Learning device, information processing device, substrate processing system, learning method, recipe determination method, and learning program

Abstract

In the present invention, a data acquisition unit acquires data that indicates the relationship between: processing conditions for processing to dry a substrate by sublimation using a processing liquid that contains a sublimation agent and a solvent, and a molecular descriptor for the sublimation agent and / or the solvent; and substrate processing results. A data compensation unit, using a prediction algorithm for predicting physical values from molecular descriptors, compensates for defects in physical values of the sublimation agent and / or the solvent in the data acquired by the data acquisition unit. A results model generation unit, by machine-learning the data for which the physical values have been compensated by the data compensation unit, generates a results model for predicting the substrate processing results.

Description

Learning device, information processing device, substrate processing system, learning method, recipe determination method, and learning program

[0001] The present invention relates to a learning device, an information processing device, a substrate processing system, a learning method, a recipe determination method, and a learning program.

[0002] In recent years, the finer patterns formed on substrates have tended to weaken their strength. In this case, when a liquid applied to a substrate is dried, the pattern is more likely to collapse due to the surface tension acting between the liquid and the surface of the pattern formed on the substrate. Therefore, in order to satisfactorily dry a substrate on which a fine pattern is formed, sublimation drying is sometimes performed on the substrate. Sublimation drying is a technique for drying a liquid applied to the surface of a substrate by converting it into a solid and then changing the phase from solid to gas.

[0003] For example, in the sublimation drying method described in Patent Document 1, a treatment liquid in which cyclohexanone oxime is dissolved in a solvent is supplied to the surface of a substrate on which a pattern has been formed, and a liquid film of the treatment liquid is formed on the surface of the substrate. The treatment liquid has a concentration adjusted to suit the treatment conditions, such as the type of substrate or rotation speed. Next, the liquid film of the treatment liquid is solidified to form a solidified film of cyclohexanone oxime. The solidified film is then removed from the surface of the substrate by sublimation.

[0004] Japanese Patent Application Laid-Open No. 2021-9988

[0005] In sublimation drying, a processing solution in which a sublimation agent and a solvent are appropriately combined depending on the substrate to be processed is used. When the sublimation agent or solvent is changed, the processing conditions for the substrate must also be changed. However, it is costly to confirm the optimal processing conditions for each combination of sublimation agent and solvent in advance through experiments, etc.

[0006] Therefore, the present inventors have investigated the possibility of searching for optimal processing conditions for any combination of sublimation agent and solvent by using a prediction model generated by machine learning. However, according to the findings of the present inventors, it has been found that the physical properties of sublimation agents or solvents used in the field of sublimation drying have not often been investigated. Therefore, it is difficult to search for optimal processing conditions using a prediction model.

[0007] An object of the present invention is to provide a learning device, an information processing device, a substrate processing system, a learning method, a recipe determination method, and a learning program that are capable of generating a predictive model for searching for optimal processing conditions for a substrate even when there are missing physical property values ​​for a sublimation agent or solvent.

[0008] A learning device according to one aspect of the present invention includes a data acquisition unit that acquires data indicating the relationship between processing conditions in a process of sublimating and drying a substrate using a processing liquid containing a sublimator and a solvent, molecular descriptors of at least one of the sublimator and the solvent, and the processing results of the substrate; a data completion unit that uses a prediction algorithm to predict physical property values ​​from the molecular descriptors to complete missing physical property values ​​of at least one of the sublimator and the solvent in the data acquired by the data acquisition unit; and a result model generation unit that generates a result model that predicts the processing results of the substrate by machine learning the data whose physical property values ​​have been completed by the data completion unit.

[0009] An information processing device according to another aspect of the present invention is an information processing device that uses a result model generated by the above-mentioned learning device, and includes a model acquisition unit that acquires the result model, a sublimation agent determination unit that determines a sublimation agent to be used for sublimation drying of a substrate, a solvent determination unit that determines a solvent to be used for sublimation drying of a substrate, a condition selection unit that selects processing conditions, and a result prediction unit that uses the result model acquired by the model acquisition unit to predict a processing result of a substrate from the sublimation agent determined by the sublimation agent determination unit, the solvent determined by the solvent determination unit, and the processing conditions selected by the condition selection unit.

[0010] A substrate processing system according to yet another aspect of the present invention includes the above-described information processing apparatus.

[0011] A learning method according to yet another aspect of the present invention includes acquiring data showing the relationship between processing conditions in a process of sublimating and drying a substrate using a processing liquid containing a sublimator and a solvent, molecular descriptors of at least one of the sublimator and the solvent, and the processing results of the substrate; complementing missing physical property values ​​of at least one of the sublimator and the solvent in the acquired data using a prediction algorithm that predicts physical property values ​​from the molecular descriptors; and generating a result model that predicts the processing results of the substrate by machine learning the data with the complemented physical property values, and is executed by a processor.

[0012] A recipe determination method according to yet another aspect of the present invention is a recipe determination method that uses a result model generated by the above-described learning method, and includes acquiring the result model, determining a sublimation agent to be used for sublimation drying of a substrate, determining a solvent to be used for sublimation drying of the substrate, selecting process conditions, and using the acquired result model to predict a process result of the substrate from the determined sublimation agent, the determined solvent, and the selected process conditions, and is executed by a processor.

[0013] A learning program according to yet another aspect of the present invention causes a processor to execute the following processes: acquiring data indicating the relationship between processing conditions in a process of sublimating and drying a substrate using a processing liquid containing a sublimator and a solvent, molecular descriptors of at least one of the sublimator and the solvent, and the processing results of the substrate; complementing missing physical property values ​​of at least one of the sublimator and the solvent in the acquired data using a prediction algorithm that predicts physical property values ​​from the molecular descriptors; and generating a result model that predicts the processing results of the substrate by machine learning the data with the complemented physical property values.

[0014] According to the present invention, even when the physical property values ​​of the sublimation agent or solvent are missing, a prediction model can be generated to search for the optimum processing conditions for the substrate.

[0015] FIG. 1 is a diagram showing an example of the configuration of a substrate processing system according to an embodiment of the present invention. FIG. 2 is a diagram showing an example of the configuration of an information processing device. FIG. 3 is a diagram showing an example of a reference sublimator database pre-stored in a database storage device. FIG. 4 is a diagram showing an example of a reference solvent database pre-stored in a database storage device. FIG. 5 is a diagram showing an example of a result database stored in a database storage device. FIG. 6 is a block diagram showing the functional configuration of the learning device of FIG. 1. FIG. 7 is a diagram showing an example of a processing sublimator database generated by a sublimator specifying unit. FIG. 8 is a diagram showing an example of a processing solvent database generated by a solvent specifying unit. FIG. 9 is a diagram showing an example of a complementary sublimator database generated by a sublimator complementing unit. FIG. 10 is a diagram showing an example of a complementary solvent database generated by a solvent complementing unit. FIG. 11 is a diagram showing an example of an integrated database generated by a database generating unit. FIG. 12 is a diagram showing an example of a learning database generated by a processing liquid calculating unit. FIG. 13 is a block diagram showing the functional configuration of the information processing device of FIG. 1. FIG. 14 is a schematic diagram showing the interior of the substrate processing device as viewed horizontally. FIG. 15 is a flowchart showing an example of the flow of a learning process. FIG. 16 is a flowchart showing an example of the flow of a learning process. FIG. 17 is a flowchart showing an example of the flow of a recipe determination process. FIG. 18 is a process diagram for explaining an example of substrate processing by the substrate processing apparatus.

[0016] 1. Substrate Processing System A substrate processing system according to an embodiment of the present invention will now be described with reference to the drawings. In the following description, the term "substrate" refers to a semiconductor substrate (wafer), a substrate for an FPD (Flat Panel Display) such as a liquid crystal display device or an organic EL (Electro Luminescence) display device, a substrate for an optical disk, a substrate for a magnetic disk, a substrate for a magneto-optical disk, a substrate for a photomask, a ceramic substrate, or a substrate for a solar cell. FIG. 1 is a diagram showing an example of the configuration of a substrate processing system according to an embodiment of the present invention. As shown in FIG. 1, a substrate processing system 500 includes a substrate processing apparatus 100, a database storage device 200, an information processing device 300, and a learning device 400.

[0017] The substrate processing apparatus 100, the database storage device 200, the information processing device 300, and the learning device 400 are connected to a network 501, and are capable of transmitting and receiving data to and from each other. The network 501 may be, for example, a local area network (LAN) or a wide area network (WAN). Alternatively, the network 501 may be the Internet. The network 501 may be connected via a wired connection or a wireless connection. Furthermore, the substrate processing apparatus 100 and the information processing device 300 may be connected via a dedicated network instead of the network 501.

[0018] The substrate processing apparatus 100 performs a series of processes to form a solid or liquid film on the substrate by supplying a processing liquid to the substrate under predetermined processing conditions, and then removes the solid or liquid film from the substrate and dries the substrate. Specifically, the substrate processing apparatus 100 applies a processing liquid, which is a mixture of a sublimation agent and a solvent at a predetermined concentration, to the surface of the substrate on which a pattern has been formed, and then performs a series of processes to deposit the sublimation agent on the substrate by rotating the substrate and evaporating the solvent. The substrate processing apparatus 100 then sprays an inert gas onto the substrate to sublimate and dry it.

[0019] The database storage device 200 includes a large-capacity storage device such as a server. The database storage device 200 pre-stores a reference sublimation agent database indicating molecular descriptors (hereinafter simply referred to as descriptors) and physical property values ​​of reference sublimation agents. The reference sublimation agents are general sublimation agents that are not necessarily used for sublimation drying of substrates. The database storage device 200 also pre-stores a reference solvent database indicating descriptors and physical property values ​​of reference solvents. The reference solvents are general solvents that are not necessarily used for sublimation drying of substrates. The database storage device 200 also stores a result database indicating the relationship between combinations of sublimation agents and solvents actually used for sublimation drying of substrates, combinations of multiple substrate processing conditions, and processing results.

[0020] Information processing device 300 is configured, for example, by a general-purpose computer. Fig. 2 is a diagram showing an example of the configuration of information processing device 300. As shown in Fig. 2, information processing device 300 includes a CPU (central processing unit) 310, a RAM (random access memory) 320, a ROM (read-only memory) 330, a storage device 340, an operation unit 350, a display device 360, an input / output I / F (interface) 370, and a bus 380. CPU 310, RAM 320, ROM 330, storage device 340, operation unit 350, display device 360, and input / output I / F 370 are connected to bus 380.

[0021] The RAM 320 is used as a work area for the CPU 310. The ROM 330 stores a system program. The storage device 340 includes a storage medium such as a hard disk or semiconductor memory, and stores a recipe determination program for executing the recipe determination process described below. The recipe determination program may be stored in the ROM 330 or another external storage device.

[0022] The operation unit 350 is an input device such as a keyboard, a mouse, or a touch panel. A user can give predetermined instructions to the information processing device 300 by operating the operation unit 350. The display device 360 ​​is a display device such as a liquid crystal display device, and displays a GUI (Graphical User Interface) or the like for receiving instructions from the user. The input / output I / F 370 is connected to a network 501.

[0023] The learning device 400 is configured, for example, by a general-purpose computer. Therefore, the learning device 400 has a configuration basically similar to that of the information processing device 300 except for the following points, and therefore a detailed description of the configuration of the learning device 400 will be omitted. The storage device of the learning device 400 stores a learning program for executing the learning process described below. The learning program may be stored in the ROM of the learning device 400 or in another external storage device. Note that the information processing device 300 and the learning device 400 may be configured with the same hardware. In this case, the programs installed on the information processing device 300 and the learning device 400 are different. Therefore, the processes executed by the information processing device 300 and the learning device 400 are different from each other.

[0024] 2. Database Storage Device Fig. 3 is a diagram showing an example of a reference sublimation agent database stored in advance in the database storage device 200. As shown in Fig. 3, the reference sublimation agent database 201 shows a plurality of descriptors and a plurality of physical property values ​​for each of a plurality of reference sublimation agents. In the reference sublimation agent database 201, there are no missing descriptors or physical property values ​​for each reference sublimation agent.

[0025] Specific examples of the descriptor of the reference sublimation agent include RDKit, Mordred, or fingerprint (Morgan or MACCS Keys, etc.).Specific examples of the physical properties of the sublimation agent include crystal structure, lattice constant, vapor pressure, molecular weight, density, molar volume, melting point, glass transition temperature, boiling point, heat of evaporation, heat of fusion, heat of sublimation, viscosity, HSP (Hansen Solubility Parameter), chemical potential, hydration energy, partition coefficient, interfacial free energy with the substrate, refractive index, dielectric constant, polarizability, dipole moment, HOMO / LUMO, or HOMO / LUMO gap.

[0026] Fig. 4 is a diagram showing an example of a reference solvent database pre-stored in the database storage device 200. As shown in Fig. 4, the reference solvent database 202 shows a plurality of descriptors and a plurality of physical property values ​​for each of a plurality of reference solvents. In the reference solvent database 202, there are no missing descriptors or physical property values ​​for each reference solvent.

[0027] Specific examples of the descriptor of the reference solvent include RDKit, Mordred, or fingerprint (Morgan or MACCS Keys, etc.). Specific examples of the physical properties of the reference solvent include vapor pressure, molecular weight, density, molar volume, melting point, glass transition temperature, boiling point, heat of evaporation, heat of fusion, heat of sublimation, viscosity, HSP, chemical potential, hydration energy, partition coefficient, surface tension, interfacial free energy with the substrate, refractive index, dielectric constant, polarizability, dipole moment, HOMO / LUMO, or HOMO / LUMO gap.

[0028] 5 is a diagram showing an example of a result database stored in the database storage device 200. As shown in Fig. 5, the result database 203 shows the relationship between the combination of sublimation agent and solvent used in the actual sublimation drying of a substrate, the combination of multiple substrate processing conditions, and the processing result. In this example, the processing result is the collapse rate of a pattern formed on the surface of the substrate, and is evaluated by, for example, image analysis of the substrate.

[0029] Specific examples of substrate processing conditions include the rotation speed of the substrate, the temperature of the solution, the temperature of the substrate, the temperature inside the chamber described below, the humidity inside the chamber, the processing time, the flow rate of the inert gas, the flow rate of the processing liquid, the air flow inside the chamber, the ejection position of each nozzle described below, the residence time of each nozzle, the oscillation behavior of each nozzle, or the pressure inside the chamber.

[0030] The result database 203 may be updated by adding new data each time sublimation drying of a substrate is performed. In this case, the data used in the sublimation drying and the processing results may be added to the result database 203 by a user. Alternatively, some data, such as the sublimation agent, solvent, or processing conditions, may be added directly to the result database 203 from the substrate processing apparatus 100 of FIG. 1 .

[0031] 3. Learning Device Fig. 6 is a block diagram showing the functional configuration of the learning device 400 of Fig. 1. As shown in Fig. 6, the learning device 400 includes, as functional units, a reference sublimation agent acquisition unit 401, a sublimation agent model generation unit 402, a reference solvent acquisition unit 403, a solvent model generation unit 404, a result acquisition unit 405, a sublimation agent identification unit 406, a solvent identification unit 407, a sublimation agent complementation unit 408, a solvent complementation unit 409, a database generation unit 410, a treatment liquid calculation unit 411, a result model generation unit 412, and an optimization unit 413. The functional units of the learning device 400 are realized by the CPU of the learning device 400 executing a learning program. Some or all of the functional units of the learning device 400 may be realized by hardware such as electronic circuits.

[0032] 3 from the database storage device 200. The reference sublimation agent acquisition unit 401 may acquire the reference sublimation agent database 201 from a server or the like provided outside the substrate processing system 500. The sublimation agent model generation unit 402 generates a sublimation agent model that predicts the physical property values ​​of the sublimation agent from the descriptors of the sublimation agent by training a predetermined machine learning model using the descriptors of the reference sublimation agent in the reference sublimation agent database 201 acquired by the reference sublimation agent acquisition unit 401 as explanatory variables and the physical property values ​​of the reference sublimation agent as objective variables.

[0033] 4 from the database storage device 200. The reference solvent acquisition unit 403 may acquire the reference solvent database 202 from a server or the like provided outside the substrate processing system 500. The solvent model generation unit 404 generates a solvent model that predicts the physical property values ​​of the solvent from the solvent descriptors by training a predetermined machine learning model using the descriptors of the reference solvents in the reference solvent database 202 acquired by the reference solvent acquisition unit 403 as explanatory variables and the physical property values ​​of the reference solvent as objective variables.

[0034] 5 from the database storage device 200. The sublimation agent specifying unit 406 specifies the sublimation agent used in the actual sublimation drying of the substrate, and specifies the descriptors and physical property values ​​of the sublimation agent, based on the result database 203 obtained by the result obtaining unit 405. In this example, the sublimation agent specifying unit 406 generates a processing sublimation agent database indicating the descriptors and physical property values ​​of the specified sublimation agent.

[0035] FIG. 7 is a diagram showing an example of the processing sublimator database generated by the sublimator identification unit 406. As shown in FIG. 7, the processing sublimator database 204 shows descriptors and physical property values ​​for each of a plurality of sublimators used in the sublimation drying of actual substrates. The processing sublimator database 204 may be generated by a user operating the operation unit of the learning device 400 to input the descriptors and physical property values ​​of the sublimators used in the sublimation drying of actual substrates into the sublimator identification unit 406. Here, the physical property values ​​of sublimators used in the sublimation drying of actual substrates are often not investigated. Therefore, the processing sublimator database 204 may contain missing physical property values ​​for each sublimator.

[0036] The solvent identification unit 407 identifies the solvent actually used in the sublimation drying of the substrate, and also identifies the descriptors and physical property values ​​of the solvent, based on the result database 203 acquired by the result acquisition unit 405. In this example, the solvent identification unit 407 generates a processing solvent database indicating the descriptors and physical property values ​​of the identified solvent.

[0037] FIG. 8 is a diagram showing an example of a process solvent database generated by the solvent identification unit 407. As shown in FIG. 8, the process solvent database 205 shows descriptors and physical property values ​​for each of a plurality of solvents actually used in the sublimation drying of substrates. The process solvent database 205 may be generated by a user operating the operation unit of the learning device 400 to input the descriptors and physical property values ​​of the solvents actually used in the sublimation drying of substrates into the solvent identification unit 407. Here, the physical property values ​​of solvents actually used in the sublimation drying of substrates are often not investigated. Therefore, the process solvent database 205 may contain missing physical property values ​​for each solvent.

[0038] The sublimation agent complementing unit 408 predicts the physical property values ​​of each sublimation agent from the descriptor of the sublimation agent in the processing sublimation agent database 204 generated by the sublimation agent identifying unit 406, using the sublimation agent model generated by the sublimation agent model generating unit 402. Furthermore, the sublimation agent complementing unit 408 complements missing physical property values ​​in the processing sublimation agent database 204, using the predicted physical property values ​​of the sublimation agent. In this example, the sublimation agent complementing unit 408 generates a complementary sublimation agent database in which the physical property values ​​of the sublimation agents are complemented. FIG. 9 is a diagram showing an example of the complementary sublimation agent database generated by the sublimation agent complementing unit 408. As shown in FIG. 9, the complementary sublimation agent database 206 does not lack any descriptors or physical property values ​​of each sublimation agent.

[0039] The solvent complementing unit 409 uses the solvent model generated by the solvent model generating unit 404 to predict the physical property values ​​of each solvent in the processing solvent database 205 generated by the solvent identifying unit 407 from the descriptors of the solvent. The solvent complementing unit 409 also uses the predicted physical property values ​​of the solvent to complement missing physical property values ​​in the processing solvent database 205. In this example, the solvent complementing unit 409 generates a complementary solvent database in which the physical property values ​​of the solvents are complemented. FIG. 10 is a diagram showing an example of the complementary solvent database generated by the solvent complementing unit 409. As shown in FIG. 10, the complementary solvent database 207 has no missing descriptors or physical property values ​​for each solvent.

[0040] The database generating unit 410 generates an integrated database. Fig. 11 is a diagram showing an example of the integrated database generated by the database generating unit 410. As shown in Fig. 11 , the integrated database 208 integrates the result database 203 of Fig. 5 acquired by the result acquiring unit 405, the complementary sublimation agent database 206 of Fig. 9 generated by the sublimation agent complementing unit 408, and the complementary solvent database 207 of Fig. 10 generated by the solvent complementing unit 409.

[0041] The treatment liquid calculation unit 411 calculates the descriptors and physical property values ​​of a treatment liquid obtained by dissolving a sublimation agent in a solvent from the descriptors and physical property values ​​of the sublimation agent and the descriptors and physical property values ​​of the solvent in the integrated database 208. In this example, the treatment liquid calculation unit 411 generates a training database using the calculated descriptors and physical property values ​​of the treatment liquid. FIG. 12 is a diagram showing an example of the training database generated by the treatment liquid calculation unit 411. As shown in FIG. 12, in the training database 209, the descriptors and physical property values ​​of the treatment liquid calculated by the treatment liquid calculation unit 411 are added to the integrated database 208.

[0042] A specific example of a descriptor of the treatment liquid is the result of an arithmetic operation of descriptors of the same type for the sublimation agent and the solvent. A specific example of a physical property value of the treatment liquid is the result of an arithmetic operation of descriptors of the same type for the sublimation agent and the solvent. In these arithmetic operations, weights may be assigned to the descriptors or physical property values ​​of the sublimation agent or the solvent based on the concentration ratio between the sublimation agent and the solvent. Further, other specific examples of physical property values ​​of the treatment liquid include the solubility of the sublimation agent in the solvent, the χ parameter or HSP of the sublimation agent in the solvent, the interfacial free energy between the solvent and the sublimation agent, or the heat of solution when the sublimation agent is dissolved in the solvent.

[0043] The result model generation unit 412 generates a result model that predicts the processing results of a substrate from the combination of a sublimant and a solvent and the processing conditions of the substrate, based on the training database 209 generated by the database generation unit 410. Specifically, the result model is generated by training a predetermined machine learning model using the descriptors and physical property values ​​of the sublimant, solvent, and processing liquid in the training database 209 and the processing conditions of the substrate as explanatory variables, and the processing conditions of the substrate as objective variables.

[0044] Before the result model is generated, the optimization unit 413 optimizes the explanatory variables used to generate the result model. Specifically, the optimization unit 413 calculates the importance of the explanatory variables. The importance may be calculated using a SHAP (Shapely Additive exPlanations) value or the importance of an extra-tree feature. In this case, the optimized result model is generated by the result model generation unit 412. Alternatively, the optimization unit 413 may optimize the explanatory variables after the result model is generated, and cause the result model generation unit 412 to re-learn using the optimized explanatory variables. In this case, the result model generated by the result model generation unit 412 is updated to a result model that can predict processing results with higher accuracy.

[0045] 13 is a block diagram showing the functional configuration of the information processing device 300 of FIG. 1. As shown in FIG. 13, the information processing device 300 includes, as functional units, a model acquisition unit 301, a sublimation agent determination unit 302, a solvent determination unit 303, a condition selection unit 304, a result prediction unit 305, an evaluation unit 306, and a presentation unit 307. The functional units of the information processing device 300 are realized by the CPU 310 of the information processing device 300 of FIG. 2 executing a recipe determination program. Some or all of the functional units of the information processing device 300 may be realized by hardware such as electronic circuits.

[0046] The model acquisition unit 301 acquires a result model generated by the result model generation unit 412 of the learning device 400 in FIG. 6 . The sublimation agent determination unit 302 determines a sublimation agent to be used for sublimation drying of a substrate based on a user's specification. The solvent determination unit 303 determines a solvent to be used for sublimation drying of a substrate based on a user's specification. The user can specify a sublimation agent and a solvent appropriate for the substrate to be processed in the sublimation agent determination unit 302 and the solvent determination unit 303, respectively, by operating the operation unit 350 in FIG. 2 .

[0047] The condition selection unit 304 selects processing conditions for performing a series of processes on a substrate from among a plurality of processing conditions. The processing conditions include a combination of a plurality of conditions, such as the rotation speed of the substrate, the temperature of the solution, and the flow rate of the inert gas during the series of processes. The processing conditions are selected using known methods such as Bayesian optimization, steepest descent method, or genetic algorithm. The result prediction unit 305 predicts the processing results of the substrate from the sublimator determined by the sublimator determination unit 302, the solvent determined by the solvent determination unit 303, and the processing conditions selected by the condition selection unit 304, using the result model acquired by the model acquisition unit 301.

[0048] The evaluation unit 306 evaluates a score indicating the quality of the substrate processing result predicted by the result prediction unit 305. In this example, a higher score is assigned to a lower pattern collapse rate. The evaluation unit 306 also identifies the processing conditions selected by the condition selection unit 304 when the processing result to which the highest score was assigned is obtained. When one or more processing results to which a score equal to or greater than a predetermined threshold is assigned are obtained, the evaluation unit 306 may identify one or more processing conditions selected by the condition selection unit 304 when the one or more processing results were obtained.

[0049] The presentation unit 307 presents to the user the processing conditions identified by the evaluation unit 306. The presentation of the processing conditions may be performed, for example, by displaying the processing conditions on the display device 360 ​​of Fig. 2. This allows the user to visually check the processing conditions displayed on the display device 360 ​​and set optimal processing conditions for sublimation drying of the substrate in the substrate processing apparatus 100 of Fig. 1.

[0050] 5. Substrate Processing Apparatus Figure 14 is a horizontal schematic diagram of the interior of the substrate processing apparatus 100. As shown in Figure 14, the substrate processing apparatus 100 includes a processing unit 2 and a control device 3. The control device 3 is capable of communicating with the information processing apparatus 300 of Figure 1 and controls the processing unit 2 to sublimate and dry the substrate W under processing conditions set by a user. The processing unit 2 includes a box-shaped chamber 4, a spin chuck 10 that holds one substrate W horizontally in the chamber 4 and rotates the substrate W about a vertical rotation axis A1 passing through the center of the substrate W, and a cylindrical processing cup 21 that surrounds the spin chuck 10 about the rotation axis A1.

[0051] The spin chuck 10 includes a disk-shaped spin base 12 held in a horizontal position, a plurality of chuck pins 11 that hold the substrate W in a horizontal position above the spin base 12, a spin shaft 13 that extends downward from the center of the spin base 12, and a spin motor 14 that rotates the spin shaft 13 to rotate the spin base 12 and the plurality of chuck pins 11.

[0052] The processing cup 21 includes a plurality of guards 24 that receive the processing liquid discharged outward from the substrate W, a plurality of cups 23 that receive the processing liquid guided downward by the plurality of guards 24, and a cylindrical outer wall member 22 that surrounds the plurality of guards 24 and the plurality of cups 23. The plurality of guards 24 can be individually raised and lowered by a guard lifting unit 27.

[0053] Processing unit 2 includes a chemical solution nozzle 31 that discharges a chemical solution, a rinse solution nozzle 35 that discharges a rinse solution, a processing solution nozzle 39 that discharges a processing solution, and a substitution solution nozzle 43 that discharges a substitution solution. Chemical solution nozzle 31, rinse solution nozzle 35, processing solution nozzle 39, and substitution solution nozzle 43 can be independently moved horizontally within chamber 4 by nozzle movement units 34, 38, 42, and 46 provided corresponding to each nozzle.

[0054] The processing liquid nozzle 39 is connected to a processing liquid pipe 40 that guides the processing liquid to the processing liquid nozzle 39. When a processing liquid valve 41 provided in the processing liquid pipe 40 is opened, the processing liquid is continuously discharged downward from the discharge port of the processing liquid nozzle 39. The processing liquid is a mixed liquid containing a sublimation agent and a solvent that is soluble in the sublimation agent.

[0055] The processing liquid nozzle 39 is connected to a nozzle moving unit 42. The nozzle moving unit 42 moves the processing liquid nozzle 39 in at least one of the vertical and horizontal directions. The nozzle moving unit 42 moves the processing liquid nozzle 39 horizontally between a processing position where the processing liquid ejected from the processing liquid nozzle 39 is supplied to the upper surface of the substrate W, and a standby position where the processing liquid nozzle 39 is positioned around the processing cup 21 in a plan view.

[0056] The processing unit 2 includes a disk-shaped blocking member 51 arranged above the spin chuck 10. The blocking member 51 includes a disk portion 52 arranged horizontally above the spin chuck 10. The blocking member 51 is supported horizontally by a cylindrical support shaft 53 extending upward from the center of the disk portion 52. The center line of the disk portion 52 is arranged on the rotation axis A1 of the substrate W. The lower surface of the disk portion 52 corresponds to the lower surface 51L of the blocking member 51. The lower surface 51L of the blocking member 51 is parallel to the upper surface of the substrate W and has an outer diameter equal to or greater than the diameter of the substrate W.

[0057] The blocking member 51 is connected to a blocking member lifting unit 54 that vertically raises and lowers the blocking member 51. The blocking member lifting unit 54 moves the blocking member 51 to any position between an upper position (the position shown in FIG. 14 ) and a lower position.

[0058] A center nozzle 55 is disposed in a through-hole that passes vertically through the center of the blocking member 51. The center nozzle 55 moves up and down together with the blocking member 51. The center nozzle 55 is connected to an upper gas pipe 56 that guides inert gas to the center nozzle 55. The substrate processing apparatus 100 includes an upper temperature regulator 59 that heats or cools the inert gas discharged from the center nozzle 55. When an upper gas valve 57 disposed in the upper gas pipe 56 is opened, the inert gas is continuously discharged downward from the discharge port of the center nozzle 55 at a flow rate that corresponds to the opening of a flow rate adjustment valve 58 that changes the flow rate of the inert gas. The inert gas discharged from the center nozzle 55 is nitrogen gas.

[0059] 15 and 16 are flowcharts showing an example of the flow of the learning process. The learning process is performed by a CPU included in the learning device 400 as the CPU executes a learning program. The learning process of FIGS. 15 and 16 will be described below with reference to the learning device 400 of FIG. 6 and the various databases of FIGS. 3 to 5 and 7 to 12.

[0060] 3 from the database storage device 200 (step S1). Next, the sublimation agent model generation unit 402 generates a sublimation agent model that predicts the physical property values ​​of the sublimation agent from the descriptors of the sublimation agent, based on the reference sublimation agent database 201 acquired in step S1 (step S2).

[0061] 4 from the database storage device 200 (step S3). Next, the solvent model generation unit 404 generates a solvent model that predicts the physical property values ​​of the solvent from the solvent descriptors based on the reference solvent database 202 obtained in step S3 (step S4).

[0062] 5 from the database storage device 200 (step S5). Next, the sublimation agent specifying unit 406 specifies the sublimation agent used in the sublimation drying, its descriptors, and its physical property values ​​based on the result database 203 obtained in step S5 (step S6). Next, the sublimation agent specifying unit 406 generates the processing sublimation agent database 204 in FIG. 7 based on the sublimation agent, descriptors, and physical property values ​​specified in step S6 (step S7).

[0063] Furthermore, the solvent identification unit 407 identifies the solvent used in the sublimation drying, the descriptors, and the physical property values ​​of the solvent based on the result database 203 acquired in step S5 (step S8). Subsequently, the solvent identification unit 407 generates the processing solvent database 205 shown in FIG. 8 based on the solvent, descriptors, and physical property values ​​identified in step S8 (step S9). Steps S6 and S7 and steps S8 and S9 may be executed first or simultaneously. Furthermore, steps S1 and S2, steps S3 and S4, and steps S5 to S9 may be executed first or simultaneously.

[0064] Next, the sublimation agent complementing unit 408 predicts the physical property values ​​of each sublimation agent from the descriptor of the sublimation agent in the processing sublimation agent database 204 generated in step S7 using the sublimation agent model generated in step S2 (step S10). Subsequently, the sublimation agent complementing unit 408 complements missing physical property values ​​in the processing sublimation agent database 204 using the physical property values ​​of the sublimation agent predicted in step S10 (step S11). Thereafter, the sublimation agent complementing unit 408 generates the complemented sublimation agent database 206 shown in FIG. 9 in which the physical property values ​​of the sublimation agents have been complemented (step S12).

[0065] Furthermore, the solvent complementing unit 409 uses the solvent model generated in step S4 to predict the physical property values ​​of each solvent from the descriptors of the solvent in the processing solvent database 205 generated in step S9 (step S13). Subsequently, the solvent complementing unit 409 complements missing physical property values ​​in the processing solvent database 205 using the solvent property values ​​predicted in step S13 (step S14). Thereafter, the solvent complementing unit 409 generates the complementary solvent database 207 shown in FIG. 10 in which the physical property values ​​of the solvents have been complemented (step S15). Steps S10 to S12 and steps S13 to S15 may be performed one after the other, or may be performed simultaneously.

[0066] Next, the database generation unit 410 generates the integrated database 208 shown in Fig. 11 by integrating the complementary sublimation agent database 206 generated in step S12 and the complementary solvent database 207 generated in step S13 (step S16). Subsequently, the treatment liquid calculation unit 411 calculates the descriptors and physical property values ​​of the treatment liquid obtained by dissolving a sublimation agent in a solvent based on the integrated database 208 generated in step S16 (step S17). Thereafter, the treatment liquid calculation unit 411 generates the training database 209 shown in Fig. 12 using the descriptors and physical property values ​​of the treatment liquid calculated in step S17 (step S18).

[0067] Next, the optimization unit 413 optimizes the explanatory variables in the training database 209 generated in step S18 (step S19). The explanatory variables include the descriptors and physical property values ​​of each of the sublimation agent, solvent, and treatment liquid, as well as the substrate treatment conditions. Next, the result model generation unit 412 generates a result model based on the training database 209 generated in step S18 and whose explanatory variables have been optimized in step S19 (step S20). This completes the training process. Instead of or in addition to optimizing the explanatory variables in step S19, the optimization unit 413 may cause the result model generation unit 412 to perform re-training using the optimized explanatory variables after step S20.

[0068] 7. Recipe Determination Processing Fig. 17 is a flowchart showing an example of the flow of the recipe determination processing. The recipe determination processing is processing that is executed by CPU 310 of information processing device 300 (Fig. 2) as CPU 310 executes a recipe determination program. The recipe determination processing of Fig. 17 will be described below with reference to information processing device 300 of Fig. 13 .

[0069] First, the model acquisition unit 301 acquires the result model generated in step S20 of the learning process (step S31). Furthermore, the sublimation agent determination unit 302 determines the sublimation agent to be used for sublimation drying of the substrate W based on the user's designation (step S32). The solvent determination unit 303 determines the solvent to be used for sublimation drying of the substrate based on the user's designation (step S33). Steps S31 to S34 may be executed either first or simultaneously.

[0070] Next, the condition selection unit 304 selects processing conditions for performing a series of processes on the substrate W from among the plurality of processing conditions (step S34). Subsequently, the result prediction unit 305 predicts the processing result of the substrate W from the sublimation agent determined in step S32, the solvent determined in step S33, and the processing conditions selected in step S34, using the result model acquired in step S31 (step S35). Thereafter, the evaluation unit 306 evaluates the score indicating the quality of the processing result of the substrate W predicted in step S35 (step S36).

[0071] Next, the evaluation unit 306 determines whether the processing conditions have been searched for a predetermined number of times (step S37). If the processing conditions have not been searched for the predetermined number of times, the process returns to step S34. In this case, the processing conditions are reselected by the condition selection unit 304 in step S34. As described above, the processing conditions are selected using Bayesian optimization, steepest descent, a genetic algorithm, or the like, so that the optimal processing conditions can be selected in a relatively short time. Thereafter, steps S35 to S37 are executed sequentially.

[0072] When the processing conditions have been searched for a predetermined number of times in step S37, appropriate processing conditions are identified based on the scores evaluated in step S36 (step S38). In step S38, the processing conditions that resulted in the processing result with the highest score may be identified as appropriate processing conditions. Alternatively, the processing conditions that resulted in the processing result with a score equal to or greater than a threshold may be identified as appropriate processing conditions.

[0073] Next, the presentation unit 307 presents the processing conditions identified in step S38 (step S39), and the recipe determination process ends. In step S39, the identified processing conditions may be displayed on the display device 360 ​​of FIG. 2. Furthermore, if multiple processing conditions are identified in step S38, the processing conditions may be displayed on the display device 360 ​​in descending order of the scores of the processing results. In this case, the user can easily recognize the processing conditions that are likely to be more appropriate.

[0074] 8. Substrate Processing Figure 18 is a process diagram illustrating an example of substrate processing by the substrate processing apparatus 100. The substrate processing in Figure 18 will be described below with reference to the substrate processing apparatus 100 in Figure 14. When a substrate W is processed by the substrate processing apparatus 100, the substrate W is loaded into the chamber 4 and held by the spin chuck 10. Thereafter, the guard lifting unit 27 raises at least one guard 24 from the lower position to the upper position. The surface of the substrate W is the surface on which devices such as transistors and capacitors are formed, and has a pattern formed thereon.

[0075] First, the spin motor 14 is driven to start rotating the substrate W (step S41). Next, a chemical supplying step is performed in which a chemical solution is supplied to the upper surface of the substrate W to form a chemical solution film covering the entire upper surface of the substrate W (step S42). Specifically, the chemical solution nozzle 31 is moved from the standby position to the processing position, discharges the chemical solution for a predetermined period of time, and then returns to the standby position. The chemical solution discharged from the chemical solution nozzle 31 collides with the upper surface of the rotating substrate W, and then flows outward along the upper surface of the substrate W due to centrifugal force.

[0076] Next, a rinse liquid supply step is performed in which pure water, an example of a rinse liquid, is supplied to the upper surface of the substrate W to rinse away the chemical liquid on the substrate W (step S43). Specifically, the rinse liquid nozzle 35 moves from the standby position to the processing position, discharges the rinse liquid for a predetermined period of time, and then returns to the standby position. The pure water discharged from the rinse liquid nozzle 35 collides with the upper surface of the rotating substrate W, and then flows outward along the upper surface of the substrate W due to centrifugal force.

[0077] Next, a substitute liquid supply step is performed in which a substitute liquid that is miscible with both the rinse liquid and the processing liquid is supplied to the upper surface of the substrate W to replace the pure water on the substrate W with the substitute liquid (step S44). Specifically, the substitute liquid nozzle 43 moves from the standby position to the processing position, discharges the substitute liquid for a predetermined period of time, and then returns to the standby position. The substitute liquid discharged from the substitute liquid nozzle 43 collides with the upper surface of the substrate W and then flows outward along the upper surface of the substrate W due to centrifugal force. The pure water on the substrate W is replaced with the substitute liquid discharged from the substitute liquid nozzle 43. As a result, a liquid film of the substitute liquid is formed that covers the entire upper surface of the substrate W.

[0078] Next, a processing liquid supply step is performed in which the processing liquid is supplied to the upper surface of the substrate W to form a liquid film of the processing liquid on the substrate W (step S45). Specifically, the nozzle moving unit 42 moves the processing liquid nozzle 39 from the standby position to the processing position. Thereafter, the processing liquid valve 41 is opened, and the processing liquid nozzle 39 starts to discharge the processing liquid. Before the discharge of the processing liquid starts, the guard lifting unit 27 may vertically move at least one guard 24 to switch the guard 24 that receives the liquid discharged from the substrate W. When a predetermined time has elapsed since the processing liquid valve 41 was opened, the processing liquid valve 41 is closed, and the discharge of the processing liquid is stopped. Thereafter, the nozzle moving unit 42 moves the processing liquid nozzle 39 to the standby position.

[0079] The processing liquid discharged from the processing liquid nozzle 39 collides with the upper surface of the substrate W rotating at the liquid supply speed, and then flows outward along the upper surface of the substrate W due to centrifugal force. The substitute liquid on the substrate W is replaced with the processing liquid discharged from the processing liquid nozzle 39. This forms a liquid film of the processing liquid that covers the entire upper surface of the substrate W. While the processing liquid nozzle 39 is discharging the processing liquid, the nozzle moving unit 42 moves the landing position of the processing liquid on the upper surface of the substrate W so that the landing position passes through the center and the outer periphery.

[0080] Next, a film thickness reduction step is performed (step S46) in which a portion of the processing liquid on the substrate W is removed to reduce the film thickness (thickness of the liquid film) of the processing liquid on the substrate W while maintaining a state in which the entire upper surface of the substrate W is covered with a liquid film of the processing liquid. Specifically, the spin motor 14 rotates the substrate W with the blocking member 51 in the lower position. The processing liquid on the substrate W is discharged outward from the substrate W by centrifugal force even after the discharge of the processing liquid has stopped. Therefore, the thickness of the liquid film of the processing liquid on the substrate W decreases. When a certain amount of the processing liquid on the substrate W has been discharged, the amount of processing liquid discharged from the substrate W per unit time decreases to zero or approximately zero.

[0081] Next, a solidification forming step is performed in which the processing liquid on the substrate W is solidified to form a solidification containing a sublimation agent on the substrate W (step S47). Specifically, with the blocking member 51 in the lower position, the spin motor 14 rotates the substrate W. Furthermore, the upper gas valve 57 is opened to start discharging nitrogen gas from the center nozzle 55. In the solidification forming step, evaporation of the processing liquid is promoted, and a portion of the processing liquid on the substrate W evaporates. Therefore, the concentration of the sublimation agent gradually increases, and the film thickness of the processing liquid gradually decreases.

[0082] Next, a sublimation step is performed in which the solidified material on the substrate W is sublimated and removed from the upper surface of the substrate W (step S48). Specifically, with the blocking member 51 in the lower position, the spin motor 14 rotates the substrate W. Furthermore, the upper gas valve 57 is opened to start discharging nitrogen gas from the center nozzle 55. When the sublimation step is completed, the spin motor 14 stops, and the rotation of the substrate W is stopped (step S49). This completes the substrate processing.

[0083] 9. Effects In the learning device 400 according to this embodiment, the result acquisition unit 405, the sublimator identification unit 406, and the solvent identification unit 407 respectively acquire the result database 203, the processing sublimator database 204, and the processing solvent database 205. That is, data indicating the relationship between the processing conditions and at least one of the descriptors of the sublimator and the solvent, and the processing results of the substrate W is acquired.

[0084] Here, the physical property values ​​of the sublimation agent or solvent used in the sublimation drying of the substrate W are often not investigated. Therefore, the physical property value of the sublimation agent or solvent may be missing from the data. Even in such cases, the sublimation agent complementing unit 408 or the solvent complementing unit 409 predicts the physical property value from the descriptor of at least one of the sublimation agent and the solvent in the data, and complements the missing physical property value.

[0085] According to this configuration, the result model generation unit 412 can perform machine learning on the data with the supplemented physical property values ​​to generate a result model that predicts the processing results of the substrate W. This makes it possible to search for optimal processing conditions based on the processing results of the substrate W. Therefore, even if there is a deficiency in the physical property values ​​of the sublimation agent or solvent, it is possible to generate a result model for searching for optimal processing conditions for the substrate W.

[0086] Vapor pressure, melting point, heat of solution, or Hansen solubility parameter are often missing as physical property values ​​of a sublimation agent used in the sublimation drying of a substrate W. Therefore, the physical property values ​​of the sublimation agent supplemented by the sublimation agent supplementing unit 408 include at least one of vapor pressure, melting point, heat of solution, and Hansen solubility parameter. This makes it possible to easily generate a result model for searching for optimal processing conditions for a substrate W.

[0087] The vapor pressure, Hansen solubility parameter, viscosity, or surface tension is often missing as a physical property value of a solvent used in sublimation drying of a substrate W. Therefore, the physical property values ​​of the solvent supplemented by the solvent supplementing unit 409 include at least one of the vapor pressure, Hansen solubility parameter, viscosity, and surface tension. This makes it possible to easily generate a result model for searching for optimal processing conditions for the substrate W.

[0088] The processing liquid calculation unit 411 calculates the descriptors and physical property values ​​of the processing liquid based on the descriptors and physical property values ​​of the sublimation agent and the descriptors and physical property values ​​of the solvent. The result model predicts the processing results of the substrate W based further on the descriptors and physical property values ​​of the processing liquid calculated by the processing liquid calculation unit 411. In this case, the result model for searching for optimal processing conditions for the substrate W can be more appropriately generated based on the descriptors and physical property values ​​of the processing liquid.

[0089] The solubility of the sublimation agent in the solvent is often missing from the physical property values ​​of the processing liquid used for sublimation drying of the substrate W. Therefore, the physical property values ​​of the processing liquid include the solubility of the sublimation agent in the solvent. This makes it possible to easily generate a result model for searching for optimal processing conditions for the substrate W.

[0090] The result model generation unit 412 may perform machine learning on the processing results of the substrate W using at least some of the processing conditions, the descriptors and physical property values ​​of the sublimator, and the descriptors and physical property values ​​of the solvent as explanatory variables. In this case, a result model that predicts the processing results of the substrate W can be easily generated by machine learning. In this example, the explanatory variables are optimized by the optimization unit 413 based on the importance of the explanatory variables. The result model generation unit 412 performs machine learning on the explanatory variables optimized by the optimization unit 413. Therefore, the result model can be efficiently generated.

[0091] The sublimation agent model generation unit 402 generates a sublimation agent model for predicting physical property values ​​by machine learning the relationship between the descriptors and physical property values ​​of the reference sublimation agent in the reference sublimation agent database 201. The sublimation agent complementation unit 408 predicts physical property values ​​from the descriptors of the sublimation agent using the sublimation agent model generated by the sublimation agent model generation unit 402, and complements missing physical property values ​​of the sublimation agent in the processing sublimation agent database 204 acquired by the sublimation agent identification unit 406 using the predicted physical property values. In this case, the physical property values ​​of the sublimation agent used for sublimation drying of the substrate W can be easily predicted based on the relationship between the descriptors and physical property values ​​of the reference sublimation agent, and the missing physical property values ​​can be complemented.

[0092] The solvent model generation unit 404 generates a solvent model for predicting physical property values ​​by machine learning the relationship between the descriptors and physical property values ​​of the reference solvent in the reference solvent database 202. The solvent complementation unit 409 predicts physical property values ​​from the solvent descriptors using the solvent model generated by the solvent model generation unit 404, and uses the predicted physical property values ​​to complement missing physical property values ​​of the solvent in the processing solvent database 205 acquired by the solvent identification unit 407. In this case, the physical property values ​​of the solvent used for sublimation drying of the substrate W can be easily predicted based on the relationship between the descriptors and physical property values ​​of the reference solvent, and the missing physical property values ​​can be complemented.

[0093] In the information processing device 300 according to this embodiment, the processing result of the substrate W is predicted from the sublimation agent, solvent, and processing conditions used in the sublimation drying of the substrate W, based on the result model generated by the learning device 400. Therefore, it is possible to search for optimal processing conditions for the substrate W based on the predicted processing result of the substrate W. Here, the condition selection unit 304 selects processing conditions using Bayesian optimization, steepest descent method, or genetic algorithm. In this case, it is possible to efficiently search for optimal processing conditions for the substrate W.

[0094] The evaluation unit 306 evaluates the quality of the processing result of the substrate W predicted by the result prediction unit 305. The presentation unit 307 presents the processing conditions corresponding to the processing result evaluated as good by the evaluation unit 306. In this case, the user can easily recognize the processing conditions under which good processing results were obtained. This makes it possible to easily search for the optimal processing conditions for the substrate W.

[0095] According to the substrate processing system 500 of this embodiment, the substrate W can be sublimated and dried under optimal processing conditions in the substrate processing apparatus 100 based on the processing result of the substrate W predicted by the information processing apparatus 300.

[0096] 10. Other Embodiments (1) In the above embodiment, the learning device 400 includes the treatment liquid calculation unit 411 and the optimization unit 413, but the embodiment is not limited to this. If a resultant model is generated without using molecular descriptors and physical property values ​​of the treatment liquid, the learning device 400 does not need to include the treatment liquid calculation unit 411. Furthermore, if optimization of explanatory variables in machine learning is not performed, the learning device 400 does not need to include the optimization unit 413.

[0097] (2) In the above embodiment, the sublimation agent identification unit 406, the solvent identification unit 407, the sublimation agent complement unit 408, the solvent complement unit 409, the database generation unit 410, and the treatment liquid calculation unit 411 generate databases as appropriate, but the embodiment is not limited to this. The sublimation agent identification unit 406, the solvent identification unit 407, the sublimation agent complement unit 408, the solvent complement unit 409, the database generation unit 410, or the treatment liquid calculation unit 411 may not generate a database. In this case, the integrated database 208 is not generated, and therefore the learning device 400 does not include the database generation unit 410.

[0098] (3) In the above embodiment, the information processing device 300 includes the evaluation unit 306 and the presentation unit 307, but the embodiment is not limited to this. In a case where a user determines whether the processing result of the substrate W predicted by the result prediction unit 305 is good or bad, the information processing device 300 does not need to include the evaluation unit 306 and the presentation unit 307.

[0099] (4) In the above embodiment, the condition selection unit 304 selects processing conditions using Bayesian optimization, steepest descent, or a genetic algorithm, but the embodiment is not limited to this. If the information processing device 300 has sufficient processing speed, the condition selection unit 304 may randomly select processing conditions.

[0100] 11. Correspondence between each element of the claims and each part of the embodiment Examples of correspondence between each element of the claims and each element of the embodiment are described below, but the present invention is not limited to the following examples. Various other elements having the configuration or function described in the claims can also be used as each element of the claims.

[0101] In the above-described embodiments, the result acquisition unit 405, the sublimation agent identification unit 406, and the solvent identification unit 407 are examples of data acquisition units, the sublimation agent complementation unit 408 and the solvent complementation unit 409 are examples of data complementation units, and the result model generation unit 412 is an example of a result model generation unit. The learning device 400 is an example of a learning device, the treatment liquid calculation unit 411 is an example of a treatment liquid calculation unit, the optimization unit 413 is an example of an optimization unit, the sublimation agent model generation unit 402 is an example of a sublimation agent model generation unit, and the sublimation agent complementation unit 408 is an example of a sublimation agent complementation unit.

[0102] The solvent model generation unit 404 is an example of a solvent model generation unit, the solvent complementation unit 409 is an example of a solvent complementation unit, the information processing device 300 is an example of an information processing device, the model acquisition unit 301 is an example of a model acquisition unit, the sublimation agent determination unit 302 is an example of a sublimation agent determination unit, the solvent determination unit 303 is an example of a solvent determination unit, the condition selection unit 304 is an example of a condition selection unit, the result prediction unit 305 is an example of a result prediction unit, the evaluation unit 306 is an example of an evaluation unit, the presentation unit 307 is an example of a presentation unit, and the substrate processing system 500 is an example of a substrate processing system.

[0103] 12. Summary of Embodiments (Item 1) A learning device according to item 1 includes: a data acquisition unit that acquires data indicating the relationship between processing conditions for a process of sublimating and drying a substrate using a processing liquid containing a sublimator and a solvent, molecular descriptors of at least one of the sublimator and the solvent, and a processing result of the substrate; a data complementation unit that complements missing physical property values ​​of at least one of the sublimator and the solvent in the data acquired by the data acquisition unit using a prediction algorithm that predicts physical property values ​​from the molecular descriptor; and a result model generation unit that generates a result model that predicts the processing result of the substrate by machine learning the data whose physical property values ​​have been complemented by the data complementation unit.

[0104] In this learning device, the data acquisition unit acquires data indicating the relationship between the processing conditions, molecular descriptors of at least one of the sublimation agent and the solvent, and the processing results of the substrate. Here, the physical property values ​​of the sublimation agent or solvent used in the sublimation drying of the substrate are often not investigated. Therefore, the data may lack physical property values ​​for the sublimation agent or the solvent. Even in such cases, the data interpolation unit predicts the physical property values ​​from the molecular descriptors of at least one of the sublimation agent and the solvent in the data, thereby interpolating the missing physical property values.

[0105] According to this configuration, the result model generation unit can generate a result model that predicts the processing results of the substrate by machine learning the data with the supplemented physical property values. This makes it possible to search for optimal processing conditions based on the processing results of the substrate. Therefore, even if the physical property values ​​of the sublimation agent or solvent are missing, it is possible to generate a result model for searching for optimal processing conditions for the substrate.

[0106] (Item 2) In the learning device described in item 1, the physical property values ​​of the sublimation agent complemented by the data complementing unit may include at least one of vapor pressure, melting point, heat of solution, and Hansen solubility parameter.

[0107] In many cases, the physical properties of the sublimation agent used in the sublimation drying of the substrate, such as vapor pressure, melting point, heat of fusion, or Hansen solubility parameter, are missing. Even in such cases, the above configuration makes it possible to easily generate a result model for searching for optimal processing conditions for the substrate.

[0108] (Item 3) In the learning device described in item 1 or 2, the physical property values ​​of the solvent supplemented by the data supplementation unit may include at least one of vapor pressure, Hansen solubility parameter, viscosity, and surface tension.

[0109] In many cases, the physical properties of the solvent used in the sublimation drying of the substrate, such as vapor pressure, Hansen solubility parameter, viscosity, or surface tension, are missing. Even in such cases, the above configuration makes it possible to easily generate a result model for searching for optimal processing conditions for the substrate.

[0110] (4) The learning device described in any one of paragraphs 1 to 3 further includes a processing liquid calculation unit that calculates molecular descriptors and physical property values ​​of the processing liquid based on molecular descriptors and physical property values ​​of the sublimation agent and molecular descriptors and physical property values ​​of the solvent, and the result model may predict the processing results of the substrate further based on the molecular descriptors and physical property values ​​of the processing liquid calculated by the processing liquid calculation unit.

[0111] In this case, a result model for searching for optimal processing conditions for a substrate can be more appropriately generated based on the molecular descriptors and physical property values ​​of the processing liquid.

[0112] (Item 5) In the learning device described in Item 4, the physical property values ​​of the treatment liquid may include the solubility of the sublimation agent in the solvent.

[0113] The solubility of the sublimation agent in the solvent is often missing as a physical property of the processing liquid used for the sublimation drying of the substrate. Even in such cases, the above-described configuration makes it possible to easily generate a result model for searching for the optimal processing conditions for the substrate.

[0114] (6) In the learning device described in any one of paragraphs 1 to 5, the result model generation unit may perform machine learning using processing conditions, molecular descriptors and physical property values ​​of the sublimation agent, and at least some of the molecular descriptors and physical property values ​​of the solvent as explanatory variables, and the processing results of the substrate as objective variables.

[0115] In this case, a result model that predicts the processing results of the substrate can be easily generated by machine learning.

[0116] (Clause 7) The learning device described in clause 6 may further include an optimization unit that optimizes the explanatory variables based on the importance of the explanatory variables, and the result model generation unit may perform machine learning on the explanatory variables optimized by the optimization unit.

[0117] In this case, the resulting model can be generated efficiently.

[0118] (Item 8) The learning device described in any one of items 1 to 7 may further include a sublimation model generation unit that generates a sublimation model that predicts physical property values ​​by machine learning the relationship between molecular descriptors and physical property values ​​of a reference sublimation agent, and the data complementation unit may include a sublimation agent complementation unit that predicts physical property values ​​from the molecular descriptors of the sublimation agent using the sublimation agent model generated by the sublimation model generation unit, and complements missing physical property values ​​of the sublimation agent in the data acquired by the data acquisition unit using the predicted physical property values.

[0119] In this case, the physical property values ​​of the sublimation agent used for the sublimation drying of the substrate can be easily predicted based on the relationship between the molecular descriptors and the physical property values ​​of the reference sublimation agent, and the missing physical property values ​​can be compensated for.

[0120] (Item 9) The learning device described in any one of items 1 to 8 may further include a solvent model generation unit that generates a solvent model that predicts physical property values ​​by machine learning the relationship between molecular descriptors and physical property values ​​of a reference solvent, and the data complementation unit may include a solvent complementation unit that predicts physical property values ​​from the molecular descriptors of the solvent using the solvent model generated by the solvent model generation unit and complements missing physical property values ​​of the solvent in the data acquired by the data acquisition unit using the predicted physical property values.

[0121] In this case, the physical property values ​​of the solvent used for sublimation drying of the substrate can be easily predicted based on the relationship between the molecular descriptors and the physical property values ​​of the reference solvent, and missing physical property values ​​can be compensated for.

[0122] (Clause 10) The information processing device according to clause 10 is an information processing device that uses a result model generated by the learning device described in any one of clauses 1 to 9, and includes: a model acquisition unit that acquires the result model; a sublimation agent determination unit that determines a sublimation agent to be used for sublimation drying of a substrate; a solvent determination unit that determines a solvent to be used for sublimation drying of a substrate; a condition selection unit that selects processing conditions; and a result prediction unit that uses the result model acquired by the model acquisition unit to predict a processing result of a substrate from the sublimation agent determined by the sublimation agent determination unit, the solvent determined by the solvent determination unit, and the processing conditions selected by the condition selection unit.

[0123] In this information processing device, the processing results of a substrate are predicted from the sublimation agent, solvent, and processing conditions used in the sublimation drying of the substrate based on the result model generated by the learning device. Therefore, it is possible to search for optimal processing conditions for the substrate based on the predicted processing results of the substrate.

[0124] (Item 11) In the information processing device described in item 10, the condition selection unit may select the processing conditions using Bayesian optimization, a steepest descent method, or a genetic algorithm.

[0125] In this case, the optimum processing conditions for the substrate can be efficiently searched for.

[0126] (Clause 12) The information processing device described in clause 10 or 11 may further include an evaluation unit that evaluates the quality of the substrate processing result predicted by the result prediction unit, and a presentation unit that presents processing conditions corresponding to the processing result evaluated as good by the evaluation unit.

[0127] In this case, the user can easily recognize the processing conditions under which good processing results were obtained, thereby easily searching for the optimum processing conditions for the substrate.

[0128] (13th Item) A substrate processing system according to a 13th item includes the information processing apparatus according to any one of the 10th to 12th items.

[0129] In this substrate processing system, the substrate can be sublimated and dried under optimal processing conditions based on the substrate processing results predicted by the information processing device.

[0130] (Item 14) A learning method according to item 14 includes: acquiring data showing the relationship between processing conditions in a process of sublimating and drying a substrate using a processing liquid containing a sublimator and a solvent, molecular descriptors of at least one of the sublimator and the solvent, and a processing result of the substrate; complementing missing physical property values ​​of at least one of the sublimator and the solvent in the acquired data using a prediction algorithm that predicts physical property values ​​from the molecular descriptors; and generating a result model that predicts the processing result of the substrate by machine learning the data with the complemented physical property values, and is executed by a processor.

[0131] According to this learning method, a processor performs machine learning on data with supplemented physical property values, thereby generating a result model that predicts the processing results of a substrate. This makes it possible to search for optimal processing conditions based on the processing results of the substrate. Therefore, even if the physical property values ​​of the sublimation agent or solvent are missing, it is possible to generate a result model for searching for optimal processing conditions for a substrate.

[0132] (Item 15) A recipe determination method according to item 15 is a recipe determination method that uses a result model generated by the learning method described in item 14, and includes the steps of: acquiring the result model; determining a sublimation agent to be used for sublimation drying of a substrate; determining a solvent to be used for sublimation drying of a substrate; selecting processing conditions; and predicting a processing result of a substrate from the determined sublimation agent, the determined solvent, and the selected processing conditions using the acquired result model, and is executed by a processor.

[0133] According to this recipe determination method, the processing result of a substrate is predicted from the sublimation agent, solvent, and processing conditions used in the sublimation drying of the substrate based on the result model generated by the above-described learning method. Therefore, it is possible to search for optimal processing conditions for the substrate based on the predicted processing result of the substrate.

[0134] (Item 16) A learning program according to Item 16 causes a processor to execute the following processes: acquiring data showing the relationship between processing conditions in a process of sublimating and drying a substrate using a processing liquid containing a sublimator and a solvent, molecular descriptors of at least one of the sublimator and the solvent, and the processing results of the substrate; complementing missing physical property values ​​of at least one of the sublimator and the solvent in the acquired data using a prediction algorithm that predicts physical property values ​​from the molecular descriptors; and generating a result model that predicts the processing results of the substrate by machine learning the data with the complemented physical property values.

[0135] According to this learning program, a processor performs machine learning on data with supplemented physical property values ​​to generate a result model that predicts the processing results of a substrate. This makes it possible to search for optimal processing conditions based on the processing results of the substrate. Therefore, even if the physical property values ​​of the sublimation agent or solvent are missing, it is possible to generate a result model for searching for optimal processing conditions for the substrate.

Claims

1. A learning device comprising: a data acquisition unit that acquires data showing the relationship between processing conditions in a process of sublimating and drying a substrate using a processing liquid containing a sublimation agent and a solvent, molecular descriptors of at least one of the sublimation agent and the solvent, and a processing result of the substrate; a data complementation unit that complements missing physical property values ​​of at least one of the sublimation agent and the solvent in the data acquired by the data acquisition unit using a prediction algorithm that predicts physical property values ​​from the molecular descriptor; and a result model generation unit that generates a result model that predicts the processing result of the substrate by machine learning the data whose physical property values ​​have been complemented by the data complementation unit.

2. The learning device according to claim 1, wherein the physical property values ​​of the sublimation agent supplemented by the data supplementing section include at least one of vapor pressure, melting point, heat of solution, and Hansen solubility parameter.

3. A learning device according to claim 1 or 2, wherein the physical property values ​​of the solvent supplemented by the data supplementing section include at least one of vapor pressure, Hansen solubility parameter, viscosity, and surface tension.

4. A learning device as described in any one of claims 1 to 3, further comprising a processing liquid calculation unit that calculates molecular descriptors and physical property values ​​of the processing liquid based on molecular descriptors and physical property values ​​of the sublimation agent and molecular descriptors and physical property values ​​of the solvent, wherein the result model predicts the processing results of the substrate based further on the molecular descriptors and physical property values ​​of the processing liquid calculated by the processing liquid calculation unit.

5. The learning device according to claim 4, wherein the physical property values ​​of the processing liquid include the solubility of the sublimation agent in the solvent.

6. A learning device described in any one of claims 1 to 5, wherein the result model generation unit performs machine learning using processing conditions, molecular descriptors and physical property values ​​of the sublimation agent, and at least a portion of the molecular descriptors and physical property values ​​of the solvent as explanatory variables, and the processing results of the substrate as objective variables.

7. The learning device according to claim 6, further comprising an optimization unit that optimizes the explanatory variables based on the importance of the explanatory variables, wherein the result model generation unit performs machine learning on the explanatory variables optimized by the optimization unit.

8. A learning device as described in any one of claims 1 to 7, further comprising a sublimation model generation unit that generates a sublimation model that predicts physical property values ​​by machine learning the relationship between molecular descriptors and physical property values ​​of a reference sublimation agent, wherein the data complementation unit includes a sublimation agent complementation unit that predicts physical property values ​​from the molecular descriptors of the sublimation agent using the sublimation agent model generated by the sublimation agent model generation unit and complements missing physical property values ​​of the sublimation agent in the data acquired by the data acquisition unit using the predicted physical property values.

9. A learning device as described in any one of claims 1 to 8, further comprising a solvent model generation unit that generates a solvent model that predicts physical property values ​​by machine learning the relationship between molecular descriptors and physical property values ​​of a reference solvent, wherein the data complementation unit includes a solvent complementation unit that predicts physical property values ​​from the molecular descriptors of the solvent using the solvent model generated by the solvent model generation unit and complements missing physical property values ​​of the solvent in the data acquired by the data acquisition unit using the predicted physical property values.

10. An information processing device that uses a result model generated by a learning device according to any one of claims 1 to 9, comprising: a model acquisition unit that acquires the result model; a sublimation agent determination unit that determines a sublimation agent to be used for sublimation drying of a substrate; a solvent determination unit that determines a solvent to be used for sublimation drying of the substrate; a condition selection unit that selects processing conditions; and a result prediction unit that uses the result model acquired by the model acquisition unit to predict processing results of a substrate from the sublimation agent determined by the sublimation agent determination unit, the solvent determined by the solvent determination unit, and the processing conditions selected by the condition selection unit.

11. The information processing device according to claim 10, wherein the condition selection section selects the processing conditions using Bayesian optimization, a steepest descent method or a genetic algorithm.

12. An information processing device as described in claim 10 or 11, further comprising an evaluation unit that evaluates the quality of the substrate processing result predicted by the result prediction unit, and a presentation unit that presents processing conditions corresponding to the processing result evaluated as good by the evaluation unit.

13. A substrate processing system comprising the information processing device according to any one of claims 10 to 12.

14. A learning method executed by a processor, comprising: acquiring data showing a relationship between processing conditions in a process of sublimating and drying a substrate using a processing liquid containing a sublimation agent and a solvent, molecular descriptors of at least one of the sublimation agent and the solvent, and a processing result of the substrate; complementing missing physical property values ​​of at least one of the sublimation agent and the solvent in the acquired data using a prediction algorithm that predicts physical property values ​​from the molecular descriptors; and generating a result model that predicts the processing result of the substrate by machine learning the data with the complemented physical property values.

15. A recipe determination method using a result model generated by the learning method of claim 14, comprising the steps of: acquiring the result model; determining a sublimation agent to be used for sublimation drying of the substrate; determining a solvent to be used for sublimation drying of the substrate; selecting process conditions; and predicting a substrate process result from the determined sublimation agent, the determined solvent and the selected process conditions using the acquired result model, the recipe determination method being executed by a processor.

16. A learning program that causes a processor to execute the following processes: a process of acquiring data indicating the relationship between processing conditions in a process of sublimating and drying a substrate using a processing liquid containing a sublimation agent and a solvent, and molecular descriptors of at least one of the sublimation agent and the solvent, and the processing results of the substrate; a process of complementing missing physical property values ​​of at least one of the sublimation agent and the solvent in the acquired data using a prediction algorithm that predicts physical property values ​​from the molecular descriptors; and a process of generating a result model that predicts the processing results of the substrate by machine learning using the data with the complemented physical property values.

Images