device according to claim 3.

5. The optimization processing unit: Calculate the amount of change in the manufacturing conditions that maximizes the yield by using at least one of an optimization algorithm that does not use derivatives, a local search method, a simulated annealing method, a tabu search method, and a genetic algorithm.

2. The manufacturing condition optimization device according to claim 1 .

6. The yield estimation unit includes: When estimating the probability that the product will pass the inspection by the inspection equipment from the amount of change in quality, the probability that the product will pass the inspection by the inspection equipment is estimated based on the quality of a predetermined number of products most recently manufactured.

3. The manufacturing condition optimization device according to claim 2.

7. The yield estimation unit includes: determining a probability distribution of the quality of the product from the amount of change in the quality of the product; Estimating the probability that the product will pass inspection by the inspection equipment based on the probability distribution.

3. The manufacturing condition optimization device according to claim 2.

8. The probability distribution is a normal distribution, a binomial distribution, or a Poisson distribution.

8. The manufacturing condition optimization device according to claim 7.

9. a regression model generating unit that generates a regression model for estimating a change in quality of the product from the quality of the product at a predetermined timing, using a change in manufacturing conditions from the manufacturing conditions at the predetermined timing as an explanatory variable, based on correspondence data between a change in manufacturing conditions in the past and a change in quality of the product due to the change in the manufacturing conditions Regression model generator.

10. A program for causing a computer to function as the manufacturing condition optimization device according to any one of claims 1 to 8 or as the regression model generation device according to claim 9.

11. A manufacturing condition optimization method for a manufacturing condition optimization device, comprising: a step of estimating a yield of the product having a quality that passes an inspection by an inspection facility based on a quality of the product at a predetermined time when a manufacturing condition of the product by the manufacturing facility is changed at the predetermined time; Calculating the amount of change from the manufacturing conditions at the predetermined timing so as to maximize the yield. Method for optimizing manufacturing conditions.

12. A regression model generation method for a regression model generation device, comprising: A step of generating a regression model for estimating a change in quality of the product from the quality of the product at a predetermined timing, using a change in manufacturing conditions from the manufacturing conditions at the predetermined timing as an explanatory variable, based on correspondence data between a change in manufacturing conditions in the past and a change in quality of the product due to the change in the manufacturing conditions Regression model generation methods.

Images