Crystal growth process optimization control system and method based on machine learning

Abstract

The invention relates to the technical field of crystal growth, and discloses a machine learning-based crystal growth process optimization control system and method, and the system comprises a difference detection module which is used for obtaining a crystal contour of a current crystal growth image, and obtaining a crystal growth difference degree; the model establishment module is used for acquiring historical crystal growth data and establishing a growth morphological parameter prediction model according to the historical crystal growth data; the morphology prediction module is used for inputting the current crystal growth difference degree into a growth morphology parameter prediction model to obtain a predicted growth morphology parameter of the current crystal; and the temperature setting module is used for determining the temperature required by growth according to the predicted growth morphological parameters of the current crystal and correcting the temperature required by growth according to the historical crystal growth data. The growth morphology of the crystal is predicted by establishing a growth morphology parameter prediction model, so that the optimal temperature required by crystal growth is obtained, and the crystal growth quality is improved.

Description

Technical Field

[0001] The present application relates to the field of crystal growth technology, and more specifically, to a crystal growth process optimization control system and method based on machine learning. Background Art

[0002] With the industrial development of integrated circuit science and technology and modern information technology based on silicon materials, the derived technical products have been integrated into all aspects of human social development and become an important symbol of a country's comprehensive national economic strength. With the continuous enhancement of national strength and the rapid development of the national economy, my country has become one of the regions with the most intensive investment in the international semiconductor chip manufacturing industry, and has become the hotbed of the fastest growth and most dynamic technological innovation in the global semiconductor chip manufacturing industry.

[0003] At present, the technology of crystal growth equipment used for crystal preparation is relatively mature. However, in actual crystal preparation, actual equipment control still relies heavily on manual experience. There is a defect of insufficient effectiveness of crystal growth control, and it is impossible to maintain the optimal temperature during the crystal growth process, which affects the quality of crystal growth. Summary of the invention

[0004] The present invention provides a crystal growth process optimization control system and method based on machine learning, which is used to solve the problem that the crystal growth control effectiveness in the prior art is insufficient and the optimal temperature cannot be maintained during the crystal growth process, including: A difference detection module is used to obtain a crystal contour of a current crystal growth image, determine a shape regularity of the crystal according to the crystal contour, and obtain a first crystal difference; Obtaining a crystal profile of a previous preset period, calculating distance values ​​in various directions between the current crystal profile and the crystal profile of the previous preset period, obtaining differences between each distance value and an average distance value, determining a second crystal difference according to the differences between each distance value and the average distance value, and multiplying the first crystal difference by the second crystal difference to obtain a crystal growth difference; A model building module is used to obtain historical crystal growth data and build a growth morphology parameter prediction model based on the historical crystal growth data; A morphology prediction module is used to input the current crystal growth difference into the growth morphology parameter prediction model to obtain the predicted growth morphology parameters of the current crystal; The temperature setting module is used to determine the required growth temperature according to the predicted growth morphology parameters of the current crystal, and to correct the required growth temperature according to the historical crystal growth data.

[0005] Furthermore, the difference detection module determines the crystal growth difference according to the current crystal growth image, including: Obtain the current crystal growth image, and perform edge detection on the current crystal growth image based on the Canny edge detection algorithm; Extracting the crystal contour according to the edge detection result, determining the shape regularity of the crystal according to the crystal contour, and obtaining the first crystal difference; Obtaining a crystal profile of a previous preset period, calculating distance values ​​in various directions between the current crystal profile and the crystal profile of the previous preset period, determining differences between each distance value and an average distance value according to the distance values, normalizing the sum of the differences between each distance value and the average distance value, and obtaining a second crystal difference; The first crystal difference is multiplied by the second crystal difference to obtain the crystal growth difference.

[0006] Further, determining the shape regularity of the crystal according to the crystal contour includes: Obtain the center point position of the crystal contour, and generate a number of radial lines starting from the center point position of the crystal contour and ending at the contour edge at a preset angle; The lengths of the radiation lines are counted, and the number of radiation lines with unequal lengths is determined as the shape regularity of the crystal.

[0007] Furthermore, the model building module builds a growth morphology parameter prediction model based on historical crystal growth data, including: Determine historical crystal growth differences and index parameters and corresponding crystal growth morphology parameters according to historical crystal growth data, wherein the index parameters include liquid surface temperature, crystal rotation speed, and crucible rotation speed; A training sample set is established according to the crystal growth difference and index parameters of the historical crystal growth process and the corresponding crystal growth morphological parameters, an initial growth morphological parameter prediction model is established, and the initial growth morphological parameter prediction model is trained according to the training sample set to obtain a trained growth morphological parameter prediction model.

[0008] Furthermore, the temperature setting module determines the required growth temperature according to the predicted growth morphology parameters of the current crystal, including: Obtaining standard growth morphology parameters, and calculating the difference between the growth morphology parameters of the current crystal and the standard growth morphology parameters; Determine whether the difference between the growth morphology parameter of the current crystal and the standard growth morphology parameter is greater than a first preset threshold value, and if the difference between the growth morphology parameter of the current crystal and the standard growth morphology parameter is greater than the first preset threshold value, set the required growth temperature to the first temperature; If the difference between the growth morphology parameter of the current crystal and the standard growth morphology parameter is less than or equal to the first preset threshold, then determining whether the difference between the growth morphology parameter of the current crystal and the standard growth morphology parameter is greater than a second preset threshold; If the difference between the growth morphology parameter of the current crystal and the standard growth morphology parameter is greater than a second preset threshold, setting the growth temperature to the second temperature; If the difference between the growth morphology parameter of the current crystal and the standard growth morphology parameter is less than or equal to the second preset threshold, the temperature required for growth is set to the third temperature.

[0009] Furthermore, the temperature setting module corrects the temperature required for growth according to historical crystal growth data, including: Determine the growth rate of each period in the historical crystal growth process according to the historical crystal growth data, and draw a historical growth rate curve according to the growth rate of each period in the historical crystal growth process; Obtain a preset sliding time window, and divide the historical growth rate curve according to the preset sliding time window to obtain a plurality of sub-rate curves; Calculate the absolute value of the slope of the sub-speed curve, and draw a slope curve according to the absolute value of the slope of the sub-speed curve; Divide the slope curve into several curve segments, count the number of peak points in the curve segments, multiply the maximum slope value in the curve segments by the number of peak points to obtain the curve characteristic value; Each curve segment is clustered according to the characteristic value of the curve, and the temperature required for growth is corrected according to the cluster center of each curve segment.

[0010] Furthermore, clustering each curve segment according to the curve characteristic value includes: A curve feature data set is established according to the curve feature values, and k initial clustering centers of the curve feature data set are randomly selected; Calculate the Euclidean distance from the curve characteristic value in the curve characteristic data set to the initial cluster center, and divide each curve characteristic value into corresponding partitions according to the Euclidean distance from the curve characteristic value in the curve characteristic data set to the initial cluster center; Calculate the average value of the curve characteristic value in each partition, and re-determine the cluster center according to the average value of the curve characteristic value in each partition; Repeat the above steps until the cluster center no longer changes or the number of iterations reaches the preset iteration threshold, and obtain k final cluster centers.

[0011] Furthermore, the correction of the growth required temperature according to the cluster center of each curve segment includes: Obtaining a growth rate curve of the current crystal growth process, and calculating a current curve characteristic value according to the growth rate curve of the current crystal growth process; Calculate the crystal growth similarity according to the current curve characteristic value and the curve characteristic value of the cluster center, calculate the ratio of the crystal growth similarity to the preset allowable similarity, and obtain the temperature correction parameter; The temperature required for growth is multiplied by the temperature correction parameter to obtain the final temperature required for growth.

[0012] Further, the crystal growth similarity is calculated according to the current curve characteristic value and the curve characteristic value of the cluster center, including: The crystal growth similarity is calculated according to the crystal growth similarity calculation formula, and the crystal growth similarity calculation formula is specifically: , in, is the crystal growth similarity, is the number of curve segments, is the curve eigenvalue of the cluster center where the i-th curve segment is located, is the curve characteristic value of the i-th curve segment of the current growth rate curve.

[0013] In order to achieve the above object, the present invention also provides a crystal growth process optimization control method based on machine learning, comprising: Acquire a crystal contour of the current crystal growth image, determine the shape regularity of the crystal according to the crystal contour, and obtain a first crystal difference; Obtaining a crystal profile of a previous preset period, calculating distance values ​​in various directions between the current crystal profile and the crystal profile of the previous preset period, obtaining differences between each distance value and an average distance value, determining a second crystal difference according to the differences between each distance value and the average distance value, and multiplying the first crystal difference by the second crystal difference to obtain a crystal growth difference; Obtain historical crystal growth data, and establish a growth morphology parameter prediction model based on the historical crystal growth data; Inputting the current crystal growth difference into the growth morphology parameter prediction model to obtain the predicted growth morphology parameters of the current crystal; The required growth temperature is determined based on the predicted growth morphology parameters of the current crystal, and the required growth temperature is corrected based on historical crystal growth data.

[0014] The beneficial effects of the present invention are: By applying the above technical scheme, the present invention calculates the crystal growth difference by collecting the crystal growth image, detects the degree of deviation of the crystal during the growth process in real time, and establishes a growth morphology parameter prediction model to predict the growth morphology of the current crystal based on the deviation degree. According to the prediction result, the optimal temperature required for crystal growth is obtained to improve the crystal growth quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 The overall structure diagram of a crystal growth process optimization control system based on machine learning proposed in an embodiment of the present invention is shown; Figure 2 A flow chart of a method for optimizing and controlling a crystal growth process based on machine learning proposed in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0018] The present application embodiment provides a crystal growth process optimization control system based on machine learning, such as Figure 1 As shown, it includes: a difference detection module, which is used to obtain the crystal contour of the current crystal growth image, determine the shape regularity of the crystal according to the crystal contour, and obtain a first crystal difference; obtain the crystal contour of the previous preset period, calculate the distance values ​​in each direction between the current crystal contour and the crystal contour of the previous preset period, obtain the difference between each distance value and the average distance value, determine the second crystal difference according to the difference between each distance value and the average distance value, and multiply the first crystal difference by the second crystal difference to obtain the crystal growth difference; a model establishment module, which is used to obtain historical crystal growth data and establish a growth morphology parameter prediction model according to the historical crystal growth data; a morphology prediction module, which is used to input the current crystal growth difference into the growth morphology parameter prediction model to obtain the predicted growth morphology parameters of the current crystal; a temperature setting module, which is used to determine the required growth temperature according to the predicted growth morphology parameters of the current crystal, and correct the required growth temperature according to the historical crystal growth data.

[0019] In this embodiment, when crystal growth is carried out, the crystal growth image is collected in real time and the crystal contour of the crystal growth image is extracted by the canny edge detection algorithm, the first crystal difference is obtained by calculating the shape regularity of the crystal contour, the deviations in multiple directions of the current crystal contour and the crystal contour of the previous preset period are compared to obtain the difference between each distance value and the average distance value, the sum of the differences between each distance value and the average distance value is normalized to obtain the second crystal difference, thereby calculating the crystal growth difference, and establishing a machine learning model through historical crystal growth data to predict the growth morphological parameters of the crystal, the growth morphological parameters specifically being the diameter, volume and length change of the crystal, the required temperature for the current crystal growth is set through the growth morphological parameters, and the required growth temperature is corrected according to the historical crystal growth data, so that the crystal can always maintain the optimal temperature during the growth process, thereby improving the crystal growth quality.

[0020] In some embodiments of the present application, determining the shape regularity of a crystal based on the crystal contour includes: obtaining the center point position of the crystal contour, generating a plurality of radial lines starting from the center point position of the crystal contour at a preset angle and ending at the edge of the contour; counting the length values ​​of the radial lines, and determining the number of radial lines with unequal length values ​​as the shape regularity of the crystal.

[0021] In this embodiment, the shape regularity of the crystal is obtained by calculating the length of the radiation line from the center point of the crystal contour to the edge of the contour and according to the number of radiation lines with unequal length values.

[0022] In some embodiments of the present application, the model building module establishes a growth morphology parameter prediction model based on historical crystal growth data, including: determining historical crystal growth differences and index parameters and corresponding crystal growth morphology parameters based on historical crystal growth data, the index parameters including liquid surface temperature, crystal rotation speed, and crucible rotation speed; establishing a training sample set based on the crystal growth differences and index parameters of the historical crystal growth process and the corresponding crystal growth morphology parameters, establishing an initial growth morphology parameter prediction model, training the initial growth morphology parameter prediction model based on the training sample set, and obtaining a trained growth morphology parameter prediction model.

[0023] In this embodiment, the crystal growth difference of the crystal at each moment and the liquid surface temperature, crystal rotation speed, and crucible rotation speed are obtained through historical crystal growth data and a training sample set is established. The initial growth morphology parameter prediction model is trained through the training sample set to obtain a trained growth morphology parameter prediction model.

[0024] In some embodiments of the present application, the temperature setting module determines the required growth temperature based on the predicted growth morphological parameters of the current crystal, including: obtaining standard growth morphological parameters, calculating the difference between the growth morphological parameters of the current crystal and the standard growth morphological parameters; judging whether the difference between the growth morphological parameters of the current crystal and the standard growth morphological parameters is greater than a first preset threshold value, if the difference between the growth morphological parameters of the current crystal and the standard growth morphological parameters is greater than the first preset threshold value, setting the required growth temperature to the first temperature; if the difference between the growth morphological parameters of the current crystal and the standard growth morphological parameters is less than or equal to the first preset threshold value, judging whether the difference between the growth morphological parameters of the current crystal and the standard growth morphological parameters is greater than a second preset threshold value; if the difference between the growth morphological parameters of the current crystal and the standard growth morphological parameters is greater than the second preset threshold value, setting the required growth temperature to the second temperature; if the difference between the growth morphological parameters of the current crystal and the standard growth morphological parameters is less than or equal to the second preset threshold value, setting the required growth temperature to the third temperature.

[0025] In this embodiment, a standard growth morphology parameter database is established to obtain the standard growth morphology parameters of the crystal at each moment. The required growth temperature is set by the difference between the growth morphology parameters of the current crystal and the standard growth morphology parameters at the corresponding moment, thereby realizing dynamic control of the crystal growth temperature.

[0026] In some embodiments of the present application, the temperature setting module corrects the temperature required for growth according to historical crystal growth data, including: determining the growth rate of each time period in the historical crystal growth process according to the historical crystal growth data, and drawing a historical growth rate curve according to the growth rate of each time period in the historical crystal growth process; obtaining a preset sliding time window, and dividing the historical growth rate curve according to the preset sliding time window to obtain a number of sub-rate curves; calculating the absolute value of the slope of the sub-rate curve, and drawing the slope curve according to the absolute value of the slope of the sub-rate curve; dividing the slope curve into a number of curve segments, counting the number of peak points in the curve segment, and multiplying the maximum slope value in the curve segment by the number of peak points to obtain the curve characteristic value; clustering each curve segment according to the curve characteristic value, and correcting the temperature required for growth according to the cluster center of each curve segment.

[0027] In this embodiment, the change in the growth rate of the crystal during the growth process is obtained through the number of peak points in the curve segment and the maximum slope value, and the temperature required for growth is corrected according to the change in the growth rate to make the growth process of the crystal more stable.

[0028] In some embodiments of the present application, clustering each curve segment according to the curve characteristic value includes: establishing a curve characteristic data set according to the curve characteristic value, randomly selecting k initial clustering centers of the curve characteristic data set; calculating the Euclidean distance from the curve characteristic value in the curve characteristic data set to the initial clustering center, and dividing each curve characteristic value into corresponding partitions according to the Euclidean distance from the curve characteristic value in the curve characteristic data set to the initial clustering center; calculating the average value of the curve characteristic value in each partition, and re-determining the clustering center according to the average value of the curve characteristic value in each partition; repeatedly iterating the above steps until the clustering center no longer changes or the number of iterations reaches a preset iteration threshold, and obtaining k final clustering centers.

[0029] In this embodiment, each curve segment is clustered based on the k-means clustering algorithm to obtain the cluster center of each curve segment, and the cluster center is used as the representative curve feature value of each curve segment.

[0030] In some embodiments of the present application, the temperature required for growth is corrected according to the cluster center of each curve segment, including: obtaining a growth rate curve of the current crystal growth process, and calculating a current curve characteristic value according to the growth rate curve of the current crystal growth process; calculating the crystal growth similarity according to the current curve characteristic value and the curve characteristic value of the cluster center, calculating the ratio of the crystal growth similarity to a preset allowable similarity, and obtaining a temperature correction parameter; multiplying the temperature required for growth by the temperature correction parameter to obtain the final growth required temperature.

[0031] In some embodiments of the present application, calculating the crystal growth similarity according to the current curve characteristic value and the curve characteristic value of the cluster center includes: calculating the crystal growth similarity according to a crystal growth similarity calculation formula, wherein the crystal growth similarity calculation formula is specifically, , in, is the crystal growth similarity, is the number of curve segments, is the curve eigenvalue of the cluster center where the i-th curve segment is located, is the curve characteristic value of the i-th curve segment of the current growth rate curve.

[0032] In this embodiment, the crystal growth similarity is calculated by the current curve eigenvalue and the curve eigenvalue of the cluster center. The crystal growth similarity represents the similarity between the current crystal growth rate change and the change of its corresponding cluster center. The temperature required for growth is corrected by the crystal growth similarity to make the crystal growth process more stable.

[0033] Based on the same technical concept, such as Figure 2 As shown, the present invention also provides a crystal growth process optimization control method based on machine learning, comprising: S101, obtaining a crystal contour of a current crystal growth image, determining a shape regularity of the crystal according to the crystal contour, and obtaining a first crystal difference; obtaining a crystal contour of a previous preset period, calculating distance values ​​in various directions between the current crystal contour and the crystal contour of the previous preset period, obtaining a difference between each distance value and an average distance value, determining a second crystal difference according to the difference between each distance value and the average distance value, and multiplying the first crystal difference by the second crystal difference to obtain a crystal growth difference; S102, acquiring historical crystal growth data, and establishing a growth morphology parameter prediction model according to the historical crystal growth data; S103, inputting the current crystal growth difference into a growth morphology parameter prediction model to obtain predicted growth morphology parameters of the current crystal; S104, determining the required growth temperature according to the predicted growth morphology parameters of the current crystal, and correcting the required growth temperature according to historical crystal growth data.

[0034] By applying the above technical solutions, the present invention uses a difference detection module to obtain the crystal contour of the current crystal growth image to obtain the crystal growth difference; a model building module to obtain historical crystal growth data and establish a growth morphology parameter prediction model based on the historical crystal growth data; a morphology prediction module to input the current crystal growth difference into the growth morphology parameter prediction model to obtain the predicted growth morphology parameters of the current crystal; a temperature setting module to determine the required growth temperature based on the predicted growth morphology parameters of the current crystal, and to correct the required growth temperature based on the historical crystal growth data. By establishing a growth morphology parameter prediction model to predict the crystal growth morphology, the optimal temperature required for crystal growth is obtained, thereby improving the crystal growth quality.

[0035] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present invention can be implemented by hardware, or by software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present invention.

[0036] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A crystal growth process optimization control system based on machine learning, characterized in that: include: A difference detection module is used to obtain a crystal contour of a current crystal growth image, determine a shape regularity of the crystal according to the crystal contour, and obtain a first crystal difference; Obtaining a crystal profile of a previous preset period, calculating distance values ​​in various directions between the current crystal profile and the crystal profile of the previous preset period, obtaining differences between each distance value and an average distance value, determining a second crystal difference according to the differences between each distance value and the average distance value, and multiplying the first crystal difference by the second crystal difference to obtain a crystal growth difference; A model building module is used to obtain historical crystal growth data and build a growth morphology parameter prediction model based on the historical crystal growth data; A morphology prediction module is used to input the current crystal growth difference into the growth morphology parameter prediction model to obtain the predicted growth morphology parameters of the current crystal; The temperature setting module is used to determine the required growth temperature according to the predicted growth morphology parameters of the current crystal, and to correct the required growth temperature according to the historical crystal growth data.

2. The crystal growth process optimization control system based on machine learning according to claim 1, characterized in that: Determining the shape regularity of the crystal according to the crystal contour includes: Obtain the center point position of the crystal contour, and generate a number of radial lines starting from the center point position of the crystal contour and ending at the contour edge at a preset angle; The lengths of the radiation lines are counted, and the number of radiation lines with unequal lengths is determined as the shape regularity of the crystal.

3. The crystal growth process optimization control system based on machine learning according to claim 1, characterized in that: The model building module builds a growth morphology parameter prediction model based on historical crystal growth data, including: Determine historical crystal growth differences and index parameters and corresponding crystal growth morphology parameters according to historical crystal growth data, wherein the index parameters include liquid surface temperature, crystal rotation speed, and crucible rotation speed; A training sample set is established according to the crystal growth difference and index parameters of the historical crystal growth process and the corresponding crystal growth morphological parameters, an initial growth morphological parameter prediction model is established, and the initial growth morphological parameter prediction model is trained according to the training sample set to obtain a trained growth morphological parameter prediction model.

4. The crystal growth process optimization control system based on machine learning according to claim 1, characterized in that: The temperature setting module determines the required growth temperature according to the predicted growth morphology parameters of the current crystal, including: Obtaining standard growth morphology parameters, and calculating the difference between the growth morphology parameters of the current crystal and the standard growth morphology parameters; Determine whether the difference between the growth morphology parameter of the current crystal and the standard growth morphology parameter is greater than a first preset threshold value, and if the difference between the growth morphology parameter of the current crystal and the standard growth morphology parameter is greater than the first preset threshold value, set the required growth temperature to the first temperature; If the difference between the growth morphology parameter of the current crystal and the standard growth morphology parameter is less than or equal to the first preset threshold, then determining whether the difference between the growth morphology parameter of the current crystal and the standard growth morphology parameter is greater than a second preset threshold; If the difference between the growth morphology parameter of the current crystal and the standard growth morphology parameter is greater than a second preset threshold, setting the growth temperature to the second temperature; If the difference between the growth morphology parameter of the current crystal and the standard growth morphology parameter is less than or equal to the second preset threshold, the temperature required for growth is set to the third temperature.

5. The crystal growth process optimization control system based on machine learning according to claim 4 is characterized in that: The temperature setting module corrects the temperature required for growth according to historical crystal growth data, including: Determine the growth rate of each period in the historical crystal growth process according to the historical crystal growth data, and draw a historical growth rate curve according to the growth rate of each period in the historical crystal growth process; Obtain a preset sliding time window, and divide the historical growth rate curve according to the preset sliding time window to obtain a plurality of sub-rate curves; Calculate the absolute value of the slope of the sub-speed curve, and draw a slope curve according to the absolute value of the slope of the sub-speed curve; Divide the slope curve into several curve segments, count the number of peak points in the curve segments, multiply the maximum slope value in the curve segments by the number of peak points to obtain the curve characteristic value; Each curve segment is clustered according to the characteristic value of the curve, and the temperature required for growth is corrected according to the cluster center of each curve segment.

6. The crystal growth process optimization control system based on machine learning according to claim 5, characterized in that: The clustering of each curve segment according to the curve characteristic value comprises: A curve feature data set is established according to the curve feature values, and k initial clustering centers of the curve feature data set are randomly selected; Calculate the Euclidean distance from the curve characteristic value in the curve characteristic data set to the initial cluster center, and divide each curve characteristic value into corresponding partitions according to the Euclidean distance from the curve characteristic value in the curve characteristic data set to the initial cluster center; Calculate the average value of the curve characteristic value in each partition, and re-determine the cluster center according to the average value of the curve characteristic value in each partition; Repeat the above steps until the cluster center no longer changes or the number of iterations reaches the preset iteration threshold, and obtain k final cluster centers.

7. The crystal growth process optimization control system based on machine learning according to claim 5, characterized in that: The method of correcting the temperature required for growth according to the cluster center of each curve segment includes: Obtaining a growth rate curve of the current crystal growth process, and calculating a current curve characteristic value according to the growth rate curve of the current crystal growth process; Calculate the crystal growth similarity according to the current curve characteristic value and the curve characteristic value of the cluster center, calculate the ratio of the crystal growth similarity to the preset allowable similarity, and obtain the temperature correction parameter; The temperature required for growth is multiplied by the temperature correction parameter to obtain the final temperature required for growth.

8. The crystal growth process optimization control system based on machine learning according to claim 7, characterized in that: The crystal growth similarity is calculated based on the current curve eigenvalue and the curve eigenvalue of the cluster center, including: The crystal growth similarity is calculated according to the crystal growth similarity calculation formula, and the crystal growth similarity calculation formula is specifically: ， in, is the crystal growth similarity, is the number of curve segments, is the curve eigenvalue of the cluster center where the i-th curve segment is located, is the curve characteristic value of the i-th curve segment of the current growth rate curve.

9. A crystal growth process optimization control method based on machine learning, characterized in that: include: Acquire a crystal contour of the current crystal growth image, determine the shape regularity of the crystal according to the crystal contour, and obtain a first crystal difference; Obtaining a crystal profile of a previous preset period, calculating distance values ​​in various directions between the current crystal profile and the crystal profile of the previous preset period, obtaining differences between each distance value and an average distance value, determining a second crystal difference according to the differences between each distance value and the average distance value, and multiplying the first crystal difference by the second crystal difference to obtain a crystal growth difference; Obtain historical crystal growth data, and establish a growth morphology parameter prediction model based on the historical crystal growth data; Inputting the current crystal growth difference into the growth morphology parameter prediction model to obtain the predicted growth morphology parameters of the current crystal; The required growth temperature is determined based on the predicted growth morphology parameters of the current crystal, and the required growth temperature is corrected based on historical crystal growth data.

Images