Method and system for detecting anomalies in the manufacturing of single crystals

Abstract

This invention provides a method and system for detecting abnormalities in single crystal manufacturing that can eliminate defects caused by operators overlooking anomalies, and furthermore, since the detection is automated, it eliminates the need for operators to check, leading to labor savings. [Solution] The present invention provides a method for detecting abnormalities in the production of a single crystal 2, which includes a heating and melting step of heating and melting a single crystal raw material placed in a crucible 4 with a heating source 5 to produce a raw material melt 3, and a pulling step of bringing a seed crystal into contact with the raw material melt 3 and producing a single crystal 2 by a rotational pulling method, wherein in the pulling step, image data of the portion in contact with the seed crystal in the raw material melt 3, or image data of the single crystal 2, obtained by an image sensor, is processed by a computing device incorporating a pre-trained model, and abnormalities in the production of the single crystal are detected from the results of the calculation processing. The single crystal production abnormality detection system of the present invention is a system that implements the single crystal production abnormality detection method of the present invention.

Description

Technical Field

[0001] The present invention relates to a method for detecting abnormalities in the production of single crystals and an abnormality detection system.

Background Art

[0002] The pulling method, also called the Chokralski method, is a method of growing a single crystal by heating and melting a raw material for a single crystal in a crucible to form a raw material melt, then bringing a seed crystal into contact with the melt and pulling it up while rotating. This method has the advantage that since the single crystal does not come into contact with the crucible and no mechanical strain is generated by the crucible, it is easy to produce high-quality and large-diameter crystals. Therefore, single crystals such as silicon (Si), germanium (Ge), gallium arsenide (GaAs), yttrium aluminum garnet (Y3Al5O 12 ;YAG), sapphire (Al2O3), lithium niobate (LiNbO3;LN), lithium tantalate (LiTaO3;LT), bismuth germanate (Bi4Ge3O 12 ;BGO), etc. are produced by this method.

[0003] By the way, in the growth of a single crystal by the pulling method, first, a seeding operation of bringing a seed crystal into contact with a raw material melt is performed, and then crystal growth is started. First, a neck portion having the same thickness as the seed crystal is formed. Next, a cone portion is formed by gradually expanding the diameter, and then a straight body portion with a large diameter is formed to a desired distance. Finally, a tail portion is formed by reducing the diameter in the direction opposite to the cone portion.

[0004] As a device for automatically growing single crystals using the pulling method, for example, the single crystal growth device described in Patent Document 1 is known as prior art. In the single crystal growth device described in Patent Document 1, an image processing device derives information regarding the shape of the exposed portion of the single crystal from the image information output by an image sensor. Based on this shape information, the single crystal growth device of Patent Document 1 calculates the length of the portion of the single crystal that is in the melt, or the distance between the lower end of the portion of the single crystal in the melt and the bottom of the crucible, and outputs a sound when the distance between the lower end of the crystal and the bottom of the crucible falls below a certain level.

[0005] This device calculates the radius of the melt surface and the distance from the melt surface to the bottom of the seed crystal, performs various calculations based on the output of a weight sensor and an image processing device, and controls heaters, rotary lifting devices, display devices, sound output devices, etc., according to the calculated distance and the calculation results. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Patent No. 6953912 [Overview of the project] [Problems that the invention aims to solve]

[0007] In the crystal growth process of single crystals using the pulling method, it is ideal if the crystal growth proceeds smoothly and good quality products are produced without any problems. However, there is a certain probability that an anomaly may occur during crystal growth. This anomaly during crystal growth may prevent crystal growth from continuing altogether, or even if crystal growth occurs, cracks or other defects may develop during cooling and removal, resulting in unsatisfactory products.

[0008] Such abnormalities during crystal growth would be ideal if they could be detected numerically, but they cannot be detected from measured values ​​alone. Therefore, when it is necessary to determine abnormalities during crystal growth visually or through images, methods such as operators visually checking for abnormalities, or template matching (where a predetermined range is recognized by the equipment, and the presence or absence of abnormalities is checked based on that) have been used.

[0009] For example, after a seeding operation in which a seed crystal is brought into contact with the molten raw material, the temperature of the molten raw material may be outside the optimal temperature range (the optimal temperature is slightly higher than the melting point of the molten raw material). If the temperature of the molten raw material is too high, the temperature of the seed crystal will exceed its melting point, causing the seed crystal to melt and the contact between the molten raw material and the seed crystal to be broken. Conversely, if the temperature of the molten raw material is too low, supercooling will occur around the area where the seed crystal was in contact, causing the solidified area to spread.

[0010] In the former case, the contact between the molten raw material and the seed crystal is broken, making it impossible to continue crystal growth. In the latter case, even if crystal growth can continue, the solidified area created by supercooling has poor crystallinity, which leads to poor crystallinity in the crystals that are formed afterward, increasing the probability of cracking after cooling.

[0011] Another example is the formation of the cone portion after the neck portion has formed. Growth ridges (here, raised areas that form along the facet planes) appear in a specific direction of the crystal structure. Ideally, these growth ridges should extend regularly in one direction, but if they branch due to foreign matter or secondary nucleation, the crystal becomes polycrystalline and cracks after cooling.

[0012] Traditionally, operators would visually inspect for such abnormalities and, if found, repeat the crystal growth process. However, visual inspection of abnormalities was not easy, and even skilled operators sometimes missed them.

[0013] Furthermore, Reference Document 1 does not describe a method for determining whether or not there is an abnormality in the crystal growth of a single crystal using a single crystal growth apparatus. For this reason, the single crystal growth apparatus described in Patent Document 1 continues to grow the single crystal even if an abnormality occurs during crystal growth.

[0014] The present invention aims to solve the above problems by providing a method and system for detecting abnormalities in single crystal manufacturing that eliminate defects caused by operators overlooking abnormalities, and furthermore, by automating the determination so that operators do not need to check, thus leading to labor savings. [Means for solving the problem]

[0015] As a result of diligent research, the inventors of the present invention have found that the above problem can be solved by monitoring the inside of a single crystal in a single crystal manufacturing process, in which the crystal is grown in the neck, cone, and straight body portions after seeding and finally separated, using an image sensor, processing the obtained image data with a computing device, and automatically detecting the presence or absence of abnormalities based on the results, thereby completing the present invention. The present invention is as follows. [1] A method for detecting anomalies in the production of a single crystal, comprising a heating and melting step of heating and melting raw materials for single crystals placed in a crucible with a heat source to produce a raw material melt, and a pulling step of bringing a seed crystal into contact with the raw material melt and producing a single crystal by a rotational pulling method, A method for detecting an anomaly in the manufacturing of a single crystal, characterized in that, in the pulling-up process, image data of the portion in contact with the raw material molten liquid or image data of the single crystal obtained by an image sensor is processed by a computing device incorporating a pre-trained model, and an anomaly in the manufacturing of the single crystal is detected from the result of the calculation. [2] The method for detecting an anomaly in the production of a single crystal according to [1] above, characterized in that the pulling step includes a first step of bringing the seed crystal into contact with the raw material melt and pulling up the neck portion by the rotational pulling method. [3] The method for detecting an anomaly in the manufacturing of a single crystal according to [2] above, characterized in that the pulling step includes a second step of pulling up a cone portion that expands the diameter. [4] The method for detecting an anomaly in the manufacturing of a single crystal according to [3] above, characterized in that the pulling step includes a third step of pulling up the straight body portion which has the maximum diameter. [5] The method for detecting an anomaly in the production of a single crystal according to [4] above, characterized in that the pulling step includes a fourth step of separating the pulled single crystal from the raw material melt. [6] The method for detecting anomalies in the manufacturing of single crystals according to any one of [1] to [5] above, characterized in that the pre-trained model is formed by deep learning. [7] A method for detecting an anomaly in the manufacturing of a single crystal according to any one of the above [1] to [6], wherein the single crystal is an oxide single crystal. [8] The method for detecting anomalies in the manufacturing of a single crystal according to [5] above, characterized in that all steps, including the first to fourth steps, are performed automatically. [9] An anomaly detection system for single crystal production, comprising a heating and melting step of heating and melting raw materials for single crystals placed in a crucible with a heating source to produce a raw material melt, and a pulling step of bringing a seed crystal into contact with the raw material melt and producing a single crystal by a rotational pulling method, Equipped with a computing unit that incorporates a pre-trained model, The calculation device performs calculations on image data of the portion of the raw material molten that is in contact with the seed crystal, or on image data of the single crystal, obtained by the image sensor during the pulling process, and detects an abnormality in the manufacturing of the single crystal from the results of the calculations. This is a single crystal manufacturing anomaly detection system.

[10] The abnormality detection system for manufacturing a single crystal according to [9] above, characterized in that the pulling step includes a first step of bringing the seed crystal into contact with the raw material melt and pulling up the neck portion by the rotational pulling method.

[11] The abnormality detection system for manufacturing a single crystal according to

[10] above, characterized in that the pulling step includes a second step of pulling up a cone portion that expands the diameter.

[12] The abnormality detection system for manufacturing a single crystal according to

[11] above, characterized in that the pulling step includes a third step of pulling up a straight body portion that has the maximum diameter.

[13] The pulling-up process includes a fourth process of separating the pulled single crystal from the raw material melt, and is characterized in that it is the single crystal manufacturing abnormality detection system according to

[12] above.

[14] The pre-trained model is formed by deep learning, and is characterized in that it is the single crystal manufacturing abnormality detection system according to any one of [9] to

[13] above.

[15] The single crystal is an oxide single crystal, and is characterized in that it is the single crystal manufacturing abnormality detection system according to any one of [9] to

[14] above.

[16] All processes including the first to fourth processes are automatically executed, and is characterized in that it is the single crystal manufacturing abnormality detection system according to

[13] above.

Effect of the Invention

[0016] According to the present invention, it is possible to eliminate defects caused by overlooking operator abnormalities, and furthermore, since it is automatically determined, there is no need for an operator to confirm, leading to labor saving, and it is possible to provide a single crystal manufacturing abnormality detection method and a single crystal manufacturing abnormality detection system.

Brief Description of the Drawings

[0017] [Figure 1] FIG. 1 is a schematic diagram showing an example of a single crystal manufacturing apparatus. [Figure 2] FIG. 2 is a schematic diagram of an example of a single crystal manufacturing abnormality detection system incorporated in the single crystal manufacturing apparatus of FIG. 1. [Figure 3] FIG. 3 is a schematic diagram of a single crystal produced by the rotational pulling method. [Figure 4] FIG. 4(a) is a schematic diagram when the seed crystal contacts the raw material melt, FIG. 4(b) is a schematic diagram when a solidified part occurs after the seed crystal contacts the raw material melt, and FIG. 4(c) is a schematic diagram when the seed crystal separates from the raw material melt. [Figure 5] FIG. 5(a) is a schematic diagram of the cone part of the single crystal viewed from the side, FIG. 5(b) is a schematic diagram of the cone part of the single crystal viewed from above, and FIG. 5(c) is a schematic diagram when the cone part of the single crystal with a branched growth edge is viewed from above. [Modes for carrying out the invention]

[0018] The manufacturing of single crystals includes a heating and melting step in which the raw material for single crystals placed in a crucible is heated and melted by a heat source to produce a raw material melt, and a pulling step in which a seed crystal is brought into contact with the raw material melt and a single crystal is manufactured by a rotational pulling method. The manufacturing of single crystals will be explained in detail with reference to Figure 1. Figure 1 is a schematic diagram showing an example of a single crystal manufacturing apparatus. In the single crystal manufacturing apparatus 1, the single crystal raw material is placed in a crucible 4 and heated by a heat source 5. The heat source 5 may be something like a resistance heater, or if the crucible 4 is conductive, it may be something like high-frequency heating. By heating, the temperature of the single crystal raw material is raised and melted to form a raw material melt 3. The temperature of the raw material melt 3 may be measured indirectly by placing a thermocouple 6 at the bottom of the crucible, or, although not shown in the figure, the temperature of the raw material melt may be measured directly from above using a radiation thermometer or the like. Alternatively, this temperature data may be input to a control unit 13 and PID control may be performed. A PLC (Programmable Logic Controller) or a personal computer can be used as the control unit 13.

[0019] Next, a pulling process is carried out to produce a single crystal by the rotational pulling method, in which a seed crystal is brought into contact with the molten raw material 3. This pulling process may include a first step of bringing the seed crystal into contact with the molten raw material 3 and pulling up the neck portion by the rotational pulling method. In addition to the first step, this pulling process may also include a second step of pulling up the cone portion that expands the diameter. In addition to the first and second steps, this pulling process may also include a third step of pulling up the straight body portion that becomes the maximum diameter. In addition to the first, second, and third steps, this pulling process may also include a fourth step of separating the pulled single crystal from the molten raw material. Through these steps, the single crystal is pulled up. An example of a single crystal produced by the rotational pulling method is shown in Figure 3. The control unit 13 can also control the single crystal pulling process. A PLC or personal computer can be used for control. The input data may be the crystal weight measured by the load cell 11, or the diameter of the single crystal calculated from the image obtained from the image sensor 12. The output is adjusted by the heating source 5, and PID control or similar methods can be used for the input data.

[0020] [Method for detecting anomalies in single crystal manufacturing] The following describes a method for detecting anomalies in the manufacturing of a single crystal according to one embodiment of the present invention. According to the method for detecting anomalies in the manufacturing of a single crystal according to one embodiment of the present invention, in any of the first to fourth steps, the image data obtained by the image sensor 12 is used by the arithmetic unit in the control unit 13 to determine whether or not there is an anomaly in the manufacturing of the single crystal from the image data using a pre-learned model, and an anomaly in the manufacturing of the single crystal can be automatically detected. This makes it easy to detect anomalies in single crystal manufacturing, even for operators who are not highly skilled, and also prevents operators from overlooking problems. Furthermore, the automated detection reduces the burden on operators.

[0021] In the first step, the contact area where the raw material molten liquid 3 and the seed crystal are in contact, and the surrounding area, are monitored by the image sensor 12. The image data obtained by the image sensor 12 is used by the arithmetic unit in the control unit 13 to determine whether there is an abnormality in the single crystal manufacturing process based on the image data, using a pre-trained model. If there is an abnormality in the single crystal manufacturing process, it can be automatically detected. If the temperature of the raw material melt 3 is not at the appropriate temperature after the seeding operation in which the seed crystal is brought into contact with the raw material melt 3, abnormalities will occur in the production of the single crystal. Images are prepared in advance for the following situations: when the temperature of the raw material melt 3 is at an appropriate temperature (see Figure 4(a)), when solidification occurs due to a low temperature of the raw material melt (see Figure 4(b)), and when contact between the raw material melt and the seed crystal is broken due to a high temperature of the raw material melt (see Figure 4(c)). Deep learning is then performed on these images. In this case, the goal of image classification by learning is to extract regularities, i.e., image features, for classifying "normal" and "abnormal" from the images of the raw material melt 3 at an appropriate temperature, when solidification occurs due to a low temperature, and when contact between the raw material melt and the seed crystal is broken due to a high temperature, so that all images of the seed crystal in contact with the raw material melt 3 can be correctly classified as "normal" or "abnormal". The observation data in learning consists of pairs of input data and corresponding labels. For example, if the input data is an image of the raw material melt 3 at an appropriate temperature, the label will be the class information "normal". If the input data is an image where solidification has occurred due to a low temperature of the raw material melt, the label will be the class information "abnormal". If the input data is an image where contact between the raw material melt and the seed crystal has been broken due to a high temperature of the raw material melt, the label will also be the class information "abnormal". Then, the model (for example, a multilayer neural network) is optimized so that when an image of raw material melt 3 at an appropriate temperature is input, it outputs "normal", when an image where solidification has occurred due to a low temperature of the raw material melt is input, it outputs "abnormal", and when an image where contact between the raw material melt and the seed crystal has been broken due to a high temperature is input, it outputs "abnormal". This optimization of the model is called "learning". Based on the learning results, it is possible to determine with a high probability whether or not there is an abnormality in the manufacturing of single crystals. In addition, an alarm sound may be sounded after an abnormality in the manufacturing of single crystals is detected.

[0022] In the second step, the growth ridges formed on the single crystal, particularly on the cone portion (see Figures 5(a) and (b)), are monitored by an image sensor 13. The image data obtained by the image sensor 13 is processed by a computing device incorporating a pre-trained model. From the results of the processing, abnormalities in the growth ridges are automatically detected, thereby detecting abnormalities in the single crystal manufacturing process. If foreign matter or secondary nucleation occurs during the growth of the cone, an anomaly occurs where the growth ridge branches (see Figure 5(c)). Images of normal growth ridges (see Figure 5(b)) and images of abnormal growth ridges (see Figure 5(c)) are prepared in advance, and training is performed using a deep learning method. Specifically, the model is optimized so that when an image of normal growth ridges is input to the model, it outputs "normal," and when an image of normal growth ridges is input to the model, it outputs "abnormal." Based on these training results, the presence or absence of growth ridge anomalies can be determined with a high probability, and as a result, anomalies in single crystal manufacturing can be detected with a high probability.

[0023] It is preferable that a pre-trained model capable of determining the presence or absence of growth ridge anomalies with high probability be formed using deep learning. In particular, the shapes of the growth ridges mentioned above are never the same, and each single crystal has a different shape. Therefore, no branched anomalies have the same shape, and it is difficult to obtain a high accuracy rate when trying to determine them using a fixed form such as template matching. However, it is possible to obtain a high accuracy rate by determining them using a method such as deep learning.

[0024] While convolutional neural networks (CNNs), known to be advantageous for image processing, are generally preferred as deep learning models, they are not the only options.

[0025] The single crystal may be an oxide single crystal. Oxide single crystals tend to have a higher melting point compared to other materials. Therefore, the light from inside the furnace during crystal growth is too strong, and although light shields are used to reduce the light, it is not easy for the operator to confirm this visually. Therefore, by using the image obtained by the image sensor 12, edges and other details can be displayed in more detail than by visual inspection.

[0026] The system is characterized by calculating the width of the solidified area around the contact point from images of the contact point where the raw material molten 3 and the seed crystal are in contact, using a calculation device, and automatically detecting abnormalities based on the width of the solidified area. Since the solidified area is at a lower temperature than the molten area, it will appear slightly darker in the image. A reference length may be determined within the image data, or it may be determined by the number of pixels, and the width of the solidified area is calculated from the result. For example, programming software such as Python may be used to automatically calculate the length from the image data. The width of the solidified area is preferably 4 mm or less. More preferably 2 mm or less.

[0027] This system is characterized by the fact that all processes, including the first to fourth steps, are executed automatically. Because all processes are executed automatically, labor savings can be expected. Furthermore, variations due to operator error can be reduced, leading to the expectation of stable operating rates and yields.

[0028] The method for detecting anomalies in the manufacturing of a single crystal according to one embodiment of the present invention is an example of the method for detecting anomalies in the manufacturing of a single crystal according to the present invention, and does not limit the method for detecting anomalies in the manufacturing of a single crystal according to the present invention.

[0029] [Anomaly detection system for single crystal manufacturing] The following describes an anomaly detection system for single crystal manufacturing according to one embodiment of the present invention. The present invention relates to an anomaly detection system for single crystal manufacturing, which includes a heating and melting step in which single crystal raw materials placed in a crucible are heated and melted by a heating source to produce a raw material molten liquid, and a pulling step in which a seed crystal is brought into contact with the raw material molten liquid and a single crystal is manufactured by a rotational pulling method. The single crystal manufacturing anomaly detection system according to one embodiment of the present invention comprises a computing device into which a pre-trained model is incorporated, and the computing device performs computational processing on image data of the portion in contact with the raw material molten liquid obtained by an image sensor or image data of the single crystal during the pulling step. The single crystal manufacturing anomaly detection system according to one embodiment of the present invention then detects an anomaly in single crystal manufacturing from the results of the computational processing. This makes it easy to detect anomalies in single crystal manufacturing, even for operators who are not highly skilled, and also prevents operators from overlooking problems. Furthermore, the automated detection reduces the burden on operators.

[0030] The manufacturing of single crystals has been described in the method for detecting anomalies in the manufacturing of single crystals according to one embodiment of the present invention, so the description of the manufacturing of single crystals will be omitted. Furthermore, the calculation device in the system for detecting anomalies in the manufacturing of single crystals according to one embodiment of the present invention has been described in the method for detecting anomalies in the manufacturing of single crystals according to one embodiment of the present invention, so the description of the calculation device will be omitted.

[0031] The single crystal manufacturing anomaly detection system of one embodiment of the present invention is an example of the single crystal manufacturing anomaly detection system of the present invention, and is not limited to the single crystal manufacturing anomaly detection system of the present invention. [Examples]

[0032] The present invention will be described in more detail below with reference to examples and comparative examples, but the present invention is not limited to these examples.

[0033] [Example 1] Lithium tantalate single crystal raw materials were placed in a crucible, heated and melted in a high-frequency induction heating furnace, and a 6-inch diameter lithium tantalate single crystal was produced by the rotary pulling method. A schematic diagram of the manufacturing apparatus used to produce the lithium tantalate single crystal is shown in Figure 1. The lithium tantalate manufacturing process was carried out using this apparatus, with the following four steps performed automatically. The process involves contacting a lithium tantalate seed crystal with a lithium tantalate molten raw material, followed by the first step of pulling up the neck portion, the second step of pulling up the cone portion which expands the diameter, the third step of pulling up the straight body portion which has the maximum diameter, and the fourth step of separating the pulled-up single crystal from the raw material molten material. Finally, the lithium tantalate single crystal is cooled and removed. This series of processes (steps 1 through 4) was considered one lifting operation, and a total of 491 lifting operations were carried out.

[0034] During the series of processes, from the first to the fourth step, the inside of the furnace was monitored by an image sensor 12 installed on the top of the apparatus. The images acquired by the image sensor 12 were processed within the anomaly detection system shown in Figure 2. The anomaly detection system incorporated a computer as a control unit 13, which processed the video from the image sensor 12 in real time. When an anomaly was detected, it was detected, and an alarm sound was emitted by the sound generator 14. Furthermore, YOLOv5 (a model that can perform object detection from the results of deep learning) was implemented in the programming software Python for anomaly detection. For training, 500 images of seedling anomalies and 200 images of growth ridge anomalies were prepared and used for pre-training.

[0035] Out of 491 pull-ups, seeding abnormalities occurred 82 times in the first process. These abnormalities included the formation of a solidified area around the seed crystal in 70 instances, and a loss of contact between the molten raw material and the seed crystal in 12 instances. The formation of solidified areas was determined by observing the neck of the lithium tantalate single crystal after removal, checking for any sections with a diameter wider than the set diameter. When these abnormalities were checked during the pull-up process using an abnormality detection system, 75 out of 82 instances were detected, achieving a detection rate of 91.5%.

[0036] [Example 2] Following the first step in Example 1, abnormalities in the growth ridges of the cone portion were also confirmed in the second step.

[0037] Out of 491 pull-ups, abnormalities in the growth ridges of the cone occurred 41 times during the second process. These growth ridge abnormalities were also observed and determined after the lithium tantalate single crystal was extracted. When these abnormalities were checked during the pull-up process using the abnormality detection system, 39 out of 41 were detected, achieving a detection rate of 95.1%.

[0038] The results of the above examples are shown in the table below. [Table 1] [Explanation of Symbols]

[0039] 1. Single crystal manufacturing apparatus 2 Single crystals 2S Seed Section 2N Neck section 2C Cone Section 2B Straight body part 2T Tail section 2R growth ridge 3 Raw material melt 4 Crucibles 5 Heating source 6 Thermocouples 11 Load Cells 12 Image sensors 13 Control Unit 14. Sound Generator SS solidification part Abnormalities in AR growth ridges

Claims

1. A method for detecting abnormalities in the production of a single crystal, comprising a heating and melting step of heating and melting raw materials for single crystals placed in a crucible with a heat source to produce a raw material melt, and a pulling step of bringing a seed crystal into contact with the raw material melt and producing a single crystal by a rotational pulling method, A method for detecting an anomaly in the manufacturing of a single crystal, characterized in that, in the pulling-up process, image data of the portion in contact with the raw material molten liquid or image data of the single crystal obtained by an image sensor is processed by a computing device incorporating a pre-trained model, and an anomaly in the manufacturing of the single crystal is detected from the result of the calculation.

2. The method for detecting an anomaly in the production of a single crystal according to claim 1, characterized in that the pulling step includes a first step of bringing the seed crystal into contact with the raw material molten liquid and pulling up the neck portion by the rotational pulling method.

3. The method for detecting anomalies in the manufacturing of a single crystal according to claim 2, characterized in that the pulling-up step includes a second step of pulling up a cone portion that expands the diameter.

4. The method for detecting anomalies in the manufacturing of a single crystal according to claim 3, characterized in that the pulling-up step includes a third step of pulling up the straight body portion which has the maximum diameter.

5. The method for detecting an anomaly in the production of a single crystal according to claim 4, characterized in that the pulling step includes a fourth step of separating the pulled single crystal from the raw material melt.

6. The method for detecting anomalies in the manufacturing of single crystals according to claim 1, characterized in that the pre-trained model is formed by deep learning.

7. The method for detecting anomalies in the manufacturing of a single crystal according to claim 1, wherein the single crystal is an oxide single crystal.

8. The method for detecting anomalies in the manufacturing of a single crystal according to claim 5, characterized in that all steps, including the first to fourth steps, are performed automatically.

9. An anomaly detection system for single crystal manufacturing, comprising a heating and melting step of heating and melting single crystal raw materials placed in a crucible with a heat source to produce a raw material melt, and a pulling step of bringing a seed crystal into contact with the raw material melt to manufacture a single crystal by a rotational pulling method, Equipped with a computing unit that incorporates a pre-trained model, The calculation device performs calculations on image data of the portion of the raw material molten that is in contact with the seed crystal, or on image data of the single crystal, obtained by the image sensor during the pulling process, and detects an abnormality in the manufacturing of the single crystal from the results of the calculations. This is a single crystal manufacturing anomaly detection system.

10. The abnormality detection system for single crystal manufacturing according to claim 9, characterized in that the pulling step includes a first step of bringing the seed crystal into contact with the raw material molten liquid and pulling up the neck portion by the rotational pulling method.

11. The abnormality detection system for single crystal manufacturing according to claim 10, characterized in that the pulling-up step includes a second step of pulling up a cone portion that expands the diameter.

12. The abnormality detection system for single crystal manufacturing according to claim 11, characterized in that the pulling-up step includes a third step of pulling up the straight body portion which has the maximum diameter.

13. The single crystal manufacturing anomaly detection system according to claim 12, characterized in that the pulling step includes a fourth step of separating the pulled single crystal from the raw material melt.

14. The anomaly detection system for single crystal manufacturing according to claim 9, characterized in that the pre-trained model is formed by deep learning.

15. The single crystal manufacturing anomaly detection system according to claim 9, wherein the single crystal is an oxide single crystal.

16. The single crystal manufacturing anomaly detection system according to claim 13, characterized in that all processes, including the first to fourth processes, are performed automatically.

Images