Design optimization device, design optimization method, program, and recording medium

Abstract

To select parameters that contribute to the optimization of manufactured products from a vast and complex manufacturing parameter space without relying on implicit knowledge or trial and error to optimize a device design.SOLUTION: A design optimization device 1 includes an optimal manufacturing environment determination unit 11 that determines an optimal manufacturing environment within a manufacturing device when objective variables representing optimal indices of manufactured products manufactured by the manufacturing device are optimized and a manufacturing parameter selection unit 12 that selects one or more manufacturing parameters that contribute to realizing an optimal manufacturing environment from among a predetermined plurality of manufacturing parameters for the manufacturing device.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to a design optimization apparatus, a design optimization method, a program, and a recording medium. [Background technology]

[0002] In recent years, with the improvement of computer calculation performance, numerical simulation technology and machine learning technology have been rapidly developed. These technologies are being applied to the high-speed optimization of process conditions of manufacturing equipment and parameters related to equipment design and manufacturing (for example, see Patent Documents 1 to 3). [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] N. Dropka and M. Holena, “Optimization of magnetically driven directional solidification of silicon using artificial neural networks and Gaussian process models,” J Cryst Growth, vol. 471, pp. 53.61, Aug. 2017, doi: 10.1016 / j.jcrysgro.2017.05.007. [Non-Patent Document 2] X. Liu et al., “Data-Driven Optimization and Experimental Validation for the Lab-Scale Mono-Like Silicon Ingot Growth by Directional Solidification,” ACS Omega, vol. 7, no. 8, pp. 6665.6673, Mar. 2022, doi: 10.1021 / acsomega.1c06018. [Non-Patent Document 3] W. Yu et al., “Geometrical design of a crystal growth system guided by a machine learning algorithm,” CrystEngComm, vol. 23, no. 14, pp. 2695.2702, Apr. 2021, doi: 10.1039 / d1ce00106j. Summary of the Invention [Problem to be solved by the invention]

[0004] However, there are still two major challenges, especially in material manufacturing process technology. The first is that the space of parameters that characterize the manufacturing process (hereinafter referred to as "manufacturing parameters") is vast and complex. In addition to the manufacturing parameters that can be adjusted by changing the values, such as the heater temperature and the moving speed of the parts, which contribute to the quality of the material and manufacturing efficiency, there are also those related to the design of the equipment, such as the position and size of the equipment parts (i.e., those that cannot be adjusted by changing the values ​​after the equipment is installed). Furthermore, there are various variations in the manufacturing process for each material, and the manufacturing parameters that control each process also interact with each other in a complex manner, making it difficult to handle them in a unified manner. The second challenge is the cost in terms of time and money. In order to optimize the quality and manufacturing efficiency of the product, the manufacturing process development involves repeated cycles of producing the material for the product, evaluating the produced material, and verifying the manufacturing process conditions according to the evaluation results. Usually, the success or failure of such a cycle depends heavily on tacit knowledge, such as the experience and proficiency of the designer and operator. However, as mentioned in the first issue, the parameter space that characterizes the manufacturing process is vast and complex, so conventional methods that rely on the designer's tacit knowledge and trial-and-error processes require a lot of time and money. The technologies disclosed in Patent Documents 1 to 3 do not adequately solve these issues.

[0005] The present disclosure has been made in consideration of these circumstances, and its purpose is to select parameters that contribute to optimizing the product from a vast and complex manufacturing parameter space and to optimize equipment design without relying on tacit knowledge or trial and error. [Means for solving the problem]

[0006] In order to solve the above problems, a design optimization device of one embodiment of the present invention includes an optimal manufacturing environment determination unit that determines an optimal manufacturing environment in the manufacturing device when a target variable representing an optimal index of a product manufactured by the manufacturing device is optimized, and a manufacturing parameter selection unit that selects one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from a predetermined number of manufacturing parameters related to the manufacturing device.

[0007] In an embodiment, the design optimization apparatus may include a manufacturing parameter optimization unit that optimizes the manufacturing parameters selected by the manufacturing parameter selection unit.

[0008] In an embodiment, the optimal manufacturing environment determination unit may perform machine learning using the manufacturing environment in the manufacturing equipment and the objective variables of the product as training data.

[0009] In one embodiment, the optimal manufacturing environment determination unit may execute a simulation to obtain a target variable of a product that is realized when a manufacturing environment within the device is given, in order to create training data.

[0010] In one embodiment, the manufacturing parameter selection unit performs a parameter analysis on a response variable of a product using a plurality of predetermined manufacturing parameters related to a manufacturing device as explanatory variables.

[0011] In one embodiment, the parametric analysis may be a multiple regression analysis.

[0012] In an embodiment, the manufacturing parameter optimization unit may perform machine learning using the manufacturing parameters selected by the manufacturing parameter selection unit and the manufacturing environment within the manufacturing equipment as training data.

[0013] Another aspect of the present invention is a design optimization method, which includes an optimal manufacturing environment determination step of determining an optimal manufacturing environment in the manufacturing equipment when a target variable representing an optimal index of a product manufactured by the manufacturing equipment is optimized, and a manufacturing parameter selection step of selecting one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing equipment.

[0014] Yet another aspect of the present invention is a program that causes a computer to execute a design optimization method including an optimal manufacturing environment determination step of determining an optimal manufacturing environment in the manufacturing equipment when a target variable representing an optimal index of a product manufactured by the manufacturing equipment is optimized, and a manufacturing parameter selection step of selecting one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing equipment.

[0015] Yet another aspect of the present invention is a recording medium, which records a program including an optimal manufacturing environment determination step of determining an optimal manufacturing environment in the manufacturing apparatus when a target variable representing an optimal index of a product manufactured by the manufacturing apparatus is optimized, and a manufacturing parameter selection step of selecting one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing apparatus.

[0016] Yet another embodiment of the present invention is a design optimization apparatus. This design optimization apparatus includes an optimal boundary condition determination unit that determines optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a part to which an objective variable and a structural variable or a process variable are reflected, a parameter selection unit that optimizes the boundary conditions so that the objective function is the best within a range set based on a simulation result, and a parameter optimization unit that models the relationship between the optimal boundary condition determined by the optimal boundary condition determination unit as the objective function and the process variable and the structural variable as explanatory variables by machine learning, and optimizes the process variable and the structural variable so as to approach the optimal boundary condition determined by the optimal boundary condition determination unit. The optimal boundary condition determination unit executes machine learning using the boundary conditions and the objective function such as a physical quantity to be optimized as teacher data.

[0017] Yet another aspect of the present invention is a design optimization method. This method includes an optimal boundary condition determination step of determining optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a part reflecting an objective variable and a structural variable or a process variable, a parameter selection step of optimizing the boundary conditions so that the objective function is the best within a range set based on a simulation result, and a parameter optimization step of modeling the relationship between the optimal boundary condition determined in the optimal boundary condition determination step as the objective function and the process variable and the structural variable as explanatory variables by machine learning, and optimizing the process variable and the structural variable so as to approach the optimal boundary condition determined in the optimal boundary condition determination step. The optimal boundary condition determination step performs machine learning using the boundary conditions and the objective function such as a physical quantity to be optimized as training data.

[0018] Yet another aspect of the present invention is a program. This program causes a computer to execute an optimal boundary condition determination step of determining optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a portion to which an objective variable and a structural variable or a process variable are reflected, a parameter selection step of optimizing the boundary conditions so that the objective function is the best within a range set based on a simulation result, and a parameter optimization step of modeling the relationship between the optimal boundary condition determined in the optimal boundary condition determination step as the objective function and the process variable and the structural variable as explanatory variables by machine learning, and optimizing the process variable and the structural variable so as to approach the optimal boundary condition determined in the optimal boundary condition determination step. The optimal boundary condition determination step executes machine learning using the boundary conditions and the objective function such as a physical quantity to be optimized as teacher data.

[0019] Yet another aspect of the present invention is a recording medium. This recording medium records a program for causing a computer to execute the following steps: an optimal boundary condition determination step for determining optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a portion to which an objective variable and a structural variable or a process variable are reflected; a parameter selection step for optimizing the boundary conditions so that the objective function is the best within a range set based on a simulation result; and a parameter optimization step for modeling the relationship between the optimal boundary condition determined in the optimal boundary condition determination step as the objective function and the process variable and the structural variable as explanatory variables by machine learning, and optimizing the process variable and the structural variable so as to approach the optimal boundary condition determined in the optimal boundary condition determination step. In the optimal boundary condition determination step, machine learning is executed using the boundary conditions and the objective function such as a physical quantity to be optimized as teacher data.

[0020] Any combination of the above components, and conversion of the present disclosure into a method, device, system, recording medium, computer program, etc. are also valid aspects of the present disclosure. Effect of the Invention

[0021] According to the present disclosure, it is possible to select parameters that contribute to optimizing the product from a vast and complex manufacturing parameter space and optimize the device design without relying on tacit knowledge or trial and error. [Brief description of the drawings]

[0022] [Figure 1] FIG. 1 is a functional block diagram of a design optimization device according to a first embodiment. [Diagram 2] FIG. 1 is a schematic diagram of crystal growth of polycrystalline Si in a crystal growth furnace equipped with a crucible. [Diagram 3] FIG. 1 shows an example of 13 nodes defined on the wall of a crucible and the temperature distribution within the crucible. [Figure 4] FIG. 4 shows an example of a temperature progression given to one of the 13 nodes of FIG. [Diagram 5] FIG. 2 is a schematic diagram of a crucible model for performing a simulation. [Figure 6] FIG. 13 is a diagram showing a dislocation density distribution obtained by simulation when a certain temperature profile is followed. [Figure 7] FIG. 7 is a diagram showing the transition over time of the solid-liquid interface shape obtained by simulation when following the same temperature profile as in FIG. 6. [Figure 8] 4A to 4C are diagrams showing examples of various temperature distribution transitions for the 13 nodes in FIG. 3. [Figure 9] FIG. 1 is a schematic diagram showing the structure of a DNN model. [Figure 10] FIG. 13 is a diagram showing a comparison between the simulation results and the predicted results of the DNN model regarding the dislocation density distribution. [Figure 11] This figure shows a comparison between the simulation results and the predicted results of the DNN model regarding the time progression of the solid-liquid interface shape. [Figure 12] FIG. 13 shows a YY plot for evaluating the prediction accuracy of the DNN model for all test data regarding dislocation density. [Figure 13] FIG. 13 shows a YY plot for evaluating the prediction accuracy of the DNN model for all test data regarding crystallization time. [Figure 14] FIG. 13 is a diagram showing the transition of optimal temperature distribution on the side wall of the crucible obtained by applying a genetic algorithm. [Figure 15] FIG. 13 is a diagram showing a comparison of dislocation density distribution before and after optimization. [Figure 16] FIG. 13 is a diagram showing a comparison of the results before and after optimization of the crystal growth time. [Figure 17] FIG. 4 is a diagram showing manufacturing parameters of the polycrystalline silicon manufacturing apparatus in the present embodiment. [Figure 18] This is a YY plot of ΔTtop. [Figure 19] 4 is a YY plot of ΔTwall. [Figure 20] This is a YY plot for ΔTbottom. [Figure 21] FIG. 13 is a heat map showing the standard multiple regression coefficient matrix of the constructed multiple regression model. [Figure 22] FIG. 4 is a functional block diagram of a design optimization device according to a second embodiment. [Figure 23] FIG. 1 is a schematic diagram showing the structure of a DNN model. [Figure 24] FIG. 13 is a diagram showing the time progression of the vertical motion of a crucible used in the training data. [Diagram 25] FIG. 10 shows a comparison of simulation results and DNN model predictions for temperatures at 13 nodes on the crucible wall. [Figure 26] This is a parity plot. [Figure 27] FIG. 13 is a diagram showing the Pareto surface of ΔTwall and ΔTtop obtained by applying a genetic algorithm. [Figure 28] FIG. 13 is a diagram showing the Pareto surface of ΔTwall and ΔTbottom obtained by applying a genetic algorithm. [Figure 29] FIG. 13 is a diagram showing a comparison of the time progression of the vertical coordinate of the crucible obtained as an optimal solution by the above machine learning with the one before optimization. [Diagram 30]This figure shows a comparison of the optimal temperature distribution transition (Target) and the simulation results (GA) using the equipment structure and crucible movement speed obtained as the optimal solution by applying a genetic algorithm, with regard to the dislocation density distribution. [Diagram 31] This figure shows a comparison of the optimum temperature distribution transition (Target) and the simulation results using the device structure and crucible moving speed obtained as the optimum solution by applying a genetic algorithm, regarding the crystal growth time distribution (solid-liquid interface shape transition). [Diagram 32] This figure shows a comparison of the optimal temperature distribution (Optimal_temperature_transtion) and the simulation results (GA) using the equipment structure and crucible movement speed obtained as the optimal solution by applying a genetic algorithm, regarding the temperature transition of the crucible side wall. [Diagram 33] This figure shows a comparison between the simulation results and the predicted results of the DNN model 52 regarding the temperature distribution transition of the crucible side wall under optimal conditions obtained by the genetic algorithm. [Diagram 34] This is a scatter plot of the polynomial coefficients associated with the crucible movement speed selected as the optimal solution, for the teacher data (Teacher_data) and the optimal solutions (Optimal_data) obtained by the genetic algorithm. [Diagram 35] This is a scatter plot of the polynomial coefficients associated with the crucible movement speed selected as the optimal solution, for the teacher data (Teacher_data) and the optimal solutions (Optimal_data) obtained by the genetic algorithm. [Diagram 36] This is a scatter plot of the polynomial coefficients associated with the crucible movement speed selected as the optimal solution, for the teacher data (Teacher_data) and the optimal solutions (Optimal_data) obtained by the genetic algorithm. [Figure 37] 13 is a flowchart showing a processing procedure of a design optimization method according to a third embodiment. [Figure 38] FIG. 13 is a functional block diagram of a design optimization device according to a sixth embodiment. [Figure 39] 13 is a flowchart showing a processing procedure of a design optimization method according to a seventh embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0023] Hereinafter, the present disclosure will be described based on preferred embodiments with reference to the drawings. In the embodiments and modified examples, identical or equivalent components and parts are given the same reference numerals, and duplicated descriptions are omitted as appropriate. In addition, the dimensions of the parts in each drawing are enlarged or reduced as appropriate for ease of understanding. In addition, in each drawing, some elements that are not important for explaining the embodiments are omitted. In addition, terms including ordinal numbers such as first, second, etc. are used to explain various components, but these terms are used only for the purpose of distinguishing one component from other components, and the components are not limited by these terms.

[0024] In the following specific embodiments, a typical example will be described focusing on a polycrystalline silicon device and its manufacturing process for manufacturing polycrystalline silicon (silicon) by a directional solidification (DS) method, but it goes without saying that the present disclosure is not limited to this example.

[0025] In the DS method, after all the raw silicon is melted, a temperature gradient is applied to the melt to cause nucleation at the bottom of the crucible. The crucible is then lowered while maintaining the temperature gradient, causing the crystal to solidify and grow upward. The DS method is used in the pseudo-single crystal method, in which a seed crystal placed in advance at the bottom of the crucible grows, and in the production of High Performance (HP) polycrystalline silicon, in which the enlargement of crystal grains is suppressed by laying fine seed crystals over the entire surface.

[0026] In the DS method, nucleation occurs randomly. As a result, the size and crystal orientation of the crystal grains are also roughly random. Therefore, controlling this is considered important for producing high-quality ingots.

[0027] When considering applications to solar cells, it is desirable for the dislocation density of polycrystalline Si ingots grown by the DS method to be as small as possible. This is because the presence of high dislocation density regions at the top of the Si ingot or metal contamination at the bottom will cause a deterioration in the conversion efficiency of polycrystalline Si solar cells. In other words, with polycrystalline Si, the dislocation density of the Si ingot is one of the optimal indicators of the quality of the product.

[0028] Furthermore, from the viewpoint of shortening the manufacturing time and increasing the efficiency of the manufacturing process, it is desirable to complete the crystal growth in as short a time as possible. In other words, in the case of polycrystalline silicon, the crystal growth completion time is one of the optimal indicators for manufacturing efficiency.

[0029] Hereinafter, in this specification, indices for which optimization is aimed at for a product, such as dislocation density and crystal growth completion time (product quality / performance, manufacturing efficiency, manufacturing cost, etc.), are generally referred to as "objective variables." The objective of this disclosure is to identify and select, from among various manufacturing parameters, manufacturing parameters that contribute to optimizing the objective variables. A further objective of this disclosure is to determine equipment design conditions, manufacturing control conditions, process conditions, etc. that optimize the selected manufacturing parameters.

[0030] In the following embodiment, two steps are described: a step of determining an optimal manufacturing environment (first step), and a step of selecting and optimizing one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment (second step). However, the present disclosure is not limited to this, and can be expanded to three or more steps. In this case, the optimal manufacturing parameters obtained in the second step are fixed, and another set of manufacturing parameters that affect the same manufacturing environment as in the second step can be set, and parameter selection can be performed from there in a similar manner. Alternatively, another manufacturing environment can be set using the results of the second step, and parameter selection can be performed in a similar manner.

[0031] [First embodiment] 1 is a functional block diagram of a design optimization apparatus 1 according to a first embodiment. The design optimization apparatus 1 includes an optimal manufacturing environment determination unit 11 and a manufacturing parameter selection unit 12.

[0032] The optimal manufacturing environment determination unit 11 determines the optimal manufacturing environment in the manufacturing equipment when a target variable representing an optimal index of a product manufactured by the manufacturing equipment is optimized. The manufacturing parameter selection unit 12 selects one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing equipment.

[0033] Hereinafter, the operation of the design optimization apparatus 1 will be described by taking as an example a case where the product is polycrystalline Si and the manufacturing apparatus is a polycrystalline Si manufacturing apparatus using the DS method.

[0034] 2 is a schematic diagram showing the crystal growth of polycrystalline silicon in a crystal growth furnace equipped with a crucible 100 of a polycrystalline silicon manufacturing apparatus. This crystal growth furnace is equipped with three heaters 101, 102, and 103 whose temperatures can be set. The wall 104 of the crucible 100 is made of quartz (SiO2). The crucible stand 105 that supports the crucible 100 is equipped with a vertical movement mechanism.

[0035] The growth process of Si crystals in a crystal furnace is explained below. First, the seed crystal is placed in a crucible coated with a release agent (Si3N4) after the surface damage layer is removed with a HF:HNO3 (1:6) solution. After semiconductor-grade Si raw material is filled on top of this seed crystal, the temperature inside the crystal growth furnace is controlled by heaters 101, 102, and 103 of the crystal growth furnace, and all the raw material and the top 2 cm of the seed crystal are melted. Then, the crucible 100 is pulled down to solidify in one direction, and an ingot is grown. The shape of the solid-liquid interface in the initial crystal growth process is convex in the z direction of the figure.

[0036] [Determining the optimal manufacturing environment] It is known that in polycrystalline silicon manufacturing using the DS method, the temperature environment around the crystal has a large effect on the crystal quality and crystal growth time. Therefore, in order to optimize the objective variables of the product, the Si crystal, it is necessary to optimize the temperature environment around the crystal as the manufacturing environment. Therefore, the optimal manufacturing environment determination unit 11 determines the optimal manufacturing environment (temperature distribution and its time transition) in the manufacturing equipment (crystal growth furnace) when the objective variables (dislocation density and crystal growth time) of the product (polycrystalline silicon) are optimized.

[0037] In this example, 13 nodes are defined around the crucible 100 (specifically, the wall of the crucible 100), and the temperature of each node is changed over time. The dislocation density distribution and the time transition of crystal growth are calculated by simulation. 10 , P 11 , P 12 , P 13 3 shows an example of the temperature distribution in the crucible 100. However, since the crucible 100 is axially symmetrical with respect to the central axis AX (the same applies below), only the left half of the crucible 100 is shown in FIG.

[0038] FIG. 4 shows the 13 nodes P1 to P 13 The following shows an example of the temperature transition given to one of the nodes. Such a temperature profile is given to each node at 30 temperature points every 20 minutes for 600 minutes from the start of crystal growth.

[0039] Figure 5 is a schematic diagram of a model of the crucible 100 for carrying out the above simulation. In this model, it is assumed that crystal growth is carried out in an Ar gas atmosphere, and unsteady calculations are carried out by giving time-series temperature boundary conditions to all four sides of the model. In the Si ingot region, 62 x 108 = 6696 mesh coordinates (r, z) are given.

[0040] Figure 6 shows the dislocation density distribution obtained by simulation when a certain temperature profile is followed. Figure 7 shows the time transition of the solid-liquid interface shape obtained by simulation when the same temperature profile is followed. The time transition of this solid-liquid interface shape provides information on the crystal growth time. Therefore, by analyzing Figures 6 and 7, the average and maximum dislocation density and the crystal growth completion time, which are the objective variables of polycrystalline Si, can be obtained.

[0041] FIG. 8 shows examples of various temperature distribution transitions for 13 nodes P1 to P13.

[0042] The optimum manufacturing environment determination unit 11 executes machine learning based on the above simulation results to determine the optimum manufacturing environment (temperature distribution and its time transition) in the manufacturing equipment (crystal growth furnace) when the objective variables (dislocation density and crystal growth time) of the product (polycrystalline Si) are optimized. Specifically, the optimum manufacturing environment determination unit 11 executes machine learning using the temperature distribution of the 13 nodes defined on the wall of the crucible 100 and the dislocation density and crystal growth time at each coordinate in the crucible 100 as teacher data. As a result, the optimum manufacturing environment (temperature distribution and its time transition) in the manufacturing equipment (crystal growth furnace) when the objective variables (dislocation density and crystal growth time) are optimized is determined.

[0043] In this embodiment, the optimal manufacturing environment determination unit 11 performs machine learning using a DNN (deep neural network) model. Fig. 9 is a schematic diagram showing the structure of a DNN model 51 in this embodiment. The DNN model 51 is composed of a total of seven layers, including an input layer, five hidden layers, and an output layer. The hidden layer is composed of 150 nodes x 5 layers.

[0044] The input to the DNN model 51 is the temperature profile (T1, T2, T3, ... T4) at 30 times (i.e., 13 x 30 = 390 temperature points) at the 13 nodes on the wall of the crucible 100 selected above. 389 , T 390) and mesh coordinates (r, z) of multiple points (62 × 108 points) in the Si ingot region. For the temperature profile, a total of 30 time points are selected for each node, every 20 minutes for 600 minutes from the start of crystal growth.

[0045] The output from the DNN model 51 is the dislocation density log 10 N(r,z) and solidification time t(r,z). Dislocation density is output in common logarithm.

[0046] The optimal manufacturing environment determination unit 11 performs a simulation using the model of FIG. 7 to create training data for the DNN model 51. This simulation is performed with 134 sets of data for 390 (13 nodes × 30 time points) input combinations. The training data created in this way is 134 sets of Si ingot regions (62 × 108 points). These sets are randomly divided into 127 cases for learning and 7 cases for testing. Therefore, in order to match the coordinates (r, z), a data set of (134 × 62 × 108 pieces) is required for learning the DNN model 51.

[0047] Fig. 10 shows a comparison between the simulation results and the prediction results of the DNN model 51 regarding the dislocation density distribution at the end of cooling. Fig. 11 shows a comparison between the simulation results and the prediction results of the DNN model 51 regarding the time transition of the solid-liquid interface shape. The contour lines in Fig. 11 correspond to the transition of the solid-liquid interface shape during crystal growth.

[0048] As can be seen from Figures 10 and 11, the DNN model 51 is well trained and can instantly predict the solidification time, dislocation density, and interface shape during crystal growth in the entire region of the final ingot that represents the properties of the Si ingot.

[0049] 12 shows a YY plot for evaluating the prediction accuracy of the DNN model 51 for all test data with respect to the dislocation density. FIG. 13 shows a YY plot for evaluating the prediction accuracy of the DNN model 51 for all test data with respect to the crystallization time.

[0050] As can be seen from Figures 12 and 13, both the dislocation density distribution and the crystallization time can be predicted with extremely high accuracy. Furthermore, a simulation of a 10-hour process requires approximately 7 hours (25,200 seconds) of calculation time. In contrast, the constructed DNN model51 can output similar results in approximately 800 μs. In other words, by constructing a machine learning model, it is possible to obtain similar results approximately 31.5 million times faster than with simulation.

[0051] In this example, the optimum manufacturing environment determination unit 11 applies a genetic algorithm (GA) as a numerical optimization method to obtain the optimum manufacturing environment (optimum temperature distribution transition of the crucible wall).

[0052] The optimal temperature distribution transition on the crucible side wall obtained by applying the genetic algorithm is shown in Fig. 14. It can be seen from Fig. 14 that the optimal temperature distribution transition includes a temperature distribution transition in which cooling proceeds in the early stage of growth, while the temperature rises in the later stage of growth.

[0053] Fig. 15 shows a comparison of the results of the dislocation density distribution before and after optimization, and Fig. 16 shows a comparison of the results of the crystal growth time before and after optimization.

[0054] Figure 15 shows that the high dislocation density region at the top of the ingot that existed before optimization has disappeared after optimization, and Figure 16 shows that the shape of the solid-liquid interface at the end of crystal growth has been improved from a downward convex shape to a nearly flat shape.

[0055] [Selection of manufacturing parameters]

[0056] When there are many manufacturing parameters related to the shape of the device to be optimized, the number of simulations required to clarify the interactions between the parameters increases. To solve this problem, optimization can be performed by narrowing down the manufacturing parameters that have a large contribution to the objective variable from among the many manufacturing parameters, thereby allowing the optimization to proceed efficiently in terms of time. However, if the manufacturing parameters are narrowed down simply by using a simulation, it is necessary to redraw the drawing and redo the mesh division, which increases the cost in terms of time. To solve this problem, the manufacturing parameter selection unit 12 of the design optimization device 1 selects one or more manufacturing parameters that contribute to realizing an optimal manufacturing environment from among the manufacturing parameters related to the manufacturing device by performing a parameter analysis on the objective variable of the product using a predetermined number of manufacturing parameters related to the manufacturing device as explanatory variables.

[0057] For ease of understanding, the following description focuses on the manufacturing parameters that determine the internal structure of the furnace tube from among the many manufacturing parameter candidates, but it goes without saying that there are also manufacturing parameters related to the exterior of the furnace tube, for example.

[0058] 17 shows the manufacturing parameters of the polycrystalline silicon manufacturing apparatus in this embodiment. There are seven types of manufacturing parameters that are considered to have a large contribution to the temperature distribution around the crystal: d t : Core tube width r c :crucible stand radius t c : crucible base thickness d i :Insulation width t i : Insulation thickness r s : Susceptor diameter t g : Graphite base thickness

[0059] Table 1 is an L18 orthogonal array created by allocating these seven manufacturing parameters. [Table 1]

[0060] Table 2 shows the ranges of variation for the seven manufacturing parameters mentioned above. [Table 2]

[0061] The manufacturing parameter selection unit 12 performs parameter analysis on the objective variable of the product using a plurality of predetermined manufacturing parameters related to the manufacturing equipment as explanatory variables. The parameter analysis method may be any suitable data analysis method, such as multiple regression analysis, random forest regression, gradient boosting, SHAP, RIME, etc. In the following, a description will be given focusing on a method using multiple regression analysis as a representative example, but it goes without saying that the present disclosure is not limited thereto.

[0062] The manufacturing parameter selection unit 12 obtains the temperature distribution transition around the crucible by performing a non-steady calculation under the same temperature boundary conditions using a model in which each parameter is changed from the model shown in Figure 5. After that, the root mean squared error (RMSE) is calculated for the 13 node temperature profiles in Figure 3 at the top, side wall, and bottom of the crucible, and the average values ​​of the RMSE at each node of the top, side wall, and bottom (respectively, ΔT top , ΔT wall , ΔT bottom Then, each parameter is used as an explanatory variable, and ΔT top , ΔT wall , ΔT bottom A multiple regression model is constructed with the above as the objective variable. This multiple regression model is used to perform multivariate analysis to determine the degree of contribution of each parameter to the temperature distribution around the crystal.

[0063] Table 3 shows the results of calculating the objective variables after running 18 simulations. [Table 3] The contents listed in each No. in Table 3 correspond to the parameter values ​​listed in each No. in Table 1.

[0064] 18 to 20 are YY plots created to evaluate the prediction accuracy of the constructed multiple regression model. top Regarding ΔT wall Regarding ΔT bottom From the R2 Score (coefficient of determination) values ​​in Figures 18 to 20, it can be seen that the obtained multiple regression model has a certain degree of accuracy. Therefore, it is considered that the responsiveness of the objective variable to the explanatory variables (i.e., the contribution of each explanatory variable to the objective variable) can be seen generally using the magnitude of the regression coefficient of the multiple regression equation.

[0065] Figure 21 shows the standard multiple regression coefficient matrix of the constructed multiple regression model in a heat map. Standard multiple regression coefficient matrix. The values ​​in the heat map are ΔT top , ΔT wall , ΔT bottom is the regression coefficient of each manufacturing parameter term in the multiple regression equation.

[0066] From Figure 21, the radius of the crucible base r c and the graphite base thickness t g It can be seen that the response to is smaller than that of other explanatory variables. Therefore, in this embodiment, the five manufacturing parameters other than these two manufacturing parameters (core tube width d t , crucible base thickness t c , insulation width d i , insulation thickness t i , susceptor diameter r s ) can be selected as a manufacturing parameter that contributes to realizing an optimal manufacturing environment.

[0067] As described above, according to this embodiment, it is possible to realize an apparatus that selects parameters that contribute to optimizing a product from a vast and complex manufacturing parameter space without relying on tacit knowledge or trial and error.

[0068] [Second embodiment] 22 is a functional block diagram of a design optimization device 2 according to the second embodiment. The design optimization device 2 includes an optimal manufacturing environment determination unit 11, a manufacturing parameter selection unit 12, and a manufacturing parameter optimization unit 13. That is, the design optimization device 2 includes the manufacturing parameter optimization unit 13 in addition to the configuration of the design optimization device 1 in FIG.

[0069] The manufacturing parameter optimization unit 13 optimizes the manufacturing parameters selected by the manufacturing parameter selection unit 12 .

[0070] Here, the five manufacturing parameters (core tube width d t , crucible base thickness t c , insulation width d i , insulation thickness t i , susceptor diameter r s An embodiment in which the above-mentioned optimization is performed will be described.

[0071] [Optimization of manufacturing parameters] The manufacturing parameter optimization unit 13 optimizes the manufacturing parameters (core tube width d t , crucible base thickness t c , insulation width d i , insulation thickness t i , susceptor diameter r s ) and the objective variable of the product (polycrystalline Si) are used as training data to perform machine learning. As a result, the manufacturing parameters selected by the manufacturing parameter selection unit (core tube width d t , crucible base thickness t c , insulation width d i , insulation thickness t i , susceptor diameter r s ) is optimized.

[0072] In this embodiment, the manufacturing parameter optimization unit 13 performs machine learning using a DNN model. Fig. 23 is a schematic diagram showing the structure of the DNN model 52 of this embodiment. The DNN model 52 is composed of a total of seven layers, including an input layer, five hidden layers, and an output layer. The hidden layer is composed of five layers of 100 nodes.

[0073] The input to the DNN model 52 is the five manufacturing parameters (core tube width d t , crucible base thickness t c , insulation width d i , insulation thickness t i , susceptor diameter r s ), experimental time t, and vertical coordinate of the crucible z(t).

[0074] The output from the DNN model 52 is 13 nodes P1, P2, P3, P4, P5, P6, P7, P8, P9, P10, P11, P12, P13, P14, P15, P16, P17, P18, P19, P20, P21, P22, P23, P24, P25, P26, P27, P28, P29, P30, P 10 , P 11 , P 12 , P 13 Temperatures T1(t), T2(t), T3(t), T4(t), T5(t), T6(t), T7(t), T8(t), T9(t), T 10 (t), T 11 (t), T 12 (t), T 13 (t). By using the experimental time as one of the input layers, it is possible to obtain the temperatures at 13 locations at any time.

[0075] The manufacturing parameter optimization unit 13 performs a simulation using a partial model to create training data for the DNN model 52. The created training data is randomly divided into 135 cases for learning and 15 cases for testing. This means that a (150×46×13) data set is required for learning the DNN model 52 in order to associate the experimental time t, the vertical coordinate z of the crucible, and each node position.

[0076] Figure 24 shows the time progression of the vertical movement of the crucible used in the training data. The horizontal axis is the movement time, and the vertical axis is the vertical position. The initial position indicated by a circle and the final position indicated by a star are fixed.

[0077] Figure 25 shows temperatures T1(t), T2(t), T3(t), T4(t), T5(t), T6(t), T7(t), T8(t), T9(t), T 10 (t), T 11 (t), T 12 (t), T 13 Re (t): The simulation results are compared with the predicted results of the DNN model 52. The horizontal axis represents the elapsed time from the start of the experiment, and the vertical axis represents the absolute temperature.

[0078] The parity plot is shown in Fig. 26. This allows us to verify the prediction accuracy of the temperature distribution transition of the 13 nodes.

[0079] As can be seen from Figures 25 and 26, the DNN model 52 is sufficiently trained and can instantly predict the optimal temperature distribution transition around the crucible. In addition, a simulation of a 14-hour process requires a calculation time of about 6 hours (21,600 seconds). In contrast, the constructed DNN model 52 can output a similar result in about 800 μs. In other words, by constructing a machine learning model, a similar result can be obtained about 27 million times faster.

[0080] In this example, the manufacturing parameter optimization unit 13 applies a genetic algorithm (GA) to optimize the manufacturing parameters.

[0081] Figure 27 shows the ΔT wall , ΔT top Figure 28 shows the Pareto surface of ΔT wall , ΔT bottom27 and 28 show the Pareto surface of the above. In the DS method, the crystal growth proceeds with the vertical temperature gradient as the driving force, so the crucible side wall temperature is considered to be important. The manufacturing parameter optimization unit 13 selects the parameters at the positions indicated by the stars in Figs. 27 and 28 as the optimal solution.

[0082] The manufacturing parameter optimization unit 13 further executes a simulation using the device structure and the crucible moving speed derived from the above-mentioned set of optimal parameters. Table 4 shows the parameters related to the device structure obtained by optimization. [Table 4]

[0083] FIG. 29 shows a comparison of the time progression of the vertical coordinate of the crucible obtained as the optimal solution by the above machine learning with that before optimization.

[0084] FIG. 30 shows a comparison of the optimal temperature distribution transition (Target) and the simulation results (GA) using the equipment structure and crucible movement speed obtained as the optimal solution by applying a genetic algorithm, with regard to the dislocation density distribution.

[0085] Figure 31 shows a comparison of the optimal temperature distribution transition (Target) and the simulation results (GA) using the equipment structure and crucible movement speed obtained as the optimal solution by applying a genetic algorithm, regarding the crystal growth time distribution (solid-liquid interface shape transition).

[0086] As can be seen from FIG. 30, regarding dislocation density, the Target and GA are roughly consistent in terms of the absence of a high-dislocation region in the upper part of the ingot and the shape of the low-dislocation region in the center.

[0087] As can be seen from FIG. 31, regarding the crystal growth time distribution, there is a large difference between Target and GA in terms of the crystal growth start height and the crystal growth completion time.

[0088] FIG. 32 shows a comparison of the optimal temperature distribution (Optimal_temperature_transition) and the simulation results (GA) using the equipment structure and crucible movement speed obtained as the optimal solution by applying a genetic algorithm, with regard to the temperature transition of the crucible side wall.

[0089] From Fig. 32, it can be seen that under the conditions obtained by the genetic algorithm, the temperature drop at the bottom of the crucible at the beginning of crystal growth is greater than that of the optimal temperature distribution transition. In this respect, it can be seen that there is some room for improvement in the prediction accuracy of the optimal temperature distribution transition in this embodiment. In the following, the prediction accuracy of the DNN model of this embodiment will be considered, and in order to further improve the accuracy, the cause of the difference between the genetic algorithm and the optimal temperature distribution transition will be considered.

[0090] FIG. 33 shows a comparison between the simulation results and the prediction results of the DNN model 52 regarding the temperature distribution transition of the crucible side wall under the optimal conditions obtained by the genetic algorithm.

[0091] As can be seen from Fig. 33, the simulation results show that the temperature drop at the bottom of the crucible side wall is large in the early stage of crystal growth. On the other hand, the DNN model 52 predicts a temperature that is about 100 K higher than the simulation results. This shows that the DNN model 52 has some issues with the prediction accuracy in the early stage of crystal growth.

[0092] In order to verify the prediction accuracy of the DNN model 52, Fig. 34, Fig. 35, and Fig. 36 show scatter plots of the teacher data (Teacher_data) and the optimal solutions (Optimal_data) obtained by the genetic algorithm for the polynomial coefficients associated with the crucible movement speed selected as the optimal solution. Note that the high linearity of the distribution of the polynomial coefficients is believed to be due to the constraints on the crucible movement start position and crucible movement end position added when generating the conditions for the time transition of the vertical coordinate.

[0093] 34 to 36, it can be seen that the conditions obtained as the optimal solution by the genetic algorithm are conditions in the area not included in the teacher data. In other words, the temperature distribution transition predicted by the DNN model 52 based on the conditions obtained as the optimal solution by the genetic algorithm was obtained by extrapolating to the DNN model 52, which is thought to have resulted in low prediction accuracy.

[0094] Based on the above analysis, the discrepancy between the optimal temperature distribution and the results obtained based on the genetic algorithm is believed to be mainly due to the prediction accuracy of the DNN model itself and the influence of other device parameters.

[0095] The prediction accuracy of the DNN model is expected to improve by adding conditions not included in the current training data to the training data, which will improve the generalization performance of the DNN model. Also, the equipment parameters are expected to improve by optimizing the heater structure on the outside of the furnace tube, for example, to achieve a temperature distribution closer to the optimal temperature distribution transition.

[0096] As described above, according to this embodiment, it is possible to realize an apparatus that selects parameters that contribute to optimizing the product from a vast and complex manufacturing parameter space and optimizes the apparatus design without relying on tacit knowledge or trial and error.

[0097] [Third embodiment] 37 is a flowchart showing the procedure of a design optimization method according to the third embodiment. This method includes an optimal manufacturing environment determination step S1 and a manufacturing parameter selection step S2.

[0098] The optimal manufacturing environment determination step S1 determines an optimal manufacturing environment in the manufacturing equipment when a target variable representing an optimal index of a product manufactured by the manufacturing equipment is optimized. The manufacturing parameter selection step S2 selects one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing equipment.

[0099] According to this embodiment, it is possible to select parameters that contribute to optimizing the product from a vast and complex manufacturing parameter space, without relying on tacit knowledge or trial and error, and to execute a method of optimizing the equipment design in a step-by-step manner.

[0100] [Fourth embodiment] The fourth embodiment is a computer program. Fig. 37 shows the processing procedure of this program. This program causes a computer to execute a design optimization method including an optimal manufacturing environment determination step S1 for determining an optimal manufacturing environment in the manufacturing equipment when a target variable representing an optimal index of a product manufactured by the manufacturing equipment is optimized, and a manufacturing parameter selection step S2 for selecting one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing equipment.

[0101] According to this embodiment, a method for selecting parameters that contribute to optimizing the manufactured product from a vast and complex manufacturing parameter space and optimizing the equipment design can be implemented as a computer program without relying on tacit knowledge or trial and error.

[0102] [Fifth embodiment] The fifth embodiment is a recording medium for recording a computer program. Fig. 37 shows the processing procedure of this program. This program causes a computer to execute a design optimization method including an optimal manufacturing environment determination step S1 for determining an optimal manufacturing environment in the manufacturing equipment when a target variable representing an optimal index of a product manufactured by the manufacturing equipment is optimized, and a manufacturing parameter selection step S2 for selecting one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing equipment.

[0103] According to this embodiment, it is possible to record a computer program that selects parameters that contribute to optimizing the product from a vast and complex manufacturing parameter space and optimizes the device design without relying on tacit knowledge or trial and error.

[0104] In the above examples, a crystal growth furnace has been described. In this case, the objective variables were dislocation density and crystal growth time. The manufacturing environment to be optimized was the temperature distribution of the crucible wall and its change over time. The manufacturing parameters to be optimized were the component shape and size of the crystal furnace. However, the technology disclosed herein can be applied to manufacturing equipment and manufacturing processes related to various types of products other than crystal growth furnaces, such as those shown below.

[0105] [Sixth embodiment] 38 is a functional block diagram of a design optimization device 3 according to the sixth embodiment. The design optimization device 3 includes an optimal boundary condition determination unit 21, a parameter selection unit 22, and a parameter optimization unit .

[0106] The optimal boundary condition determination unit 21 determines optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a portion reflecting the objective variable and the structural variable or the process variable. The parameter selection unit 22 optimizes the boundary conditions so that the objective function is the best within a range set based on the simulation results. The parameter optimization unit 23 uses the optimal boundary conditions determined by the optimal boundary condition determination unit 21 as the objective function and the process variables and structural variables as explanatory variables, models the relationship between them by machine learning, and optimizes the process variables and structural variables so as to approach the optimal boundary conditions determined by the optimal boundary condition determination unit 21.

[0107] [Determination of optimal boundary conditions] The boundary conditions in this embodiment are a closed system located between the objective function such as the physical quantity to be optimized and the part to which the objective variable and the structural variable or process variable are reflected, and include boundary conditions such as the time change of the temperature distribution around the heating target such as the wall of the crucible, the elasticity around the user's seat, the temperature around the area surrounding the chip, battery, and peripheral elements, and the space surrounding the user. These boundary conditions, like the manufacturing environment described above, are elements that need to be optimized to optimize the objective variable. The optimal boundary condition determination unit 21 executes machine learning using the boundary conditions and the objective function such as the physical quantity to be optimized as training data. As a result, the optimal boundary conditions when the objective function is optimized are determined.

[0108] The configurations and operations of the parameter selection unit 22 and the parameter optimization unit 23 are similar to those of the manufacturing parameter selection unit 12 and the manufacturing parameter optimization unit 13 in the above-described embodiment.

[0109] According to this embodiment, parameters that contribute to optimization are selected from a vast and complex parameter space based on boundary conditions, and optimization can be realized from the inside, that is, from the side closest to the target.

[0110] [Seventh embodiment] 39 is a flowchart showing the procedure of the design optimization method according to the seventh embodiment. This method includes an optimal boundary condition determination step S11, a parameter selection step S12, and a parameter optimization step S13.

[0111] In the optimal boundary condition determination step S11, optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a part reflecting the objective variable and the structural variable or the process variable are determined. In the parameter selection step S12, the boundary conditions are optimized so that the objective function is the best within a range set based on the simulation results. In the parameter optimization step S13, the optimal boundary conditions determined in the optimal boundary condition determination step S11 are used as the objective function, and the process variables and structural variables are used as explanatory variables, and the relationship between the two is modeled by machine learning, and the process variables and structural variables are optimized so as to approach the optimal boundary conditions determined in the optimal boundary condition determination step S11.

[0112] [Determination of optimal boundary conditions] The boundary conditions in this embodiment are a closed system located between the objective function such as the physical quantity to be optimized and the part to which the objective variable and the structural variable or process variable are reflected, and include boundary conditions such as the time change of the temperature distribution around the heating target such as the wall of the crucible, the elasticity around the user's seat, the temperature around the area surrounding the chip, battery, and peripheral elements, and the space surrounding the user. These boundary conditions are also elements that need to be optimized to optimize the objective variable, just like the manufacturing environment described above. In the optimal boundary condition determination step S11, machine learning is performed using the boundary conditions and the objective function such as the physical quantity to be optimized as training data. As a result, the optimal boundary conditions when the objective function is optimized are determined.

[0113] The configurations and operations of the parameter selection step S12 and the parameter optimization step S13 are similar to the configurations and operations of the manufacturing parameter selection step S2 and the manufacturing parameter optimization step S3 in the above-described embodiment.

[0114] According to this embodiment, parameters that contribute to optimization are selected from a vast and complex parameter space based on boundary conditions, and optimization can be realized from the inside, that is, from the side closest to the target.

[0115] [Eighth embodiment] The eighth embodiment is a computer program. FIG. 39 shows the processing procedure of this program. This program causes a computer to execute an optimal boundary condition determination step S11 for determining optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a part to which an objective variable and a structural variable or a process variable are reflected, a parameter selection step S21 for optimizing the boundary conditions so that the objective function is the best within a range set based on a simulation result, and a parameter optimization step S23 for modeling the relationship between the optimal boundary condition determined in the optimal boundary condition determination step S11 as the objective function and the process variable and structural variable as explanatory variables by machine learning, and optimizing the process variable and structural variable so as to approach the optimal boundary condition determined in the optimal boundary condition determination step. The optimal boundary condition determination step S11 executes machine learning using the boundary conditions and the objective function such as a physical quantity to be optimized as teacher data.

[0116] According to this embodiment, a method can be implemented as a computer program in which parameters that contribute to optimization are selected from a vast and complex parameter space based on boundary conditions, and optimization is achieved from the inside, i.e., from the side closest to the target object.

[0117] [Ninth embodiment] The ninth embodiment is a recording medium for recording a computer program. FIG. 39 shows the processing procedure of this program. This program causes a computer to execute an optimal boundary condition determination step S11 for determining optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a part to which an objective variable and a structural variable or a process variable are reflected, a parameter selection step S21 for optimizing the boundary conditions so that the objective function is the best within a range set based on a simulation result, and a parameter optimization step S23 for modeling the relationship between the optimal boundary condition determined in the optimal boundary condition determination step S11 as the objective function and the process variable and structural variable as explanatory variables by machine learning, and optimizing the process variable and structural variable so as to approach the optimal boundary condition determined in the optimal boundary condition determination step. The optimal boundary condition determination step S11 executes machine learning using the boundary conditions and the objective function such as a physical quantity to be optimized as teacher data.

[0118] According to this embodiment, parameters that contribute to optimization can be selected from a vast and complex parameter space based on boundary conditions, and a computer program can be recorded that realizes optimization from the inside, i.e., from the side closest to the target object.

[0119] [Application Example 1] Optimization of tandem solar cell device structure The present disclosure is applicable to the optimization of a tandem solar cell device structure. In the first to fifth embodiments, the objective variable is the solar cell conversion efficiency. The manufacturing environment to be optimized is the optical characteristics of the boundary (reflectance, transmittance, wavelength dependence of scattering) and the electrical characteristics of the boundary (band alignment). The manufacturing parameters to be optimized are the material (substance, doping concentration, etc.), structure (film thickness, surface structure, etc.), and process conditions.

[0120] In the sixth to ninth embodiments, the objective variable is the solar cell conversion efficiency. The boundary conditions are the optical characteristics of the boundary (reflectance, transmittance, wavelength dependency of scattering) and the electrical characteristics of the boundary (band alignment). The process variables and structural variables are the material (substance, doping concentration, etc.), the structure (film thickness, surface structure, etc.), and the process conditions.

[0121] [Application Example 2] Optimizing heat generation in semiconductor chips The present disclosure is applicable to optimizing heat generation in semiconductor chips. In the first to fifth embodiments, the objective variable is the chip temperature. The manufacturing environment to be optimized is the temperature of the area surrounding the chip and peripheral elements. The manufacturing parameters to be optimized are the surrounding structure (material, arrangement, structure, etc. of the heat dissipation member).

[0122] In the sixth to ninth embodiments, the objective variable is the chip temperature. The boundary condition is the temperature of the area surrounding the chip and peripheral elements. The process and structural variables are the surrounding structure (material, arrangement, structure, etc. of the heat dissipation member).

[0123] [Application example 3] Optimizing heat generation in smartphones The present disclosure is applicable to heat generation countermeasures for smartphones. In the first to fifth embodiments, the objective variable is the chip / battery temperature. The manufacturing environment to be optimized is the temperature of the area surrounding the chip / battery and peripheral elements. The manufacturing parameters to be optimized are the surrounding structure (material, arrangement, structure, etc. of the heat dissipation member).

[0124] In the sixth to ninth embodiments, the objective variable is the chip / battery temperature. The boundary condition is the temperature of the area surrounding the chip / battery and peripheral elements. The process variable / structural variable is the surrounding structure (material, arrangement, structure, etc. of the heat dissipation member).

[0125] [Application Example 4] Infrared heating, flash lamp The present disclosure is applicable to infrared heating and flash lamps. In the first to fifth embodiments, the objective variable is the characteristics of the heating target. The manufacturing environment to be optimized is the time change in temperature distribution around the heating target. The manufacturing parameters to be optimized are the process conditions and equipment that realize the time change in temperature distribution around the heating target.

[0126] In the sixth to ninth embodiments, the objective variable is the characteristic of the heated object. The boundary condition is the time change of the temperature distribution around the periphery of the heated object. The process variable and structural variable are the process conditions and the device that realizes the time change of the temperature distribution around the periphery of the heated object.

[0127] [Application Example 5] Optimization of automobile ride comfort The present disclosure is applicable to optimizing the ride comfort of an automobile. In the first to fifth embodiments, the objective variable is the ride comfort. The manufacturing environment to be optimized is the elastic-plastic properties of the seat periphery. The manufacturing parameters to be optimized are the vehicle body structure, etc.

[0128] In the sixth to ninth embodiments, the objective variable is the ride comfort, the boundary condition is the elastic-plastic property of the seat periphery, and the process variable / structural variable is the vehicle body structure, etc.

[0129] [Application Example 6] Air conditioner air outlet optimization The present disclosure is applicable to air conditioner outlet optimization. In the first to fifth embodiments, the objective variable is human comfort. The manufacturing environment to be optimized is the boundary conditions of the space surrounding the person. The manufacturing parameters to be optimized are the arrangement and structure of the outlet.

[0130] In the sixth to ninth embodiments, the objective variable is the comfort of the person. The boundary conditions are the boundary conditions of the space surrounding the person. The process variables and structural variables are the arrangement and structure of the air outlets.

[0131] [Application Example 7] Resin molding / injection molding The present disclosure is applicable to resin molding and injection molding. In the first to fifth embodiments, the objective variables are shape followability and resin characteristics. The manufacturing environment to be optimized is the time change of stress distribution. The manufacturing parameter to be optimized is a device that realizes the optimal time change of stress distribution.

[0132] In the sixth to ninth embodiments, the objective variables are shape followability and resin characteristics. The boundary conditions are time changes in stress distribution. The process variables and structural variables are devices that realize optimal time changes in stress distribution.

[0133] [Application Example 8] Resin / foam molding The present disclosure is applicable to resin and foam molding. In the first to fifth embodiments, the objective variables are shape followability and resin properties. The manufacturing environment to be optimized is the time change in temperature distribution. The manufacturing parameters to be optimized are the process conditions that realize the optimal time change in temperature distribution.

[0134] In the sixth to ninth embodiments, the objective variables are shape followability and resin characteristics. The boundary conditions are time changes in temperature distribution. The process variables and structure variables are process conditions that realize optimal time changes in temperature distribution.

[0135] [Application Example 9] Cooking utensils The present disclosure is applicable to cooking utensils. In the first to fifth embodiments, the objective variable is delicious food. The production environment to be optimized is the temperature change over time of the surface or inside of the food. The production parameter to be optimized is a device that realizes the optimal temperature change over time of the surface or inside of the food.

[0136] In the sixth to ninth embodiments, the objective variable is delicious food. The boundary conditions are the temperature changes over time on the surface and inside of the food. The process variables and structure variables are devices that realize the optimal temperature changes over time on the surface and inside of the food.

[0137] [Application Example 10] Speaker placement in movie theaters The present disclosure is applicable to speaker arrangement in movie theaters. In the first to fifth embodiments, the objective variable is good acoustics. The manufacturing environment to be optimized is the wavelength dependency of the spatial boundary condition. The manufacturing parameter to be optimized is the speaker arrangement that realizes the optimal wavelength dependency of the spatial boundary condition.

[0138] In the sixth to ninth embodiments, the objective variable is good acoustics. The boundary condition is the wavelength dependency of the spatial boundary condition. The process variable and structural variable are the speaker arrangement that realizes the wavelength dependency of the optimal spatial boundary condition.

[0139] [Application Example 11] Lighting arrangement The present disclosure is applicable to lighting arrangements. In the first to fifth embodiments, the objective variable is good visibility. The manufacturing environment to be optimized is the wavelength dependency of the spatial boundary condition (e.g., window glass). The manufacturing parameter to be optimized is the device that realizes the wavelength dependency of the optimal spatial boundary condition.

[0140] In the sixth to ninth embodiments, the objective variable is good visibility. The boundary condition is the wavelength dependency of the spatial boundary condition (e.g., window glass). The process variable / structural variable is a device that realizes the wavelength dependency of the optimal spatial boundary condition.

[0141] [Each aspect of the present disclosure] The following summarizes each aspect of the present disclosure: A design optimization device according to an aspect of the present disclosure includes an optimal manufacturing environment determination unit that determines an optimal manufacturing environment in the manufacturing device when a target variable representing an optimal index of a product manufactured by the manufacturing device is optimized, and a manufacturing parameter selection unit that selects one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing device.

[0142] According to this embodiment, it is possible to realize an apparatus that selects parameters that contribute to optimizing the product from a vast and complex manufacturing parameter space and optimizes the apparatus design without relying on tacit knowledge or trial and error.

[0143] In one embodiment, the design optimization apparatus includes a manufacturing parameter optimization unit that optimizes the manufacturing parameters selected by the manufacturing parameter selection unit.

[0144] In one embodiment, the optimal manufacturing environment determination unit performs machine learning using the manufacturing environment within the manufacturing equipment and the objective variable of the product as training data.

[0145] In one embodiment, the optimum manufacturing environment determination unit executes a simulation to obtain a target variable of a product that is realized when a manufacturing environment within the device is given, in order to create training data.

[0146] In one embodiment, the manufacturing parameter selection unit performs a parameter analysis on a target variable of a product using a plurality of predetermined manufacturing parameters related to a manufacturing device as explanatory variables.

[0147] In one embodiment, the parametric analysis is a multiple regression analysis.

[0148] In one aspect, the manufacturing parameter optimization unit performs machine learning using the manufacturing parameters selected by the manufacturing parameter selection unit and the manufacturing environment within the manufacturing equipment as training data.

[0149] Another aspect of the present invention is a design optimization method, which includes an optimal manufacturing environment determination step of determining an optimal manufacturing environment in the manufacturing equipment when a target variable representing an optimal index of a product manufactured by the manufacturing equipment is optimized, and a manufacturing parameter selection step of selecting one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing equipment.

[0150] According to this aspect, it is possible to select parameters that contribute to optimizing the product from a vast and complex manufacturing parameter space, without relying on tacit knowledge or trial and error, and to execute a method of optimizing the equipment design in a step-by-step manner.

[0151] Yet another aspect of the present invention is a program that causes a computer to execute a design optimization method including an optimal manufacturing environment determination step of determining an optimal manufacturing environment in the manufacturing equipment when a target variable representing an optimal index of a product manufactured by the manufacturing equipment is optimized, and a manufacturing parameter selection step of selecting one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing equipment.

[0152] According to this aspect, a method for selecting parameters that contribute to optimizing the product from a vast and complex manufacturing parameter space and optimizing the device design without relying on tacit knowledge or trial and error can be implemented as a computer program.

[0153] Yet another aspect of the present invention is a recording medium having recorded thereon a program including an optimal manufacturing environment determination step of determining an optimal manufacturing environment in the manufacturing apparatus when a target variable representing an optimal index of a product manufactured by the manufacturing apparatus is optimized, and a manufacturing parameter selection step of selecting one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing apparatus.

[0154] According to this embodiment, it is possible to record a computer program that selects parameters that contribute to optimizing the product from a vast and complex manufacturing parameter space and optimizes the device design without relying on tacit knowledge or trial and error.

[0155] Yet another embodiment of the present invention is a design optimization apparatus. This design optimization apparatus includes an optimal boundary condition determination unit that determines optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a part to which an objective variable and a structural variable or a process variable are reflected, a parameter selection unit that optimizes the boundary conditions so that the objective function is the best within a range set based on a simulation result, and a parameter optimization unit that models the relationship between the optimal boundary condition determined by the optimal boundary condition determination unit as the objective function and the process variable and the structural variable as explanatory variables by machine learning, and optimizes the process variable and the structural variable so as to approach the optimal boundary condition determined by the optimal boundary condition determination unit. The optimal boundary condition determination unit executes machine learning using the boundary conditions and the objective function such as a physical quantity to be optimized as teacher data.

[0156] According to this embodiment, parameters that contribute to optimization are selected from a vast and complex parameter space based on the boundary conditions, and optimization can be realized from the inside, that is, from the side closest to the target.

[0157] Yet another aspect of the present invention is a design optimization method. This method includes an optimal boundary condition determination step of determining optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a part reflecting an objective variable and a structural variable or a process variable, a parameter selection step of optimizing the boundary conditions so that the objective function is the best within a range set based on a simulation result, and a parameter optimization step of modeling the relationship between the optimal boundary condition determined in the optimal boundary condition determination step as the objective function and the process variable and the structural variable as explanatory variables by machine learning, and optimizing the process variable and the structural variable so as to approach the optimal boundary condition determined in the optimal boundary condition determination step. The optimal boundary condition determination step performs machine learning using the boundary conditions and the objective function such as a physical quantity to be optimized as teacher data.

[0158] According to this embodiment, parameters that contribute to optimization are selected from a vast and complex parameter space based on the boundary conditions, and optimization can be realized from the inside, that is, from the side closest to the target.

[0159] Yet another aspect of the present invention is a program. This program causes a computer to execute an optimal boundary condition determination step of determining optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a portion to which an objective variable and a structural variable or a process variable are reflected, a parameter selection step of optimizing the boundary conditions so that the objective function is the best within a range set based on a simulation result, and a parameter optimization step of modeling the relationship between the optimal boundary condition determined in the optimal boundary condition determination step as the objective function and the process variable and the structural variable as explanatory variables by machine learning, and optimizing the process variable and the structural variable so as to approach the optimal boundary condition determined in the optimal boundary condition determination step. The optimal boundary condition determination step executes machine learning using the boundary conditions and the objective function such as a physical quantity to be optimized as teacher data.

[0160] According to this aspect, a method can be implemented as a computer program in which parameters that contribute to optimization are selected from a vast and complex parameter space based on boundary conditions, and optimization is achieved from the inside, i.e., from the side closest to the target object.

[0161] Yet another embodiment of the present invention is a recording medium, which performs machine learning using as training data the following steps: an optimal boundary condition determination step for determining optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a portion reflecting an objective variable and a structural variable or a process variable, a parameter selection step for optimizing the boundary conditions so that the objective function is the best within a range set based on a simulation result, and a parameter optimization step for modeling the relationship between the optimal boundary condition determined in the optimal boundary condition determination step as the objective function and the process variable and the structural variable as explanatory variables by machine learning, and optimizing the process variable and the structural variable so as to approach the optimal boundary condition determined in the optimal boundary condition determination step.

[0162] According to this aspect, parameters that contribute to optimization can be selected from a vast and complex parameter space based on boundary conditions, and a computer program can be recorded that realizes optimization from the inside, i.e., from the side closer to the target object.

[0163] The present disclosure has been described above based on the embodiments. These embodiments are merely illustrative, and it will be understood by those skilled in the art that various modifications are possible in the combination of each component and each treatment process, and that such modifications are also within the scope of the present invention.

[0164] For example, the design optimization device of the above embodiment executes machine learning in two steps: when determining the optimal manufacturing environment when optimizing the objective variable of the product, and when optimizing the selected manufacturing parameters. However, the present invention is not limited to this, and the design optimization device of the modified example may further divide the steps and execute machine learning in three or more steps.

[0165] The above describes the embodiments and modifications. When understanding the technical ideas that abstract the embodiments and modifications, the technical ideas should not be interpreted as being limited to the contents of the embodiments and modifications. The above-mentioned embodiments and modifications are merely illustrative examples, and many design changes such as changes, additions, and deletions of components are possible. In the embodiments, the contents in which such design changes are possible are emphasized by adding the notation "embodiment". However, design changes are also permitted even in contents without such notation. [Explanation of symbols]

[0166] 1. Design optimization device, 2. Design optimization device, 3. Design optimization device, 11. Optimal manufacturing environment determination unit, 12. Manufacturing parameter selection section, 13. Manufacturing parameter optimization section, 21. Optimal boundary condition determination part, 22 Parameter selection section, 23 Parameter optimization part, 51··DNN model, 52··DNN model, 100...crucible, 101 Heater, 102 Heater, 103 Heater, 104 Wall, 105...crucible stand, AX: central axis of the crucible, S1: determining the optimal manufacturing environment; S2: Selecting manufacturing parameters; S11: Determining optimal boundary conditions; S12: A step of selecting parameters; S13··Parameter optimization step.

Claims

1. an optimal manufacturing environment determination unit that determines an optimal manufacturing environment in the manufacturing device when a response variable representing an optimal index of a product manufactured by the manufacturing device is optimized; a manufacturing parameter selection unit that selects one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing device; A design optimization device comprising:

2. 2. The design optimization apparatus according to claim 1, further comprising a manufacturing parameter optimization unit that optimizes the manufacturing parameters selected by the manufacturing parameter selection unit.

3. 3. The design optimization device according to claim 1, wherein the optimal manufacturing environment determination unit performs machine learning using the manufacturing environment in the manufacturing device and a target variable of the product as training data.

4. 4. The design optimization device according to claim 3, wherein the optimal manufacturing environment determination unit executes a simulation to determine a target variable of the product that is realized when a manufacturing environment in the device is given, in order to create training data.

5. 3. The design optimization apparatus according to claim 1, wherein the manufacturing parameter selection unit performs a parameter analysis on a target variable of the product using a plurality of predetermined manufacturing parameters related to the manufacturing apparatus as explanatory variables.

6. 6. The design optimization apparatus according to claim 5, wherein the parameter analysis is a multiple regression analysis.

7. 3. The design optimization device according to claim 2, wherein the manufacturing parameter optimization unit performs machine learning using the manufacturing parameters selected by the manufacturing parameter selection unit and a manufacturing environment within the manufacturing device as training data.

8. an optimal manufacturing environment determination step for determining an optimal manufacturing environment in the manufacturing equipment when a response variable representing an optimal index of a product manufactured by the manufacturing equipment is optimized; a manufacturing parameter selection step of selecting one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing device; A design optimization method comprising:

9. an optimal manufacturing environment determination step for determining an optimal manufacturing environment in the manufacturing equipment when a response variable representing an optimal index of a product manufactured by the manufacturing equipment is optimized; and a manufacturing parameter selection step of selecting, from among a plurality of predetermined manufacturing parameters related to the manufacturing equipment, one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment.

10. an optimal manufacturing environment determination step for determining an optimal manufacturing environment in the manufacturing equipment when a response variable representing an optimal index of a product manufactured by the manufacturing equipment is optimized; A recording medium having a program recorded thereon, the program including a manufacturing parameter selection step of selecting one or more manufacturing parameters that contribute to realizing the optimal manufacturing environment from among a plurality of predetermined manufacturing parameters related to the manufacturing apparatus.

11. an optimal boundary condition determination unit that determines optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a portion reflecting an objective variable and a structural variable or a process variable; a parameter selection unit that optimizes the boundary conditions so as to maximize the objective function within a range that is set based on a simulation result; a parameter optimization unit that uses the optimal boundary conditions determined by the optimal boundary condition determination unit as an objective function and process variables and structural variables as explanatory variables to model the relationship between the two through machine learning, and optimizes the process variables and structural variables so as to approach the optimal boundary conditions determined by the optimal boundary condition determination unit; Equipped with The design optimization device is characterized in that the optimal boundary condition determination unit performs machine learning using the boundary conditions and an objective function such as a physical quantity to be optimized as training data.

12. an optimal boundary condition determination step of determining optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a portion reflecting an objective variable and a structural variable or a process variable; a parameter selection step of optimizing the boundary conditions so as to maximize the objective function within a range set based on a simulation result; a parameter optimization step of modeling the relationship between the optimal boundary conditions determined in the optimal boundary condition determination step as an objective function and the process variables and the structural variables as explanatory variables by machine learning, and optimizing the process variables and the structural variables so as to approach the optimal boundary conditions determined in the optimal boundary condition determination step; Including, The design optimization method is characterized in that the optimal boundary condition determination step performs machine learning using the boundary conditions and an objective function such as a physical quantity to be optimized as training data.

13. an optimal boundary condition determination step of determining optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a portion reflecting an objective variable and a structural variable or a process variable; a parameter selection step of optimizing the boundary conditions so as to maximize the objective function within a range set based on a simulation result; a parameter optimization step of modeling the relationship between the optimal boundary conditions determined in the optimal boundary condition determination step as an objective function and the process variables and the structural variables as explanatory variables by machine learning, and optimizing the process variables and the structural variables so as to approach the optimal boundary conditions determined in the optimal boundary condition determination step; Run the following on your computer: The program is characterized in that the optimal boundary condition determination step performs machine learning using the boundary conditions and an objective function such as a physical quantity to be optimized as training data.

14. an optimal boundary condition determination step of determining optimal boundary conditions of a closed system located between an objective function such as a physical quantity to be optimized and a portion reflecting an objective variable and a structural variable or a process variable; a parameter selection step of optimizing the boundary conditions so as to maximize the objective function within a range set based on a simulation result; a parameter optimization step of modeling the relationship between the optimal boundary conditions determined in the optimal boundary condition determination step as an objective function and the process variables and the structural variables as explanatory variables by machine learning, and optimizing the process variables and the structural variables so as to approach the optimal boundary conditions determined in the optimal boundary condition determination step; Run the following on your computer: A recording medium having a program recorded thereon, characterized in that the optimal boundary condition determination step performs machine learning using the boundary conditions and an objective function such as a physical quantity to be optimized as training data.

Images