Proposal device, proposal method, and program

Abstract

To propose a molecular structure of an organic polymer material physical property values that simultaneously satisfy multiple target values.SOLUTION: A proposal device comprising a control unit for proposing a molecular structure of an organic polymer material is provided, the proposal device being configured to predict multiple physical property values based on a molecular structure of an organic polymer material, and output a molecular structure of the organic polymer material with the multiple physical property values that satisfy target values.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present disclosure relates to a proposal device, a proposal method, and a program. [Background technology]

[0002] There is a technique for proposing a substance with improved specific properties using a machine learning model. For example, Patent Document 1 discloses a method for selecting a surrogate property correlated with thermal conductivity from a data set, constructing a trained model that outputs the surrogate property for the molecular structure using a plurality of molecular structures in the data set and the surrogate property for each of the plurality of molecular structures, transferring the trained model to construct a retrained model that outputs the thermal conductivity for the molecular structure, and predicting the thermal conductivity for an arbitrary molecular structure using the retrained model. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2020-95310 A Summary of the Invention [Problem to be solved by the invention]

[0004] To generate a machine learning model, it is necessary to collect a sufficient amount of training data, but it is difficult to actually create a large number of organic polymer materials and measure their physical properties through experiments, etc. Therefore, it is not easy to propose a molecular structure for an organic polymer material that simultaneously satisfies target values ​​for multiple physical properties.

[0005] The present disclosure provides a technique for proposing a molecular structure of an organic polymer material that simultaneously satisfies target values ​​for multiple physical properties. [Means for solving the problem]

[0006] A proposal device according to a first aspect of the present disclosure is a proposal device having a control unit that proposes a molecular structure of an organic polymer material, the control unit predicting a plurality of physical property values ​​based on the molecular structure, and outputting the molecular structure in which all of the plurality of physical property values ​​satisfy target values.

[0007] According to the first aspect of the present disclosure, it is possible to propose a molecular structure of an organic polymer material in which a plurality of physical property values ​​simultaneously satisfy target values.

[0008] A second aspect of the present disclosure is a proposed device according to the first aspect, wherein the control unit predicts at least one of the plurality of physical property values ​​based on a trained machine learning model.

[0009] A third aspect of the present disclosure is a proposed device relating to the second aspect, wherein the control unit generates learning data including the plurality of physical property values ​​calculated by molecular simulation, and creates or updates the machine learning model by learning the learning data.

[0010] A fourth aspect of the present disclosure is a proposed device related to the third aspect, wherein the control unit calculates a descriptor based on a molecular structure in which monomer molecular structures are repeatedly connected and the number of atoms is within a predetermined range, and generates the learning data including the descriptor.

[0011] A fifth aspect of the present disclosure is a proposed device relating to the third or fourth aspect, wherein the control unit repeatedly executes outputting the molecular structure, generating the learning data, and updating the machine learning model.

[0012] A sixth aspect of the present disclosure is a proposed device according to any one of the first to fifth aspects, wherein the plurality of physical property values ​​include a physical property value related to a dielectric characteristic and a physical property value related to thermal expansion.

[0013] A seventh aspect of the present disclosure is the proposed device according to the sixth aspect, wherein the physical property value relating to the dielectric characteristic is a dielectric constant or a dielectric loss tangent.

[0014] An eighth aspect of the present disclosure is a proposed device according to the sixth or seventh aspect, wherein the physical property value relating to the dielectric characteristics is calculated using a dielectric relaxation function based on time series data of the dipole moment of the organic polymer material.

[0015] A ninth aspect of the present disclosure is a proposed device according to the eighth aspect, wherein the physical property value relating to the dielectric characteristics is calculated using a dielectric relaxation function based on time series data obtained by extracting a predetermined frequency band from time series data of the dipole moment of the organic polymer material.

[0016] A tenth aspect of the present disclosure is the proposed device according to the sixth aspect, wherein the physical property value relating to thermal expansion is a linear expansion coefficient.

[0017] An eleventh aspect of the present disclosure is a proposed device according to any one of the first to tenth aspects, wherein the molecular structure of the organic polymer material comprises polyimide, polyolefin, polyvinyl, polyacrylic, polyester, polyurethane, polyurea, polycarbonate, polysulfone, polyamide, polyhaloolefin, polystyrene, polyketone or polyimine.

[0018] In a proposal method according to a twelfth aspect of the present disclosure, a control unit of a proposal device that proposes a molecular structure of an organic polymer material predicts a plurality of physical property values ​​based on the molecular structure, and outputs the molecular structure in which all of the plurality of physical property values ​​satisfy target values.

[0019] A program according to a thirteenth aspect of the present disclosure causes a control unit of a proposal device that proposes a molecular structure of an organic polymer material to execute a process of predicting a plurality of physical property values ​​based on the molecular structure, and outputting the molecular structure in which all of the plurality of physical property values ​​satisfy target values. [Brief description of the drawings]

[0020] [Figure 1] FIG. 2 is a block diagram showing an example of a hardware configuration of the proposed device. [Diagram 2]1 is a flowchart illustrating an example of a proposed method. [Diagram 3] 13 is a flowchart illustrating an example of a learning data generation process. [Figure 4] 1 is a graph showing an example of the calculation accuracy of the dielectric constant. [Diagram 5] 1 is a graph showing an example of calculation accuracy of a linear expansion coefficient. [Figure 6] 1 is a graph showing an example of the prediction accuracy of the dielectric constant. [Figure 7] 1 is a graph showing an example of prediction accuracy of a linear expansion coefficient. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] Hereinafter, each embodiment will be described with reference to the accompanying drawings. In this specification and the drawings, components having substantially the same functional configurations are denoted by the same reference numerals, and redundant description will be omitted.

[0022] [Embodiment] An embodiment of the present disclosure is a proposal apparatus for proposing a molecular structure of an organic polymer material, for example, an organic polymer used in the manufacture of low-frequency substrates.

[0023] Materials for low-frequency circuit boards are required to have low dielectric properties and low thermal expansion. It is generally known that there is a trade-off between dielectric properties and thermal expansion. Although it is possible to actually measure the physical properties related to dielectric properties and thermal expansion, it is very costly to actually create organic polymer materials and measure their physical properties. Therefore, it is not easy to find materials that have low dielectric properties and low thermal expansion based on experimental results.

[0024] In addition, the number of known materials that meet both low dielectric properties and low thermal expansion is limited, so even if a machine learning model is generated based on the results of experiments using known materials, there is a limit to how much the prediction accuracy can be improved due to insufficient training data.

[0025] In this embodiment, the object is to propose a molecular structure of an organic polymer material in which multiple physical property values ​​simultaneously satisfy target values. To this end, in this embodiment, a physical property value related to dielectric properties and a physical property value related to thermal expansion are predicted based on a trained machine learning model, and a molecular structure in which both of these predicted values ​​satisfy the target values ​​is output. This machine learning model is trained using training data including physical property values ​​calculated by molecular simulation. It has been confirmed that the calculated values ​​by molecular simulation have a high positive correlation with experimental values, so the machine learning model generated based on these calculated values ​​has high prediction accuracy.

[0026] According to one aspect, the present embodiment makes it possible to efficiently propose a molecular structure of an organic polymer material in which a plurality of physical property values ​​simultaneously satisfy target values, thereby shortening the lead time in the development of an organic polymer material and a low-frequency substrate manufactured using the same.

[0027] <Hardware configuration> Fig. 1 is a block diagram showing an example of a hardware configuration of a proposed device 100 in this embodiment. As shown in Fig. 1, the proposed device 100 includes a processor 101, a memory 102, an auxiliary storage device 103, an operation device 104, a display device 105, a communication device 106, and a drive device 107. Each piece of hardware of the proposed device 100 is connected to each other via a bus 108.

[0028] The processor 101 includes various arithmetic devices such as a CPU (Central Processing Unit), etc. The processor 101 reads out various programs installed in the auxiliary storage device 103 onto the memory 102 and executes them.

[0029] The memory 102 has a main storage device such as a ROM (Read Only Memory), a RAM (Random Access Memory), etc. The processor 101 and the memory 102 form a so-called computer (hereinafter also referred to as a "control unit"), and the processor 101 executes various programs read onto the memory 102, whereby the computer realizes various functions.

[0030] The auxiliary storage device 103 (hereinafter also referred to as a “storage unit”) stores various programs and various data used when the processor 101 executes the various programs.

[0031] The operation device 104 is an operation device that allows a user of the proposal device 100 to perform various operations. The display device 105 is a display device that displays the results of various processes executed by the proposal device 100.

[0032] The communication unit 106 is a communication device for communicating with an external device via a communication network.

[0033] The drive device 107 is a device for setting the storage medium 109. The storage medium 109 here includes media that store information optically, electrically, or magnetically, such as a CD-ROM, a flexible disk, a magneto-optical disk, etc. The storage medium 109 may also include semiconductor memory that stores information electrically, such as a ROM, a flash memory, etc.

[0034] The various programs to be installed in the auxiliary storage device 103 are installed, for example, by setting the distributed storage medium 109 in the drive device 107 and reading out the various programs stored in the storage medium 109 by the drive device 107. Alternatively, the various programs to be installed in the auxiliary storage device 103 may be installed by being downloaded from a network via the communication device 106.

[0035] <Proposal method flow> 2 is a flowchart showing an example of the flow of the proposing method executed by the proposing device 100 in this embodiment. The proposing method in this embodiment is a method for proposing a molecular structure of an organic polymer material.

[0036] In step S1, the control unit of the proposing device 100 stores options for the molecular structure of the proposed organic polymer material (hereinafter also referred to as "candidate structures") in the storage unit. In this embodiment, the candidate structures are extracted from a publicly available virtual polymer library or the like. The candidate structures may be generated by the proposing device 100 based on a technique such as machine learning. The number of candidate structures is not limited, but in this embodiment, it is set to about 1 million structures.

[0037] In this embodiment, the organic polymeric material is intended to be a homopolymer, but may also be a copolymer. In this embodiment, the candidate structures may include polyimide, polyolefin, polyvinyl, polyacrylic, polyester, polyurethane, polyurea, polycarbonate, polysulfone, polyamide, polyhaloolefin, polystyrene, polyketone, or polyimine. The candidate structures are not limited to these structures and may include a variety of structures.

[0038] In step S2, the control unit of the proposed device 100 calculates a descriptor based on the candidate structure, and stores the descriptor in the storage unit in association with the candidate structure. For example, the descriptor disclosed in Reference 1 below can be used. The descriptor may be calculated based on a molecular structure in which monomer molecular structures are repeatedly connected and in which the number of atoms falls within a predetermined range.

[0039] [Reference 1] Huan Doan Tran, Chiho Kim, Lihua Chen, Anand Chandrasekaran, Rohit Batra, Shruti Venkatram, Deepak Kamal, Jordan P. Lightstone, Rishi Gurnani, Pranav Shetty, Manav Ramprasad, Julia Laws, Madeline Shelton, Rampi Ramprasad, "Machine-learning predictions of polymer properties with Polymer Genome", Journal of Applied Physics, vol. 128, 171104 (2020)

[0040] Specifically, the control unit generates a SMILES of an oligomer (or polymer) having a number of atoms in a predetermined range by repeating the SMILES of the monomers contained in the candidate structure. The range of the number of atoms may be arbitrarily adjusted depending on the amount of calculation and prediction accuracy, etc., but may be set to about 250 atoms (e.g., a range from 225 to 275 atoms). Then, the control unit calculates a descriptor of the candidate structure based on the SMILES of the oligomer (or polymer).

[0041] The minimum number of atoms is preferably 100 or more, more preferably 150 or more, particularly preferably 200 or more, and most preferably 225 or more. The maximum number of atoms is preferably 1000 or less, more preferably 700 or less, particularly preferably 400 or less, and most preferably 275 or less.

[0042] Conventionally, in machine learning models that predict the physical properties of substances, descriptors are generated on a monomer basis. In this case, the prediction accuracy can vary greatly depending on the number of atoms in the monomer. By generating molecular structures of oligomers (or polymers) with approximately the same number of atoms and calculating the descriptors based on the molecular structures, it is expected that the prediction accuracy of the machine learning model will improve.

[0043] In step S3, the control unit of the proposal device 100 selects a molecular structure of an organic polymer material (hereinafter also referred to as a "learning structure") for generating learning data. The learning structure may be arbitrarily selected from organic polymer materials whose physical property values ​​are known. The learning structure may be selected from the candidate structures stored in step S1. The number of learning structures may be a sufficient amount for learning a machine learning model (hereinafter also referred to as a "prediction model"). In this embodiment, the number of learning structures is, for example, about 250 structures.

[0044] In step S4, the control unit of the proposed device 100 generates learning data for generating a prediction model based on the learning structure selected in step S3. The learning data includes a descriptor indicating the molecular structure of the organic polymer material, and calculated values ​​of the physical properties of the organic polymer material. The physical properties of the organic polymer material include a physical property value related to dielectric properties and a physical property value related to thermal expansion. In this embodiment, the physical property value related to dielectric properties is, for example, a dielectric constant or a dielectric loss tangent. The physical property value related to thermal expansion is, for example, a linear expansion coefficient.

[0045] <Learning data generation process> 3 is a flowchart showing an example of the learning data generation process, which corresponds to step S4 in FIG.

[0046] In step S4-1, the control unit of the proposal device 100 reads out the descriptor of the learning structure from the storage unit. The descriptor of the learning structure is the descriptor calculated in step S2 and associated with the candidate structure.

[0047] In step S4-2, the control unit of the proposed device 100 calculates a physical property value related to the dielectric property based on the learning structure. The physical property value related to the dielectric property can be calculated by molecular simulation. In this embodiment, the dielectric constant is calculated.

[0048] The physical property values ​​related to the dielectric characteristics are calculated using a dielectric relaxation function derived based on the results of molecular dynamics calculations. However, the dielectric relaxation function derived by molecular dynamics calculations may have low accuracy. One of the reasons for this is that the time series data of the dipole moment calculated by molecular dynamics calculations has gentle fluctuations in the low frequency components. In this embodiment, a high-pass filter is used to extract a predetermined frequency band from the time series data of the dipole moment, and the dielectric relaxation function is derived based on the filtered time series data. This configuration improves the accuracy of the dielectric relaxation function, and also improves the accuracy of the physical property values ​​calculated using the dielectric relaxation function.

[0049] Specifically, the control unit of the proposed device 100 calculates physical property values ​​related to dielectric characteristics as follows. First, the control unit calculates the dipole moment of the learning structure at a predetermined time interval by molecular dynamics calculation. This generates time series data of the dipole moment. Next, the control unit designs a high-pass filter. The cutoff frequency of the high-pass filter is set to the inverse of the length of the dielectric relaxation function. The length of the dielectric relaxation function can be arbitrarily determined by the user. The length of the dielectric relaxation function is set to 100 nanoseconds, for example. Therefore, the cutoff frequency is 10 MHz.

[0050] Next, the control unit extracts a frequency band equal to or higher than the cutoff frequency from the time series data of the dipole moment. This generates filtered time series data of the dipole moment. Next, the control unit calculates a dielectric relaxation function based on the filtered time series data of the dipole moment. Then, the control unit performs a Fourier transform on the dielectric relaxation function to calculate a frequency-dependent complex dielectric constant. The control unit may calculate a frequency-dependent dielectric tangent, complex electrical conductivity, resistivity, or the like.

[0051] In step S4-3, the control unit of the proposed device 100 calculates the physical property value related to thermal expansion based on the learning structure. The physical property value related to thermal expansion can be calculated by molecular simulation. In this embodiment, the linear expansion coefficient is calculated.

[0052] Note that steps S4-2 and S4-3 may be executed in any order. For example, steps S4-2 and S4-3 may be executed in parallel, step S4-2 may be executed after step S4-3, or step S4-3 may be executed after step S4-2.

[0053] In step S4-4, the control unit of the proposal device 100 associates the physical property values ​​related to the dielectric properties calculated in step S4-2 and the physical property values ​​related to the thermal expansion calculated in step S4-3 with the descriptor of the learning structure calculated in step S2, thereby generating learning data related to the learning structure.

[0054] The processes from step S4-1 to step S4-4 are repeatedly executed for all learning structures selected in step S3, thereby generating learning data including descriptors and calculated values ​​for all learning structures selected in step S3.

[0055] Fig. 4 is a graph showing an example of the calculation accuracy of the dielectric constant. The graph shown in Fig. 4 plots the dielectric constants of various molecular structures with the horizontal axis representing the experimental value and the vertical axis representing the calculated value. Note that Fig. 4 targets dielectric constants of 10 MHz or more. As shown in Fig. 4, the calculated value of the dielectric constant by molecular simulation has a positive correlation with the experimental value.

[0056] Fig. 5 is a graph showing an example of the calculation accuracy of the linear expansion coefficient. The graph shown in Fig. 5 plots the linear expansion coefficients (ppm / K) of various molecular structures, with the horizontal axis representing the experimental values ​​and the vertical axis representing the calculated values. As shown in Fig. 5, the calculated values ​​of the linear expansion coefficient by molecular simulation have a positive correlation with the experimental values.

[0057] Returning to FIG. 2, the explanation will be given. In step S5, the control unit of the proposed device 100 generates a prediction model by learning the learning data generated in step S4. The prediction model includes a machine learning model that predicts a physical property value related to dielectric properties based on a molecular structure (hereinafter also referred to as a "dielectric property prediction model"), and a machine learning model that predicts a physical property value related to thermal expansion based on a molecular structure (hereinafter also referred to as a "thermal expansion prediction model"). The control unit of the proposed device 100 stores the learned dielectric property prediction model and thermal expansion prediction model in a storage unit.

[0058] The dielectric property prediction model is a machine learning model that learns the relationship between the molecular structure descriptor and the dielectric constant. The dielectric property prediction model takes the molecular structure descriptor as input and outputs a predicted value of the dielectric constant. The structure of the dielectric property prediction model is not limited, but for example, a LightGBM (Gradient Boosting Machine) or an ensemble decision tree can be used.

[0059] The thermal expansion prediction model is a machine learning model that learns the relationship between the molecular structure descriptor and the linear expansion coefficient. The thermal expansion prediction model takes the molecular structure descriptor as input and outputs a predicted value of the linear expansion coefficient. The structure of the thermal expansion prediction model is not limited, but, like the dielectric property prediction model, LightGBM or an ensemble decision tree can be used.

[0060] FIG. 6 is a graph showing an example of the prediction accuracy of the dielectric constant. The graph shown in FIG. 6 plots the dielectric constants of various molecular structures with the horizontal axis representing the calculated values ​​and the vertical axis representing the predicted values. Note that the black circles in the graph represent the learning data, and the white circles represent the verification data. In the graph shown in FIG. 6, the coefficient of determination R2 is about 0.8. As shown in FIG. 6, it can be seen that the dielectric constant can be predicted with high accuracy according to the dielectric property prediction model.

[0061] FIG. 7 is a graph showing an example of prediction accuracy of the linear expansion coefficient. The graph shown in FIG. 7 plots the linear expansion coefficients (ppm / K) of various molecular structures, with the horizontal axis representing the calculated values ​​and the vertical axis representing the predicted values. As in FIG. 6, the black circles in the graph represent learning data, and the white circles represent verification data. In the graph shown in FIG. 7, the coefficient of determination R2 is about 0.8. As shown in FIG. 7, it can be seen that the linear expansion coefficient can be predicted with high accuracy according to the thermal expansion prediction model.

[0062] Returning to FIG. 2, in step S6, the control unit of the proposed device 100 predicts the physical property values ​​of each candidate structure based on the trained prediction model. Specifically, the control unit first reads out the trained dielectric property prediction model and thermal expansion prediction model from the storage unit. Next, the control unit reads out the descriptor of the candidate structure for which the physical property values ​​are to be predicted from the storage unit. The candidate structure for which the physical property values ​​are to be predicted is any one of the candidate structures other than the trained structure selected in step S3.

[0063] Next, the control unit inputs the read-out descriptors of the candidate structure to a trained dielectric property prediction model. The dielectric property prediction model predicts a dielectric constant based on the input descriptors and outputs the predicted value. The control unit also inputs the descriptors of the candidate structure to a trained thermal expansion prediction model. The thermal expansion prediction model predicts a linear expansion coefficient based on the input descriptors and outputs the predicted value.

[0064] In step S7, the control unit of the proposed device 100 outputs proposal information indicating candidate structures in which all of the physical property values ​​predicted in step S6 satisfy the target values. The target values ​​include a target value for the linear expansion coefficient and a target value for the dielectric constant. The control unit may accept an input of the target values ​​in response to a user's operation. The control unit may also read out a predetermined target value from the storage unit.

[0065] Specifically, the control unit compares the predicted value of the dielectric constant predicted in step S6 with the target value for the dielectric constant. The control unit also compares the predicted value of the linear expansion coefficient predicted in step S6 with the target value for the linear expansion coefficient. If the predicted value of the dielectric constant satisfies the target value for the dielectric constant and the predicted value of the linear expansion coefficient satisfies the target value for the linear expansion coefficient, the control unit includes information about the candidate structure in the proposal information.

[0066] The proposed information includes information indicating the candidate structure and predicted values ​​of physical property values ​​(dielectric constant and linear expansion coefficient) related to the candidate structure. The proposed information may include information regarding the candidate structure that does not satisfy the target value. The proposed information may be sorted in ascending order of at least one of the dielectric constant and the linear expansion coefficient.

[0067] The control unit may display the proposed information on the display device 105. By referring to the proposed information displayed on the display device 105, the user can learn the molecular structure of an organic polymer material whose physical property values ​​related to dielectric characteristics and thermal expansion both satisfy the target values ​​(in other words, which achieves both low dielectric characteristics and low thermal expansion). The user can create an organic polymer material based on the proposed information and measure the physical property values ​​through experiments. The user can also manufacture a low-frequency substrate using the created organic polymer material.

[0068] The user can determine whether or not to update the prediction model based on the proposed information. Specifically, the user can instruct the display device 105 to consider whether the proposed information displayed on the display device 105 includes a promising molecular structure, generate new learning data in the vicinity of the molecular structure that the user considers to be promising, and update the prediction model. In this case, the user executes an update operation for the prediction model. The update operation for the prediction model may be performed by selecting a learning structure for generating new learning data.

[0069] In step S8, the control unit of the proposal device 100 determines whether to update the prediction model. Whether to update the prediction model can be determined, for example, based on whether the user of the proposal system 1 executes an update operation for the prediction model. If the user executes an update operation for the prediction model, the control unit determines to update the prediction model. On the other hand, if the user does not execute an update operation for the prediction model, the control unit determines not to update the prediction model.

[0070] If it is determined that the prediction model is to be updated (YES), the control unit of the proposal device 100 returns the process to step S3. On the other hand, if it is determined that the prediction model is not to be updated (NO), the control unit of the proposal device 100 ends the proposal method.

[0071] After returning the process to step S3, the control unit of the proposal device 100 selects a learning structure for generating new learning data. The learning structure may be determined based on the proposed structure output in step S7. For example, the user may select any number of learning structures from among the candidate structures in the vicinity of the proposed structure. The number of learning structures selected in step S3 from the second time onwards is arbitrary, but for example, several tens to a hundred structures may be selected.

[0072] Thereafter, the control unit of the proposal device 100 executes the processes from step S4 to step S8. Specifically, the control unit of the proposal device 100 creates learning data based on the learning structure selected in step S3 from the second time onwards, and updates the prediction model by learning the learning data. Then, the control unit predicts multiple physical property values ​​for the candidate structure based on the updated prediction model, and outputs a proposed structure in which all of the multiple physical property values ​​satisfy the target values.

[0073] In this way, the control unit of the proposal device 100 repeatedly executes outputting the proposed structure, generating learning data, and updating the prediction model until updating of the prediction model is completed. For example, by generating new learning data in the vicinity of a molecular structure that the user considers promising among the proposed molecular structures, a prediction model with improved prediction accuracy for molecular structures similar to the molecular structure can be obtained.

[0074] <Modification> In this embodiment, a configuration has been described in which a physical property value related to dielectric properties and a physical property value related to thermal expansion are predicted based on a trained machine learning model. At least one of the physical property values ​​used in the proposal may be predicted based on a trained machine learning model, and the other physical property values ​​may be obtained by other methods. For example, the other methods may include calculating the physical property values ​​by molecular simulation, or reading out known physical property values ​​stored in a database and using the read values ​​as they are or by interpolating them.

[0075] <Summary> As described above, according to one embodiment of the present disclosure, it is possible to propose a molecular structure of an organic polymer material in which multiple physical property values ​​simultaneously satisfy target values. For example, the proposal device 100 predicts multiple physical property values ​​based on the molecular structure of an organic polymer material, and outputs a molecular structure in which all of the multiple physical property values ​​satisfy target values. Therefore, according to one embodiment of the present disclosure, it is possible to propose a molecular structure of an organic polymer material in which multiple physical property values ​​simultaneously satisfy target values.

[0076] The proposed device 100 may predict at least one of the multiple physical property values ​​based on a trained machine learning model. The proposed device 100 may generate training data including multiple physical property values ​​calculated by molecular simulation, and create or update a machine learning model by learning the training data. It is very costly to actually create an organic polymer material and measure the physical property values ​​through experiments. According to an embodiment of the present disclosure, training data can be generated at low cost through molecular simulation.

[0077] The proposed device 100 may calculate a descriptor based on a molecular structure in which monomer molecular structures are repeatedly connected and the number of atoms falls within a predetermined range, and generate training data including the descriptor. When a descriptor calculated on a monomer basis is used, the prediction accuracy varies depending on the number of atoms, but it has been found that prediction accuracy is improved when a descriptor calculated from a molecular structure of an oligomer (or polymer) in which the number of atoms falls within a predetermined range is used. According to one embodiment of the present disclosure, physical property values ​​related to an organic polymer material can be predicted with high accuracy.

[0078] The proposing device 100 may repeatedly execute outputting the molecular structure, generating the learning data, and updating the machine learning model. For example, a user can generate learning data in the vicinity of a promising molecular structure while referring to the proposed molecular structure, thereby further improving the accuracy of the prediction model. According to an embodiment of the present disclosure, it is possible to predict the physical property values ​​of an organic polymer material with high accuracy.

[0079] The plurality of physical property values ​​may include a physical property value related to a dielectric property and a physical property value related to thermal expansion. The physical property value related to the dielectric property may be a dielectric constant or a dielectric loss tangent. The physical property value related to thermal expansion may be a linear expansion coefficient. As a material for a low-frequency substrate, a substance that satisfies low dielectric properties and low thermal expansion is required, but it is known that there is a trade-off between the dielectric property and the thermal expansion. According to an embodiment of the present disclosure, it is possible to propose a molecular structure of an organic polymer material that can be used as a material for a low-frequency substrate.

[0080] The physical property values ​​related to the dielectric properties may be calculated using a dielectric relaxation function based on time series data of the dipole moment of the organic polymer material. The physical property values ​​related to the dielectric properties may be calculated using a dielectric relaxation function based on time series data obtained by extracting a predetermined frequency band from the time series data of the dipole moment of the organic polymer material. The dielectric relaxation function derived by molecular simulation may have low accuracy. One of the reasons for this is that the dipole moment calculated by molecular simulation has gradual fluctuations in the low frequency components. According to one embodiment of the present disclosure, the physical property values ​​related to the dielectric properties can be calculated with high accuracy.

[0081] [supplement] Each function of the above-described embodiments can be realized by one or more processing circuits. Here, the term "processing circuit" in this specification includes a processor programmed to execute each function by software, such as a CPU (Central Processing Unit) or GPU (Graphics Processing Unit) implemented by an electronic circuit, and an ASIC (Application Specific Integrated Circuit), DSP (Digital Signal Processor), FPGA (Field Programmable Gate Array), conventional circuit module, and other devices designed to execute each function described above.

[0082] Although the embodiments have been described above, it will be understood that various changes in form and details are possible without departing from the spirit and scope of the claims. [Explanation of symbols]

[0083] 100 Proposed device

Claims

1. A proposal device having a control unit that proposes a molecular structure of an organic polymer material, The control unit is Predicting a plurality of physical properties based on the molecular structure; outputting the molecular structure in which all of the plurality of physical property values ​​satisfy target values; Proposed device.

2. The control unit is predicting at least one of the plurality of physical property values ​​based on a trained machine learning model; The proposal device according to claim 1 .

3. The control unit is generating learning data including the plurality of physical property values ​​calculated by molecular simulation; Creating or updating the machine learning model by learning the learning data. The proposal device according to claim 2 .

4. The control unit is Calculating a descriptor based on a molecular structure in which a monomer molecular structure is repeatedly connected and the number of atoms is within a predetermined range; generating the training data including the descriptors; The proposal device according to claim 3 .

5. The control unit is outputting the molecular structure; generating the training data; updating the machine learning model; and Repeatedly execute The proposal device according to claim 3 .

6. The plurality of physical properties include a physical property value related to a dielectric property and a physical property value related to thermal expansion. A proposal device according to any one of claims 1 to 5.

7. The physical property value related to the dielectric characteristic is a dielectric constant or a dielectric loss tangent. The proposal device according to claim 6.

8. The physical property value regarding the dielectric characteristics is calculated using a dielectric relaxation function based on time series data of the dipole moment of the organic polymer material. The proposal device according to claim 6.

9. The physical property value related to the dielectric characteristics is calculated using a dielectric relaxation function based on time-series data obtained by extracting a predetermined frequency band from time-series data of the dipole moment of the organic polymer material. The proposal device according to claim 8.

10. The physical property value related to thermal expansion is a linear expansion coefficient. The proposal device according to claim 6.

11. The molecular structure of the organic polymer material includes polyimide, polyolefin, polyvinyl, polyacrylic, polyester, polyurethane, polyurea, polycarbonate, polysulfone, polyamide, polyhaloolefin, polystyrene, polyketone or polyimine; A proposal device according to any one of claims 1 to 5.

12. A control unit of a proposal device for proposing a molecular structure of an organic polymer material, Predicting a plurality of physical properties based on the molecular structure; outputting the molecular structure in which all of the plurality of physical property values ​​satisfy target values; The suggested way to perform the process.

13. A control unit of a proposal device for proposing a molecular structure of an organic polymer material, Predicting a plurality of physical properties based on the molecular structure; outputting the molecular structure in which all of the plurality of physical property values ​​satisfy target values; A program for executing a process.

Images