Design assistance system, design assistance method, and design assistance program

Abstract

The present invention assists the design of manufacturing conditions in a process for manufacturing an SiC single crystal by a sublimation method. Provided is a design assistance system for assisting the design of manufacturing conditions applied in a process for manufacturing an SiC single crystal by a sublimation method, the system comprising: a first storage part for storing a first trained model trained by using first training data including manufacturing conditions at each time that have been input to a simulation device replicating the process, and first item data indicating the state in the furnace at each time that has been simulated by inputting the manufacturing conditions at each time; a prediction part for predicting the first item data indicating the state in the furnace at each time by inputting manufacturing conditions at each time for prediction to the first trained model; and a display part for displaying the predicted first item data.

Description

Design support system, design support method, and design support program

[0001] The present disclosure relates to a design support system, a design support method, and a design support program.

[0002] Sublimation is a known method for producing SiC single crystals. To grow SiC single crystals in a desired growth process using sublimation, it is necessary to appropriately change various production conditions, such as heater output and heater position, over time.

[0003] For this reason, conventionally, when designing manufacturing conditions, simulations have been carried out on the growth process of SiC single crystals.

[0004] Japanese Patent Application Laid-Open No. 2018-169818

[0005] Shunta Harada et al., "Design of high-quality SiC crystal growth conditions using machine learning," Materia, Vol. 59, No. 3 (2020), pp. 145-152

[0006] On the other hand, simulating the growth process of a SiC single crystal requires enormous calculation costs, making it difficult to smoothly design manufacturing conditions.

[0007] The present disclosure assists in the design of manufacturing conditions in a process for manufacturing SiC single crystals by sublimation.

[0008] A first aspect of the present disclosure is a design support system that supports the design of manufacturing conditions applied in a process for manufacturing SiC single crystals by sublimation, comprising: a first storage unit that stores a first trained model trained using first training data including manufacturing conditions for each time input into a simulation device that reproduces the process and a first item of data indicating the state inside the furnace at each time simulated by inputting the manufacturing conditions for each time; a prediction unit that predicts the first item of data indicating the state inside the furnace at each time by inputting the manufacturing conditions for each time for prediction into the first trained model; and a display unit that displays the predicted first item of data.

[0009] A second aspect of the present disclosure is a design support system according to the first aspect, wherein the first storage unit stores an Ith trained model trained using Ith training data including the manufacturing conditions at each time and the coordinates of each position in the furnace, and a first item of data indicating the state inside the furnace at each position in the furnace at each time, simulated by inputting the manufacturing conditions at each time; the prediction unit predicts the first item of data indicating the state inside the furnace at each position in the furnace at each time by inputting the manufacturing conditions at each time for prediction and the coordinates of each position in the furnace into the Ith trained model; and the display unit displays the predicted first item of data.

[0010] A third aspect of the present disclosure is a design support system according to the second aspect, wherein the I trained model is generated by updating the model parameters of the I model so as to reduce the calculation result using a loss function capable of calculating the error between a first item of data output from the I model when the manufacturing conditions at each time and the coordinates of each position in the furnace contained in the I training data are input into the I model, and a first item of data corresponding to the manufacturing conditions at each time and the coordinates of each position in the furnace contained in the I training data.

[0011] A fourth aspect of the present disclosure is a design support system according to the second aspect, wherein the I trained model is generated by updating the model parameters of the I model so as to reduce the calculation result in a loss function capable of calculating the error between data obtained by applying an image processing filter to a first item of data output from the I model when the manufacturing conditions at each time and the coordinates of each position in the furnace contained in the I training data are input into the I model, and data obtained by applying an image processing filter to a first item of data corresponding to the manufacturing conditions at each time and the coordinates of each position in the furnace contained in the I training data.

[0012] A fifth aspect of the present disclosure is the design support system according to the second aspect, wherein the I trained model is generated by updating the model parameters of the I model so that the calculation result is smaller in a loss function that can weight and add the error between the first item of data output from the I model and the first item of data corresponding to the manufacturing conditions at each time and the coordinates of each position in the furnace, which is included in the I training data, and the error between the data obtained by applying an image processing filter to the first item of data output from the I model and the data obtained by applying an image processing filter to the first item of data corresponding to the manufacturing conditions at each time and the coordinates of each position in the furnace, which is included in the I training data.

[0013] A sixth aspect of the present disclosure is a design support system according to any one of the second to fifth aspects, wherein the first item of data indicating the state inside the furnace at each time and at each position inside the furnace includes any one of the temperature at each time and at each position inside the furnace, the gas concentration at each time and at each position inside the furnace, the gas flow rate at each time and at each position inside the furnace, and the raw material void fraction at each time and at each position inside the furnace.

[0014] A seventh aspect of the present disclosure is the design support system according to the first aspect, wherein the first item of data indicating the state inside the furnace at each time includes either the amount of raw material sublimated at each time or the amount of SiC single crystal growth at each time.

[0015] An eighth aspect of the present disclosure is a design support system described in any of the second to sixth aspects, comprising a second storage unit that stores a second trained model trained using second training data including a first item of data indicating the state inside the furnace at each time at each position in the furnace, simulated by inputting manufacturing conditions at each time into a simulation device that reproduces the process, and a second item of data correlating with the first item of data; the prediction unit inputs the first item of data indicating the state inside the furnace at each time at each position in the furnace, predicted by inputting the manufacturing conditions at each time for prediction and the coordinates of each position in the furnace into the first trained model, into the second trained model, and predicts the second item of data indicating the state inside the furnace at each time at each position in the furnace; and the display unit further displays the predicted second item of data.

[0016] A ninth aspect of the present disclosure is the design support system according to the eighth aspect, wherein the II trained model is generated by updating the model parameters of the II model so as to reduce the calculation result of a loss function capable of calculating the error between a second item of data output from the II model when a first item of data included in the II training data is input to the II model, and a second item of data included in the II training data that corresponds to the input first item of data.

[0017] A tenth aspect of the present disclosure is the design support system according to the ninth aspect, wherein the combination of the first item of data and the second item of data included in the II learning data includes either a combination of the temperature at each position in the furnace at each time and the raw material void ratio at each position in the furnace at each time, or a combination of the raw material sublimation amount at each time and the SiC single crystal growth amount at each time.

[0018] An eleventh aspect of the present disclosure is the design support system according to the tenth aspect, wherein the amount of raw material sublimation at each time is calculated based on the raw material void fraction at each time at each position in the furnace.

[0019] A twelfth aspect of the present disclosure is a design support system according to the first or second aspect, further comprising a search unit that, when a change is received for a portion of the data of the first item predicted by the prediction unit, searches for manufacturing conditions for each time period used for the prediction in order to predict the data of the first item after the change, and the display unit displays the searched manufacturing conditions for each time period used for the prediction.

[0020] A thirteenth aspect of the present disclosure is a design support system according to the eighth aspect, further comprising a search unit that, when a change is received for a portion of the data of the second item predicted by the prediction unit, searches for manufacturing conditions for each time period used for the prediction in order to predict the data of the second item after the change, and the display unit displays the searched manufacturing conditions for each time period used for the prediction.

[0021] A fourteenth aspect of the present disclosure is the design support system according to the first or second aspect, wherein the trained model I is a recurrent neural network.

[0022] A fifteenth aspect of the present disclosure is the design support system according to the eighth aspect, wherein the II trained model is a recurrent neural network.

[0023] A sixteenth aspect of the present disclosure is the design support system according to the fourteenth or fifteenth aspect, wherein the recurrent neural network is any one of an RNN, an LSTM, a GRU, a Seq2Seq, a Bidirectional RNN, an Attention mechanism-enabled Seq2Seq, or a Transformer.

[0024] A seventeenth aspect of the present disclosure is a design support method, wherein a computer of a design support system that supports the design of manufacturing conditions applied in a process for manufacturing SiC single crystals by sublimation performs the following steps: reading out from a first storage unit a first trained model that has been trained using first training data that includes manufacturing conditions at each time input into a simulation device that reproduces the process and first item of data indicating the state inside the furnace at each time at each position within the furnace, simulated by inputting the manufacturing conditions at each time; predicting the first item of data indicating the state inside the furnace at each time at each position within the furnace by inputting the manufacturing conditions at each time for prediction into the first trained model; and displaying the predicted first item of data.

[0025] An eighteenth aspect of the present disclosure is a design support program, which causes a computer of a design support system that supports the design of manufacturing conditions applied in a process for manufacturing SiC single crystals by sublimation, to perform the following steps: reading from a first storage unit a first trained model that has been trained using first training data that includes manufacturing conditions for each time input into a simulation device that reproduces the process and a first item of data indicating the state inside a furnace at each time, simulated by inputting the manufacturing conditions for each time; predicting the first item of data indicating the state inside a furnace at each time by inputting the manufacturing conditions for each time for prediction into the first trained model; and displaying the predicted first item of data.

[0026] According to the present disclosure, it is possible to assist in the design of manufacturing conditions in a process for manufacturing SiC single crystals by sublimation.

[0027] FIG. 1 is a first diagram showing an example of the system configuration of a design support system. FIG. 2A is a diagram showing an overview of a manufacturing apparatus. FIG. 2B is a diagram for explaining manufacturing conditions and the state inside a furnace in the manufacturing apparatus. FIG. 2C is a diagram showing the relationship between time changes in manufacturing conditions in the manufacturing apparatus and time changes in information indicating the state inside the furnace. FIG. 2D is a diagram showing a specific example of the relationship between time changes in manufacturing conditions in the manufacturing apparatus and time changes in information indicating the state inside the furnace. FIG. 3 is a diagram showing an example of the hardware configuration of a simulation apparatus, a learning apparatus, and a design support apparatus. FIG. 4 is a diagram showing a specific example of processing by the simulation apparatus. FIG. 5 is a diagram showing an example of the functional configuration of the learning apparatus. FIG. 6 is a first diagram showing an example of learning data. FIG. 7 is a second diagram showing an example of learning data. FIG. 8 is a third diagram showing an example of learning data. FIG. 9 is a fourth diagram showing an example of learning data. FIG. 10 is an example of a flowchart showing the flow of learning data generation processing and learning processing. FIG. 11 is a first diagram showing a specific example of learning processing by the learning unit. FIG. 12 is a second diagram showing a specific example of learning processing by the learning unit. FIG. 13 is a third diagram showing a specific example of learning processing by the learning unit. FIG. 14 is a fourth diagram showing a specific example of learning processing by the learning unit. FIG. 15 is a first diagram showing an example of the functional configuration of a design support apparatus. FIG. 16 is a diagram showing variations of prediction processing by the prediction unit. FIG. 17 is a first diagram showing a specific example of prediction processing by the prediction unit. FIG. 18 is a second diagram showing a specific example of prediction processing by the prediction unit. FIG. 19 is a third diagram showing a specific example of prediction processing by the prediction unit. FIG. 20 is a first diagram showing an example of a design support screen. FIG. 21 is a second diagram showing an example of a design support screen. FIG. 22 is a third diagram showing an example of a design support screen. FIG. 23 is a fourth diagram showing an example of a design support screen. FIG. 24 is a first flowchart showing the flow of design support processing. FIG. 25 is a second diagram showing an example of the functional configuration of a design support apparatus. FIG. 26 is a fourth diagram showing an example of a design support screen. FIG. 27 is a second flowchart showing the flow of design support processing. FIG. 28 is a second diagram illustrating an example of the system configuration of the design support system.

[0028] Hereinafter, each embodiment will be described with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configurations are designated by the same reference numerals, and redundant description will be omitted.

[0029] First Embodiment <System Configuration of Design Support System> First, a description will be given of the system configuration of a design support system according to the first embodiment. Fig. 1 is a first diagram showing an example of the system configuration of a design support system.

[0030] As shown in FIG. 1, a design support system 100 according to the first embodiment includes a simulation device 110, a learning device 120, and a design support device .

[0031] Prerequisites for reproducing the target manufacturing apparatus 10 are set in the simulation apparatus 110. In the first embodiment, the target manufacturing apparatus 10 is an apparatus for manufacturing SiC single crystals by sublimation deposition.

[0032] By setting the preconditions, changes over time in various manufacturing conditions are input to the simulation device 110 that replicates the manufacturing device 10. As a result, the simulation device 110 executes a simulation and predicts the state inside the furnace at each time.

[0033] The prerequisites set in the simulation device 110, the manufacturing conditions input to the simulation device 110, and the state inside the furnace output from the simulation device 110 will be described in detail later.

[0034] A learning program is installed in the learning device 120, and by executing the program, the learning device 120 functions as a learning data generation unit 121 and a learning unit 122.

[0035] The learning data generation unit 121 acquires the manufacturing conditions input when the simulation is performed by the simulation device 110. The learning data generation unit 121 acquires information indicating the state inside the furnace output when the simulation is performed by the simulation device 110. The learning data generation unit 121 generates various combinations of learning data, with the acquired manufacturing conditions as input data and the acquired information indicating the state inside the furnace as correct answer data, and stores the data in the learning data storage unit 123 for each precondition.

[0036] The learning unit 122 performs model learning processing using each piece of learning data stored in the learning data storage unit 123, and generates each trained model.

[0037] A design support program is installed in the design support device 130 , and by executing the program, the design support device 130 functions as a display unit 131 and a prediction unit 132 .

[0038] The display unit 131 provides a design support screen to the user 140. In response to providing the design support screen, the display unit 131 acquires the following to the prediction unit 132: preconditions input by the user 140; prediction items, which are items to be predicted among information indicating the state inside the furnace; and manufacturing conditions. The display unit 131 acquires prediction results predicted by the prediction unit 132 in response to the notification of the preconditions, prediction items, and manufacturing conditions, and displays them to the user 140.

[0039] The prediction unit 132 reads out a trained model that is generated by the learning device 120 and stored in a trained model storage unit 133, which is an example of a storage unit, and that corresponds to the prerequisites, prediction items, and manufacturing conditions notified by the display unit 131. The prediction unit 132 executes the read trained model by inputting the manufacturing conditions notified by the display unit 131 into the trained model. The prediction unit 132 notifies the display unit 131 of information indicating the state inside the furnace that corresponds to the prediction items output from the trained model by executing the read trained model.

[0040] As described above, the design support system 100 according to the first embodiment: Generates training data by performing a simulation using a simulation device 110 that replicates a manufacturing device for manufacturing SiC single crystals by sublimation. Generates a trained model (surrogate model) by using the generated training data to learn the relationship between the manufacturing conditions and information indicating the state inside the furnace for each precondition. Uses the generated trained model to predict information indicating the state inside the furnace from the manufacturing conditions.

[0041] As a result, according to the first embodiment, it is possible to predict the growth process of a SiC single crystal at low calculation cost, and to support the smooth design of manufacturing conditions in the process of manufacturing a SiC single crystal by sublimation.

[0042] <Outline of Manufacturing Apparatus> Next, an outline of manufacturing apparatus 10 (an apparatus for manufacturing SiC single crystals by sublimation) reproduced by simulation apparatus 110 will be described. Fig. 2A is a diagram showing an outline of the manufacturing apparatus, showing a state in which a part of the sublimation furnace of manufacturing apparatus 10 is cut away to expose the inside.

[0043] 2A , the sublimation furnace of the manufacturing apparatus 10 is composed of a graphite crucible 210, and a seed crystal 220 is attached to the ceiling. SiC powder 231, which is a raw material, is deposited on the bottom of the graphite crucible 210, and the SiC powder 231 is heated from the outside of the graphite crucible 210 by a heater (not shown), for example, to about 2300° C. As a result, the SiC powder 231 sublimes and rises within the graphite crucible 210. The rising sublimation gas 232 is recrystallized in the seed crystal 220, and a SiC single crystal 233 is produced. Generally, it takes about 100 to 300 hours to produce the SiC single crystal 233.

[0044] Next, a description will be given of the relationship between the manufacturing conditions of manufacturing apparatus 10 and the state inside the sublimation furnace (hereinafter simply referred to as the "state inside the furnace") when manufacturing apparatus 10 manufactures SiC single crystal 233. Fig. 2B is a diagram for explaining the manufacturing conditions in the manufacturing apparatus and the state inside the furnace.

[0045] As shown in FIG. 2B , when manufacturing apparatus 10 manufactures SiC single crystal 233, the manufacturing conditions of manufacturing apparatus 10 include the output of heater 240 that heats SiC powder 231 from outside graphite crucible 210 and the position of heater 240.

[0046] Changing the output of the heater 240 changes the amount of sublimation per unit time of the SiC powder 231. Changing the position of the heater 240 changes the vertical position to which the SiC powder 231 is heated. In the manufacturing apparatus 10, the output and position of the heater 240 are controlled at each time so that the growth process of the SiC single crystal 233 becomes the desired growth process.

[0047] 2B(a) shows the position of the heater 240 at the start of production. When the heater 240 heats the SiC powder 231 for a certain period of time at the position shown in FIG. 2B(a), and some of the SiC powder 231 sublimes, the manufacturing apparatus 10 changes the position of the heater 240.

[0048] 2B(b) shows a state in which, after a certain time has elapsed since the start of production, the position of heater 240 has been lowered from the position at the start of production. FIG. 2B(b) shows a state in which, after a certain time has elapsed, a region has been created in which the raw material porosity has changed due to partial sublimation of SiC powder 231, and the growth amount of SiC single crystal 233 has increased.

[0049] Next, a description will be given of the relationship between the time changes in the manufacturing conditions of manufacturing apparatus 10 and the time changes in the information indicating the state inside the furnace when manufacturing apparatus 10 manufactures SiC single crystal 233. Fig. 2C is a diagram showing the relationship between the time changes in the manufacturing conditions of the manufacturing apparatus and the time changes in the information indicating the state inside the furnace.

[0050] In FIG. 2C, graph 251 schematically shows the change in heater output over time during production of SiC single crystal 233, with the horizontal axis representing time and the vertical axis representing heater output.

[0051] In FIG. 2C, graph 252 schematically shows the change in heater position over time during production of SiC single crystal 233, with the horizontal axis representing time and the vertical axis representing heater position.

[0052] On the other hand, in FIG. 2C, graph 253 schematically shows the change over time in the amount of sublimation of SiC powder 231 during production of SiC single crystal 233, with the horizontal axis representing time and the vertical axis representing the amount of raw material sublimation.

[0053] In FIG. 2C, graph 254 schematically shows the change over time in the growth amount of SiC single crystal 233 during production of SiC single crystal 233, with the horizontal axis representing time and the vertical axis representing the growth amount of SiC single crystal.

[0054] In manufacturing apparatus 10, adjusting the change over time of heater output shown in graph 251 and the change over time of heater position shown in graph 252 controls the change over time of the source material sublimation amount shown in graph 253. As a result, the change over time of the SiC single crystal growth amount shown in graph 254 is controlled.

[0055] Next, a specific example will be used to explain the relationship between the time change of the manufacturing conditions in the manufacturing apparatus 10 and the time change of the information indicating the state inside the furnace. Fig. 2D is a diagram showing a specific example of the relationship between the time change of the manufacturing conditions in the manufacturing apparatus and the time change of the information indicating the state inside the furnace.

[0056] In the example of Figure 2D, the following will be explained: - Changes over time in manufacturing conditions in the manufacturing apparatus 10: changes over time in the position of the heater 240; - Changes over time in information indicating the state inside the furnace: changes over time in the amount of raw material sublimated at each position inside the furnace.

[0057] 2D, (a-1) to (a-4) and (b-1) to (b-4) represent the amount of raw material sublimated in the region indicated by the dotted line 260 at the bottom where the SiC powder 231 is deposited in the graphite crucible 210 that constitutes the sublimation furnace of the manufacturing apparatus 10. In (a-1) to (a-4) and (b-1) to (b-4), red indicates that the SiC powder 231 has completely sublimated, and dark blue indicates that the SiC powder 231 has not sublimated. Light blue indicates that a portion of the SiC powder 231 has sublimated, and the lighter the concentration, the greater the amount that has sublimated.

[0058] In Fig. 2D, time passes from (a-1) to (a-4). Similarly, in Fig. 2D, time passes from (b-1) to (b-4).

[0059] Among these, (a-1) to (a-4) represent the amount of raw material sublimated in the region indicated by the dotted line 260 when the position of the heater 240 is changed over time as shown in the graph 252_1.

[0060] On the other hand, (b-1) to (b-4) show the amount of raw material sublimated in the region indicated by the dotted line 260 when the position of the heater 240 is changed over time as shown in the graph 252_2.

[0061] In this way, the time-dependent change in the information indicating the state inside the furnace also changes depending on how the manufacturing conditions are changed over time. As a result, the growth process of the SiC single crystal also changes. For this reason, in order to achieve a desired growth process of the SiC single crystal, it is necessary to appropriately change the manufacturing conditions over time.

[0062] <Hardware Configuration of Simulation Apparatus, Learning Apparatus, and Design Support Apparatus> Next, we will explain the hardware configurations of the simulation apparatus 110, learning apparatus 120, and design support apparatus 130 that make up the design support system 100. Note that the simulation apparatus 110, learning apparatus 120, and design support apparatus 130 have roughly the same hardware configuration, so they will be explained together here using FIG. 3.

[0063] 3 is a diagram showing an example of the hardware configuration of a simulation device, a learning device, and a design support device. As shown in Fig. 3, the simulation device 110, the learning device 120, and the design support device 130 each include a processor 301, a memory 302, an auxiliary storage device 303, an I / F (Interface) device 304, a communication device 305, and a drive device 306. The hardware components of the simulation device 110, the learning device 120, and the design support device 130 are connected to each other via a bus 307.

[0064] The processor 301 has various computing devices such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), etc. The processor 301 executes various programs (e.g., a simulation program, a learning program, a design support program, etc.) by reading them into the memory 302.

[0065] The memory 302 has a main storage device such as a read-only memory (ROM) or a random access memory (RAM). The processor 301 and the memory 302 form a so-called computer, and the processor 301 executes various programs read onto the memory 302, thereby enabling the computer to realize various functions.

[0066] The auxiliary storage device 303 stores various programs and various data used when the various programs are executed by the processor 301. For example, the training data storage unit 123 and the trained model storage unit 133 are realized in the auxiliary storage device 303.

[0067] The I / F device 304 is a connection device for connecting an operation device 311, which is an example of a user interface device, and a display device 312. The communication device 305 is a communication device for communicating with an external device via a network (not shown).

[0068] The drive device 306 is a device for loading a recording medium 313. The recording medium 313 here includes media that record information optically, electrically, or magnetically, such as a CD-ROM, a flexible disk, a magneto-optical disk, etc. The recording medium 313 may also include semiconductor memory that records information electrically, such as a ROM, a flash memory, etc.

[0069] The various programs to be installed in the auxiliary storage device 303 are installed, for example, by setting the distributed recording medium 313 in the drive device 306 and reading the various programs recorded on the recording medium 313 by the drive device 306. Alternatively, the various programs to be installed in the auxiliary storage device 303 may be installed when downloaded from a network via the communication device 305.

[0070] <Specific Example of Processing by Simulation Apparatus> Next, a description will be given of a specific example of processing by the simulation apparatus 110. Fig. 4 is a diagram showing a specific example of processing by the simulation apparatus.

[0071] As described above, in order to reproduce the target manufacturing apparatus 10, various preconditions are set in the simulation apparatus 110. As shown in Fig. 4, the various preconditions set in the simulation apparatus 110 include: sublimation furnace structure; arrangement of SiC powder as raw material; thermal conductivity, specific heat, and emissivity, which are material property values ​​of each member constituting the sublimation furnace; type of graphite constituting the graphite crucible; type of SiC powder as raw material; and the like.

[0072] Note that the example in Figure 4 shows only a portion of the various preconditions set in the simulation device 110, and it is assumed that many more preconditions other than those in the example in Figure 4 are set in the simulation device 110.

[0073] As described above, when performing a simulation, various manufacturing conditions are input to the simulation device 110. As shown in Fig. 4, the manufacturing conditions input to the simulation device 110 include heater output, heater position, etc.

[0074] The example of FIG. 4 merely shows a part of the manufacturing conditions input to the simulation device 110, and manufacturing conditions other than those shown in the example of FIG. 4 may be input to the simulation device 110.

[0075] As described above, the simulation device 110 predicts information indicating the state inside the furnace through simulation. As shown in FIG. 4 , the information indicating the state inside the furnace predicted by the simulation device 110 includes the temperature distribution inside the furnace, the gas concentration distribution inside the furnace, the gas flow rate inside the furnace, the SiC single crystal growth amount, the raw material void fraction, the raw material sublimation amount, etc. Among these, the temperature distribution inside the furnace, the gas concentration distribution inside the furnace, the gas flow rate inside the furnace, the raw material void fraction, etc. are examples of first item data indicating the state inside the furnace at each position inside the furnace and each time. Furthermore, when deriving second item data correlated with the first item data from the first item data, combinations of the first item data and the second item data include combinations of the temperature distribution inside the furnace and the raw material void fraction, and combinations of the raw material sublimation amount and the SiC single crystal growth amount.

[0076] The raw material porosity refers to the sublimation rate of the SiC raw material at each position in the furnace. The raw material sublimation amount is the total value of the sublimation amounts at each position in the furnace calculated from the sublimation rate at each position. The growth amount of the SiC single crystal is expressed by the thickness or volume of the SiC single crystal.

[0077] <Functional Configuration of Learning Device> Next, a detailed description will be given of the functional configuration of the learning device 120. Fig. 5 is a diagram showing an example of the functional configuration of the learning device.

[0078] As described above, the learning device 120 functions as a learning data generation unit 121 and a learning unit 122. Of these, the learning data generation unit 121 has a prerequisite selection unit 510, an input data selection unit 520, a supervised data selection unit 530, and a combination unit 500. The learning unit 122 has a learning data reading unit 540, a first model 550_1 to an m-th model 550_m, and comparison and modification units 560_1 to 560_m (where m is an integer of 2 or greater).

[0079] The precondition selection unit 510 receives the preconditions input when the simulation is executed by the simulation device 110. When a simulation is executed under different preconditions, each precondition is input.

[0080] The input data selection unit 520 receives the manufacturing conditions input when the simulation device 110 executes the simulation.

[0081] The correct data selection unit 530 receives information indicating the state inside the reactor output during the simulation by the simulation device 110 .

[0082] The combination unit 500 combines the preconditions input by the precondition selection unit 510, the manufacturing conditions input by the input data selection unit 520, and information indicating the state inside the furnace input by the correct data selection unit 530. The combination unit 500 separates the learning data, in which the combined manufacturing conditions and information indicating the state inside the furnace are used as input data and correct data, respectively, for each precondition and stores them in the learning data storage unit 123.

[0083] The learning data reading unit 540 reads out the learning data stored in the learning data storage unit 123. The learning data reading unit 540 inputs the input data included in the read learning data to one of the first model 550_1 to the m-th model 550_m. The learning data reading unit 540 inputs the correct answer data included in the read learning data to one of the comparison and modification units 560_1 to 560_m.

[0084] The first model 550_1 to the m-th model 550_m receive input data from the learning data reading unit 540 and output output data.

[0085] The comparison and modification units 560_1 to 560_m compare the output data output from each of the first model 550_1 to the m-th model 550_m with the correct answer data input by the learning data reading unit 540. The comparison and modification units 560_1 to 560_m update the model parameters of the first model 550_1 to the m-th model 550_m, respectively, according to the comparison results.

[0086] As a result, the learning unit 122 generates the first trained model to the mth trained model.

[0087] The first to m-th trained models are configured, for example, by a recurrent neural network (RNN). However, the algorithms configuring the first to m-th trained models are not limited to RNN. The first to m-th trained models may also be configured by other recurrent neural networks, such as: a long short-term memory (LSTM), a gated recurrent unit (GRU), a Seq2Seq, a bidirectional RNN, a Seq2Seq with an attention mechanism, and a transformer.

[0088] <Specific Examples of Learning Data> (1) First Learning Data Next, a description will be given of a specific example of the first learning data generated by the learning data generating unit 121. Fig. 6 is a first diagram showing an example of learning data.

[0089] In FIG. 6, the input data of the first learning data 610A, the first learning data 610B, ... includes: １ ~t ｎ Heater output W(t １ ) ~ W(t ｎ ), t as the change in heater position over time １ ~t ｎ The heater position H(t １ ) ~ H(t ｎ ), and each time t as a coordinate at a specific cross section in the furnace １ ~t ｎ (x １ , y １ ) ~ (x ｍ , y ｍ ), and

[0090] In FIG. 6, the correct data of the first learning data 610A, the first learning data 610B, ... includes the following: １ Each coordinate (X １ , Y １ ) ~ (X ｋ , Y ｋ ) raw material void ratio S１１ (t １ ) ~ S ｋｋ (t １ ) from time t ｎ Each coordinate (X １ , Y １ ) ~ (X ｋ , Y ｋ ) raw material void ratio S １１ (t ｎ ) ~ S ｋｋ (t ｎ ), and

[0091] 6 is learning data acquired by executing a simulation after "precondition A" is set in the simulation device 110. The first learning data 610B shown in FIG. 6 is learning data acquired by executing a simulation after "precondition B" is set in the simulation device 110.

[0092] (2) Second Learning Data Next, a description will be given of a specific example of the second learning data generated by the learning data generating unit 121. Fig. 7 is a second diagram showing an example of the learning data.

[0093] In FIG. 7, the input data of the second learning data 710A, the second learning data 710B, ... includes: １ ~t ｎ Heater output W(t １ ) ~ W(t ｎ ), t as the change in heater position over time １ ~t ｎ The heater position H(t １ ) ~ H(t ｎ ), and each time t as a coordinate at a specific cross section in the furnace １ ~t ｎ (x １ , y １ ) ~ (x ｍ , y ｍ ), and

[0094] In FIG. 7, the correct data of the second learning data 710A, the second learning data 710B, ... includes the following: １ Each coordinate (X １ , Y １ ) ~ (X ｋ , Y ｋ ) furnace temperature T １１ (t １ ) ~ T ｋｋ (t １ ) from time t ｎ Each coordinate (X １ , Y １ ) ~ (X ｋ , Y ｋ ) furnace temperature T １１ (t ｎ ) ~ T ｋｋ (t ｎ ), and

[0095] 7 is learning data acquired by executing a simulation after "precondition A" is set in the simulation device 110. Second learning data 710B shown in FIG. 7 is learning data acquired by executing a simulation after "precondition B" is set in the simulation device 110.

[0096] (3) Third Learning Data Next, a specific example of the third learning data generated by the learning data generating unit 121 will be described. Fig. 8 is a third diagram showing an example of the learning data.

[0097] In FIG. 8, the input data of the third learning data 810A, the third learning data 810B, ... includes the following: １ Each coordinate (X １ , Y １ ) ~ (X ｋ , Y ｋ ) furnace temperature T １１ (t １ ) ~ T ｋｋ (t １ ) from time t ｎ Each coordinate (X １ , Y １ ) ~ (X ｋ , Yｋ ) furnace temperature T １１ (t ｎ ) ~ T ｋｋ (t ｎ ), and

[0098] In FIG. 8, the correct data of the third learning data 810A, the third learning data 810B, ... includes the following: １ Each coordinate (X １ , Y １ ) ~ (X ｋ , Y ｋ ) raw material void ratio S １１ (t １ ) ~ S ｋｋ (t １ ) from time t ｎ Each coordinate (X １ , Y １ ) ~ (X ｋ , Y ｋ ) raw material void ratio S １１ (t ｎ ) ~ S ｋｋ (t ｎ ), and

[0099] 8 is learning data acquired by executing a simulation after "precondition A" is set in the simulation device 110. The third learning data 810B shown in FIG. 8 is learning data acquired by executing a simulation after "precondition B" is set in the simulation device 110.

[0100] (4) Fourth Learning Data Next, a specific example of the fourth learning data generated by the learning data generating unit 121 will be described. Fig. 9 is a fourth diagram showing an example of the learning data.

[0101] In FIG. 9, the input data of fourth learning data 910A, fourth learning data 910B, . . . stores the change in the amount of raw material sublimation over time.

[0102] The time change of the amount of sublimation of the raw material is: Time change of the raw material porosity, time t １ Each coordinate (X １ , Y １ ) ~ (Xｋ , Y ｋ ) raw material void ratio S １１ (t １ ) ~ S ｋｋ (t １ ) from time t ｎ Each coordinate (X １ , Y １ ) ~ (X ｋ , Y ｋ ) raw material void ratio S １１ (t ｎ ) ~ S ｋｋ (t ｎ ), and the time change of the amount of sublimation of the raw material is calculated based on １ ~t ｎ The amount of raw material sublimated at each time SV (t １ ) ~ SV(t ｎ ), and

[0103] In FIG. 9, the correct data of the fourth learning data 910A, the fourth learning data 910B, . . . are: １ ~t ｎ The SiC single crystal growth amount GV (t １ ) ~ GV(t ｎ ), and

[0104] 9 is learning data acquired by executing a simulation after "precondition A" is set in the simulation device 110. The fourth learning data 910B shown in FIG. 9 is learning data acquired by executing a simulation after "precondition B" is set in the simulation device 110.

[0105] <Flow of Learning Data Generation Process and Learning Process> Next, a description will be given of the flow of the learning data generation process and learning process performed by the learning device 120. Fig. 10 is a flowchart showing the flow of the learning data generation process and learning process.

[0106] 10( a), when the learning data generation process is started, in step S1001, the learning device 120 acquires information indicating the manufacturing conditions and the state inside the furnace for each precondition from the simulation device 110. Note that, for simplicity of explanation, a case will be described here in which information indicating the manufacturing conditions and the state inside the furnace is acquired when a simulation is executed after precondition A is set.

[0107] In step S1002, the learning device 120 selects input data to be used as learning data from the information indicating the manufacturing conditions and the state inside the furnace obtained from the simulation device 110.

[0108] In step S1003, the learning device 120 selects correct answer data to be used as learning data from the information indicating the manufacturing conditions and the state inside the furnace acquired from the simulation device 110.

[0109] In step S1004, the learning device 120 combines the selected input data and the selected supervised data to generate learning data, and stores the generated learning data in the learning data storage unit 123. At this time, if the supervised data is two-dimensional data and the input data is one-dimensional data, the learning device 120 includes coordinate data in the input data.

[0110] As shown in FIG. 10B, when the learning process starts, in step S1011, the learning device 120 reads out learning data from the learning data storage unit 123.

[0111] In step S1012, the learning device 120 performs a learning process on the model using the learning data.

[0112] In step S1013, the learning device 120 generates a trained model and stores it in the trained model storage unit 133 of the design support device 130.

[0113] 10A does not mention the order in which the learning data generation processes for the first to fourth learning data are executed, the order in which the learning data generation processes are executed is arbitrary. For example, the learning data generation process for the fourth learning data may be executed after the learning data generation process for the third learning data is executed.

[0114] Similarly, although the order in which the learning processes for the first to fourth models are executed is not mentioned in Fig. 10(b), the order in which the learning processes are executed may be arbitrary. For example, the learning process for the third model may be executed first, followed by the learning process for the fourth model.

[0115] <Specific Example of Learning Process> Next, a specific example of the learning process performed by the learning unit 122 will be described. As described with reference to Fig. 6 to Fig. 9 , in the first embodiment, the learning data generation unit 121 generates fourth learning data 910A from first learning data 610A for "prerequisite A."

[0116] Therefore, when explaining a specific example of the learning process by the learning unit 122, we will explain the case where learning processes for the first model to the fourth model are performed using the first learning data 610A to the fourth learning data 910A, respectively, for ``precondition A.''

[0117] (1) Learning Process of First Model First, as a specific example of the learning process by the learning unit 122, a case will be described in which the learning process is performed on the first model 550_1 using the first learning data 610A for the prerequisite condition A. Fig. 11 is a first diagram showing a specific example of the learning process by the learning unit.

[0118] 11, the first model 550_1 is a model that predicts the raw material void fraction from the manufacturing conditions. The learning unit 122 uses the following data included in the input data of the first learning data 610A: a change in heater output over time: W(t), a change in heater position over time: H(t), and a coordinate at a specific cross section inside the furnace: (x １ , y １ ) ~ (x ｍ , y ｍ ), at time t １are sequentially input to the first model 550_1. As a result, output data is sequentially output from the first model 550_1 and notified to the filter processing unit 1110 and the comparison and modification unit 560_1, respectively. The filter processing unit 1110 applies an image processing filter to each piece of output data, thereby enabling the learning unit 122 to perform learning processing that increases the correlation between output data at adjacent coordinates. The filter processing unit 1110 applies, as the image processing filter, for example, a function of a difference approximation filter used in general image processing. A difference approximation filter is also called a differential filter, and includes a Sobel filter, a Laplacian filter, etc.

[0119] The comparison and change unit 560_1 receives the raw material void ratio at each coordinate at each time, which is included in the correct answer data of the first learning data 610A: (X １ , Y １ , S １１ (t)) to (X ｋ , Y ｋ , S ｋｋ (t)) and the raw material void ratio at each coordinate at each time included in the correct data of the first learning data 610A: (X １ , Y １ , S １１ (t)) to (X ｋ , Y ｋ , S ｋｋ (t)), where the raw material void ratio at each coordinate at each time when the image processing filter is applied by the filter processing unit 1111 is １ are input sequentially.

[0120] The comparison and change unit 560_1 compares the output data sequentially output from the first model 550_1 with the correct answer data to calculate an error. The comparison and change unit 560_1 also compares the output data sequentially output from the filter processing unit 1110 after application of the image processing filter with the correct answer data sequentially output from the filter processing unit 1111 after application of the image processing filter to calculate an error. The comparison and change unit 560_1 has a loss function capable of calculating the error. The comparison and change unit 560_1 weights and adds the two types of errors and uses the error obtained by the weighted addition as the calculation result of the loss function. The comparison and change unit 560_1 generates a first trained model by updating the model parameters of the first model 550_1 so as to reduce the calculation result. The first trained model is stored in the trained model storage unit 133, which functions as a first storage unit.

[0121] By performing the learning process using the output data after applying the image processing filter, it is possible to prevent the porosity at adjacent coordinates from becoming unnaturally discontinuous values. Note that the learning process may be configured to be performed without using the output data after applying the image processing filter.

[0122] (2) Learning Process of Second Model Next, as a specific example of the learning process by the learning unit 122, a case will be described in which the learning process is performed on the second model 550_2 using the second learning data 710A for the prerequisite condition A. Fig. 12 is a second diagram showing a specific example of the learning process by the learning unit.

[0123] 12, the second model 550_2 is a model that predicts the temperature inside the furnace from the manufacturing conditions. The learning unit 122 uses the following data included in the input data of the second learning data 710A: a change in heater output over time: W(t), a change in heater position over time: H(t), and coordinates at a specific cross section inside the furnace: (x １ , y １ ) ~ (x ｍ , y ｍ ), at time t １11 , and input them to the second model 550_2 in sequence. As a result, output data is sequentially output from the second model 550_2 and notified to the filter processing unit 1210 and the comparison and modification unit 560_2, respectively. The filter processing unit 1210 applies an image processing filter to each piece of output data, which enables the learning unit 122 to perform learning processing that increases the correlation between output data at adjacent coordinates. The filter processing unit 1210 applies, as the image processing filter, the same image processing filter as that applied to the filter processing unit 1110 in FIG. 11 .

[0124] The comparison and change unit 560_2 receives the following: the furnace temperature at each coordinate at each time included in the correct answer data of the second learning data 710A: (X １ , Y １ , T １１ (t)) to (X ｋ , Y ｋ , T ｋｋ (t)) and the furnace temperature at each coordinate at each time included in the correct data of the second learning data 710A: (X １ , Y １ , T １１ (t)) to (X ｋ , Y ｋ , T ｋｋ (t)), and the temperature inside the furnace at each coordinate at each time when the image processing filter is applied by the filter processing unit 1211 is １ are input sequentially.

[0125] The comparison and modification unit 560_2 compares the output data sequentially output from the second model 550_2 with the correct answer data to calculate an error. The comparison and modification unit 560_2 also compares the output data sequentially output from the filter processing unit 1210 after application of the image processing filter with the correct answer data sequentially output from the filter processing unit 1211 after application of the image processing filter to calculate an error. The comparison and modification unit 560_2 has a loss function capable of calculating the error. The comparison and modification unit 560_2 weights and adds the two types of errors and uses the error obtained by the weighted addition as the calculation result of the loss function. The comparison and modification unit 560_2 generates a second trained model by updating the model parameters of the second model 550_2 so as to reduce the calculation result. The second trained model is stored in the trained model storage unit 133, which functions as a second storage unit.

[0126] By performing the learning process using the output data after applying the image processing filter, it is possible to prevent the porosity at adjacent coordinates from becoming unnaturally discontinuous values. Note that the learning process may be configured to be performed without using the output data after applying the image processing filter.

[0127] (3) Learning Process of Third Model Next, as a specific example of the learning process by the learning unit 122, a case will be described in which learning process is performed on the third model 550_3 using the third learning data 810A for the prerequisite condition A. Fig. 13 is a third diagram showing a specific example of the learning process by the learning unit.

[0128] As shown in FIG. 13 , the third model 550_3 is a model that predicts the raw material void fraction from the furnace temperature. The learning unit 122 uses the following data included in the input data of the third training data 810A: furnace temperature at each coordinate at each time: (X １ , Y １ , T １１ (t)) to (X ｋ , Y ｋ , T ｋｋ (t)) at time t １ , and input them to the third model 550_3 in sequence. As a result, output data is output in sequence from the third model 550_3.

[0129] The comparison and change unit 560_3 receives the following data included in the correct answer data of the third training data 810A: raw material void ratio at each coordinate at each time: (X １ , Y １ , S １１ (t)) to (X ｋ , Y ｋ , S ｋｋ (t)), is time t １ are input sequentially.

[0130] As a result, the comparison and change unit 560_3 compares the output data sequentially output from the third model 550_3 and calculates an error. The comparison and change unit 560_3 has a loss function capable of calculating the error, and updates the model parameters of the third model 550_3 so as to reduce the error, which is the calculation result of the loss function, thereby generating a third trained model. The third trained model is stored in the trained model storage unit 133, which functions as a third storage unit.

[0131] (4) Learning Process of Fourth Model Next, as a specific example of the learning process by the learning unit 122, a case will be described in which learning process is performed on the fourth model 550_4 using the fourth training data 910A for the prerequisite condition A. Fig. 14 is a fourth diagram showing a specific example of the learning process by the learning unit.

[0132] 14, the fourth model 550_4 executes a prediction model that predicts the SiC single crystal growth amount from the raw material sublimation amount. The learning unit 122 calculates the raw material sublimation amount at each time: SV(t) included in the input data of the fourth training data 910A, １ , and inputs them to the fourth model 550_4 in sequence. As a result, output data is output in sequence from the fourth model 550_4.

[0133] The comparison and change unit 560_4 receives the following data included in the correct answer data of the fourth training data 910A: SiC single crystal growth amount at each time: GV(t), １ are input sequentially.

[0134] As a result, the comparison and change unit 560_4 compares the output data sequentially output from the fourth model 550_4 and calculates an error. The comparison and change unit 560_4 has a loss function capable of calculating the error, and updates the model parameters of the fourth model 550_4 so as to reduce the error, which is the calculation result of the loss function, thereby generating a fourth trained model. The fourth trained model is stored in the trained model storage unit 133, which functions as a fourth storage unit.

[0135] <Functional Configuration of Design Support Apparatus> Next, a detailed description will be given of the functional configuration of the design support apparatus 130. Fig. 15 is a first diagram showing an example of the functional configuration of the design support apparatus.

[0136] As described above, the design support device 130 functions as the display unit 131 and the prediction unit 132. Of these, the display unit 131 has a setting unit 1511, a manufacturing condition input unit 1512, and a prediction result output unit 1513. The prediction unit 132 has a trained model reading unit 1521 and a trained model executing unit 1522.

[0137] The setting unit 1511 sets the preconditions and information corresponding to the prediction items that the user 140 should check among the information indicating the state inside the reactor in the trained model reading unit 920.

[0138] The manufacturing condition input unit 1512 accepts manufacturing conditions input by the user 140 in response to the display of the design support screen on the display unit 131. The manufacturing condition input unit 1512 notifies the learned model executing unit 1522 of the manufacturing conditions whose input has been accepted.

[0139] The prediction result output unit 1513 acquires information indicating the state inside the furnace predicted by the learned model executing unit 1522 in response to the manufacturing conditions being notified to the learned model executing unit 1522 by the manufacturing condition input unit 1512. The prediction result output unit 1513 displays the acquired information indicating the state inside the furnace to the user 140 using a design support screen.

[0140] The trained model reading unit 1521 reads a trained model from the trained model storage unit 133 based on information corresponding to the prerequisites and prediction items set by the setting unit 1511.

[0141] The trained model reading unit 1521 notifies the trained model executing unit 1522 of the trained model that has been read.

[0142] The trained model executing unit 1522 inputs the manufacturing conditions notified by the manufacturing condition input unit 1512 into the trained model notified by the trained model reading unit 1521. As a result, the trained model executing unit 1522 operates the trained model and predicts information indicating the state inside the furnace.

[0143] The trained model execution unit 1522 notifies the prediction result output unit 1513 of information indicating the state inside the reactor predicted by the trained model.

[0144] In this way, design support system 100 according to the first embodiment uses a trained model to predict information indicating the state inside a furnace when manufacturing a SiC single crystal by sublimation. As a result, design support system 100 according to the first embodiment can predict information indicating the state inside a furnace at low calculation cost, thereby supporting the smooth design of manufacturing conditions in the process of manufacturing a SiC single crystal by sublimation.

[0145] <Variations of Prediction Processing> Next, variations of the prediction processing when the prediction unit 132 performs the prediction processing using the first to fourth trained models will be described. Fig. 16 is a diagram showing variations of the prediction processing by the prediction unit.

[0146] Figure 16(a) shows a prediction process that predicts the temperature change over time at each coordinate in the furnace: (X, Y, T(t)) by inputting the following into the second trained model generated by the learning process shown in Figure 12: - Change over time in heater output: W(t), - Change over time in heater position: H(t).

[0147] Figure 16(b) shows a prediction process in which the second trained model generated by the learning process shown in Figure 12 is input with the following as input: - Time change in heater output: W(t), - Time change in heater position: H(t), to predict - Time change in temperature at each coordinate in the furnace: (X, Y, T(t)), and the third trained model generated by the learning process shown in Figure 13 is input with the following as input: - Time change in temperature at each coordinate in the furnace: (X, Y, T(t)), to predict - Time change in raw material void ratio at each coordinate in the furnace: (X, Y, S(t)).

[0148] Figure 16(c) shows a prediction process that predicts the time change in raw material void ratio at each coordinate in the furnace: (X, Y, S(t)) by inputting the following into the first trained model generated by the learning process shown in Figure 11: - Time change in heater output: W(t), - Time change in heater position: H(t).

[0149] FIG. 16( d ) shows a prediction process for calculating the change over time in the amount of sublimation of raw material: SV(t) based on the predicted change over time in the raw material void ratio at each coordinate in the furnace: (X, Y, S(t)) after the prediction process shown in FIG. 16( b ) is executed.

[0150] FIG. 16( e) shows a prediction process for calculating the change over time in the amount of sublimation of raw material: SV(t) based on the predicted change over time in the raw material void ratio at each coordinate in the furnace: (X, Y, S(t)) after the prediction process shown in FIG. 16( c) is executed.

[0151] Figure 16(f) shows a prediction process in which, after executing the prediction process shown in Figure 16(d), the predicted time change in raw material sublimation amount: SV(t) is input into the fourth trained model generated by the learning process shown in Figure 14, thereby predicting the time change in SiC single crystal growth amount: GV(t).

[0152] Figure 16(g) shows a prediction process in which, after executing the prediction process shown in Figure 16(e), the predicted change over time in the amount of raw material sublimation: SV(t) is input into the fourth trained model generated by the learning process shown in Figure 16, thereby predicting the change over time in the amount of SiC single crystal growth: GV(t).

[0153] <Specific Example of Prediction Process by Prediction Unit> Next, a specific example of prediction process by the prediction unit 132 will be described. As described using Fig. 16 , in the first embodiment, the prediction unit 132 executes seven types of prediction processes (a) to (g) using first to fourth trained models. Therefore, as specific examples of the seven types of prediction processes by the prediction unit 132, prediction processes using the first to fourth trained models will be described.

[0154] (1) Prediction Process Using First Trained Model First, a specific example of the prediction process using the first trained model will be described. Fig. 17 is a first diagram illustrating a specific example of the prediction process by the prediction unit.

[0155] As shown in FIG. 17 , the trained model executing unit 1522 reads out the “prediction model that predicts raw material void fraction from manufacturing conditions” as the first trained model, and calculates the following prediction data: Time change in heater output: W(t), Time change in heater position: H(t), Coordinates at a specific cross section inside the furnace: (x １ , y １ ) ~ (x ｍ , y ｍ ), at time t １ As a result, the "prediction model that predicts raw material void ratio from manufacturing conditions," which is an example of the first trained model, is １ , Y １ , S １１ (t)) to (X ｋ , Y ｋ , S ｋｋ (t)) at time t １ The output is sequentially from

[0156] The sublimation amount calculation unit 1710 calculates the time change SV(t) in the raw material sublimation amount by summing the raw material void fractions at each coordinate at each time, which are sequentially output from the trained model execution unit 1522, for all coordinates.

[0157] (2) Prediction Process Using Second and Third Trained Models Next, a specific example of the prediction process using the second and third trained models will be described. Fig. 18 is a second diagram illustrating a specific example of the prediction process by the prediction unit.

[0158] As shown in FIG. 18 , the trained model executing unit 1522 reads out a “prediction model that predicts the furnace temperature from the manufacturing conditions” as the second trained model, and calculates the following prediction data: Time change in heater output: W(t), Time change in heater position: H(t), Coordinates at a specific cross section inside the furnace: (x １ , y １ ) ~ (x ｍ , y ｍ ), at time t １ As a result, the "prediction model that predicts the furnace temperature from the manufacturing conditions," which is an example of the second trained model, is １ , Y １ , T １１ (t)) to (X ｋ , Y ｋ , T ｋｋ (t)) at time t １ The output is sequentially from

[0159] As shown in FIG. 18 , the trained model executing unit 1522 reads out a “prediction model for predicting raw material void fraction from furnace temperature” as the third trained model, and calculates the furnace temperature at each coordinate at each time: (X １ , Y １ , T １１ (t)) to (X ｋ , Y ｋ , T ｋｋ (t)) at time t １ As a result, the "prediction model that predicts raw material void ratio from furnace temperature", which is an example of the third trained model, is １ , Y １ , S １１ (t)) to (X ｋ , Y ｋ , S ｋｋ (t)) at time t １ The output is sequentially from

[0160] The sublimation amount calculation unit 1710 calculates the time change SV(t) in the amount of sublimation of raw material by summing up the raw material void fractions at each coordinate at each time, which are sequentially output from a "prediction model that predicts raw material void fraction from furnace temperature," which is an example of a third trained model.

[0161] (3) Prediction Processing Using the Fourth Trained Model Next, a specific example of the prediction processing using the fourth trained model will be described. Fig. 19 is a third diagram illustrating a specific example of the prediction processing by the prediction unit.

[0162] As shown in FIG. 19 , the trained model executing unit 1522 reads out a “prediction model for predicting the SiC single crystal growth amount from the raw material sublimation amount” as the fourth trained model, and calculates the time change in the raw material sublimation amount: SV(t) at time t １ As a result, the fourth trained model, "a prediction model that predicts the SiC single crystal growth amount from the raw material sublimation amount", is １ The output is sequentially from

[0163] <Explanation of Design Support Screen> Next, the design support screen displayed by the display unit 131 will be described.

[0164] (1) Explanation of Design Support Screen Part 1 Fig. 20 is a first diagram showing an example of a design support screen. As shown in Fig. 20, a design support screen 2000 includes a prerequisite condition input area 2010, a prediction item input area 2020, a manufacturing condition input area 2030, and an in-furnace state display area 2040.

[0165] As shown in Fig. 20 , input fields for inputting prerequisites are provided in the prerequisite input area 2010. The example of Fig. 20 shows a state in which input fields for inputting: Data for specifying the sublimation furnace structure, Data for specifying the arrangement of the SiC powder as the raw material, Thermal conductivity, specific heat, and emissivity, which are material property values ​​of each member constituting the sublimation furnace, Data indicating the type of graphite constituting the graphite crucible, and The type of SiC powder as the raw material are displayed.

[0166] 20 , the prediction item input area 2020 has input fields for inputting prediction items when predicting information indicating the state inside the furnace using any of the first to fourth trained models. The example of Fig. 20 shows a state in which the following items have been input: furnace temperature, raw material void fraction.

[0167] As shown in Fig. 20 , the manufacturing condition input area 2030 has an input field for inputting manufacturing conditions when predicting information indicating the state inside the furnace using any of the first to fourth trained models. The example of Fig. 20 shows how the following have been input: - Time change in heater output; - Time change in heater position. Note that, as shown in the example of Fig. 20 , in the first embodiment, the time change in heater output and the time change in heater position are configured to be input using graphs 2031 and 2032.

[0168] As shown in FIG. 20 , the furnace status display area 2040 displays the prediction results when the "furnace temperature" and "raw material void ratio," which are the prediction items input in the prediction item input area 2020, are predicted using: - the second trained model; - the third trained model. As indicated by reference numerals 2041 and 2042, the furnace temperature and raw material void ratio are displayed as distributions at specific cross sections within the furnace at each time. The user 140 can visually check the furnace temperature and raw material void ratio at any time by using the slider bar 2043.

[0169] (2) Explanation of Design Support Screen Part 2 Figure 21 is a second diagram showing an example of the design support screen. The difference from Figure 20 is that only the raw material void fraction is input in the prediction item input area 2020. Therefore, in the example of Figure 21, the in-furnace state display area 2040 displays the prediction result when the "raw material void fraction" is predicted using the first trained model.

[0170] (3) Explanation of Design Support Screen Part 3 Figure 22 is a third diagram showing an example of the design support screen. The difference from Figure 20 is that the furnace temperature, raw material void fraction, raw material sublimation amount, and SiC single crystal growth amount are input in the prediction item input area 2020. Therefore, in the example of Figure 22, the furnace status display area 2040 displays: - the prediction result of "furnace temperature" predicted using the second trained model; - the prediction result of "raw material void fraction" and "raw material sublimation amount" predicted using the third trained model (see graph 2044); and - the prediction result of "SiC single crystal growth amount" predicted using the fourth trained model (see graph 2045).

[0171] (4) Explanation of Design Support Screen Part 4 Figure 23 is a fourth diagram showing an example of the design support screen. The difference from Figure 20 is that the raw material void ratio, raw material sublimation amount, and SiC single crystal growth amount are input in the prediction item input area 2020. Therefore, in the example of Figure 23, the furnace state display area 2040 displays the prediction results of the "raw material void ratio" and the "raw material sublimation amount" (see graph 2044) predicted using the first trained model, and the prediction result of the "SiC single crystal growth amount" (see graph 2045) predicted using the fourth trained model.

[0172] <Flow of Design Support Processing> Next, a description will be given of the flow of design support processing by the design support device 130. Fig. 24 is a first flowchart showing the flow of the design support processing.

[0173] In step S2401, the design support device 130 sets preconditions.

[0174] In step S2402, the design support device 130 sets prediction items.

[0175] In step S2403, the design support device 130 reads out a trained model corresponding to the set preconditions and prediction items.

[0176] In step S2404, the design support device 130 accepts input of manufacturing conditions.

[0177] In step S2405, the design support device 130 executes the trained model read out in step S2403 using the manufacturing conditions received in step S2404 as input data, thereby predicting information indicating the state inside the furnace for the set prediction items.

[0178] In step S2406, the design support system 130 displays information indicating the predicted state inside the reactor for the prediction item as a prediction result.

[0179] In step S2407, the design support device 130 determines whether or not a change instruction to the manufacturing conditions has been issued. If it is determined in step S2407 that a change instruction to the manufacturing conditions has been issued (YES in step S2407), the process returns to step S2404.

[0180] On the other hand, if it is determined in step S2407 that there is no instruction to change the manufacturing conditions (NO in step S2407), the process proceeds to step S2408.

[0181] In step S2408, the design support device 130 determines whether or not a change instruction for the prediction item has been issued. If it is determined in step S2408 that a change instruction for the prediction item has been issued (YES in step S2408), the process returns to step S2402.

[0182] On the other hand, if it is determined in step S2408 that there is no instruction to change the prediction item (NO in step S2408), the process proceeds to step S2409.

[0183] In step S2409, the design support device 130 determines whether or not a precondition change instruction has been issued. If it is determined in step S2409 that a precondition change instruction has been issued (YES in step S2409), the process returns to step S2401.

[0184] On the other hand, if it is determined in step S2409 that there is no instruction to change the prerequisites (NO in step S2409), the design support process ends.

[0185] <Summary> As is clear from the above description, the design support system 100 according to the first embodiment: - performs a simulation using a simulation device 110 that replicates a manufacturing device that manufactures SiC single crystals by sublimation. As a result, training data is generated for various combinations of input data and correct answer data. - trained models are generated by learning the relationship between the input data and correct answer data for each of the generated training data. - predicts information indicating the state inside a furnace from the manufacturing conditions using one or more of the generated trained models.

[0186] As a result, according to the first embodiment, it is possible to predict information indicating the state inside the furnace at low calculation cost, and to support the smooth design of manufacturing conditions in the process of manufacturing SiC single crystals by sublimation.

[0187] [Second embodiment] In the above first embodiment, a case was described in which, when input of manufacturing conditions is received on the design support screen, information indicating the state inside the furnace is predicted and information indicating the predicted state inside the furnace is displayed.

[0188] In contrast, in the second embodiment, when a prediction result is displayed by predicting information indicating the state inside the furnace on a design support screen and a user inputs a request to change the prediction result, a search is made for manufacturing conditions that will result in the changed prediction result. As a result, according to the second embodiment, it is possible to propose to the user manufacturing conditions that will realize the prediction result requested by the user. The second embodiment will be described below, focusing on the differences from the first embodiment.

[0189] <Functional Configuration of Design Support Apparatus> First, the functional configuration of a design support apparatus according to a second embodiment will be described. FIG. 25 is a second diagram illustrating an example of the functional configuration of the design support apparatus. As shown in FIG. 25, a design support apparatus 2500 according to the second embodiment functions as a display unit 2510 and a prediction unit 2520. Of these, the display unit 2510 includes a setting unit 1511, a manufacturing condition input unit 1512, a prediction result output unit 1513, a change request acquisition unit 2511, and a proposed condition output unit 2512. The prediction unit 2520 includes a trained model reading unit 1521, a trained model execution unit 1522, a trained model execution unit 2521, an error calculation unit 2522, and a manufacturing condition search unit 2523.

[0190] Of the units included in the display unit 2510, the units from the setting unit 1511 to the prediction result output unit 1513 have already been described in the first embodiment using FIG. 15, and therefore description thereof will be omitted here.

[0191] The change request acquisition unit 2511 accepts a change request input by the user 140 in response to the prediction result output unit 1513 displaying information indicating the state inside the reactor to the user 140 using the design support screen. The change request refers to a request to change part of the information indicating the state inside the reactor. The change request acquisition unit 2511 notifies the error calculation unit 2522 of the output data changed in accordance with the change request.

[0192] The proposed condition output unit 2512 acquires the manufacturing conditions (referred to as proposed conditions) proposed from the manufacturing condition search unit 2523 in response to the change request acquisition unit 2511 notifying the error calculation unit 2522 of the changed information indicating the state inside the furnace. The proposed condition output unit 2512 displays the acquired proposed conditions to the user 140.

[0193] The trained model execution unit 2521 is similar to the trained model execution unit 1522, and by inputting the manufacturing conditions notified from the manufacturing condition search unit 2523, it executes the trained model and predicts information indicating the state inside the furnace.

[0194] The error calculation unit 2522 calculates the error between the information indicating the state inside the furnace predicted by the learned model execution unit 2521 and the changed information indicating the state inside the furnace notified by the change request acquisition unit 2511. The error calculation unit 2522 notifies the manufacturing condition search unit 2523 of the calculated error.

[0195] The manufacturing condition search unit 2523 comprehensively generates manufacturing conditions under a specific constraint range. The manufacturing condition search unit 2523 notifies the learned model execution unit 2521 of the comprehensively generated manufacturing conditions. The manufacturing condition search unit 2523 acquires an error corresponding to each of the comprehensively generated manufacturing conditions from the error calculation unit 2522. The manufacturing condition search unit 2523 searches for a manufacturing condition with the smallest acquired error.

[0196] The manufacturing condition search unit 2523 notifies the searched manufacturing conditions to the proposed condition output unit 2512 as proposed conditions.

[0197] <Explanation of Design Support Screen> Next, a description will be given of the design support screen displayed by the display unit 2510. Fig. 26 is a second diagram showing an example of the design support screen.

[0198] As shown in FIG. 26, a design support screen 2600 includes a prerequisites input area 2610 , a predicted item input area 2620 , a change request input area 2630 , and a proposed conditions display area 2640 .

[0199] Of these, the precondition input area 2610 and the predicted item input area 2620 are similar to the precondition input area 2010 and the predicted item input area 2020 described with reference to FIG. 20, and therefore description thereof will be omitted here.

[0200] A change request to change part of the information indicating the state inside the furnace is input in the change request input area 2630. The example in Fig. 26 shows that "raw material sublimation amount" has been selected as information indicating the state inside the furnace, and part of the change in the raw material sublimation amount over time has been changed. As shown in Fig. 26, the change request input area 2630 displays the change in the raw material sublimation amount over time as a graph 2631, and the user 140 changes part of the change in the raw material sublimation amount over time by changing part of the graph 2631.

[0201] 26, when a change request is input in the change request input area 2630, the proposed conditions are displayed in the proposed condition display area 2640. The example of Fig. 26 shows that a graph 2641 showing the change in heater output over time and a graph 2642 showing the change in heater position over time are displayed as the proposed conditions.

[0202] <Flow of Design Support Processing> Next, the flow of the design support processing by the design support device 2500 will be described. Fig. 27 is a second flowchart showing the flow of the design support processing. Note that, in the second flowchart shown in Fig. 24, steps S2401 to S2409 are the same as steps S2401 to S2409 in the first flowchart shown in Fig. 24, and therefore description thereof will be omitted here.

[0203] In step S2701, the design support system 2500 determines whether a change request for changing part of the information indicating the state inside the reactor has been input. If it is determined in step S2701 that a change request has not been input (NO in step S2701), the design support process ends. On the other hand, if it is determined in step S2701 that a change request has been input (YES in step S2701), the process proceeds to step S2702.

[0204] In step S2702, the design support device 2500 accepts an input of a change request by the user 140 using the design support screen 2600.

[0205] In step S2703, the design support device 2500 comprehensively generates manufacturing conditions under a specific constraint range. The design support device 2500 searches for manufacturing conditions that have the smallest error from the post-change information indicating the state inside the furnace changed in response to the change request, from among the prediction results of the state inside the furnace based on the manufacturing conditions.

[0206] In step S2704, the design support device 2500 displays the manufacturing conditions with the smallest error found in step S2702 as proposed conditions to the user 140 on the design support screen 2600.

[0207] <Summary> As is clear from the above explanation, the design support system 100 according to the second embodiment: Accepts a request from the user 140 to change part of the information indicating the state inside the furnace displayed on the design support screen. Searches for manufacturing conditions that minimize the error between the information indicating the predicted state inside the furnace and the changed information indicating the state inside the furnace. Displays the found manufacturing conditions as proposed conditions on the design support screen.

[0208] In this way, according to the design support system 100 of the second embodiment, it is possible to search for manufacturing conditions to convert information indicating the state inside the furnace into desired information, thereby improving the convenience for users when designing manufacturing conditions.

[0209] [Third Embodiment] In the second embodiment, no mention is made of how to use the proposed conditions output by the design support device 2500, but the proposed conditions output by the design support device 2500 may be reflected in the manufacturing device 10. The third embodiment will be described below, focusing on the differences from the above embodiments.

[0210] <System Configuration of Design Support System> First, a description will be given of the system configuration of the design support system according to the third embodiment. Fig. 28 is a second diagram showing an example of the system configuration of the design support system.

[0211] As shown in FIG. 28, a design support system 2800 according to the third embodiment includes a manufacturing apparatus 10 , an evaluation apparatus 20 , and a design support apparatus 2500 .

[0212] Of these, the manufacturing apparatus 10 and the design support apparatus 2500 have already been described in the first or second embodiment, and therefore a description thereof will be omitted here.

[0213] Evaluation device 20 is an evaluation device for evaluating the SiC single crystal manufactured by manufacturing device 10. The evaluation results obtained by evaluation device 20 are notified to user 140.

[0214] The user 140 considers how to change the state inside the furnace based on the evaluation results notified from the evaluation device 20. The user 140 inputs the considered change request into the design support device 2500 using the design support screen 2600. Note that when inputting the change request, it is assumed that the design support screen 2600 displays information indicating the state inside the furnace that is predicted based on the current manufacturing conditions set in the manufacturing device 10.

[0215] When the user 140 inputs a change request, the proposed conditions are displayed on the design support screen 2600 of the design support device 2500, and the user 140 reflects the displayed proposed conditions in the manufacturing conditions of the manufacturing device 10.

[0216] <Summary> As is clear from the above description, the design support system 2800 according to the third embodiment includes the manufacturing apparatus 10, the evaluation apparatus 20, and the design support apparatus 2500. A change request according to the evaluation result by the evaluation apparatus 20 is input to the design support apparatus 2500, and the proposed conditions output by the design support apparatus 2500 are reflected in the manufacturing conditions in the manufacturing apparatus 10.

[0217] In this way, according to the design support system 2800 of the third embodiment, it is possible to design manufacturing conditions in accordance with events occurring at the manufacturing site, thereby improving the convenience for the user when designing manufacturing conditions.

[0218] That is, according to the third embodiment, it is possible to provide a design support system 2800 that provides support suitable for designing manufacturing conditions in a process for manufacturing SiC single crystals by sublimation deposition.

[0219] [Other Embodiments] In the first embodiment, the first training data 610A to the fourth training data 910A are exemplified as the training data, but the training data is not limited to these. For example, the training data may include the following as input data: - Time change in heater output: W(t), - Time change in heater position: H(t), and the correct answer data: - Amount of raw material sublimation at each time: SV(t). Alternatively, the training data may include the following as input data: - Time change in heater output: W(t), - Time change in heater position: H(t), and the correct answer data: - Amount of SiC single crystal growth at each time: GV(t). However, in the first embodiment, all the training data can be broadly divided into: - First training data in which the input data is the manufacturing conditions and the correct answer data is the data of the first item; and - Second training data in which the input data is the data of the first item and the correct answer data is the data of the second item. In other words, the amount of raw material sublimation and the amount of SiC single crystal growth are examples of data of the first item that indicates the state inside the furnace at each time. Note that the input data for the first learning data may include prerequisites in addition to the manufacturing conditions.

[0220] Similarly, in the first embodiment, the first model 550_1 to the fourth model 550_4 are exemplified as models, but the models are not limited to these four types. For example, a model may be included that receives the following input data: a time change in heater output: W(t), a time change in heater position: H(t), and outputs a time change in raw material sublimation amount: SV(t). Alternatively, a model may be included that receives the following input data: a time change in heater output: W(t), a time change in heater position: H(t), and outputs a time change in SiC single crystal growth amount: GV(t). However, in the first embodiment, all of the models can be broadly divided into: a first model trained using first learning data, and a second model trained using second learning data.

[0221] Similarly, in the first embodiment, four types of trained models were exemplified as trained models executed by the trained model executing unit 1522, but the number of trained models executed by the trained model executing unit 1522 is not limited to four. For example, a prediction model may be included that predicts a time change in the amount of raw material sublimation: SV(t) by inputting the following input data: a time change in heater output: W(t), a time change in heater position: H(t). Alternatively, a prediction model may be included that predicts a time change in the amount of SiC single crystal growth: GV(t) by inputting the following input data: a time change in heater output: W(t), a time change in heater position: H(t). However, in the first embodiment, all trained models can be broadly divided into: a first trained model trained using the first training data, and a second trained model trained using the second training data.

[0222] In the first embodiment, when the furnace temperature and raw material void ratio are displayed as information indicating the state inside the furnace on the design support screen 2000, an image of a color (predetermined color) corresponding to the data value of each coordinate is displayed. However, the relationship between the data value and the color (pixel value) may be configured so that the user 140 can change it as appropriate.

[0223] Specifically, the system may be configured so that a range of data values ​​can be set for each prediction item, and the pixel value of each coordinate is calculated by color-converting the data value of each coordinate according to the range.

[0224] In the first embodiment described above, the case has been described in which the simulation device 110, the learning device 120, and the design support device 130 are configured as separate devices in the design support system 100. However, in the design support system 100, the simulation device 110, the learning device 120, and the design support device 130 may be configured as an integrated device. Alternatively, any two of the devices may be configured as an integrated device.

[0225] In the first embodiment described above, the case where the furnace temperature is used as the second learning data or the third learning data is described, but instead of the furnace temperature, the gas concentration in the furnace, the gas flow rate in the furnace, etc. may also be used.

[0226] In the first embodiment, the functional configurations of the simulation device 110, the learning device 120, and the design support device 130 are merely examples, and part of the functions of any of the devices may be realized by other devices.

[0227] In the second embodiment, the change request is described as a change to a portion of the change in the amount of raw material sublimation over time. However, the change request is not limited to a change to a portion of the change in the amount of raw material sublimation over time. For example, the change may be to change the temperature at a certain position in the furnace over a specific time range. Alternatively, the change may be to change the raw material void ratio at a certain position in the furnace over a specific time range.

[0228] In the third embodiment, it has been described that when the user 140 inputs a change request, the current manufacturing conditions set in the manufacturing apparatus 10 are input by the user 140 into the design support screen 2600. However, the current manufacturing conditions input through the design support screen 2600 may be configured to be automatically input through communication with the manufacturing apparatus 10. As a result, the design support screen 2600 always displays information indicating the current state inside the furnace of the manufacturing apparatus 10.

[0229] In the third embodiment, the user 140 considers how to change the state inside the reactor based on the evaluation results notified from the evaluation device 20, and then inputs a change request into the design support device 2500 using the design support screen 2600. However, the configuration may be such that an appropriate change request is automatically made based on the evaluation results.

[0230] The present invention is not limited to the configurations described in the above embodiments, but may be combined with other elements, etc. These aspects can be changed without departing from the spirit of the present invention, and can be appropriately determined depending on the application form.

[0231] This application claims priority based on Japanese Patent Application No. 2023-185442 filed on October 30, 2023, the entire contents of which are incorporated herein by reference.

[0232] 10: Manufacturing apparatus 20: Evaluation apparatus 100: Design support system 110: Simulation apparatus 120: Learning apparatus 121: Learning data generation unit 122: Learning unit 130: Design support apparatus 131: Display unit 132: Prediction unit 210: Graphite crucible 220: Seed crystal 231: SiC powder 232: Sublimation gas 233: SiC single crystal 240: Heater 550_1: First model 550_2: Second model 550_3: Third model 550_4: Fourth model 1110: Filter processing unit 1210: Filter processing unit 1511: Setting unit 1512: Manufacturing condition input unit 1513: Prediction result output unit 1521: Learned model reading unit 1522: Trained model execution unit 1710: Sublimation amount calculation unit 2000: Design support screen 2500: Design support device 2510: Display unit 2520: Prediction unit 2511: Change request acquisition unit 2512: Proposed condition output unit 2521: Trained model execution unit 2522: Error calculation unit 2523: Manufacturing condition search unit 2600: Design support screen 2800: Design support system

Claims

1. A design support system that supports the design of manufacturing conditions applied in a process for manufacturing SiC single crystals by a sublimation method, comprising: a first storage unit that stores an Ith trained model trained using Ith training data including manufacturing conditions for each time input into a simulation device that reproduces the process and a first item of data indicating the state inside a furnace at each time simulated by inputting the manufacturing conditions for each time; a prediction unit that predicts the first item of data indicating the state inside a furnace at each time by inputting the manufacturing conditions for each time for prediction into the Ith trained model; and a display unit that displays the predicted first item of data.

2. The design support system described in claim 1, wherein the first storage unit stores an Ith trained model trained using Ith learning data including the manufacturing conditions at each time and the coordinates of each position in the furnace, and a first item of data indicating the state inside the furnace at each position in the furnace at each time, simulated by inputting the manufacturing conditions at each time, the coordinates of each position in the furnace, the prediction unit predicts the first item of data indicating the state inside the furnace at each position in the furnace at each time by inputting the manufacturing conditions at each time for prediction and the coordinates of each position in the furnace into the Ith trained model, and the display unit displays the predicted first item of data.

3. The design support system of claim 2, wherein the Ith trained model is generated by updating the model parameters of the Ith model so as to reduce the calculation result in a loss function capable of calculating the error between a first item of data output from the Ith model when the manufacturing conditions at each time and the coordinates of each position in the furnace contained in the Ith training data are input to the Ith model, and a first item of data corresponding to the manufacturing conditions at each time and the coordinates of each position in the furnace contained in the Ith training data.

4. The design support system of claim 2, wherein the I-th trained model is generated by updating the model parameters of the I-th model so as to reduce the calculation result in a loss function capable of calculating the error between data obtained by applying an image processing filter to a first item of data output from the I-th model when the manufacturing conditions at each time and the coordinates of each position in the furnace contained in the I-th training data are input to the I-th model, and data obtained by applying an image processing filter to a first item of data corresponding to the manufacturing conditions at each time and the coordinates of each position in the furnace contained in the I-th training data.

5. The design support system of claim 2, wherein the I trained model is generated by updating the model parameters of the I model so as to reduce the calculation result in a loss function capable of weighting and adding the error between the first item of data output from the I model and the first item of data corresponding to the manufacturing conditions at each time and the coordinates of each position in the furnace, which is included in the I training data, and the error between the data obtained by applying an image processing filter to the first item of data output from the I model and the data obtained by applying an image processing filter to the first item of data corresponding to the manufacturing conditions at each time and the coordinates of each position in the furnace, which is included in the I training data.

6. A design support system as claimed in any one of claims 2 to 5, wherein the first item of data indicating the state inside the furnace at each position in the furnace at each time includes any one of the temperature at each position in the furnace at each time, the gas concentration at each position in the furnace at each time, the gas flow rate at each position in the furnace at each time, and the raw material void fraction at each position in the furnace at each time.

7. The design support system according to claim 1, wherein the first item of data indicating the state inside the furnace at each said time includes either the amount of raw material sublimated at each said time or the amount of SiC single crystal growth at each said time.

8. A design support system as described in any one of claims 2 to 6, further comprising a second storage unit for storing a second trained model trained using second training data including a first item of data indicating the state inside the furnace at each position in the furnace at each time, simulated by inputting the manufacturing conditions at each time into a simulation device that reproduces the process, and a second item of data correlating with the first item of data; wherein the prediction unit inputs the first item of data indicating the state inside the furnace at each position in the furnace at each time, predicted by inputting the manufacturing conditions at each time for prediction and the coordinates of each position in the furnace into the first trained model, and predicts the second item of data indicating the state inside the furnace at each position in the furnace at each time; and the display unit further displays the predicted second item of data.

9. The design support system of claim 8, wherein the II trained model is generated by updating model parameters of the II model so as to reduce a loss function capable of calculating the error between a second item of data output from the II model when a first item of data included in the II learning data is input to the II model, and a second item of data included in the II learning data that corresponds to the input first item of data.

10. The design support system described in claim 9, wherein the combination of the first item of data and the second item of data contained in the II learning data includes any one of a combination of the temperature at each position in the furnace at each time and the raw material void ratio at each position in the furnace at each time, and a combination of the amount of raw material sublimation at each time and the amount of SiC single crystal growth at each time.

11. The design support system according to claim 10, wherein the amount of raw material sublimated at each time is calculated based on the raw material void ratio at each time at each position in the furnace.

12. A design support system as described in claim 1 or 2, further comprising a search unit that searches for manufacturing conditions for each time period for the prediction in order to predict the changed data for the first item when a change is received for a portion of the data for the first item predicted by the prediction unit, and the display unit displays the searched manufacturing conditions for each time period for the prediction.

13. A design support system as described in claim 8, further comprising a search unit that searches for manufacturing conditions for each time period for the prediction in order to predict the changed data for the second item when a change is received for a portion of the data for the second item predicted by the prediction unit, and the display unit displays the searched manufacturing conditions for each time period for the prediction.

14. A design support system as described in claim 1 or 2, wherein the I trained model is a recurrent neural network.

15. The design support system according to claim 8, wherein the II trained model is a recurrent neural network.

16. The design support system according to claim 14 or 15, wherein the recurrent neural network is any one of an RNN, an LSTM, a GRU, a Seq2Seq, a Bidirectional RNN, a Seq2Seq with an Attention mechanism, or a Transformer.

17. A design support method in which a computer of a design support system that supports the design of manufacturing conditions applied in a process for manufacturing SiC single crystals by a sublimation method executes the following steps: reading out from a first storage unit a first trained model trained using first training data including manufacturing conditions for each time input into a simulation device that reproduces the process and a first item of data indicating the state inside a furnace at each time simulated by inputting the manufacturing conditions for each time; predicting the first item of data indicating the state inside a furnace at each time by inputting the manufacturing conditions for each time for prediction into the first trained model; and displaying the predicted first item of data.

18. A design support program for causing a computer of a design support system that supports the design of manufacturing conditions applied in a process for manufacturing SiC single crystals by a sublimation method to execute the following steps: reading out from a first storage unit a first trained model trained using first training data including manufacturing conditions for each time input into a simulation device that reproduces the process and a first item of data indicating the state inside a furnace at each time simulated by inputting the manufacturing conditions for each time; predicting the first item of data indicating the state inside a furnace at each time by inputting the manufacturing conditions for each time for prediction into the first trained model; and displaying the predicted first item of data.

Images