Silicon carbide single crystal manufacturing apparatus, computing device of the same, and manufacturing method of silicon carbide single crystal

Abstract

To provide a crystal manufacturing apparatus that can predict a clogging state based on the measured result of a directly observable parameter.SOLUTION: A silicon carbide single crystal manufacturing apparatus includes: a pressure sensor 14 configured to measure a gas pressure Pobs of a supply gas 20 containing a SiC raw material gas and introduced into a heating vessel 9; and a computing device 15 configured to perform a prediction of a clogging time, which is a time until a gas introduction pipe 4 is clogged with a solid deposit 30, based on a learning model created by machine learning using data calculated from results of simulations of growing a SiC single crystal 3 and results of experiments of growing the SiC single crystal and the gas pressure Pobs measured by the pressure sensor 14.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to a silicon carbide (hereinafter, referred to as SiC) single crystal manufacturing apparatus, a computing machine therefor, and a method for manufacturing a silicon carbide single crystal. [Background technology]

[0002] One method for growing SiC single crystals is the gas growth method, in which SiC raw material gas is introduced into a SiC single crystal manufacturing apparatus (hereinafter referred to as the crystal manufacturing apparatus) and a SiC single crystal is grown on the surface of a seed crystal. In the gas growth method, the SiC raw material gas can be supplied continuously for a long period of time, so it is expected that long-length SiC single crystals can be grown. However, if solid deposits accumulate in the crystal manufacturing apparatus and the gas flow path is blocked, the supply of SiC raw material gas is interrupted and the growth conditions change, making it impossible to grow long-length SiC single crystals. For this reason, it is desirable to predict and respond to blockages in the gas flow path in the crystal manufacturing apparatus.

[0003] In order to predict blockage of the gas flow path in the crystal manufacturing device, it is necessary to repeatedly carry out experiments and monitor the blockage state. For example, in the motor drive device of Patent Document 1, in order to predict failure of the fan motor, the rotation speed of the fan motor is observed, and a reward is calculated from the time when an alarm was output and the time when the fan motor actually failed, and the reward is stored as past data. Then, based on the observation result of the rotation speed of the fan motor and the calculation result of the reward, an artificial intelligence predicts failure of the fan motor, and outputs an alarm when a failure is predicted. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2017-70125 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, in the case of the motor drive device of Patent Document 1, the observation quantity is the rotation speed of the fan motor, which can be directly observed, but in the case of a crystal production device, it is difficult to directly observe the volume of solid adhesions that block the gas flow path or the powder generated in the air that causes this.

[0006] An object of the present disclosure is to provide a crystal production apparatus capable of predicting a blocked state based on the observation results of directly observable parameters, a computer therefor, and a method for producing a silicon carbide single crystal. [Means for solving the problem]

[0007] A first aspect of the present disclosure is a SiC single crystal manufacturing apparatus, A cylindrical crucible (9) that constitutes a reaction chamber and has a hollow portion; a base (10) that is disposed within the hollow portion of the crucible, has one surface on which a seed crystal (2) for growing a SiC single crystal (3) is placed, and has one surface on which the seed crystal is placed; a gas introduction pipe (4) having a gas outlet (4a) for introducing a supply gas (20) containing a SiC raw material gas for growing the SiC single crystal on the surface of the seed crystal into the crucible from below the pedestal; a heating device (12, 13) for heating and decomposing the SiC raw material gas; a pressure sensor (14) for measuring the gas pressure (Pobs) of the supply gas; and a calculator (15) that predicts a clogging time, which is the time until the gas inlet pipe becomes clogged with solid adhesion (30), based on a learning model created by machine learning using data calculated from a simulation of the growth of the SiC single crystal and experimental results, and the gas pressure measured by the pressure sensor.

[0008] In this way, the crystal manufacturing apparatus is provided with a pressure sensor that measures the gas pressure of the supply gas. In addition, based on the fact that there is a correlation between the predicted clogging time based on a learning model obtained by machine learning through multiple simulations and taking into account the experimental results, a learning model obtained by machine learning through simulation is created. Then, the clogging time is predicted based on the gas pressure measured by the pressure sensor and the learning model. This makes it possible to predict the clogging state based on the observation results of the gas pressure, which is a directly observable parameter.

[0009] A second aspect of the present disclosure is a computer for a SiC single crystal manufacturing apparatus, The computing device includes: In a SiC single crystal manufacturing apparatus, a seed crystal (2) is placed on one surface of a pedestal (10) disposed in a hollow portion of a cylindrical crucible (9) constituting a reaction chamber, and a supply gas (20) containing a SiC raw material gas is introduced into the crucible from a gas introduction tube (4) located below the pedestal, and the SiC raw material gas is thermally decomposed and supplied to a surface of the seed crystal to grow a SiC single crystal (3), The gas pressure (Pobs) of the supply gas is measured by a pressure sensor (14) and the measurement result is input. A clogging time, which is the time until the gas inlet pipe is clogged by solid adhesion (30), is predicted based on a learning model created by machine learning using data calculated from the simulation of the SiC single crystal growth and experimental results, and the gas pressure measured by the pressure sensor.

[0010] In this way, the pressure of the supply gas is measured by a pressure sensor provided in the crystal manufacturing apparatus. In addition, based on the fact that there is a correlation between the predicted clogging time based on a learning model obtained by machine learning through multiple simulations and the actual clogging time obtained by experiment, a learning model is created by adding the experimental results to the machine learning through simulation. Then, the clogging time is predicted based on the gas pressure measured by the pressure sensor and the learning model. This makes it possible to create a computing machine that can predict the clogging state based on the observation results of the gas pressure, which is a directly observable parameter.

[0011] A third aspect of the present disclosure is a method for producing a SiC single crystal, comprising: using a SiC single crystal manufacturing apparatus, a seed crystal (2) is placed on one surface of a pedestal (10) arranged in a hollow portion of a cylindrical crucible (9) constituting a reaction chamber, and a supply gas (20) containing a SiC raw material gas is introduced into the crucible from a gas introduction tube (4) located below the pedestal, and the SiC raw material gas is thermally decomposed and supplied to a surface of the seed crystal, thereby growing a SiC single crystal (3); Measuring the gas pressure (Pobs) of the supply gas with a pressure sensor (14); and predicting a clogging time, which is the time until the gas inlet pipe becomes clogged with solid deposits (30), based on a learning model created by machine learning using data calculated from a simulation of the growth of the SiC single crystal and experimental results and the gas pressure measured by the pressure sensor.

[0012] In this way, the pressure of the supply gas is measured by a pressure sensor installed in the crystal manufacturing equipment. In addition, based on the fact that there is a correlation between the predicted clogging time based on a learning model obtained by machine learning through multiple simulations and taking into account the experimental results and the actual clogging time obtained in the experiment, a learning model obtained by machine learning through simulation is created. Then, the clogging time is predicted based on the gas pressure measured by the pressure sensor and the learning model. This makes it possible to manufacture SiC single crystals while predicting the clogging state based on the observation results of the gas pressure, a directly observable parameter.

[0013] The reference symbols in parentheses attached to each component indicate an example of the correspondence between the component and specific components described in the embodiments described below. [Brief description of the drawings]

[0014] [Figure 1] FIG. 1 is a diagram showing a schematic configuration of a crystal manufacturing apparatus according to a first embodiment. [Diagram 2] FIG. 13 is a correlation diagram plotting output values ​​of the clogging time obtained by a simulation and output values ​​of the clogging time obtained by an experiment performed under the same growth conditions. [Diagram 3] FIG. 1 is a diagram showing a schematic configuration of a crystal production apparatus used to create a simulation model. [Figure 4] FIG. 13 is a graph showing the relationship between the gas pressure Pobs and the dimension L(t) of the solid deposit obtained by simulation. [Diagram 5] FIG. 13 is a diagram showing an assumed correlation between gas pressure Pobs and occlusion time. [Figure 6] FIG. 1 is a diagram showing a learning model of machine learning using a neural network. [Figure 7] FIG. 1 is a diagram showing a schematic configuration of a crystal production apparatus used to create a learning model. [Figure 8] FIG. 13 is a graph showing the relationship between gas pressure Pobs and blocking time under various different growth conditions. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. In the following, including other embodiments described below, the same reference numerals will be used to denote the same or equivalent parts in each embodiment.

[0016] (First embodiment) The crystal manufacturing apparatus 1 shown in FIG. 1 is used for manufacturing a SiC single crystal ingot by long-length growth. This crystal manufacturing apparatus 1 has a function of predicting the clogging time of the gas flow path, and grows a SiC single crystal 3 on the surface of a seed crystal 2 while predicting the clogging time. In this specification, the "clogging time" means the period from the timing when the clogging of the gas flow path is predicted until it is blocked. The "clogging time" may be the period from when the gas flow path of the crystal manufacturing apparatus 1 is not blocked at all until it is blocked, or it may be the period from when the clogging is progressing until it is blocked.

[0017] [Configuration of crystal manufacturing apparatus 1] 1 faces the top and bottom, and supplies SiC raw material gas to the surface of seed crystal 2 made of a SiC single crystal substrate to grow SiC single crystal 3 on the surface of seed crystal 2. Crystal manufacturing apparatus 1 is equipped with a gas inlet pipe 4, a gas inlet source 5, a gas outlet 6, a vacuum vessel 7, a heat insulating material 8, a heating vessel 9, a pedestal 10, a rotary pull-up mechanism 11, first and second heating devices 12 and 13, a pressure sensor 14, and a control unit 15.

[0018] The gas introduction pipe 4 is provided at a lower position of the crystal manufacturing apparatus 1, and constitutes a pipe for introducing a supply gas 20 including various SiC raw material gases such as silane and propane from a gas introduction source 5 into the crystal manufacturing apparatus 1. The gas introduction pipe 4 is provided so that it extends from the lower side of the heating vessel 9 toward the pedestal 10 side and a part of it is disposed inside the heating vessel 9, and introduces the supply gas 20 into the heating vessel 9 from the lower side of the heating vessel 9. Here, the gas introduction pipe 4 is configured to protrude from the bottom surface of the vacuum vessel 7 toward the pedestal 10 side, but it may be configured not to protrude from the bottom surface of the vacuum vessel 7. The gas introduction pipe 4 introduces the supply gas 20 from the gas introduction source 5 into the crystal manufacturing apparatus 1 for each gas type or as a mixed gas. The supply gas 20 may contain gases other than the raw material gas, for example, a carrier gas such as H2 (hydrogen), an etching gas such as chlorosilane, and a dopant gas such as N2 (nitrogen). For this reason, the gas introduction pipe 4 may have a piping structure in which the gas species are separated and introduced into the crystal manufacturing apparatus 1, or it may have a piping structure in which some or all of the gas species are mixed together and then introduced into the crystal manufacturing apparatus 1.

[0019] In the present embodiment, the gas introduction pipe 4 is composed of a first introduction pipe 41 arranged concentrically and a second introduction pipe 42 arranged at an interval on the outer periphery thereof, and the supply gas 20 is introduced from the first introduction pipe 41. The second introduction pipe 42 may be omitted, and in the present embodiment, the supply gas 20 is not introduced from the second introduction pipe 42. Of course, the supply gas 20 containing the SiC raw material gas may be introduced from both the first introduction pipe 41 and the second introduction pipe 42, or the type of gas introduced may be different, for example, the SiC raw material gas may be introduced from the first introduction pipe 41 and the etching gas may be introduced from the second introduction pipe 42.

[0020] The outlet of the gas inlet pipe 4 at the portion through which the SiC raw material gas is introduced is referred to as the gas outlet 4a. As in this embodiment, even if the downstream ends of the first inlet pipe 41 and the second inlet pipe 42 of the supply gas 20 protrude closer to the pedestal 10 than the second inlet pipe 42, the outlet of the first inlet pipe 41 through which the SiC raw material gas is introduced is defined as the gas outlet 4a.

[0021] The gas introduction source 5 supplies a supply gas 20 containing a SiC raw material gas into the crystal manufacturing apparatus 1. In this embodiment, the gas introduction source 5 includes a SiC raw material gas source such as silane or propane, but when a carrier gas, an etching gas, or a dopant gas is used, an appropriate gas source to be used is provided.

[0022] Although not shown, the gas introduction source 5 is equipped with a heating device for controlling the temperature of the supply gas 20 and a flow rate control device for controlling the flow rate, and the temperature and flow rate of each gas can be controlled according to the growth status of the SiC single crystal 3. When other gas species are used as the supply gas 20 in addition to the SiC raw material gas, a heating device and a flow rate control device are provided for each gas species.

[0023] The gas outlet 6 discharges the waste portion of the supply gas 20 after it is supplied to the seed crystal 2, that is, the unreacted gas in the SiC raw material gas, the carrier gas, the dopant gas, etc., to the outside of the crystal manufacturing apparatus 1 as exhaust gas.

[0024] The vacuum vessel 7 is made of quartz glass or the like, has a bottomed tubular shape with a hollow space, in this embodiment a cylindrical shape, and is structured to allow the introduction and discharge of the supply gas 20. The vacuum vessel 7 also accommodates other components of the crystal manufacturing apparatus 1, and is structured to allow the pressure in the internal space to be reduced by evacuating the space. As described above, the bottom of the vacuum vessel 7 is provided with the gas introduction pipe 4 and the gas introduction source 5 for supplying the supply gas 20. The side wall of the vacuum vessel 7 is also pierced at the midpoint in the height direction, and the gas outlet 6 is disposed therein.

[0025] The heat insulating material 8 has a tubular shape having a hollow portion, and in this embodiment, a cylindrical shape, and is arranged coaxially with the vacuum vessel 7. Specifically, the heat insulating material 8 is composed of a cylindrical bottom portion 8a having a hollow portion formed therein into which the gas introduction pipe 4 is fitted, and a cylindrical side wall portion 8b extending toward the pedestal 10 along the outer circumferential surface of the bottom portion 8a. The bottom portion 8a and the side wall portion 8b have a smaller diameter than the vacuum vessel 7, and are arranged inside the vacuum vessel 7 to suppress heat transfer from the growth space inside the heat insulating material 8 to the vacuum vessel 7 side. The heat insulating material 8 is composed of, for example, graphite, but the surface may be coated with a high melting point metal carbide such as TaC (tantalum carbide) or NbC (niobium carbide) to make it difficult to be thermally etched. The heat insulating material 8 is penetrated at the middle position in the height direction, and the gas exhaust port 6 is arranged therein.

[0026] The heating vessel 9 constitutes a crucible serving as a reaction chamber, and is configured to have a hollow cylindrical shape, in this embodiment, a cylindrical shape. The hollow part of the heating vessel 9 constitutes a growth space for the SiC single crystal 3 on the surface of the seed crystal 2. The heating vessel 9 is made of, for example, graphite, but the surface may be coated with a high-melting point metal carbide such as TaC or NbC to make it difficult to be thermally etched.

[0027] The heating vessel 9 is disposed so as to surround the pedestal 10, and is penetrated at a lower position, in which the gas exhaust port 6 is disposed. Although a small gap is provided between the inner peripheral surface of the heating vessel 9 and the pedestal 10, the gap is narrow, and therefore the exhaust gas is turned back downward by the pedestal 10. Therefore, as shown by the arrows in FIG. 1, the supply gas 20 introduced from the gas inlet tube 4 is supplied to the growth surface of the seed crystal 2 or the SiC single crystal 3, and then turned back by the pedestal 10, passes between the heating vessel 9 and the gas inlet tube 4, and is then discharged to the outside from the gas exhaust port 6. Then, the SiC raw material gas contained in the supply gas 20 is decomposed by the heating vessel 9 heated by the first and second heating devices 12 and 13 described later, before the supply gas 20 from the gas inlet tube 4 is introduced to the seed crystal 2.

[0028] The pedestal 10 is a member for mounting the seed crystal 2, and is, for example, disk-shaped. The seed crystal 2, which is circular and plate-shaped, is mounted on one surface of the pedestal 10 facing the growth space, and the SiC single crystal 3 is grown on the surface of the seed crystal 2. The central axis of the pedestal 10 is arranged coaxially with, for example, the central axis of the heating vessel 9 or the central axis of a shaft 11a of a rotary pull-up mechanism 11 described later. The pedestal 10 is made of, for example, graphite, but the surface may be coated with a high-melting point metal carbide such as TaC or NbC to make it difficult to be thermally etched. The pedestal 10 is connected to the shaft 11a on the surface opposite to the surface on which the seed crystal 2 is mounted, and is rotated with the rotation of the shaft 11a, and can be pulled up in the upward direction of the drawing as the shaft 11a is pulled up.

[0029] The rotary pull-up mechanism 11 rotates and pulls up the pedestal 10 via a shaft 11a made of a pipe material or the like. In this embodiment, the shaft 11a is configured in a straight line extending up and down, with one end connected to the surface of the pedestal 10 opposite to the surface to which the seed crystal 2 is attached, and the other end connected to the main body of the rotary pull-up mechanism 11. This shaft 11a is also made of, for example, graphite, but the surface may be coated with a high-melting point metal carbide such as TaC or NbC to make it difficult to be thermally etched. With this configuration, the pedestal 10, the seed crystal 2, and the SiC single crystal 3 can be rotated and pulled up, and the growth surface of the SiC single crystal 3 can be adjusted to a temperature suitable for growth while the growth surface of the SiC single crystal 3 has a desired temperature distribution.

[0030] The first and second heating devices 12 and 13 are configured with heating coils such as induction heating coils and direct heating coils, and are arranged to surround the vacuum vessel 7. In the present embodiment, the first and second heating devices 12 and 13 are configured with induction heating coils. The first and second heating devices 12 and 13 are configured to be able to independently control the temperature of the target locations, with the first heating device 12 being arranged at a position corresponding to the lower position of the heating vessel 9, and the second heating device 13 being arranged at a position corresponding to the pedestal 10. Therefore, the first heating device 12 can control the temperature of the lower portion of the heating vessel 9 to heat and decompose the SiC raw material gas. In addition, the second heating device 13 can control the temperatures around the pedestal 10, the seed crystal 2, and the SiC single crystal 3 to temperatures suitable for the growth of the SiC single crystal 3.

[0031] The pressure sensor 14 measures the pressure of the supply gas 20 containing the SiC raw material gas. The pressure sensor 14 is preferably disposed inside the gas inlet pipe 4 relative to the gas outlet 4a, that is, at a position not going out through the gas outlet 4a, to measure the pressure of the supply gas 20 containing the SiC raw material gas in the gas inlet pipe 4. More preferably, the pressure sensor 14 is disposed below one surface of the bottom 8a of the thermal insulation material 8 on the pedestal 10 side.

[0032] The pressure sensor 14 may be disposed at any position where it can measure the pressure of the supply gas 20 containing the SiC raw material gas, but in particular, what is blocked in the gas flow path is the supply path of the SiC raw material gas, which in this embodiment is the first introduction pipe 41. For this reason, the pressure sensor 14 is disposed upstream of the gas outlet 4a in the flow direction of the supply gas 20 (hereinafter simply referred to as the upstream side) in the first introduction pipe 41. More specifically, the pressure sensor 14 is disposed in the first introduction pipe 41 below one surface of the bottom part 8a on the pedestal 10 side where heat transfer is retarded by the thermal insulation material 8.

[0033] In this embodiment, the first introduction pipe 41 is surrounded by the second introduction pipe 42 downstream of the gas outlet 4a in the flow direction of the supply gas 20 (hereinafter simply referred to as the downstream side). This results in a piping structure in which various gases in the first introduction pipe 41 and the second introduction pipe 42 can merge at this position. In the case of such a structure, the pressure sensor 14 may be disposed in the second introduction pipe 42 downstream of the gas outlet 4a. However, in the first introduction pipe 41 which is particularly susceptible to clogging, it is preferable to dispose the pressure sensor 14 upstream of the gas outlet 4a where clogging is predicted, since this makes it easier to measure the gas pressure which changes depending on the clogging state.

[0034] Furthermore, the gas inlet pipe 4 may have a piping structure in which the various gases introduced from the first inlet pipe 41 and the second inlet pipe 42 are not joined together as in the present embodiment, but rather the multiple inlet pipes are separated and the various gases do not join together. In that case, the pressure sensor 14 is preferably disposed in the inlet pipe containing the SiC raw material gas.

[0035] The control unit 15 corresponds to a computing machine, and controls the flow rate and temperature of the gas supplied from the gas introduction source 5, the temperature at the lower position of the heating vessel 9 by the first and second heating devices 12, 13 and the temperature at the growth surface position of the SiC single crystal 3, and the pulling amount by the rotary pulling mechanism 11. The control unit 15 also stores a learning model based on experiments and simulations, and performs clogging time prediction based on this learning model. The learning model and clogging time prediction stored in the control unit 15 will be described later.

[0036] As described above, the crystal manufacturing apparatus 1 is configured. Next, a method for manufacturing a SiC single crystal 3 using the crystal manufacturing apparatus 1 configured in this manner will be described.

[0037] [Method for producing SiC single crystal 3] First, the seed crystal 2 is attached to one surface of the pedestal 10. The surface of the seed crystal 2 opposite to the pedestal 10, that is, the growth surface of the SiC single crystal 3, is an off-substrate having a predetermined off-angle, such as 4° or 8°, with respect to the (0001) C-plane. Next, the pedestal 10 and the seed crystal 2 are placed in the heating vessel 9. Then, the first and second heating devices 12 and 13 are controlled to provide a desired temperature distribution. That is, the temperature distribution is set so that the SiC raw material gas contained in the supply gas 20 is thermally decomposed and supplied to the surface of the seed crystal 2, and the SiC raw material gas is recrystallized on the surface of the seed crystal 2, while the sublimation rate is higher than the recrystallization rate in the heating vessel 9. In this way, the temperature of the bottom of the heating vessel 9 can be set to a high temperature of 2000° C. or higher, for example, 2500° C., and the temperature of the surface of the seed crystal 2 can be set to a temperature lower than the bottom of the heating vessel 9 and suitable for recrystallization of the SiC single crystal 3, for example, about 2200° C.

[0038] Furthermore, while the vacuum vessel 7 is maintained at a desired pressure, a supply gas 20 containing a SiC raw material gas is introduced through the gas introduction tube 4. As a result, the supply gas 20 is supplied to the seed crystal 2 as shown by the arrow in FIG. 1, and based on this, a SiC single crystal 3 is grown on the surface of the seed crystal 2. Then, the rotary pull-up mechanism 11 rotates the pedestal 10, the seed crystal 2, and the SiC single crystal 3 via the shaft 11a, while pulling them up in accordance with the growth rate of the SiC single crystal 3. As a result, the height of the growth surface of the SiC single crystal 3 is kept almost constant, and it becomes possible to control the temperature distribution of the growth surface temperature with good controllability.

[0039] When growing SiC single crystal 3 in this manner, the clogging time of gas inlet pipe 4 is predicted based on a learning model stored in control unit 15. The learning model stored in control unit 15 and the clogging time prediction will now be described.

[0040] [Learning model and blockage time prediction] When creating a learning model, as in Patent Document 1, it is sufficient to perform multiple experiments, accumulate data, and model it. However, reinforcement learning, which performs multiple experiments, accumulates data, and then performs modeling, is not suitable for the growth of SiC single crystals 3. That is, in the growth process of SiC single crystals 3, the temperature must be increased from room temperature to a high temperature exceeding 2000 ° C., and after growing the SiC single crystals 3, the temperature of the crystal manufacturing apparatus 1, which has been further heated, must be lowered. The temperature increase process, growth process, and temperature decrease process all take a long time, and one manufacturing process of SiC single crystals 3 requires high costs in terms of time and money. For this reason, it is difficult to perform multiple experiments, and reinforcement learning that performs multiple experiments is not suitable for predicting the blockage of the crystal manufacturing apparatus 1. In addition, graphite parts such as the insulation material 8 and the heating container 9 are used in the crystal manufacturing apparatus 1, but the deterioration state of these parts is not constant, and reinforcement learning based on multiple experiments is not suitable for predicting the blockage time of the crystal manufacturing apparatus 1 in terms of reproducibility.

[0041] Therefore, as a method for creating a learning model without conducting multiple experiments, we verified whether it is possible to model the results of multiple simulations and perform fitting based on the results of a small number of experiments. When conducting multiple experiments and modeling the results, the SiC single crystal 3 is actually grown in the crystal manufacturing apparatus 1, and the time until the gas introduction tube 4 is clogged by solid deposits is directly observed. Therefore, when the results of multiple experiments are modeled, there is a correlation between the actual clogging time until the gas introduction tube 4 is clogged when the SiC single crystal 3 is actually grown using the crystal manufacturing apparatus 1, and the clogging predicted time based on the learning model.

[0042] However, if the results of a large number of simulations are modeled and there is no correlation with the experimental results, modeling based on the results of the simulations is meaningless. Therefore, the inventors conducted verification. As a result, they found that there is a correlation between the experimental results and the simulation results, and confirmed that it is possible to model the results of a large number of simulations and perform fitting based on the results of a small number of experiments.

[0043] Specifically, the furnace structure of the crystal manufacturing apparatus 1 shown in FIG. 1, the type of gas introduced during the growth of the SiC single crystal 3, the gas flow rate, temperature, and other conditions (hereinafter referred to as growth conditions) were determined, and the clogging time from the start of growth until the vicinity of the gas outlet 4a of the gas introduction tube 4 was clogged with solid deposits was measured. The furnace structure refers to the structure of the crystal manufacturing apparatus 1, including the dimensions and thicknesses of the vacuum vessel 7, the heat insulating material 8, the heating vessel 9, and the like. In addition, using a simulation model having the same structure as the crystal manufacturing apparatus 1 used in the experiment, simulation values ​​were obtained for the temperature, gas flow rate, and amount of solid generated at each coordinate point. When the amount of solid generated reaches the radius of the gas outlet 4a in the gas introduction tube 4, it is assumed that the gas outlet 4a is clogged with solid deposits generated on the inner wall surfaces on both sides of the central axis of the gas introduction tube 4. For this reason, the time from the start of growth until the solid deposits reach the radius of the gas outlet 4a was determined as the clogging time.

[0044] As a result, the results shown in Figure 2 were obtained. Each point in Figure 2 is plotted with the output value of the clogging time obtained as an experimental result under the same growth conditions on the vertical axis and the output value of the clogging time predicted as a result of the simulation on the horizontal axis. The growth conditions for each plot are different. As can be seen by analyzing this figure, when comparing plots under different conditions, there is a tendency that as the clogging time in the experiment increases, the clogging time in the simulation also increases. Therefore, there is a certain degree of regularity in the relationship between the clogging time in the experiment and the clogging time in the simulation, and it can be said that the two are correlated.

[0045] Therefore, by performing machine learning using the data obtained by performing multiple simulations as training data and then fitting the learned model obtained by machine learning to the experimental values, it is possible to predict the blockage time with high accuracy even if only a few experiments are performed.

[0046] In addition, various parameters were examined as parameters to be used for predicting the clogging time, and it was confirmed that there is a particular correlation with the clogging time when the gas pressure on the upstream side of the location where the solid deposit adheres in the flow direction of the supply gas 20 is used as a parameter.

[0047] Specifically, as shown in FIG. 3, a simulation model was created in which a pseudo-annular solid attachment 30 was attached to the inner peripheral wall of the gas introduction tube 4, and the relationship between the dimensions of the solid attachment 30 and the gas pressure Pobs upstream of the solid attachment 30 was examined by Sim analysis. The dimension of the solid attachment 30 in the radial direction of the gas introduction tube 4 was set to L(t), and the dimension L(t) was changed to obtain the corresponding gas pressure Pobs. The dimension of the solid attachment 30 in the central axis direction of the gas introduction tube 4 was set to 23.5 mm, and the inner diameter dimension of the gas introduction tube 4 was set to a radius of 15 mm. The inner diameter dimension of the gas introduction tube 4 is an arbitrary value determined according to the dimensions of each part of the crystal production apparatus 1. The dimensions of the solid attachment 30 are values ​​empirically set based on experiments.

[0048] As a result of the simulation, when the dimension L(t) = 0 mm, 5 mm, 10 mm, and 15 mm, the gas pressure Pobs was 1.28145 [Pa], 1.34166 [Pa], 2.14737 [Pa], and 1423.02 [Pa], respectively. The simulation results are plotted as shown in Figure 4, and it can be seen that the larger the dimension L(t), the larger the gas pressure Pobs. In other words, it is recognized that there is a correlation between the dimension L(t) and the gas pressure Pobs.

[0049] On the other hand, if the dimension L(t) can be directly observed, the clogging time of the gas introduction tube 4 can be confirmed, but it is difficult to directly observe the inside of the crystal manufacturing apparatus 1 during the growth of the SiC single crystal 3. However, if the gas pressure Pobs can be measured, the dimension L(t) that is correlated with the gas pressure Pobs can be estimated, and the clogging time can be predicted from the estimated dimension L(t). Specifically, it is assumed that the clogging time of the gas introduction tube 4 due to the solid deposits 30 progresses as the gas pressure Pobs increases, so that the clogging time of the gas introduction tube 4 is assumed to become shorter as the gas pressure Pobs increases. For this reason, the clogging time of the gas introduction tube 4 is assumed to become shorter linearly or nonlinearly as the gas pressure Pobs increases. For example, as shown in FIG. 5, the clogging time of the gas introduction tube 4 is assumed to become shorter linearly as the gas pressure Pobs increases.

[0050] Therefore, if the simulation results satisfy this correlation, it can be said that the clogging time prediction using the simulation model is valid. In addition, as shown in FIG. 4, in consideration of the fact that the simulation results show a relationship in which the gas pressure Pobs increases as the dimension L(t) increases, it can be said that the clogging time prediction using the simulation model is proven to be valid. Therefore, if a pressure sensor 14 is provided to measure the gas pressure Pobs at a location that varies depending on the clogging state, and the gas pressure Pobs is measured, the dimension of the solid deposit 30 can be estimated based on the gas pressure Pobs, and the clogging time can be predicted. For example, it is preferable to provide the pressure sensor 14 upstream of a position in the gas introduction pipe 4 that is assumed to be particularly susceptible to clogging by the solid deposit 30.

[0051] Based on the above verification, it was confirmed that it is possible to create a learning model for predicting blockage time by modeling the results of many simulations and fitting them based on the results of a small number of experiments. Therefore, a learning model was created by performing machine learning on the data showing the results of many simulations.

[0052] Any method such as neural network or deep learning can be used for machine learning. When using a neural network, a learning model can be created, for example, as follows.

[0053] Specifically, a large number of simulations, for example 1000 to 10,000 times, are performed, and data on various conditions for growing the SiC single crystal 3 used in the simulation are input as input values ​​xi. For example, data on the configuration of the crystal manufacturing apparatus 1, data on the gas type used, data on the gas flow rate, data on the temperature of each part in the crystal manufacturing apparatus 1, etc. are input as data on the input values ​​xi in the input layer. Then, learning is performed by a neural network using the input simulation data as teacher data to create a learning model as shown in FIG. 6. For example, a learning model is created in which an output value f(xi) is obtained in the output layer L3 via the intermediate layer L2 for the input value xi of each node in the input layer L1. The input value xi changes depending on various conditions under which the simulation was performed, and various conditions corresponding to each of the large number of simulations are input.

[0054] In this case, the learning model may be a model having an input layer L1, an intermediate layer L2, and an output layer L3, which are enclosed by dashed lines in the figure, but it is preferable to incorporate a correction layer L4 that adds an offset between the value of the output layer L3 of the neural network and the output value of the simulation. The value of the correction layer L4 can be a function expressed as y=a×f(xi)+b, for example, based on the output value f(xi).

[0055] Then, data on various conditions obtained by performing experiments fewer times than the number of simulations, for example 30 times, is input, and fitting is performed using data on the actual blocking time. Here, as shown in FIG. 7, various sensors are provided so as to measure the state of each part of the crystal manufacturing apparatus 1 shown in FIG. 1, and an experiment is performed in which the temperature inside the crystal manufacturing apparatus 1 is raised from 20°C to 2500°C to grow a SiC single crystal 3. As various sensors, in addition to the pressure sensor 14 located upstream of the gas outlet 4a in the gas introduction pipe 4, pressure sensors 16a, 16b, flow sensors 17a, 17b, and a pyrometer 18 are provided. The pressure sensor 16a is disposed downstream of the blocking surface 31 surrounded by a dashed line that is assumed to be blocked by the solid deposit 30 in the first introduction pipe 41. The pressure sensor 16b is disposed in the growth space. The flow sensor 17a is disposed at a position where the gas flow rate of the supply gas 20 at the position where the pressure sensor 14 is disposed is measured. The flow sensor 17b is disposed at a position where the gas flow rate inside the growth space is measured. The pyrometer 18 was placed at a position to measure the temperature of the rear surface of the base 10. Various conditions obtained from the experimental conditions are input as input values ​​xi into the learning model shown in Fig. 6, and the actual clogging time data obtained during the experiment is input into the correction layer L4, thereby fine-tuning a and b in the function y = a × f (xi) + b of the correction layer L4. This allows the learning model obtained by simulation to be fitted to the experimental results, resulting in a learning model that can predict the clogging time with higher accuracy.

[0056] Once a learning model combining the experimental results with the simulation data is obtained in this manner, data on various conditions is then input as the input values ​​xi of each node in the input layer L1 when actually manufacturing the SiC single crystal 3. For example, data on the configuration of the crystal manufacturing apparatus 1, data on the type of gas used, data on the gas flow rate, data on the temperature of each part in the crystal manufacturing apparatus 1, etc. are input as the input values ​​xi. This makes it possible to obtain the clogging time predicted under those conditions as the values ​​of the output layer L3 and the correction layer L4 of the learning model.

[0057] When various different conditions are input, the occlusion time is predicted according to the conditions, and the relationship between the gas pressure Pobs and the occlusion time can be obtained based on the predicted occlusion time, for example, as shown in Fig. 8. Fig. 8 shows a case where different occlusion times are obtained when various different conditions are input as conditions A, B, and C, and the relationship between the gas pressure Pobs and the occlusion time according to each condition is obtained. Then, when the relationship between the gas pressure Pobs and the occlusion time is obtained in this way, the occlusion time from that timing to occlusion can be predicted based on the relationship and the gas pressure Pobs indicated by the measurement value of the pressure sensor 14.

[0058] As described above, the crystal manufacturing apparatus 1 is equipped with the pressure sensor 14 for measuring the gas pressure Pobs of the supply gas 20. In addition, based on the fact that there is a correlation between the predicted clogging time based on a learning model obtained by machine learning through multiple simulations and the actual clogging time obtained by experiment, a learning model is created by adding the experimental results to the machine learning through simulation. Then, the clogging time is predicted based on the gas pressure Pobs measured by the pressure sensor 14 and the learning model. This makes it possible to predict the clogging state based on the observation results of the gas pressure Pobs, a directly observable parameter.

[0059] Furthermore, the crystal manufacturing apparatus 1 of this embodiment can provide the following effects.

[0060] (1) In the present embodiment, the pressure sensor 14 is provided upstream of the gas outlet 4a of the gas introduction pipe 4 in the crystal manufacturing apparatus 1, so that the gas pressure Pobs corresponding to the blocked state can be measured. This makes it possible to predict the blocked state with higher accuracy.

[0061] (2) Because a learning model can be created through simulation, it is possible to create a learning model without conducting multiple experiments, thereby reducing the cost of experiments.

[0062] (3) By fine-tuning the learning model by combining it with the results of a small number of experiments, it is possible to predict the occlusion time with greater accuracy.

[0063] (4) Since the occlusion time changes depending on various conditions, the control unit 15 can use a learning model to calculate and output an optimal process that optimizes the growth conditions, the configuration of the crystal manufacturing apparatus 1, and the like within a feasible range with the occlusion time as the objective variable. For example, the occlusion time can be increased by calculating the occlusion time using the learning model and deriving the growth conditions and the configuration of the crystal manufacturing apparatus 1 that will increase the occlusion time.

[0064] (5) The control unit 15 may be provided with a feedback function of feeding back the predicted clogging time, calculating the growth conditions of the SiC single crystal 3 that can make the clogging time longer, outputting the growth conditions, for example, adjusting to the growth conditions. For example, the gas pressure Pobs is measured during the growth of the SiC single crystal 3 to predict the clogging time, and various conditions including the predicted clogging time and the gas pressure Pobs are fed back. In addition, a learning model is used to determine the growth conditions that will make the clogging time longer, such as at least one of various conditions including the gas type, gas flow rate, and temperature of the reaction chamber. Then, by adjusting to the determined growth conditions, the clogging time can be further extended.

[0065] (6) If the time change of the gas pressure Pobs is not the predicted change as shown in Figure 8, deterioration of the heating vessel 9 or the insulation material 8 may have occurred. These are not constant and are unpredictable changes, and are difficult to predict by simulation. Even in such a case, if various conditions including the gas pressure Pobs are fed back to find conditions that are less likely to cause clogging, various conditions that take into account changes in conditions that cannot be predicted by simulation can be obtained, making it possible to extend the clogging time.

[0066] (Other embodiments) Although the present disclosure has been described based on the above-described embodiment, it is not limited to the embodiment, and includes various modifications and modifications within the equivalent range. In addition, various combinations and forms, and other combinations and forms including only one element, more than one element, or less than one element, are also within the scope and concept of the present disclosure.

[0067] For example, in the above embodiment, as shown in FIG. 6, an example was shown in which the number of nodes in the input layer L1 of the learning model was five, and the number of layers in the intermediate layer L2 was three. However, the number of nodes in the input layer L1 and the intermediate layer L2, and the number of layers in the intermediate layer L2 can be set arbitrarily.

[0068] In addition, if the learning model is provided with a correction layer L4 that takes into account an offset or is adjusted based on experimental results, the learning model will have a higher accuracy, but it may also be a learning model created based on multiple simulations. The number of simulations and experiments performed when creating a learning model is also arbitrary. However, the more times the number of simulations, the more accurate the learning model can be obtained.

[0069] In addition, in the above embodiment, an example has been described in which a neural network is used as machine learning, but other learning methods such as deep learning, support vector machines, random forests, and k-nearest neighbor methods may also be applied.

[0070] In the above embodiment, the crystal manufacturing apparatus 1 is described as being equipped with the control unit 15 that stores the learning model, but the present disclosure can also be used as a computer that performs blockage time prediction and in which the control unit 15 is configured separately from the crystal manufacturing apparatus 1. In that case, the computer stores the learning model, acquires data indicating the measurement results from the pressure sensor 14, and predicts the blockage time based on the measurement results and the learning model. Similarly, the present disclosure can also be understood as a blockage time prediction method.

[0071] Furthermore, in the above embodiment, the crystal manufacturing apparatus 1 employs a side flow system in which the supply gas 20 is supplied to the growth surface of the SiC single crystal 3 and then discharged in the outer circumferential direction of the heating vessel 9. This is merely one example. For example, the crystal manufacturing apparatus 1 may employ an upflow system in which the supply gas passes through the outer circumferential surface of the SiC single crystal 3 or beside the pedestal 10 and then discharged further upward. Alternatively, the crystal manufacturing apparatus 1 may employ a return flow system in which the supply gas 20 is supplied to the growth surface of the SiC single crystal 3 and then returned in the same direction as the supply direction.

[0072] The control unit and the method described in the present disclosure may be realized by a dedicated computer provided by configuring a processor and a memory programmed to execute one or more functions embodied in a computer program. Alternatively, the control unit and the method described in the present disclosure may be realized by a dedicated computer provided by configuring a processor with one or more dedicated hardware logic circuits. Alternatively, the control unit and the method described in the present disclosure may be realized by one or more dedicated computers configured by a combination of a processor and a memory programmed to execute one or more functions and a processor configured with one or more hardware logic circuits. In addition, the computer program may be stored in a computer-readable non-transitory tangible recording medium as instructions executed by the computer. [Explanation of symbols]

[0073] Reference Signs List 1...crystal manufacturing apparatus, 2...seed crystal, 3...SiC single crystal, 4...gas introduction tube, 4a...gas outlet, 5...gas introduction source, 6...gas exhaust port, 7...vacuum vessel, 8...insulation material, 8a...bottom, 8b...side wall, 9...heating vessel, 10...base, 11...rotational lifting mechanism, 11a...shaft, 12...first heating device, 13...second heating device, 14...pressure sensor, 15...control unit, 20...supply gas, 30...solid deposit, 31...blocked surface, 41...first introduction tube, 42...second introduction tube

Claims

1. An apparatus for producing silicon carbide single crystals, comprising: A cylindrical crucible (9) that constitutes a reaction chamber and has a hollow portion; a base (10) disposed in the hollow portion of the crucible, having one surface on which a seed crystal (2) for growing a silicon carbide single crystal (3) is placed and on which the seed crystal is placed; a gas introduction pipe (4) having a gas outlet (4a) for introducing a supply gas (20) containing a silicon carbide raw material gas for growing the silicon carbide single crystal on the surface of the seed crystal into the crucible from below the pedestal; a heating device (12, 13) for heating and decomposing the silicon carbide raw material gas; a pressure sensor (14) for measuring the gas pressure (Pobs) of the supply gas; and a calculator (15) that predicts a clogging time, which is the time until the gas inlet pipe becomes clogged with a solid deposit (30), based on a learning model created by machine learning using data calculated from a simulation of the growth of the silicon carbide single crystal and experimental results, and the gas pressure measured by the pressure sensor.

2. 2 . The silicon carbide single crystal manufacturing apparatus according to claim 1 , wherein the pressure sensor is disposed upstream of the gas outlet in the gas inlet pipe in a flow direction of the supply gas, and measures the gas pressure in the gas inlet pipe.

3. 3 . The silicon carbide single crystal manufacturing apparatus according to claim 1 , wherein the control unit calculates and outputs an optimal process for lengthening the blocking time when growing the silicon carbide single crystal, based on the prediction of the blocking time. 4 .

4. 3. The silicon carbide single crystal manufacturing apparatus according to claim 1, wherein the control unit has a feedback function of calculating and outputting at least one condition of the type of the supply gas, the flow rate, and the temperature of the reaction chamber that lengthens the blockage time when growing the silicon carbide single crystal, based on the prediction of the blockage time.

5. A computer for a silicon carbide single crystal manufacturing apparatus, comprising: The computing device includes: The silicon carbide single crystal manufacturing apparatus is used to grow a silicon carbide single crystal (3) by placing a seed crystal (2) on one surface of a pedestal (10) arranged in a hollow portion of a cylindrical crucible (9) constituting a reaction chamber, introducing a supply gas (20) containing a silicon carbide raw material gas into the crucible from a gas introduction pipe (4) below the pedestal, and thermally decomposing the silicon carbide raw material gas and supplying it to a surface of the seed crystal, The gas pressure (Pobs) of the supply gas is measured by a pressure sensor (14) and the measurement result is input. A computing machine of a silicon carbide single crystal manufacturing apparatus that predicts a clogging time, which is the time until the gas inlet pipe becomes clogged with a solid attachment (30), based on a learning model created by machine learning using data calculated from a simulation of the growth of the silicon carbide single crystal and experimental results, and the gas pressure measured by the pressure sensor.

6. 6. The computer of the silicon carbide single crystal manufacturing apparatus according to claim 5, wherein the pressure sensor measures the gas pressure upstream of the gas outlet in the gas inlet pipe in a flow direction of the supply gas.

7. 7. The computer of the silicon carbide single crystal manufacturing apparatus according to claim 5, further comprising: a computer that calculates and outputs an optimum process for lengthening the blocking time when growing the silicon carbide single crystal, based on the prediction of the blocking time.

8. 7. The silicon carbide single crystal manufacturing apparatus according to claim 5, further comprising a feedback function for calculating and outputting at least one condition of the type of the supply gas, the flow rate, and the temperature of the reaction chamber that lengthens the blockage time when growing the silicon carbide single crystal, based on the prediction of the blockage time.

9. A method for producing a silicon carbide single crystal, comprising the steps of: Using a silicon carbide single crystal manufacturing apparatus, a seed crystal (2) is placed on one surface of a pedestal (10) arranged in a hollow portion of a cylindrical crucible (9) constituting a reaction chamber, and a supply gas (20) containing a silicon carbide raw material gas is introduced into the crucible from a gas introduction pipe (4) located below the pedestal, and the silicon carbide raw material gas is thermally decomposed and supplied to a surface of the seed crystal, thereby growing a silicon carbide single crystal (3); Measuring the gas pressure (Pobs) of the supply gas with a pressure sensor (14); and predicting a clogging time, which is the time until the gas inlet pipe is clogged with a solid deposit (30), based on a learning model created by machine learning using data calculated from a simulation of the growth of the silicon carbide single crystal and experimental results, and the gas pressure measured by the pressure sensor.

10. 10. The method for producing a silicon carbide single crystal according to claim 9, wherein the measurement by the pressure sensor measures the gas pressure upstream of the gas outlet in the gas inlet pipe in a flow direction of the supply gas.

11. 11. The method for producing a silicon carbide single crystal according to claim 9, further comprising calculating and outputting an optimum process for lengthening the closure time when growing the silicon carbide single crystal, based on the prediction of the closure time.

12. calculating and feeding back at least one condition of the type of gas, the flow rate, and the temperature of the reaction chamber of the supply gas that lengthens the closing time when growing the silicon carbide single crystal based on the prediction of the closing time; 11. The method for producing a silicon carbide single crystal according to claim 9, wherein, in growing the silicon carbide single crystal, the silicon carbide single crystal is grown based on the conditions calculated in the feedback.

Images