Method for optimizing PVT method SiC crystal growth based on machine learning

Abstract

The invention belongs to the field of PVT method crystal preparation, and provides a method for optimizing PVT method SiC crystal growth based on machine learning, and the method comprises the steps: obtaining a defect comparison table by integrating abnormal data when growth defects occur in a plurality of historical growth cycles of SiC crystals, and carrying out prediction analysis through the data of SiC crystal growth monitored in the current cycle, according to the method, growth data which can be abnormal during growth of the SiC crystal in the current period are found and integrated to obtain a predicted abnormal group, and the predicted abnormal group is compared with the defect comparison table to predict and judge the defect to be generated by the SiC crystal in the current period, so that the defect abnormity during growth of the SiC crystal can be predicted in advance, and the detection accuracy of the defect abnormity during growth of the SiC crystal is improved. SiC crystal growth can be optimized in advance conveniently, defects caused by SiC crystal growth are prevented, and the SiC crystal growth quality is improved.

Description

Technical Field

[0001] This invention belongs to the field of crystal growth optimization technology, specifically a method for optimizing SiC crystal growth using the PVT method based on machine learning. Background Technology

[0002] The PVT (Physical Vapor Transport) method is an important technique widely used in semiconductor material preparation, especially in the growth of silicon carbide (SiC) crystals. SiC, with its wide bandgap, high thermal conductivity, and high breakdown voltage, is an ideal material for high-temperature, high-power, and high-frequency electronic devices, and is widely used in LEDs, power devices, and ceramics. The PVT method is based on a solid-gas-solid transformation process. By heating the SiC source material to its evaporation temperature, it is transformed into a gaseous state, and then recrystallized in a cooling zone to form a single crystal. This method has the capability to produce high-purity, high-quality crystals.

[0003] However, although the PVT method can produce high-quality SiC crystals, defects such as dislocations and vacancies may still occur during the growth process. However, in the existing technology, there is a lack of means to predict, analyze and identify these defects. This makes it impossible to know and judge the possible defects in SiC crystal growth in advance. Without such prediction and identification analysis, it is impossible to guarantee the growth quality of SiC crystals and it is not convenient to provide a basis for subsequent optimization strategies for SiC crystal growth.

[0004] Therefore, this invention provides a method for optimizing SiC crystal growth using the PVT method based on machine learning. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0006] The technical solution adopted by this invention to solve its technical problem is: In a first aspect, the present invention provides a method for optimizing SiC crystal growth using the PVT method based on machine learning, comprising: Step 1: Based on the growth reports of SiC crystal in multiple historical growth cycles, obtain abnormal data when growth defects occur in SiC crystal in multiple historical growth cycles, and integrate the abnormal data to obtain a defect comparison table. Step 2: During the current growth cycle, monitor the growth data of SiC crystal growth in real time. Based on the processing and analysis of the growth data, identify abnormal growth data within the growth data according to the processing and analysis results, and integrate them to obtain the predicted abnormal group. Step 3: Compare and analyze the predicted anomaly group with the defect comparison table, and determine the defects that will occur in the SiC crystal in the current cycle based on the comparison and analysis results.

[0007] As a further technical solution of the present invention: the defect comparison table is generated in the following way: Abnormal data were acquired during each SiC crystal growth anomaly across multiple historical growth cycles and integrated into a defect control group. Based on multiple historical growth cycles, multiple defect control groups can be obtained. By summarizing and integrating all defect control groups, a defect control table can be obtained.

[0008] As a further technical solution of the present invention: the process of processing and analyzing the growth data is as follows: Based on any single growth data; The real-time monitored growth data are marked and connected in a two-dimensional coordinate system to obtain the growth change curve; Real-time analysis of the growth change curve is performed, specifically as follows: If the growth change curve exceeds the warning line but does not exceed the baseline, an analysis signal is generated; Based on the analyzed signal, the superboundary curve is obtained; Connect the two endpoints of the superboundary curve with a straight line to obtain the reference line; By analyzing the boundary curve and the reference line, the deviation area ratio MJ and the adjacent distance value JL are obtained; Through the formula: The linear value Xx is obtained, where s1 and s2 are preset proportionality coefficients, with s1 taking the value of 1.34 and s2 taking the value of 1.67; If the linear value Xx is greater than or equal to the linear threshold, it indicates that the superboundary curve is linear; If the linear value Xx is less than the linear threshold, it indicates that the superboundary curve is non-linear; Based on whether the out-of-bounds curve is linear or nonlinear, the abnormal time points are analyzed and obtained. If the abnormal time point is earlier than the end time point of the current growth cycle, it means that the growth data will exceed or reach the maximum standard value before the end of the current growth cycle, and the growth data will be marked as abnormal growth data. All the abnormal growth data obtained are summarized and integrated to obtain the predicted abnormality group.

[0009] As a further technical solution of the present invention: the method for obtaining the superboundary curve based on the analyzed signal is as follows: Within the coordinate system containing the growth change curve, the maximum standard value of the growth data is used as the reference value and marked on its Y-axis to obtain the reference point. A straight line parallel to the X-axis is drawn through the reference point to obtain the reference line. Below the reference line, a straight line parallel to the X-axis is drawn based on the warning value to obtain the warning line. Real-time analysis of the growth change curve is performed, specifically as follows: If the growth change curve exceeds the warning line but does not exceed the baseline, an analysis signal is generated.

[0010] As a further technical solution of the present invention: the method for obtaining the deviation area ratio MJ is as follows: Based on the analysis signal, the time point corresponding to the intersection between the growth change curve and the warning line is obtained and marked as the arrival time point; The portion of the growth change curve corresponding to the arrival time point and the time point of signal generation is marked as the super-boundary curve; Connect the two endpoints of the superboundary curve with a straight line to obtain the reference line; The area enclosed between the boundary curve and the reference line is measured to obtain the deviation area. The deviation area is then compared with the deviation area threshold to obtain the deviation area ratio, which is denoted as MJ.

[0011] As a further technical solution of the present invention: the method for obtaining the adjacent distance value JL is as follows: The portion of the cross-boundary curve that does not overlap with the reference line is marked as a non-overlapping curve. Several sampling points are set on the non-coincident curves. The shortest distance between each sampling point and the reference line is measured and the sum is averaged to obtain the average of adjacent distances. The average of adjacent distances is then compared with the length of the reference line to obtain the adjacent distance value, which is marked as JL.

[0012] As a further technical solution of the present invention: the method for obtaining the abnormal time point is as follows: Since the out-of-bounds curve is linear, the reference line is extended to intersect with the baseline, and the time point corresponding to the intersection point on the X-axis is obtained and marked as the abnormal time point. Since the superboundary curve is nonlinear, growth data at different time points on the superboundary curve are obtained and integrated to obtain multiple sets of data parameters. Based on multiple sets of data parameters, 70% of the data parameters are used as the training set and 30% of the data parameters are used as the test set. The random forest model is trained using the training set and the test set to obtain the trained random forest model. The random forest model is then used to predict abnormal time points.

[0013] As a further technical solution of the present invention: the process of determining the defects that will occur in the SiC crystal in the current period based on the comparative analysis results is as follows: The predicted abnormal group was compared with the defect control group in the defect control table to obtain the data deviation ratio PC and the data same value XT. The obtained data deviation ratio (PC) and the data equality value (XT) are processed using the formula: The matching value Pp is obtained; where W1 and W2 are preset ratio coefficients, with W1 taking the value of 3.9 and W2 taking the value of 4.62. If the matching value is greater than or equal to the matching threshold, it indicates that the predicted abnormal group and the defect control group have a high degree of matching. After comparing the predicted abnormal group with all defect control groups, the defects corresponding to all defect control groups that have a high degree of matching with the predicted abnormal group are the defects that will occur in the SiC crystal in the current period.

[0014] As a further technical solution of the present invention: the method for obtaining the data deviation ratio (PC) is as follows: Based on any single defect control group; The number of abnormal growth data in the abnormal prediction group and the number of abnormal data in the defect control group are statistically analyzed, and the difference is processed to obtain the data deviation. The data deviation is then compared with the number of abnormal data in the defect control group to obtain the data deviation ratio, which is denoted as PC.

[0015] Secondly, the present invention provides a system for optimizing SiC crystal growth using the PVT method based on machine learning, comprising: The comparison data module: Based on the growth reports of SiC crystals in multiple historical growth cycles, it obtains abnormal data when growth defects occur in SiC crystals in multiple historical growth cycles, and integrates the abnormal data to obtain a defect comparison table; Predictive analysis module: During the current growth cycle, the growth data of SiC crystal growth is monitored in real time. Based on the processing and analysis of the growth data, abnormal growth data is identified in the growth data according to the processing and analysis results, and the abnormal data is integrated to obtain the predicted abnormal group. Defect Analysis Module: Compares and analyzes the predicted anomaly groups with the defect comparison table, and determines the defects that will occur in the SiC crystal in the current cycle based on the comparison and analysis results.

[0016] The beneficial effects of this invention are as follows: First, this invention integrates abnormal data on growth defects of SiC crystals in multiple historical growth cycles to obtain a defect comparison table. Then, it uses data from monitoring SiC crystal growth in the current cycle for predictive analysis to identify abnormal growth data that may occur during SiC crystal growth in the current cycle, and integrates these data to obtain a predicted abnormality group. By comparing the predicted abnormality group with the defect comparison table, the defects that will occur in the SiC crystal in the current cycle can be predicted and judged. This invention is beneficial for predicting defects and anomalies during SiC crystal growth in advance, facilitating early optimization of SiC crystal growth, preventing defects caused by SiC crystal growth, and improving the quality of SiC crystal growth. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart illustrating the steps of a method for optimizing SiC crystal growth using the PVT method based on machine learning, as described in an embodiment of the present invention. Figure 2 This is a flowchart of a system for optimizing SiC crystal growth using the PVT method based on machine learning, as described in an embodiment of the present invention. Detailed Implementation

[0019] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0020] Example 1 like Figure 1 As shown in the embodiment of the present invention, a method for optimizing SiC crystal growth using the PVT method based on machine learning includes: Step 1: Based on the growth reports of SiC crystal in multiple historical growth cycles, obtain abnormal data when growth defects occur in SiC crystal in multiple historical growth cycles, and integrate the abnormal data to obtain a defect comparison table. Among them, the abnormal data of SiC crystal when growth defects occur in multiple historical growth cycles include, but are not limited to, data such as temperature, pressure and inert gas flow rate. Specifically, the defect comparison table is generated as follows: Abnormal data were acquired during each SiC crystal growth anomaly across multiple historical growth cycles and integrated into a defect control group. Based on multiple historical growth cycles, multiple defect control groups can be obtained. By summarizing and integrating all defect control groups, a defect control table can be obtained. For example, if a SiC crystal develops growth defects in three historical growth cycles, the abnormal data corresponding to the defects in the first historical growth cycle are temperature and pressure, the abnormal data corresponding to the defects in the second historical growth cycle are temperature and inert gas flow rate, and the abnormal data corresponding to the defects in the third historical growth cycle are temperature, pressure, and inert gas flow rate, then the generated defect control groups are (temperature, pressure), (temperature, inert gas flow rate), and (temperature, pressure, inert gas flow rate), respectively. Step 2: During the current growth cycle, monitor the growth data of SiC crystal growth in real time. Based on the processing and analysis of the growth data, identify abnormal growth data within the growth data according to the processing and analysis results, and integrate them to obtain the predicted abnormal group. Among them, the growth data and the abnormal data are the same, and also include, but are not limited to, data such as temperature, pressure and inert gas flow rate; It should be noted that abnormal data and production data are of the same type, but the difference is that abnormal data indicates that the data exceeds the standard maximum value. Specifically, the process of processing and analyzing growth data is as follows: Based on any single growth data; Real-time growth data is monitored, and a two-dimensional coordinate system is constructed with time as the X-axis and growth data as the Y-axis. The real-time monitored growth data is marked and connected in the two-dimensional coordinate system to obtain the growth change curve. Within the coordinate system containing the growth change curve, the maximum standard value of the growth data is used as the reference value and marked on its Y-axis to obtain the reference point. A straight line parallel to the X-axis is drawn through the reference point to obtain the reference line. Below the reference line, a straight line parallel to the X-axis is drawn based on the warning value to obtain the warning line. It should be noted that the warning line was set by those skilled in the art based on experience, and its purpose is to prevent growth data from exceeding the maximum standard value; Real-time analysis of the growth change curve is performed, specifically as follows: If the growth curve does not exceed the warning line during real-time monitoring, it indicates that the growth data is normal, and no operation is performed; monitoring continues. If the growth change curve exceeds the warning line but does not exceed the baseline, an analysis signal is generated; It should be noted that the meaning of the analysis signal is: the generation of the analysis signal indicates that the growth data has approached the maximum standard value and needs to be analyzed and processed to prevent the growth data from exceeding the maximum standard value and causing harm to the SiC crystal growth. Based on the analysis signal, the time point corresponding to the intersection between the growth change curve and the warning line is obtained and marked as the arrival time point; The portion of the growth change curve corresponding to the arrival time point and the time point of signal generation is marked as the super-boundary curve; Connect the two endpoints of the superboundary curve with a straight line to obtain the reference line; The area enclosed between the boundary curve and the reference line is measured to obtain the deviation area. The deviation area is then compared with the deviation area threshold to obtain the deviation area ratio, which is denoted as MJ. It should be noted that the deviation area threshold is a reference value set by those skilled in the art based on experience; The portion of the cross-boundary curve that does not overlap with the reference line is marked as a non-overlapping curve. Several sampling points are set on the non-coincident curves. The shortest distance between each sampling point and the reference line is measured, and the sum and average values ​​are obtained to get the average of adjacent distances. The average of adjacent distances is then compared with the length of the reference line to obtain the adjacent distance value, which is marked as JL. The obtained deviation area ratio MJ and adjacent distance value JL are processed using the formula: The linear value Xx is obtained, where s1 and s2 are preset proportionality coefficients, with s1 taking the value of 1.34 and s2 taking the value of 1.67; Compare the linear value Xx with the linear threshold; It should be noted that the linear threshold is a reference value set by those skilled in the art based on experience; If the linear value Xx is greater than or equal to the linear threshold, it indicates that the superboundary curve is linear; If the linear value Xx is less than the linear threshold, it indicates that the superboundary curve is non-linear; Since the outbound curve is linear, the reference line is extended to intersect the baseline, and the time point corresponding to the intersection point on the X-axis is obtained and marked as the abnormal time point (i.e. the time point when the growth data reaches the maximum standard value). Since the superboundary curve is nonlinear, growth data at different time points on the superboundary curve are obtained and integrated to obtain multiple sets of data parameters. For example, growth data at different time points on the superboundary curve are obtained, and a specific time point and its corresponding growth data are integrated into a data parameter group. For example, if the temperature is 2200 degrees Celsius at minute T, the integrated data parameter group is (T, 2200℃). Based on multiple sets of data parameters, 70% of the data parameters are used as the training set and 30% of the data parameters are used as the test set. The random forest model is trained using the training set and the test set to obtain the trained random forest model. The random forest model is then used to predict anomalous time points (i.e., the time points when the growth data reaches the maximum standard value). Compare the abnormal time points with the end time points of the current growth cycle; If the abnormal time point is earlier than the end time point of the current growth cycle, it means that the growth data will exceed or reach the maximum standard value before the end of the current growth cycle, and the growth data will be marked as abnormal growth data. If the abnormal time point is later than or equal to the end time point of the current growth cycle, it means that the growth data will not exceed or reach the maximum standard value before the end of the current growth cycle, and no operation will be performed. All the abnormal growth data obtained are summarized and integrated to obtain the predicted abnormality group; Step 3: Compare and analyze the predicted anomaly group with the defect comparison table, and determine the defects that will occur in the SiC crystal in the current period based on the comparison and analysis results; The predicted abnormal groups were compared with the defect control groups in the defect control table, specifically as follows: Based on any single defect control group; The number of abnormal growth data in the abnormal prediction group and the number of abnormal data in the defect control group were statistically analyzed, and the difference was processed to obtain the data deviation. The data deviation was then compared with the number of abnormal data in the defect control group to obtain the data deviation ratio, which was denoted as PC. If any abnormal growth data in the predicted abnormal group can be found in the defect control group, then the abnormal growth data is marked as data of the same type. If no abnormal data of the same type can be found in the defect control group, no action will be taken. The number of data of the same type is counted and compared with the number of abnormal data in the defect control group. The data with the same value is then marked as XT. The obtained data deviation ratio (PC) and the data equality value (XT) are processed using the formula: The matching value Pp is obtained; where W1 and W2 are preset ratio coefficients, with W1 taking the value of 3.9 and W2 taking the value of 4.62. Compare the matched value with the matching threshold; It should be noted that the matching threshold is a reference value set by those skilled in the art based on experience; If the matching value is greater than or equal to the matching threshold, it indicates that the predicted abnormal group and the defect control group have a high degree of matching. If the matching value is less than the matching threshold, it indicates that the predicted abnormal group and the defect control group have a low degree of matching. After comparing the predicted abnormal group with all defect control groups, the defects corresponding to all defect control groups that have a high degree of matching with the predicted abnormal group are the defects that will occur in the SiC crystal in the current period. The technical solution of this invention is as follows: First, this invention integrates abnormal data on growth defects of SiC crystals in multiple historical growth cycles to obtain a defect comparison table. Then, it performs predictive analysis on the data of SiC crystal growth monitored in the current cycle to find abnormal growth data that will occur during SiC crystal growth in the current cycle, and integrates them to obtain a predicted abnormality group. By comparing the predicted abnormality group with the defect comparison table, the defects that will occur in SiC crystals in the current cycle are predicted and judged. This invention is beneficial for predicting defects and anomalies during SiC crystal growth in advance, facilitating early optimization of SiC crystal growth, preventing defects caused by SiC crystal growth, and improving the quality of SiC crystal growth.

[0021] Example 2 like Figure 2 As shown in the embodiment of the present invention, a system for optimizing SiC crystal growth using the PVT method based on machine learning includes: The comparison data module: Based on the growth reports of SiC crystals in multiple historical growth cycles, it obtains abnormal data when growth defects occur in SiC crystals in multiple historical growth cycles, and integrates the abnormal data to obtain a defect comparison table; Predictive analysis module: During the current growth cycle, the growth data of SiC crystal growth is monitored in real time. Based on the processing and analysis of the growth data, abnormal growth data is identified in the growth data according to the processing and analysis results, and the abnormal data is integrated to obtain the predicted abnormal group. Defect Analysis Module: Compares and analyzes the predicted anomaly groups with the defect comparison table, and determines the defects that will occur in the SiC crystal in the current cycle based on the comparison and analysis results.

[0022] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing SiC crystal growth using the PVT method based on machine learning, characterized in that: include: Step 1: Based on the growth reports of SiC crystal in multiple historical growth cycles, obtain abnormal data when growth defects occur in SiC crystal in multiple historical growth cycles, and integrate the abnormal data to obtain a defect comparison table. Step 2: During the current growth cycle, monitor the growth data of SiC crystal growth in real time. Based on the processing and analysis of the growth data, identify abnormal growth data within the growth data according to the processing and analysis results, and integrate them to obtain the predicted abnormal group. Step 3: Compare and analyze the predicted anomaly group with the defect comparison table, and determine the defects that will occur in the SiC crystal in the current cycle based on the comparison and analysis results.

2. The method for optimizing SiC crystal growth using PVT based on machine learning according to claim 1, characterized in that: The defect comparison table is generated as follows: Abnormal data were acquired during each SiC crystal growth anomaly across multiple historical growth cycles and integrated into a defect control group. Based on multiple historical growth cycles, multiple defect control groups can be obtained. By summarizing and integrating all defect control groups, a defect control table can be obtained.

3. The method for optimizing SiC crystal growth using the PVT method based on machine learning according to claim 1, characterized in that: The process of processing and analyzing the growth data is as follows: Based on any single growth data; The real-time monitored growth data are marked and connected in a two-dimensional coordinate system to obtain the growth change curve; Real-time analysis of the growth change curve is performed, specifically as follows: If the growth change curve exceeds the warning line but does not exceed the baseline, an analysis signal is generated; Based on the analyzed signal, the superboundary curve is obtained; Connect the two endpoints of the superboundary curve with a straight line to obtain the reference line; By analyzing the boundary curve and the reference line, the deviation area ratio MJ and the adjacent distance value JL are obtained; Through the formula: The linear value Xx is obtained, where s1 and s2 are preset proportionality coefficients, with s1 taking the value of 1.34 and s2 taking the value of 1.67; If the linear value Xx is greater than or equal to the linear threshold, it indicates that the superboundary curve is linear; If the linear value Xx is less than the linear threshold, it indicates that the superboundary curve is non-linear; Based on whether the out-of-bounds curve is linear or nonlinear, the abnormal time points are analyzed and obtained. If the abnormal time point is earlier than the end time point of the current growth cycle, it means that the growth data will exceed or reach the maximum standard value before the end of the current growth cycle, and the growth data will be marked as abnormal growth data. All the abnormal growth data obtained are summarized and integrated to obtain the predicted abnormality group.

4. The method for optimizing SiC crystal growth using PVT based on machine learning according to claim 3, characterized in that: The method for obtaining the superboundary curve based on the analyzed signal is as follows: Within the coordinate system containing the growth change curve, the maximum standard value of the growth data is used as the reference value and marked on its Y-axis to obtain the reference point. A straight line parallel to the X-axis is drawn through the reference point to obtain the reference line. Below the reference line, a straight line parallel to the X-axis is drawn based on the warning value to obtain the warning line. Real-time analysis of the growth change curve is performed, specifically as follows: If the growth change curve exceeds the warning line but does not exceed the baseline, an analysis signal is generated.

5. The method for optimizing SiC crystal growth using PVT based on machine learning according to claim 3, characterized in that: The method for obtaining the deviation area ratio MJ is as follows: Based on the analysis signal, the time point corresponding to the intersection between the growth change curve and the warning line is obtained and marked as the arrival time point; The portion of the growth change curve corresponding to the arrival time point and the time point of signal generation is marked as the super-boundary curve; Connect the two endpoints of the superboundary curve with a straight line to obtain the reference line; The area enclosed between the boundary curve and the reference line is measured to obtain the deviation area. The deviation area is then compared with the deviation area threshold to obtain the deviation area ratio, which is denoted as MJ.

6. The method for optimizing SiC crystal growth using PVT based on machine learning according to claim 3, characterized in that: The method for obtaining the adjacent distance value JL is as follows: The portion of the cross-boundary curve that does not overlap with the reference line is marked as a non-overlapping curve. Several sampling points are set on the non-coincident curves. The shortest distance between each sampling point and the reference line is measured and the sum is averaged to obtain the average of adjacent distances. The average of adjacent distances is then compared with the length of the reference line to obtain the adjacent distance value, which is marked as JL.

7. The SiC crystal growth method based on machine learning optimization of PVT method according to claim 3, characterized in that: The method for obtaining the abnormal time points is as follows: Since the out-of-bounds curve is linear, the reference line is extended to intersect with the baseline, and the time point corresponding to the intersection point on the X-axis is obtained and marked as the abnormal time point. Since the superboundary curve is nonlinear, growth data at different time points on the superboundary curve are obtained and integrated to obtain multiple sets of data parameters. Based on multiple sets of data parameters, 70% of the data parameters are used as the training set and 30% of the data parameters are used as the test set. The random forest model is trained using the training set and the test set to obtain the trained random forest model. The random forest model is then used to predict abnormal time points.

8. The SiC crystal growth method based on machine learning optimization of PVT method according to claim 1, characterized in that: The process of determining the defects that will occur in the SiC crystal during the current period based on the comparative analysis results is as follows: The predicted abnormal group was compared with the defect control group in the defect control table to obtain the data deviation ratio PC and the data same value XT. The obtained data deviation ratio (PC) and the data equality value (XT) are processed using the formula: The matching value Pp is obtained; where W1 and W2 are preset ratio coefficients, with W1 taking the value of 3.9 and W2 taking the value of 4.

62. If the matching value is greater than or equal to the matching threshold, it indicates that the predicted abnormal group and the defect control group have a high degree of matching. After comparing the predicted abnormal group with all defect control groups, the defects corresponding to all defect control groups that have a high degree of matching with the predicted abnormal group are the defects that will occur in the SiC crystal in the current period.

9. The SiC crystal growth method based on machine learning optimization of PVT method according to claim 8, characterized in that: The data deviation ratio (PC) is obtained as follows: Based on any single defect control group; The number of abnormal growth data in the abnormal prediction group and the number of abnormal data in the defect control group are statistically analyzed, and the difference is processed to obtain the data deviation. The data deviation is then compared with the number of abnormal data in the defect control group to obtain the data deviation ratio, which is denoted as PC.

10. The SiC crystal growth method based on machine learning optimization of PVT method according to claim 8, characterized in that: The method for obtaining the data same value XT is as follows: based on any abnormal growth data in the predicted abnormal group, if the same type of abnormal data can be found in the defect control group, then the abnormal growth data is marked as the same type of data. If no abnormal data of the same type can be found in the defect control group, no action will be taken. The number of data of the same type is counted and compared with the number of abnormal data in the defect control group. The data with the same value is then marked as XT.

Images