Prediction device, learning device, and method and program therefor

Abstract

This prediction device comprises: a first prediction unit that predicts an intermediate parameter on the basis of a first machine learning model that uses, as an objective variable, an intermediate parameter that affects a quality characteristic of a product; and a second prediction unit that predicts the quality characteristic by inputting a predicted value of an intermediate parameter to a second machine learning model that includes an intermediate parameter in an explanatory variable and uses the quality characteristic as an objective variable.

Description

Prediction device, learning device, their methods, and programs

[0001] The present disclosure relates to a prediction device, a learning device, their methods, and programs.

[0002] Techniques for predicting the characteristics of a product based on machine learning are known. For example, in Patent Document 1, for a plurality of processes from generating a product from a material, a model for describing the target variable of each process is constructed using the explanatory variable and the target variable of each process, and the target variable of each process other than the final process for manufacturing the product is included in the explanatory variable of the next process to synthesize the model functions of the plurality of processes. A characteristic prediction device is disclosed.

[0003] International Publication No. 2024 / 014094

[0004] However, in the prior art, there is room for improving the prediction accuracy. For example, in the prior art, when the quality characteristics required for a product change, the model corresponding to the previous quality characteristics cannot be applied to the new quality characteristics, so it is necessary to newly construct a model corresponding to the new quality characteristics.

[0005] One aspect of the present disclosure aims to accurately predict the characteristics of a product.

[0006] The present disclosure has the following configuration.

[0007] <1> An intermediate prediction unit configured to predict the intermediate parameter based on a first machine learning model having the intermediate parameter that functions as a quality characteristic of the product as a target variable, and a second machine learning model that includes the intermediate parameter as an explanatory variable and has the quality characteristic as a target variable. A quality prediction unit configured to predict the quality characteristic by inputting the predicted value of the intermediate parameter. A prediction device comprising:

[0008] <2> The prediction device according to <1> above, wherein the intermediate parameter is selected from a plurality of parameters related to the product.

[0009] <3> The prediction device according to <2> above, wherein the plurality of parameters include information indicating the material used in the product, the physical properties of the material, the manufacturing conditions of the product, or the usage conditions of the product.

[0010] <4> The prediction device according to <2> or <3> above, wherein the intermediate parameters are selected based on the mechanism diagram relating to the quality characteristics.

[0011] <5> The prediction device according to <2> or <3> above, wherein the intermediate parameters are selected based on the results of a statistical analysis of a dataset including the plurality of parameters and the quality characteristics.

[0012] <6> The prediction device according to <2> or <3> above, wherein the intermediate parameters are selected based on the results of querying a trained language model for the intermediate parameters.

[0013] <7> The predictive apparatus according to any one of <1> to <6> above, wherein the manufactured product is a photosensitive film, the quality characteristics include a minimum development time, and the intermediate parameters include alkali penetration time and the saturated water absorption rate of the crosslinked monomer.

[0014] <8> A learning device comprising: a selection unit configured to select intermediate parameters that function in the quality characteristics of a manufactured product; an intermediate learning unit configured to generate a first machine learning model in which the intermediate parameters are the target variables; and a quality learning unit configured to generate a second machine learning model in which the intermediate parameters are the explanatory variables and the quality characteristics are the target variables.

[0015] <9> A prediction method comprising: a procedure in which a computer predicts intermediate parameters based on a first machine learning model in which intermediate parameters functioning on the quality characteristics of a manufactured product are the target variables; and a procedure in which the computer predicts the quality characteristics by inputting the predicted values ​​of the intermediate parameters into a second machine learning model in which the intermediate parameters are included as explanatory variables and the quality characteristics are the target variables.

[0016] <10> A learning method comprising: a procedure for a computer to select intermediate parameters that function in relation to the quality characteristics of a manufactured product; a procedure for generating a first machine learning model in which the intermediate parameters are the target variable; and a procedure for generating a second machine learning model in which the intermediate parameters are the explanatory variables and the quality characteristics are the target variable.

[0017] <11> A program for causing a computer to perform the following steps: predict intermediate parameters based on a first machine learning model in which intermediate parameters functioning on the quality characteristics of a manufactured product are the target variables; and predict the quality characteristics by inputting the predicted values ​​of the intermediate parameters into a second machine learning model in which the intermediate parameters are included as explanatory variables and the quality characteristics are the target variables.

[0018] <12> A program for causing a computer to perform the following steps: a procedure for selecting intermediate parameters that function in relation to the quality characteristics of a manufactured product; a procedure for generating a first machine learning model in which the intermediate parameters are the target variable; and a procedure for generating a second machine learning model in which the intermediate parameters are the explanatory variables and the quality characteristics are the target variable.

[0019] According to one aspect of this disclosure, the characteristics of a manufactured product can be predicted with high accuracy.

[0020] Figure 1 is a block diagram showing an example of the overall configuration of a characteristic prediction system. Figure 2 is a block diagram showing an example of a computer. Figure 3 is a block diagram showing an example of the functional configuration of a characteristic prediction system. Figure 4 is a diagram showing a first example of a mechanism deployment diagram. Figure 5 is a diagram showing a second example of a mechanism deployment diagram. Figure 6 is a diagram showing an example of a prediction model. Figure 7 is a flowchart showing an example of a learning process. Figure 8 is a flowchart showing an example of a prediction process. Figure 9 is a diagram showing an example of prediction accuracy.

[0021] Hereinafter, embodiments of this disclosure will be described with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant descriptions will be omitted.

[0022] [Embodiment] One embodiment of the present disclosure is an example of an information processing system for predicting the properties of a manufactured product. Hereinafter, the information processing system according to this embodiment will be referred to as the "property prediction system". In this embodiment, the manufactured product may be any object manufactured using one or more materials. The manufactured product may, for example, be a resin or a resin composition comprising a resin. The resin composition comprising a resin may, for example, be a photosensitive film. The photosensitive film may, for example, be a dry film photoresist.

[0023] Products offered as goods are manufactured to meet the quality requirements of the user. To verify whether the product meets the user's quality requirements, its quality characteristics are evaluated. During the product development phase, the product is designed to ensure its quality characteristics meet the target. Various parameters related to the product are considered during the design phase. These parameters include, for example, information about the material (e.g., material name, molecular structure, material composition ratio, etc.), material properties, manufacturing conditions, or usage conditions.

[0024] Traditionally, the quality characteristics of manufactured goods have been evaluated using prototypes. Furthermore, to reduce the number of prototypes produced, the quality characteristics of manufactured goods are sometimes predicted. In predicting quality characteristics, machine learning models are sometimes constructed based on datasets collected using manufactured goods, including prototypes.

[0025] The quality requirements users demand for manufactured goods can change. In this case, a model that corresponds to the previous quality characteristics cannot be applied to the new quality characteristics, and a new model needs to be built to address them. However, it is uncertain whether the dataset used to build the previous model can accurately predict the new quality characteristics. Furthermore, it may be necessary to collect a new dataset to build a model that predicts the new quality characteristics.

[0026] This embodiment aims to accurately predict the characteristics of a manufactured product. To this end, the characteristic prediction system according to this embodiment selects intermediate parameters that function in the quality characteristics of the manufactured product and generates a first machine learning model in which the intermediate parameters are the target variable, and a second machine learning model in which the intermediate parameters are included as explanatory variables and the quality characteristics are the target variable. The characteristic prediction system then predicts the intermediate parameters based on the first machine learning model and predicts the quality characteristics by inputting the predicted values ​​of the intermediate parameters into the second machine learning model.

[0027] In one respect, this embodiment uses a machine learning model that focuses on the functions that determine quality characteristics, thus enabling accurate prediction of the characteristics of the manufactured product. In another respect, this embodiment predicts quality characteristics through intermediate parameters that function in relation to the quality characteristics of the manufactured product, thus enabling easy adaptation to changes in the quality characteristics required of the manufactured product.

[0028] <Overall Configuration> The overall configuration of the characteristic prediction system according to this embodiment will be explained with reference to Figure 1. Figure 1 is a block diagram showing an example of the overall configuration of the characteristic prediction system.

[0029] As shown in Figure 1, the characteristic prediction system 1000 includes a learning device 10, a prediction device 20, and a terminal device 30. The learning device 10, the prediction device 20, and the terminal device 30 are connected via a communication network N to enable data communication. The communication network N may be, for example, a LAN (Local Area Network), a VPN (Virtual Private Network), or the Internet.

[0030] The learning device 10 is an example of an information processing device such as a personal computer, workstation, or server that learns a predictive model. The predictive model is an example of a machine learning model that predicts the quality characteristics of a manufactured product. The manufactured product may be, for example, a photosensitive film. The quality characteristic may be, for example, the minimum development time. The manufactured product and quality characteristic are not limited to these, and any characteristic of any manufactured product may be used as the target of prediction.

[0031] The prediction model includes a quality prediction model that predicts quality characteristics and an intermediate prediction model that predicts intermediate parameters. The prediction model may include multiple intermediate prediction models that predict different intermediate parameters. The intermediate prediction model is an example of the first machine learning model. The quality prediction model is an example of the second machine learning model.

[0032] Intermediate parameters are parameters that function in quality characteristics. If the quality characteristic is minimum development time, intermediate parameters may include, for example, the alkali penetration time of the photosensitive film and the saturated water absorption rate of the crosslinked monomer. Intermediate parameters are not limited to these and may include any parameters that function in quality characteristics.

[0033] An intermediate prediction model is a machine learning model that uses intermediate parameters as its dependent variable. The explanatory variables of an intermediate prediction model can be any variables that can predict the intermediate parameters. For example, the explanatory variables of an intermediate prediction model may include one or more parameters related to the manufactured product.

[0034] A quality prediction model is a machine learning model that includes intermediate parameters as explanatory variables and uses quality characteristics as the dependent variable. The explanatory variables for quality characteristics may include parameters other than intermediate parameters. Parameters other than intermediate parameters may include one or more parameters related to the manufactured product. Examples of parameters related to the manufactured product may include at least one of the following: information indicating the material, physical properties of the material, manufacturing conditions of the manufactured product, or usage conditions of the manufactured product.

[0035] The quality prediction model and the intermediate prediction model may be any type of machine learning model. Examples of machine learning model types include Gaussian process regression, linear regression, or random forest. The quality prediction model and the intermediate prediction model may be the same type of machine learning model, or they may be different types of machine learning models. If the prediction model includes multiple intermediate prediction models, the intermediate prediction models may be the same type of machine learning model, or they may be different types of machine learning models.

[0036] The prediction device 20 is an example of an information processing device such as a personal computer, workstation, or server that predicts the quality characteristics of a manufactured product. The prediction device 20 may predict the quality characteristics of a manufactured product based on a trained prediction model. The trained prediction model may be generated by the learning device 10.

[0037] The prediction device 20 may predict intermediate parameters based on a trained intermediate prediction model. The prediction device 20 may also predict the quality characteristics of the manufactured product by inputting the predicted values ​​of the intermediate parameters into a quality prediction model.

[0038] The terminal device 30 is an example of an information processing terminal, such as a personal computer, smartphone, or tablet, operated by a user of the characteristic prediction system 1000. The terminal device 30 may accept input of explanatory variables for the intermediate prediction model and transmit them to the prediction device 20. The terminal device 30 may also receive prediction results from the prediction device 20 and present them to the user.

[0039] The overall configuration of the characteristic prediction system 1000 shown in Figure 1 is just one example, and various system configurations are possible depending on the application and purpose. For example, one or more of the learning device 10, prediction device 20, and terminal device 30 may be included in multiple units of the characteristic prediction system 1000. For example, the characteristic prediction system 1000 may include a single information processing device that integrates the learning device 10 and the prediction device 20.

[0040] For example, the learning device 10 or the prediction device 20 may be implemented using multiple computers, or as a cloud computing service. For example, the learning device 10 and the prediction device 20 may be implemented using standalone computers. The classification of devices such as the learning device 10, prediction device 20, and terminal device 30 shown in Figure 1 is just one example.

[0041] <Hardware Configuration> The hardware configuration of the characteristic prediction system 1000 will be explained with reference to Figure 2. The learning device 10, prediction device 20, and terminal device 30 are implemented, for example, by a computer. Figure 2 is a block diagram showing an example of the computer's hardware configuration.

[0042] As shown in FIG. 2, the computer 500 includes a CPU (Central Processing Unit) 501, a ROM (Read Only Memory) 502, a RAM (Random Access Memory) 503, a HDD (Hard Disk Drive) 504, an input device 505, a display device 506, a communication I / F (Interface) 507, and an external I / F 508. The CPU 501, the ROM 502, and the RAM 503 form a so-called computer. Each hardware component of the computer 500 is interconnected via a bus line 509. Note that the input device 505 and the display device 506 may be connected to the external I / F 508 for use.

[0043] The CPU 501 is an arithmetic unit that realizes the control and functions of the entire computer 500 by reading programs and data from a storage device such as the ROM 502 or the HDD 504 onto the RAM 503 and executing processing. The computer 500 may have a GPU (Graphics Processing Unit) in addition to or instead of the CPU 501.

[0044] The ROM 502 is an example of a non-volatile semiconductor memory (storage device) that can hold programs and data even when the power is turned off. The ROM 502 functions as a main storage device that stores various programs, data, etc. necessary for the CPU 501 to execute various programs installed in the HDD 504. Specifically, the ROM 502 stores boot programs such as BIOS (Basic Input Output System) and EFI (Extensible Firmware Interface) that are executed when the computer 500 is started up, as well as data such as OS (Operating System) settings and network settings.

[0045] The RAM 503 is an example of a volatile semiconductor memory (storage device) in which programs and data are erased when the power is turned off. The RAM 503 is, for example, a DRAM (Dynamic Random Access Memory), a SRAM (Static Random Access Memory), or the like. The RAM 503 provides a work area in which various programs installed in the HDD 504 are developed when executed by the CPU 501.

[0046] The HDD 504 is an example of a non-volatile storage device that stores programs and data. Programs and data stored in the HDD 504 include an OS, which is basic software that controls the entire computer 500, and applications that provide various functions on the OS. Note that the computer 500 may use a storage device (e.g., an SSD: Solid State Drive or the like) that uses a flash memory as a storage medium instead of the HDD 504.

[0047] The input device 505 is a touch panel, operation keys or buttons, a keyboard or a mouse, a microphone that inputs sound data such as voice, etc., which are used by the user to input various signals.

[0048] The display device 506 is composed of a display such as a liquid crystal or an organic EL (Electro-Luminescence) that displays a screen, a speaker that outputs sound data such as voice, etc.

[0049] The communication I / F 507 is an interface for connecting to a communication network and enabling the computer 500 to perform data communication.

[0050] The external I / F 508 is an interface with an external device. Examples of the external device include a drive device 510 and the like.

[0051] The drive device 510 is a device for setting the recording medium 511. The recording medium 511 here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, flexible disks, and magneto-optical disks. The recording medium 511 may also include semiconductor memory that records information electrically, such as ROMs and flash memory. This allows the computer 500 to read and / or write to the recording medium 511 via the external interface 508.

[0052] The various programs to be installed on the HDD 504 are installed, for example, when the distributed recording medium 511 is set in a drive device 510 connected to an external I / F 508, and the various programs recorded on the recording medium 511 are read by the drive device 510. Alternatively, the various programs to be installed on the HDD 504 may be installed by downloading them via the communication I / F 507 from the communication network N or another network different from the communication network N.

[0053] <Functional Configuration> The functional configuration of the characteristic prediction system 1000 will be explained with reference to Figure 3. Figure 3 is a block diagram showing an example of the functional configuration of the characteristic prediction system.

[0054] <Learning Device> As shown in Figure 3, the learning device 10 includes a data storage unit 101, a selection unit 110, an intermediate learning unit 120, a quality learning unit 130, and a model output unit 140. The learning device 10 functions as the data storage unit 101, selection unit 110, intermediate learning unit 120, quality learning unit 130, and model output unit 140 when a pre-installed learning program is executed.

[0055] For example, the data storage unit 101 is implemented by the HDD 504 shown in Figure 2. For example, the selection unit 110, intermediate learning unit 120, quality learning unit 130, and model output unit 140 are implemented by a process in which a program loaded from the HDD 504 shown in Figure 2 onto the RAM 503 is executed by the CPU 501.

[0056] The data storage unit 101 stores a dataset. The dataset is a collection of training data used to build a predictive model. The training data may include multiple parameters related to the manufactured product and the ground truth values ​​of the product's quality characteristics. The ground truth values ​​of the quality characteristics may include at least one of the measured values ​​obtained from experiments using the product or prototype of the manufactured product, or the calculated values ​​obtained from simulations.

[0057] The selection unit 110 selects intermediate parameters. The selection unit 110 may select intermediate parameters from a plurality of parameters relating to the product. The selection unit 110 may select multiple intermediate parameters. The number of intermediate parameters selected by the selection unit 110 is not limited.

[0058] The selection unit 110 may select intermediate parameters based on a mechanism diagram. The selection unit 110 may select intermediate parameters based on a mechanism diagram relating to quality characteristics. A mechanism diagram is a document that describes the relationship between the characteristics of a product, the functions that affect those characteristics, and the parameters that affect those functions. The mechanism diagram may be written based on hypotheses. The selection unit 110 may accept input of intermediate parameters selected by the user of the characteristic prediction system 1000 based on the mechanism diagram.

[0059] The selection unit 110 may select intermediate parameters based on the results of statistical analysis of the dataset. The selection unit 110 may also perform statistical analysis on the dataset read from the data storage unit 101. The selection unit 110 may obtain the results of the statistical analysis of the dataset from another information processing device.

[0060] As an example, the selection unit 110 may select parameters with a high correlation coefficient with quality characteristics as intermediate parameters. For example, the selection unit 110 may select a predetermined number of parameters with high correlation coefficients as intermediate parameters. Alternatively, the selection unit 110 may select parameters with a correlation coefficient above a threshold as intermediate parameters.

[0061] As another example, the selection unit 110 may construct a machine learning model based on the dataset, using parameters related to the manufactured product as explanatory variables and quality characteristics as the objective variable, and then select parameters with high importance among the explanatory variables in that machine learning model as intermediate parameters. For example, the selection unit 110 may select a predetermined number of parameters with high importance as intermediate parameters. Alternatively, the selection unit 110 may select parameters with importance above a threshold as intermediate parameters.

[0062] The selection unit 110 may select intermediate parameters based on a trained language model. The selection unit 110 may also select intermediate parameters based on the results of querying the trained language model for intermediate parameters. For example, the selection unit 110 may query the trained language model for parameters that are effective in predicting quality characteristics.

[0063] The language model may be a large language model (LLM) trained to perform various language processing tasks, or a small language model (SLM) trained to perform a specific language processing task. The trained language model may be built into the learning device 10, or it may be held by another information processing device that can communicate via the communication network N.

[0064] The intermediate learning unit 120 generates an intermediate prediction model. The intermediate learning unit 120 may generate an intermediate prediction model based on a dataset read from the data storage unit 101. The intermediate learning unit 120 may generate an intermediate prediction model in which the intermediate parameters selected by the selection unit 110 are the target variables. The intermediate learning unit 120 may generate multiple intermediate prediction models for each of the multiple intermediate parameters selected by the selection unit 110.

[0065] The explanatory variables of the intermediate prediction model may include any variables that can predict the intermediate parameters. For example, the intermediate learning unit 120 may select the explanatory variables of the intermediate prediction model based on a mechanism diagram. The intermediate learning unit 120 may select the explanatory variables of the intermediate prediction model based on the results of statistical analysis of the dataset. The intermediate learning unit 120 may select the explanatory variables of the intermediate prediction model based on a trained language model.

[0066] The quality learning unit 130 generates a quality prediction model. The quality learning unit 130 may generate a quality prediction model based on a dataset read from the data storage unit 101. The explanatory variables of the quality prediction model may include intermediate parameters selected by the selection unit 110. The explanatory variables of the quality prediction model may include all of the multiple intermediate parameters selected by the selection unit 110.

[0067] The model output unit 140 outputs the trained prediction model. The model output unit 140 may output a prediction model that includes the intermediate prediction model generated by the intermediate learning unit 120 and the quality prediction model generated by the quality learning unit 130. The model output unit 140 may transmit the trained prediction model to the prediction device 20. The trained prediction model may be stored in the model storage unit 201 of the prediction device 20.

[0068] <Prediction Device> As shown in Figure 3, the prediction device 20 comprises a model storage unit 201, a request receiving unit 210, an intermediate prediction unit 220, a quality prediction unit 230, and a result output unit 240. The prediction device 20 functions as the model storage unit 201, request receiving unit 210, intermediate prediction unit 220, quality prediction unit 230, and result output unit 240 when a pre-installed prediction program is executed.

[0069] For example, the model storage unit 201 is implemented by the HDD 504 shown in Figure 2. For example, the request reception unit 210, the intermediate prediction unit 220, the quality prediction unit 230, and the result output unit 240 are implemented by a process in which a program loaded from the HDD 504 shown in Figure 2 onto the RAM 503 is executed by the CPU 501.

[0070] The model storage unit 201 stores a trained prediction model. The model storage unit 201 may already have a pre-trained prediction model stored in it. The prediction model stored in the model storage unit 201 may be generated by the learning device 10.

[0071] The request receiving unit 210 receives prediction requests. A prediction request is information or a signal requesting a prediction of quality characteristics. The request receiving unit 210 may receive prediction requests from the terminal device 30. The request receiving unit 210 may also receive input of prediction requests via the input device 505 of the prediction device 20.

[0072] The prediction request may include parameters related to the product to be predicted. The prediction request may also include parameters that serve as explanatory variables for the intermediate prediction model. The prediction request may also include parameters entered by the user of the characteristic prediction system 1000 into the terminal device 30.

[0073] The intermediate prediction unit 220 predicts intermediate parameters. The intermediate prediction unit 220 may predict intermediate parameters based on intermediate prediction models included in the prediction model. The intermediate prediction unit 220 may predict multiple intermediate parameters for each of the multiple intermediate prediction models included in the prediction model. The intermediate prediction unit 220 may predict intermediate parameters based on prediction requests received by the request reception unit 210. The intermediate prediction unit 220 may predict intermediate parameters by inputting the parameters included in the prediction request into the intermediate prediction model.

[0074] The quality prediction unit 230 predicts the quality characteristics of the manufactured product. The quality prediction unit 230 may predict the quality characteristics of the manufactured product based on a quality prediction model included in the prediction model. The quality prediction unit 230 may also predict the quality characteristics of the manufactured product by inputting predicted values ​​of intermediate parameters into the quality prediction model. The predicted values ​​of intermediate parameters may be predicted by the intermediate prediction unit 220.

[0075] The result output unit 240 outputs the prediction results. The prediction results may include predicted values ​​of quality characteristics. The prediction results may include the confidence level of the predicted values ​​of quality characteristics. The prediction results may include the information used for the prediction. The information used for the prediction may include predicted values ​​of intermediate parameters. The information used for the prediction may include the parameters used for predicting the intermediate parameters.

[0076] (Mechanism Exploded View) The mechanism exploded view used by the selection unit 110 will be explained with reference to Figures 4 and 5. Figures 4 and 5 are examples of mechanism exploded views.

[0077] Figure 4 shows an example of a mechanism diagram relating to the minimum development time of a photosensitive film. As shown in Figure 4, the mechanism diagram describes the relationship between the characteristics of the product, the functions that affect those characteristics, and the parameters that affect those functions. In Figure 4, the relationship between characteristics, functions, and parameters is shown in a tree format as an example, but the notation of the mechanism diagram is not limited to this.

[0078] Figure 4 shows that the minimum development time of a photosensitive film is influenced by functions such as alkali penetration into the resist, monomer water absorption, and resist micelle formation. Furthermore, Figure 4 shows that the function of alkali penetration into the resist is influenced by the polymer's LogP (octanol / water partition coefficient) and the water solubility of the cross-linked monomer; the function of monomer water absorption is influenced by the water solubility of the cross-linked monomer; and the function of resist micelle formation is influenced by the polymer's acid value.

[0079] Figure 5 shows an example of a diagram illustrating the mechanism of edge fusion in photosensitive film. Edge fusion in photosensitive film is a phenomenon in which, when a photosensitive film is slit and wound into a roll, the soft photosensitive layer seeps out from the end face of the roll over time, making it impossible to unwind.

[0080] Figure 5 shows that edge fusion in photosensitive films is influenced by functions such as the softness of the photosensitive layer, the fluidity of the photosensitive layer, and the moisture absorption of the photosensitive layer. Furthermore, Figure 5 shows that the softness of the photosensitive layer is influenced by the viscosity of the crosslinked monomer and the molecular weight of the polymer, the fluidity of the photosensitive layer is influenced by the winding tension in the manufacturing conditions and the temperature in the storage conditions, and the moisture absorption of the photosensitive layer is influenced by the humidity in the storage conditions and the water solubility of the crosslinked monomer.

[0081] (Predictive Model) A specific example of a predictive model will be explained with reference to Figure 6. Figure 6 is a diagram showing an example of a predictive model.

[0082] As shown in Figure 6, the prediction model may include two intermediate prediction models M1 (M1-1, M1-2) and one quality prediction model M2.

[0083] Intermediate prediction model M1-1 predicts alkali penetration time, which is an example of an intermediate parameter. Intermediate prediction model M1-1 uses the LogP of the polymer, the water solubility of the crosslinked monomer, the film thickness, and the development temperature as explanatory variables, and alkali penetration time as the dependent variable. Intermediate prediction model M1-1 may be constructed, for example, by Gaussian process regression.

[0084] Intermediate prediction model M1-2 predicts the saturation water absorption rate of a crosslinked monomer, which is an example of an intermediate parameter. Intermediate prediction model M1-2 uses the water solubility of the crosslinked monomer, the number of methylene groups in the crosslinked monomer, and the number of O-C-C-O partial bonds in the crosslinked monomer molecule as explanatory variables, and the saturation water absorption rate of the crosslinked monomer as the dependent variable. Intermediate prediction model M1-2 may be constructed, for example, by linear regression.

[0085] The quality prediction model M2 predicts the minimum development time, which is an example of a quality characteristic. The quality prediction model M2 uses alkali penetration time and the saturation water absorption rate of the crosslinked monomer as explanatory variables, and the minimum development time as the dependent variable. The quality prediction model M2 may be constructed, for example, using a random forest.

[0086] The predicted alkali penetration time output by the intermediate prediction model M1-1 is input to the quality prediction model M2. Similarly, the predicted saturation water absorption rate of the cross-linked monomer output by the intermediate prediction model M1-2 is input to the quality prediction model M2. Based on the input predicted alkali penetration time and the predicted saturation water absorption rate of the cross-linked monomer, the quality prediction model M2 predicts the minimum development time and outputs that predicted value.

[0087] <Processing Procedure> The characteristic prediction method performed by the characteristic prediction system 1000 will be explained with reference to Figures 7 and 8. The characteristic prediction method includes a learning process performed by the learning device 10 (see Figure 7) and a prediction process performed by the prediction device 20 (see Figure 8).

[0088] ≪Learning Process≫ Figure 7 is a flowchart of an example of the learning process. The learning process is the process of learning a predictive model. The learning process is performed by the learning device 10.

[0089] In step S1, the selection unit 110 of the learning device 10 acquires quality characteristics. The selection unit 110 may receive input of information indicating quality characteristics via the input device 505 of the learning device 10. The selection unit 110 may also receive information or signals indicating quality characteristics from the terminal device 30.

[0090] In step S2, the selection unit 110 of the learning device 10 selects intermediate parameters based on the quality characteristics acquired in step S1. The selection unit 110 may also select intermediate parameters based on a mechanism diagram relating to the quality characteristics. The selection unit 110 may also select intermediate parameters based on the results of statistical analysis of a dataset read from the data storage unit 101. The selection unit 110 may also select intermediate parameters based on the results of querying the trained language model for intermediate parameters. The selection unit 110 sends information indicating the intermediate parameters to the intermediate learning unit 120 and the quality learning unit 130.

[0091] In step S3, the intermediate learning unit 120 of the learning device 10 receives information indicating intermediate parameters from the selection unit 110. The intermediate learning unit 120 determines explanatory variables that can predict the intermediate parameters. The intermediate learning unit 120 extracts training data from the dataset stored in the data storage unit 101, including the explanatory variables of the intermediate prediction model (i.e., parameters that can predict the intermediate parameters) and the target variable (i.e., the intermediate parameters). The intermediate learning unit 120 generates an intermediate prediction model based on the extracted training data. The intermediate learning unit 120 sends the trained intermediate prediction model to the model output unit 140.

[0092] In step S4, the quality learning unit 130 of the learning device 10 receives information indicating intermediate parameters from the selection unit 110. The quality learning unit 130 extracts learning data from the dataset stored in the data storage unit 101, including explanatory variables (i.e., intermediate parameters) and target variables (i.e., quality characteristics) of the quality prediction model. Based on the extracted learning data, the quality learning unit 130 generates a quality prediction model. The quality learning unit 130 sends the trained quality prediction model to the model output unit 140.

[0093] In step S5, the model output unit 140 of the learning device 10 receives the trained intermediate prediction model from the intermediate learning unit 120. The model output unit 140 also receives the trained quality prediction model from the quality learning unit 130.

[0094] The model output unit 140 outputs the trained prediction model. The model output unit 140 transmits the trained prediction model to the prediction device 20. The prediction device 20 receives the trained prediction model from the training device 10. The prediction device 20 stores the trained prediction model in the model storage unit 201.

[0095] <Prediction Processing> Figure 8 is a flowchart of an example of prediction processing. Prediction processing is the process of predicting the quality characteristics of a manufactured product based on a trained prediction model. Prediction processing is performed by the prediction device 20.

[0096] In step S11, the user of the characteristic prediction system 1000 inputs the parameters to be predicted into the terminal device 30. The terminal device 30 receives the parameters input by the user via the input device 505. The terminal device 30 generates a prediction request that includes the input parameters. The terminal device 30 transmits the prediction request to the prediction device 20.

[0097] The prediction device 20 receives a prediction request from the terminal device 30. The request receiving unit 210 of the prediction device 20 receives the prediction request received by the prediction device 20. The request receiving unit 210 obtains the parameters to be predicted included in the prediction request. The request receiving unit 210 sends the parameters to be predicted to the intermediate prediction unit 220.

[0098] In step S12, the intermediate prediction unit 220 of the prediction device 20 receives the parameters to be predicted from the request reception unit 210. The intermediate prediction unit 220 reads the intermediate prediction model from the model storage unit 201. The intermediate prediction unit 220 inputs the explanatory variables of the intermediate prediction model included in the parameters to be predicted into the intermediate prediction model. The intermediate prediction model predicts the intermediate parameters based on the input explanatory variables and outputs the predicted values. The intermediate prediction unit 220 obtains the predicted values ​​of the intermediate parameters output from the intermediate prediction model. The intermediate prediction unit 220 sends the predicted values ​​of the intermediate parameters to the quality prediction unit 230.

[0099] Furthermore, if multiple intermediate parameters are selected during the learning process, the prediction device 20 repeatedly executes the process in step S12 for each of the multiple intermediate parameters. As a result, the intermediate prediction unit 220 predicts each of the multiple intermediate parameters and sends these predicted values ​​to the quality prediction unit 230.

[0100] In step S13, the quality prediction unit 230 of the prediction device 20 receives predicted values ​​of intermediate parameters from the intermediate prediction unit 220. If multiple intermediate parameters are selected, the quality prediction unit 230 receives predicted values ​​for each of the multiple intermediate parameters.

[0101] The quality prediction unit 230 reads the quality prediction model from the model storage unit 201. The quality prediction unit 230 inputs the predicted values ​​of intermediate parameters into the quality prediction model. Based on the input predicted values ​​of intermediate parameters, the quality prediction model predicts the quality characteristics of the manufactured product and outputs the predicted values. The quality prediction unit 230 obtains the predicted values ​​of quality characteristics output from the quality prediction model. The quality prediction unit 230 sends the predicted values ​​of quality characteristics to the result output unit 240.

[0102] In step S14, the result output unit 240 of the prediction device 20 receives predicted values ​​of quality characteristics from the quality prediction unit 230. The result output unit 240 outputs the predicted quality characteristics result. The predicted quality characteristics result includes the predicted values ​​of quality characteristics. The result output unit 240 transmits the predicted quality characteristics result to the terminal device 30.

[0103] The terminal device 30 receives the quality characteristic prediction results from the prediction device 20. The terminal device 30 presents the received quality characteristic prediction results to the user. The terminal device 30 may display the predicted values ​​included in the prediction results on the display device 506. The terminal device 30 may also display the information used for the prediction along with the predicted values.

[0104] A user of the characteristic prediction system 1000 may refer to the quality characteristic prediction results displayed on the display device 506 of the terminal device 30 and use them in the design and development of the manufactured product. For example, the user may search for parameters that satisfy the target quality characteristics by changing the parameters related to the manufactured product. For example, the user may manufacture the manufactured product using parameters that satisfy the target quality characteristics.

[0105] <Prediction Accuracy> The results of evaluating the prediction accuracy of the prediction model according to this embodiment will be explained with reference to Figure 9. Figure 9 is a diagram showing an example of prediction accuracy.

[0106] The example shown in Figure 9 is an example of predicting the quality characteristics of dry film photoresist based on the prediction model shown in Figure 6. Specifically, the prediction model for the example is a machine learning model that includes: - A Gaussian process regression model M1-1 with polymer LogP, water solubility of the crosslinked monomer, film thickness when alkali penetration time is measured, and development temperature as explanatory variables, and alkali penetration time as the objective variable; - A linear regression model M1-2 with water solubility of the crosslinked monomer, the number of methylene groups in the crosslinked monomer, and the number of partial bonds O-C-C-O in the crosslinked monomer molecule as explanatory variables, and the saturated water absorption rate of the crosslinked monomer as the objective variable; and - A random forest M2 with alkali penetration time and saturated water absorption rate of the crosslinked monomer as explanatory variables, and the minimum development time as the objective variable.

[0107] The comparative example shown in Figure 9 is an example in which quality characteristics were predicted based on a random forest that takes the explanatory variables of the intermediate prediction model shown in Figure 6 as input and outputs predicted values ​​for quality characteristics. Specifically, the prediction model M for the comparative example is a random forest with polymer LogP, water solubility of the crosslinked monomer, film thickness when alkali penetration time is measured, development temperature, number of methylene groups in the crosslinked monomer, and number of O-C-C-O in the crosslinked monomer as explanatory variables, and the minimum development time as the objective variable.

[0108] Here, the quality prediction model M2 in the example and the prediction model M in the comparative example are models trained on measurement data of the development product and predict measurement data of the mass-produced product. There are changes in materials between the development product and the mass-produced product, and there are materials not commonly used in the polymers and crosslinking monomers. Therefore, it can be said that the prediction accuracy of extrapolation is being evaluated. In other words, it can be seen that this embodiment allows for easy adaptation to material changes.

[0109] Figure 9 shows graphs plotting the combinations of measured values ​​(true) and predicted values ​​(predict) for both the comparative example and the example. The evaluation index for prediction accuracy is the coefficient of determination (R). ２ The coefficient of determination (RMS) and the root mean square error (RMSE) were used. A higher RMS indicates higher prediction accuracy. A lower RMSE indicates higher prediction accuracy.

[0110] As shown in Figure 9, in the comparative example, the coefficient of determination was 0.29 and the root mean square error was 9.38. In contrast, in the example, the coefficient of determination was 0.46 and the root mean square error was 6.26. Therefore, it was shown that the prediction model in the example can predict quality characteristics (i.e., minimum development time) with higher accuracy than the comparative example.

[0111] <Effects of the Embodiment> The prediction device 20 according to this embodiment predicts intermediate parameters based on an intermediate prediction model in which intermediate parameters that function in the quality characteristics of a manufactured product are the objective variable, and predicts the quality characteristics by inputting the predicted values ​​of the intermediate parameters into a quality prediction model in which the intermediate parameters are included as explanatory variables and the quality characteristics are the objective variable.

[0112] In one respect, this embodiment uses a machine learning model that focuses on the functions that determine quality characteristics, thus enabling accurate prediction of the characteristics of the manufactured product. In another respect, this embodiment predicts quality characteristics through intermediate parameters that function in relation to the quality characteristics of the manufactured product, thus enabling easy adaptation to changes in the quality characteristics required of the manufactured product.

[0113] The prediction device 20 may select intermediate parameters from a plurality of parameters relating to the product. The plurality of parameters may include information indicating the materials used in the product, the physical properties of the materials, the manufacturing conditions of the product, or the usage conditions of the product. In one aspect, according to this embodiment, appropriate parameters that function for quality characteristics can be selected.

[0114] The prediction device 20 may select intermediate parameters based on a mechanism diagram relating to quality characteristics. In one respect, according to this embodiment, appropriate intermediate parameters can be selected based on the function that determines the quality characteristics.

[0115] The prediction device 20 may select intermediate parameters from multiple parameters by statistically analyzing a dataset that includes multiple parameters and quality characteristics related to the manufactured product. In one aspect, according to this embodiment, appropriate intermediate parameters can be selected based on the relationship between the parameters and quality characteristics related to the manufactured product.

[0116] The prediction device 20 may select intermediate parameters by querying the trained language model for intermediate parameters. In one aspect, according to this embodiment, appropriate intermediate parameters can be selected based on a language model that has learned various literatures.

[0117] The manufactured product may be a photosensitive film. Quality characteristics may include the minimum development time. Intermediate parameters may include alkali penetration time and the saturated water absorption rate of the crosslinked monomer. In one aspect, according to this embodiment, the minimum development time of the photosensitive film can be predicted with high accuracy.

[0118] The learning device 10 selects intermediate parameters that function in relation to the quality characteristics of the manufactured product, generates an intermediate prediction model with one or more parameters related to the manufactured product as explanatory variables and the intermediate parameters as the objective variable, and generates a quality prediction model that includes the intermediate parameters as explanatory variables and the quality characteristics as the objective variable. In one aspect, according to this embodiment, it is possible to generate a prediction model that can respond to changes in the quality characteristics required of the manufactured product.

[0119] [Other Embodiments] The above describes forward problem prediction, in which intermediate parameters that function in the quality characteristics of a manufactured product are used as the objective variable in an intermediate prediction model to predict the intermediate parameters, and the predicted values ​​of the intermediate parameters are input into a quality prediction model that includes the intermediate parameters as explanatory variables and has quality characteristics as the objective variable to predict the quality characteristics. On the other hand, inverse problem prediction may be performed using the intermediate prediction model and quality prediction model of the present invention.

[0120] Specifically, intermediate parameter search points are generated randomly or with a predetermined step size, and intermediate parameters that satisfy the target quality characteristics are identified from the quality characteristics corresponding to each search point. Next, for each of the one or more parameters related to the product, a search point is generated randomly or with a predetermined step size, and by selecting the previously identified intermediate parameters from the intermediate parameters corresponding to each search point, one or more parameter groups related to the product are identified. Then, a material that matches the one or more parameter groups related to the product is selected.

[0121] [Supplement] Each function of the embodiments described above can be realized by one or more processing circuits. Hereinafter, "processing circuit" in this specification includes processors programmed to execute each function by software, such as CPUs (Central Processing Units) or GPUs (Graphics Processing Units) implemented by electronic circuits, as well as devices such as ASICs (Application Specific Integrated Circuits), DSPs (Digital Signal Processors), FPGAs (Field Programmable Gate Arrays), and conventional circuit modules designed to execute each function described above.

[0122] While embodiments of the present disclosure have been described in detail above, the embodiments disclosed herein are illustrative and not restrictive in all respects. The embodiments can be modified and improved in various ways without departing from the scope and spirit of the appended claims. The features described in the above embodiments can be combined in any way that is not inconsistent with other configurations.

[0123] This application claims priority to Japanese Patent Application No. 2024-193109, filed with the Japan Patent Office on November 1, 2024, which is incorporated herein by reference to its entire contents.

[0124] 10: Learning device 20: Prediction device 30: Terminal device 101: Data storage unit 110: Selection unit 120: Intermediate learning unit 130: Quality learning unit 140: Model output unit 201: Model storage unit 210: Request reception unit 220: Intermediate prediction unit 230: Quality prediction unit 240: Result output unit 1000: Characteristic prediction system

Claims

1. A prediction device comprising: an intermediate prediction unit configured to predict intermediate parameters based on a first machine learning model whose objective variable is an intermediate parameter that functions in the quality characteristics of a manufactured product; and a quality prediction unit configured to predict the quality characteristics by inputting predicted values ​​of the intermediate parameters into a second machine learning model which includes the intermediate parameters as explanatory variables and whose objective variable is the quality characteristics.

2. The prediction device according to claim 1, wherein the intermediate parameter is selected from a plurality of parameters relating to the product.

3. The prediction device according to claim 2, wherein the plurality of parameters include information indicating the material used in the product, the physical properties of the material, the manufacturing conditions of the product, or the usage conditions of the product.

4. The prediction device according to claim 2 or 3, wherein the intermediate parameters are selected based on a mechanism diagram relating to the quality characteristics.

5. The prediction device according to claim 2 or 3, wherein the intermediate parameters are selected based on the results of a statistical analysis of a dataset including the plurality of parameters and the quality characteristics.

6. The prediction device according to claim 2 or 3, wherein the intermediate parameters are selected based on the results of querying a trained language model for the intermediate parameters.

7. The predictive apparatus according to any one of claims 1 to 6, wherein the product is a photosensitive film, the quality characteristics include a minimum development time, and the intermediate parameters include alkali penetration time and the saturated water absorption rate of the crosslinked monomer.

8. A learning device comprising: a selection unit configured to select intermediate parameters that function in relation to the quality characteristics of a manufactured product; an intermediate learning unit configured to generate a first machine learning model in which the intermediate parameters are the target variables; and a quality learning unit configured to generate a second machine learning model in which the intermediate parameters are the explanatory variables and the quality characteristics are the target variables.

9. A prediction method comprising: a procedure in which a computer predicts intermediate parameters based on a first machine learning model in which intermediate parameters functioning on the quality characteristics of a manufactured product are the target variable; and a procedure in which the computer predicts the quality characteristics by inputting the predicted values ​​of the intermediate parameters into a second machine learning model in which the intermediate parameters are included as explanatory variables and the quality characteristics are the target variable.

10. A learning method comprising: a procedure for a computer to select intermediate parameters that function for the quality characteristics of a manufactured product; a procedure for generating a first machine learning model with the intermediate parameters as the target variable; and a procedure for generating a second machine learning model that includes the intermediate parameters as explanatory variables and has the quality characteristics as the target variable.

11. A program for causing a computer to perform the following steps: predict intermediate parameters based on a first machine learning model in which intermediate parameters functioning on the quality characteristics of a manufactured product are the target variable; and predict the quality characteristics by inputting the predicted values ​​of the intermediate parameters into a second machine learning model in which the intermediate parameters are included as explanatory variables and the quality characteristics are the target variable.

12. A program for causing a computer to perform the following steps: a procedure for selecting intermediate parameters that function in relation to the quality characteristics of a manufactured product; a procedure for generating a first machine learning model in which the intermediate parameters are the target variable; and a procedure for generating a second machine learning model in which the intermediate parameters are included as explanatory variables and the quality characteristics are the target variable.

Images