Thermal field temperature optimization method and system for crystal pulling furnace and crystal pulling furnace

Abstract

The invention provides a thermal field temperature optimization method and system for a crystal pulling furnace and the crystal pulling furnace, which are used for solving the problem of crystal defects caused by non-uniform thermal field temperature distribution of the crystal pulling furnace and performance degradation of a heater. A temperature deviation map and a power reduction index are generated by collecting thermal field temperature and heater power data, temperature asymmetry points are repaired in combination with abnormal value detection and data interpolation, and performance degradation areas and assembly error points are accurately recognized. For an area with the temperature lower than a threshold value, a power compensation scheme is generated by optimizing heating power distribution through error compensation, and a thermal field temperature simulation model is updated by using a deviation correction coefficient. And a compensation scheme is adjusted based on real-time data verification and iterative optimization so as to eliminate residual deviation and obtain stable temperature control parameters, and finally, a heater is driven to realize dynamic power adjustment so as to achieve the purpose of thermal field equalization. According to the method, the temperature uniformity of silicon crystal growth is remarkably improved, the occurrence rate of microdefects is effectively reduced, and the crystal quality and the production stability are improved.

Description

Technical Field

[0001] This invention relates to the field of thermal field temperature control technology for crystal pulling furnaces, and particularly to a method, system, and crystal pulling furnace for optimizing the thermal field temperature of crystal pulling furnaces. Background Technology

[0002] Crystal pulling furnaces are the core equipment in the crystal growth process, and the precise control of their thermal zone temperature directly determines the quality of the crystal and production efficiency. Optimizing the thermal zone temperature not only affects the uniformity of crystal growth but also impacts production costs and equipment lifespan. In industries such as semiconductors and photovoltaics, the demand for high-quality crystals is increasing, making the stability of the thermal zone temperature a key indicator for evaluating the performance of crystal pulling furnaces.

[0003] However, in actual production, the temperature distribution of the thermal field often deviates from the design target due to various factors, making it difficult to meet the requirements of high-precision crystal growth. Existing thermal field temperature control methods mostly rely on preset heating power curves or simple feedback adjustments, but these methods have significant shortcomings when dealing with complex production environments. They often fail to adapt to the performance changes of thermal field components after long-term use, and cannot effectively handle the problem of uneven temperature distribution caused by individual equipment differences. This makes local temperature fluctuations during crystal growth a bottleneck affecting crystal quality.

[0004] In optimizing the thermal field temperature, heater performance degradation and component assembly errors are two core technical challenges. Prolonged high-temperature operation of the heater can lead to localized power reduction, causing temperatures in certain areas of the thermal field to fall below expectations. For example, in crystal growth, areas near the heater edge may experience lower temperatures due to insufficient power, resulting in uneven crystal growth rates. Component assembly errors further exacerbate this problem.

[0005] Insufficient installation precision of the insulation material or heaters within the crystal pulling furnace can lead to asymmetrical temperature distribution in the thermal field, with some areas becoming overheated while others become undercooled. This asymmetry can cause stress defects within the crystal, reducing the product yield.

[0006] Therefore, how to optimize and control the temperature deviation caused by heater performance degradation and component assembly errors during crystal growth, and achieve dynamic and precise thermal field temperature control, has become a key issue in thermal field temperature optimization. Summary of the Invention

[0007] This invention provides a method, system, and crystal pulling furnace for optimizing the thermal field temperature of a crystal pulling furnace. It aims to optimize and control temperature deviations caused by heater performance degradation and component assembly errors during crystal growth, thereby achieving dynamic and precise control of the thermal field temperature of the crystal pulling furnace.

[0008] This invention provides a method for optimizing the thermal field temperature of a crystal pulling furnace, comprising: obtaining a current temperature deviation map and power reduction index based on collected thermal field temperature distribution data and heater power output data for the silicon crystal micro-defect induced zone within the crystal pulling furnace; identifying anomalies in the temperature deviation map and power reduction index using outlier detection, filling and repairing the anomalies, and confirming the heater performance degradation area and temperature asymmetry points caused by assembly errors based on the anomalies; if the temperature of the identified performance degradation area is lower than a preset threshold, analyzing the performance degradation area and optimizing the heating power distribution parameters through temperature offset calibration to obtain an adjusted power compensation scheme; obtaining temperature asymmetry point data corresponding to component assembly errors based on the power compensation scheme, determining whether the asymmetry point data overlaps with the performance degradation area, and obtaining the deviation correction coefficient for the overlapping area; reconstructing and updating the thermal field temperature simulation model after removing the data based on the deviation correction coefficient, and verifying and determining the optimized temperature distribution prediction result in conjunction with real-time data. By analyzing the deviation map and the temperature distribution prediction results, it is determined whether there is a residual deviation in the prediction results. If there is a residual deviation, the compensation scheme is adjusted by iterative optimization process to obtain the final stable temperature control parameters. The final stable temperature control parameters are used to drive the heater to perform power adjustment in real time, and the dynamic temperature balance state of the micro-defect induced region is obtained by combining thermal field equalization correction.

[0009] Preferably, real-time data on the temperature distribution of the thermal field inside the crystal pulling furnace and the power output data of the heater are collected and stored as a raw dataset. Noise is filtered from the collected signals in the raw dataset to generate a denoised dataset. The temperature deviation dataset is generated by calculating the difference between the thermal field temperature and the preset standard temperature based on the denoised dataset. If the deviation value in the temperature deviation dataset exceeds a preset threshold, spatial interpolation is performed on the deviation value to generate a temperature deviation map. The changing trend of the heater power output data is extracted from the denoised dataset and a power drop index is calculated. The power drop index is compared with a preset power threshold to determine whether there are any power anomalies. If there are any power anomalies, the temperature deviation map area corresponding to the anomaly point is marked to generate a defect-induced area distribution.

[0010] Preferably, based on a temperature deviation map and a power reduction index, outliers are identified and their distribution data is obtained by using a preset threshold; outlier detection is performed on the visualized temperature data to determine the location of outliers; outliers are repaired to generate repaired temperature distribution data; spatial correlation is calculated using the repaired temperature distribution data to determine the areas of heater performance degradation; if the spatial correlation is lower than a preset threshold, local weighted regression analysis is used to determine the areas affected by assembly errors by identifying temperature asymmetry points; the correlation between the power reduction index and the temperature distribution data is obtained, and the degree of heater performance degradation is determined based on a linear regression model; the temperature distribution data before and after repair are compared to verify the accuracy of the data repair before generating the final degradation area and asymmetry point distribution.

[0011] Preferably, if the temperature of the performance degradation area is detected to be lower than a preset threshold, the temperature value of the abnormal area is obtained from the temperature detection data, the temperature offset is calculated, and the temperature offset analysis result is obtained; the impact of the temperature offset on the power distribution is predicted based on the temperature offset analysis result, and the error compensation parameters are determined; if the error compensation parameters exceed the preset range, the heating power distribution is adjusted through an iterative optimization algorithm to obtain the optimized power distribution parameters; an initial power compensation scheme is generated using the optimized power distribution parameters to obtain the preliminary adjusted scheme; power distribution features are extracted from the preliminary adjusted scheme and classified to determine the stability of the scheme; if the stability of the scheme meets the preset conditions, the preliminary adjusted scheme is fine-tuned through a power distribution optimization method to obtain the final power distribution scheme; a region calibration command is generated based on the final power distribution scheme, and the calibrated heating power distribution result is output.

[0012] Preferably, the component assembly error data in the power compensation scheme is obtained, and an error data set is generated through a data extraction process; the temperature asymmetry point distribution in the error data set is extracted to obtain temperature distribution characteristics; the overlapping area between the temperature asymmetry points and the performance degradation area is determined based on the temperature distribution characteristics, and the degree of overlap is calculated based on the overlapping area judgment result to obtain the quantized value of the overlapping area; a deviation correction coefficient is generated based on the quantized value of the overlapping area; the power compensation scheme is adjusted using the deviation correction coefficient to obtain optimized compensation parameters; the optimized compensation parameters are used to update the component assembly error data and generate a new error data set.

[0013] Preferably, the parameters of the thermal field temperature model are adjusted by the deviation correction coefficient to obtain the optimized temperature model; the optimized temperature model is then used to reconstruct the temperature distribution and generate the initial temperature distribution data.

[0014] Ambient temperature data is acquired from a real-time data acquisition system to determine the validity of the real-time data. If the deviation between the real-time data and the initial temperature distribution data exceeds a preset threshold, the temperature distribution is adjusted through data fusion analysis to obtain a corrected temperature distribution. The corrected temperature distribution is used to predict the temperature at future time points to generate a temperature prediction result. The accuracy of the prediction result is judged by comparing the temperature prediction result with the real-time data acquisition result. If the accuracy of the prediction result is lower than the preset standard, the model update mechanism is updated to regenerate the optimized temperature model.

[0015] Preferably, the collected temperature distribution data is used to generate an initial temperature distribution prediction result; deviation data is extracted from the prediction result to generate a deviation map; based on the deviation map, a preset threshold is used to determine whether there is a residual deviation; if there is a residual deviation, the deviation location and magnitude are recorded to obtain deviation distribution information; the parameters of the compensation scheme are iteratively adjusted based on the deviation distribution information to generate an updated compensation scheme; the updated compensation scheme is used to regenerate the temperature distribution prediction result to obtain new temperature distribution data.

[0016] The deviation map is generated and the residual deviation is judged from the new temperature distribution data. If the residual deviation is lower than the preset threshold, the final stable temperature control parameter is determined. The temperature distribution prediction model is updated based on the final stable temperature control parameter to obtain the optimized prediction result.

[0017] Preferably, temperature distribution data within the silicon crystal growth cavity is acquired in real time using a sensor array to obtain a multi-point temperature matrix. If local high-temperature or low-temperature regions exist in the temperature matrix, the heater power is adjusted to obtain an optimized power allocation scheme. The heater is driven to perform power adjustment according to the optimized power allocation scheme to obtain an updated thermal field distribution state. The thermal field distribution state is compared with a preset equilibrium threshold. If the deviation exceeds the threshold, thermal field correction parameters are calculated to obtain a corrected thermal field equilibrium model. The temperature control parameters are updated according to the corrected thermal field equilibrium model to obtain the dynamic temperature equilibrium state of the micro-defect induced region. The temperature equilibrium state of the micro-defect induced region is monitored in real time, and the potential defect formation trend is predicted to obtain a defect risk assessment result. The parameters of the iterative control algorithm are adjusted according to the defect risk assessment result to obtain an optimized temperature control strategy.

[0018] This invention provides a thermal zone temperature optimization system for crystal pulling furnaces, which applies the aforementioned thermal zone temperature optimization method to control the thermal zone temperature of the crystal pulling furnace.

[0019] The data acquisition and processing module is used to collect thermal field temperature and heater power data, and generate temperature deviation maps and power reduction indicators.

[0020] An anomaly diagnosis module is used to identify anomalies based on the deviation map and power index, and to determine areas of heater performance degradation and temperature asymmetry points caused by assembly errors.

[0021] The calibration compensation module is used to calculate the temperature offset and generate a power compensation scheme when the temperature of the performance degradation region is too low.

[0022] The deviation correction module is used to determine whether the temperature asymmetry point overlaps with the performance degradation region, and to generate deviation correction coefficients based on the quantization value of the overlapping region to update the thermal field temperature model.

[0023] The iterative control module is used to perform temperature prediction and residual deviation analysis using the updated model, determine the final stable temperature control parameters through iterative optimization, and drive the heater to adjust the power to achieve balanced control of the dynamic temperature of the silicon crystal micro-defect induced region.

[0024] The present invention also provides a crystal pulling furnace, which uses the crystal pulling furnace thermal field temperature optimization method to control the temperature of the crystal pulling furnace thermal field.

[0025] The working principle and beneficial effects of this invention are as follows:

[0026] This invention addresses crystal defects caused by uneven temperature distribution in the thermal field of crystal pulling furnaces and heater performance degradation. Specifically, it collects real-time thermal field temperature and heater power data, uses a noise filtering mechanism to generate a temperature deviation map and power reduction index, and combines outlier detection and data interpolation to repair temperature asymmetry points, accurately identifying performance degradation areas and assembly error points. For areas with temperatures below a threshold, the invention optimizes heating power distribution through error compensation, generates a power compensation scheme, and updates the thermal field temperature simulation model using a deviation correction coefficient. Based on real-time data verification and iterative optimization, the invention adjusts the compensation scheme to eliminate residual deviations, obtains stable temperature control parameters, and ultimately drives the heater to achieve dynamic power adjustment, achieving thermal field equilibrium. This invention significantly improves the temperature uniformity of silicon crystal growth, effectively reduces the incidence of micro-defects, and enhances crystal quality and production stability.

[0027] Specifically, the present invention provides a method for optimizing the thermal field temperature of a crystal pulling furnace, comprising:

[0028] Based on the collected thermal field temperature distribution data and heater power output data of the silicon crystal micro-defect induced zone in the crystal pulling furnace, the current temperature deviation map and power reduction index are obtained.

[0029] Based on outlier detection, anomalies are identified in the image-based temperature data of temperature deviation map and power reduction index. The anomalies are filled and repaired, and the areas of heater performance degradation and temperature asymmetry caused by assembly errors are identified based on the anomalies.

[0030] If the temperature of the identified performance degradation region is lower than the preset threshold, the performance degradation region is analyzed by temperature offset calibration and the heating power distribution parameters are optimized to obtain an adjusted power compensation scheme.

[0031] Based on the power compensation scheme, the temperature asymmetry point data corresponding to the component assembly error is obtained, and it is determined whether the asymmetry point data overlaps with the performance degradation area to obtain the deviation correction coefficient of the overlapping area.

[0032] The thermal field temperature simulation model was reconstructed and updated after removing the deviation correction coefficient, and the optimized temperature distribution prediction results were verified and confirmed by combining real-time data.

[0033] By combining the deviation map with the temperature distribution prediction results, it is determined whether there is a residual deviation in the prediction results. If there is a residual deviation, the compensation scheme is adjusted by iterative optimization process to obtain the final stable temperature control parameters.

[0034] The heater is driven to perform power regulation in real time by using the final stable temperature control parameters, and the dynamic temperature equilibrium state of the micro-defect induced region is obtained by combining thermal field equalization correction.

[0035] Furthermore, the present invention provides a thermal zone temperature optimization system for a crystal pulling furnace, which applies the aforementioned thermal zone temperature optimization method to control the thermal zone temperature of the crystal pulling furnace.

[0036] The data acquisition and processing module is used to collect thermal field temperature and heater power data, and generate temperature deviation maps and power reduction indicators.

[0037] An anomaly diagnosis module is used to identify anomalies based on the deviation map and power index, and to determine areas of heater performance degradation and temperature asymmetry points caused by assembly errors.

[0038] The calibration compensation module is used to calculate the temperature offset and generate a power compensation scheme when the temperature of the performance degradation region is too low.

[0039] The deviation correction module is used to determine whether the temperature asymmetry point overlaps with the performance degradation region, and to generate deviation correction coefficients based on the quantization value of the overlapping region to update the thermal field temperature model.

[0040] The iterative control module is used to perform temperature prediction and residual deviation analysis using the updated model, determine the final stable temperature control parameters through iterative optimization, and drive the heater to adjust the power to achieve balanced control of the dynamic temperature of the silicon crystal micro-defect induced region.

[0041] Furthermore, the present invention also provides a crystal pulling furnace, which uses the crystal pulling furnace thermal field temperature optimization method to control the temperature of the crystal pulling furnace thermal field.

[0042] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0043] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0044] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0045] Figure 1 This is a schematic diagram of the process of the present invention.

[0046] Figure 2 This is a schematic diagram of the system of the present invention. Detailed Implementation

[0047] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0048] according to Figures 1-2 As shown, this embodiment of the invention provides a method for optimizing the thermal field temperature of a crystal pulling furnace, characterized by comprising:

[0049] Based on the collected thermal field temperature distribution data and heater power output data of the silicon crystal micro-defect induced zone in the crystal pulling furnace, the current temperature deviation map and power reduction index are obtained.

[0050] Based on outlier detection, anomalies are identified in the image-based temperature data of temperature deviation map and power reduction index. The anomalies are filled and repaired, and the areas of heater performance degradation and temperature asymmetry caused by assembly errors are identified based on the anomalies.

[0051] If the temperature of the identified performance degradation region is lower than the preset threshold, the performance degradation region is analyzed by temperature offset calibration and the heating power distribution parameters are optimized to obtain an adjusted power compensation scheme.

[0052] Based on the power compensation scheme, the temperature asymmetry point data corresponding to the component assembly error is obtained, and it is determined whether the asymmetry point data overlaps with the performance degradation area to obtain the deviation correction coefficient of the overlapping area.

[0053] The thermal field temperature simulation model was reconstructed and updated after removing the deviation correction coefficient, and the optimized temperature distribution prediction results were verified by combining real-time data.

[0054] By combining the deviation map with the temperature distribution prediction results, it is determined whether there is a residual deviation in the prediction results. If there is a residual deviation, the compensation scheme is adjusted by iterative optimization process to obtain the final stable temperature control parameters.

[0055] The heater is driven to perform power regulation in real time by using the final stable temperature control parameters, and the dynamic temperature equilibrium state of the micro-defect induced region is obtained by combining thermal field equalization correction.

[0056] This invention uses a sensor array to collect real-time thermal field temperature distribution data and heater power output data of the silicon crystal micro-defect induced zone inside the crystal pulling furnace. It performs signal denoising based on a wavelet transform algorithm to generate a denoised temperature deviation map and power degradation index. Based on the temperature deviation map and power degradation index, it identifies outliers in the visualized temperature data using an isolated forest algorithm. Kriging interpolation is used to repair these outliers and generate repaired temperature distribution data, corresponding performance degradation regions, and temperature asymmetry points. If the temperature of a performance degradation region is lower than a preset threshold, these regions are analyzed through temperature offset calibration. The least squares method is used to calculate error compensation and generate optimized heating power allocation parameters. Based on the optimized heating power allocation parameters... Data on temperature asymmetry points is acquired, and statistical thresholds are used to determine the overlap between these points and performance degradation areas, generating deviation correction coefficients for the overlapping areas. A thermal field temperature simulation model is reconstructed using the deviation correction coefficients and the finite element method, and then validated using real-time data acquisition to generate optimized temperature distribution prediction results. Based on the temperature distribution prediction results and deviation map analysis, it is determined whether there are residual deviations. If residual deviations exist, the power compensation scheme is iteratively optimized using the gradient descent method to generate final stable temperature control parameters. The obtained final stable temperature control parameters are then adjusted in real-time by driving the heater, and a dynamic temperature equilibrium state of the micro-defect induced region during silicon crystal growth is generated using a thermal field equalization correction algorithm.

[0057] Furthermore, within the crystal pulling furnace, a sensor array is used to collect real-time data on the thermal field temperature distribution and heater power output of the silicon crystal micro-defect induced zone. For example, the temperature sensor array records a core temperature of 1450.2℃ and an edge temperature of 1428.5℃, with a power output of 85.3kW. Subsequently, a Kalman filter is used as a noise filtering mechanism to process the signal. The filter gain is calculated through state equations and observation equations to obtain the smoothed current temperature deviation. For example, the map shows a core deviation of -2.1℃ and an edge deviation of +1.8℃. Based on historical averages, the power decrease index is 3.2%, thus providing a reliable basis for subsequent analysis.

[0058] Next, based on this deviation map and power degradation index, the isolated forest algorithm is used to detect outliers in the visualized temperature data; for example, a contamination rate of 0.05 is set for training the model, calculating the outlier score for each point, and identifying multiple outliers; such as coordinates. If a temperature jump occurs, these abnormal points are filled and repaired using radial basis function interpolation. Further, the performance degradation area of ​​the heater is determined to be where the output of the lower heater decays, and the temperature asymmetry points caused by assembly errors are directly related to the degradation compensation requirements. If the temperature in the performance degradation area is lower than a preset threshold, these areas are analyzed through temperature offset calibration. The least squares error compensation method is used to optimize the heating power distribution parameters, and the weight coefficients in the linear model are solved, i.e., the contribution weight of each heating power to temperature change. This yields the adjusted power compensation scheme for the upper and lower heaters, which ensures energy balance based on a physical heat conduction model.

[0059] Furthermore, temperature asymmetry point data corresponding to component assembly errors are obtained, as shown in the point set on the left. Statistical thresholds are used to determine whether these points overlap with performance degradation areas. The mean and standard deviation are calculated. If the deviation exceeds a preset value, the points overlap, and a deviation correction coefficient for the overlapping area is obtained. This deviation correction coefficient is used to bridge the logical correction between asymmetry and degradation. Anomalies are eliminated using the deviation correction coefficient, and the finite element method is used to reconstruct and update the thermal field temperature simulation model. Real-time data is used to verify the correlation coefficients and determine the core and edge temperatures of the optimized temperature distribution prediction results. This verification process enhances the model's robustness.

[0060] Next, based on the optimized temperature distribution prediction results and deviation map analysis, the remaining deviation is determined, for example, by calculating the root mean square and determining whether it exceeds a preset threshold. If it exceeds the preset threshold, gradient descent is used for iterative optimization to adjust the compensation scheme. After multiple iterations with the learning rate until convergence, the final number of power allocation parameters, including upper, lower, and asymmetric correction parameters for stable temperature control, is obtained. The above iterations ensure that the deviation is minimized. The power allocation parameters of these stable temperature control parameters are then applied in real time to drive the heater to perform power adjustment through scheme iteration. Combined with thermal field equalization correction such as Fourier transform filtering of low-frequency drift, the dynamic temperature equalization state of the micro-defect induced region during silicon crystal growth is obtained, maintaining fluctuations below a preset value to promote uniform crystal growth. The entire process forms a closed-loop adaptive control chain.

[0061] This invention acquires real-time data on thermal field temperature and heater power, uses a noise filtering mechanism to generate a temperature deviation map and power drop index, and combines outlier detection and data interpolation to repair temperature asymmetry points, accurately identifying performance degradation areas and assembly error points. For areas with temperatures below a threshold, this invention optimizes heating power distribution through error compensation, generates a power compensation scheme, and updates the thermal field temperature simulation model using a deviation correction coefficient. Based on real-time data verification and iterative optimization, this invention adjusts the compensation scheme to eliminate residual deviations, obtains stable temperature control parameters, and ultimately drives the heater to achieve dynamic power adjustment, achieving thermal field equilibrium. This invention significantly improves the temperature uniformity of silicon crystal growth, effectively reduces the incidence of micro-defects, and enhances crystal quality and production stability.

[0062] In one embodiment, real-time data on the temperature distribution of the thermal field inside the crystal pulling furnace and the power output data of the heater are collected and stored as a raw dataset; noise filtering is performed on the collected signals in the raw dataset to generate a denoised dataset.

[0063] A temperature deviation dataset is generated by calculating the difference between the thermal field temperature and the preset standard temperature based on the denoised dataset. If the deviation value in the temperature deviation dataset exceeds the preset threshold, spatial interpolation is performed on the deviation value to generate a temperature deviation map.

[0064] The trend of heater power output data is extracted from the denoised dataset and the power drop index is calculated. The power drop index is compared with the preset power threshold to determine whether there are any power anomalies. If there are power anomalies, the temperature deviation map area corresponding to the anomaly point is marked to generate the defect-induced area distribution.

[0065] In this embodiment, multiple high-precision thermocouples are spaced at key locations in the induction zone of the crystal growth chamber within the crystal pulling furnace. These locations include the upper part, lower part, and sidewall of the crystal. This enables real-time acquisition of the thermal field temperature distribution data in the micro-defect induction zone of the silicon crystal. Simultaneously, the power monitoring module acquires the heater output data, ensuring that the acquired data covers the dynamic changes during the growth process. For example, at a growth rate of 0.8 mm / min, the thermocouples acquired a temperature field peak of 1523.5℃, a valley of 1487.2℃, and a power curve that gradually decreased from an initial 5000 watts to 4800 watts.

[0066] Next, the original signal is decomposed based on the wavelet transform algorithm. First, the signal is decomposed using the db4 wavelet basis function. Then, the threshold is set to be N times the standard deviation of the signal. After removing high-frequency noise components using a soft thresholding function, the signal is reconstructed to obtain a filtered temperature deviation map. Then, the deviation between each sensor point and the target uniform temperature is calculated to form a thermal field deviation matrix. The accuracy of the deviation map is verified by finite element simulation. At the same time, an adaptive Kalman filter is introduced for the heater power output data. After iterative updates of the state equation and observation equation, a power reduction index is obtained. For example, the current power of 4820 watts has a reduction rate of 3.6% compared to the initial value of 5000 watts. By associating this index with the temperature deviation, the risk probability of oxygen precipitation defects induced by local overcooling due to insufficient power is inferred to be 15%, thus forming a closed-loop feedback mechanism. This provides a data basis for subsequent growth parameter optimization. The entire process runs automatically in the embedded control system without manual intervention.

[0067] In one embodiment, based on a temperature deviation map and a power drop index, outliers are identified by using a preset threshold and outlier distribution data is obtained; outlier detection is performed on the imaged temperature data to determine the location of outliers; and the outliers are repaired to generate repaired temperature distribution data.

[0068] Spatial correlation is calculated using the repaired temperature distribution data to determine the areas of heater performance degradation; if the spatial correlation is lower than a preset threshold, local weighted regression analysis is used to identify temperature asymmetry points and determine the areas affected by assembly errors.

[0069] The correlation between power reduction index and temperature distribution data is obtained, and the degree of heater performance degradation is determined based on a linear regression model. Temperature distribution data before and after repair are compared, and the accuracy of data repair is verified before generating the final degradation area and asymmetric point distribution.

[0070] In this embodiment, outliers are identified based on a temperature deviation map and a preset threshold for power degradation indicators, resulting in outlier distribution data. Next, the isolated forest algorithm is used to detect outliers in the visualized temperature data and determine their locations. Then, spatial interpolation is used to repair the outliers, generating repaired temperature distribution data. Spatial correlation is calculated using the repaired temperature distribution data to determine areas of heater performance degradation. If the spatial correlation is below a preset threshold, local weighted regression analysis is used to analyze temperature asymmetry points, thereby identifying areas affected by assembly errors. Finally, the correlation between the power degradation indicator and the temperature distribution data is obtained, and a linear regression model is used to determine the degree of heater performance degradation. The accuracy of the data repair is verified by comparing the temperature distribution data before and after repair, generating the final degradation area and asymmetry point distribution.

[0071] Furthermore, when processing the heater surface temperature deviation map, the temperature deviation value of each pixel is first calculated. For example, if the reference temperature is 80.0℃ and the current measured value is 75.2℃, the deviation value is 4.8℃. A deviation matrix is ​​generated based on multiple deviation values. Then, a power drop index is introduced. Assuming the initial power Pinitial = 1000W and the current power Pcurrent = 920W, the power drop index is 8.0%. The power drop index is superimposed on the deviation map as a weighting factor to amplify the detection threshold of deviation-sensitive areas. Next, an outlier detection algorithm is used to detect outliers in the imaged temperature data. Specifically, the isolated forest algorithm is used for outlier detection. First, a contamination rate is set and the imaged temperature data is trained. The input includes features such as local temperature gradients, and the model outputs an anomaly score. If the anomaly score is greater than a preset value, it is marked as an outlier. In practical application, anomaly detection analysis shows that there are 15 outliers in the center area of ​​the heater, with an average deviation of 6.5℃, while there are only 3 outliers in the outer area, with a deviation of 3.1℃. The above data indicates that the degradation in the center is more significant. Then, a bicubic interpolation algorithm is used to fill and repair these outliers, and the new values ​​after repair are calculated for the outliers. This ensures a smooth transition after the repaired outliers are filled, reducing the error rate.

[0072] Based on the repaired map, the areas of heater performance degradation are identified. A threshold is automatically calculated using a threshold segmentation algorithm, and connected regions with offsets greater than the threshold are extracted as degradation areas. For example, an elliptical region with a diameter of about 20 pixels is identified, which accounts for 5.3% of the total image area and corresponds to a power loss of about 4.2W. It can be seen that this region is positively correlated with the power reduction index, thus confirming that the degradation mechanism is material aging.

[0073] Furthermore, for temperature asymmetry points caused by assembly errors, the image symmetry index is calculated. If the image symmetry index is greater than a preset value, asymmetry detection is triggered. Principal component analysis is used to decompose the asymmetry vector, and the first two principal components are extracted to explain the variance. Multiple asymmetry points are located, such as a deviation of +3.7℃ on the left and -2.9℃ on the right. The analysis shows that the assembly gap causes uneven heat conduction. After repair, the image symmetry index drops to the normal temperature range, thus forming a complete chain from deviation quantification to degradation location, ensuring the automation of heater maintenance decisions.

[0074] In this embodiment, the image symmetry index is calculated using the following formula: ;in With the center coordinates, Total number of pixels These are the coordinates of a pixel in the image; This represents the temperature value of each pixel in the image. The coordinates of pixels Output temperature .

[0075] In one embodiment, if the temperature of the performance degradation area is detected to be lower than a preset threshold, the temperature value of the abnormal area is obtained from the temperature detection data, the temperature offset is calculated, and the temperature offset analysis result is obtained.

[0076] Based on the temperature offset analysis results, predict the impact of temperature offset on power distribution and determine the error compensation parameters; if the error compensation parameters exceed the preset range, adjust the heating power distribution through an iterative optimization algorithm to obtain the optimized power distribution parameters;

[0077] An initial power compensation scheme is generated using the optimized power allocation parameters to obtain a preliminary adjusted scheme. Power allocation features are extracted from the preliminary adjusted scheme and classified to determine the stability of the scheme. If the stability of the scheme meets the preset conditions, the preliminary adjusted scheme is fine-tuned using the power allocation optimization method to obtain the final power allocation scheme. A regional calibration command is generated based on the final power allocation scheme, and the calibrated heating power allocation result is output.

[0078] In this embodiment, if the temperature of the performance degradation area is detected to be lower than a preset threshold, the temperature value of the abnormal area is obtained from the temperature detection data; the temperature offset is calculated using a statistical analysis method to obtain the temperature offset analysis result. Based on the temperature offset analysis result, the impact of the temperature offset on power allocation is predicted using a linear regression algorithm to determine the error compensation parameters. If the error compensation parameters exceed the preset range, the heating power allocation is adjusted using an iterative optimization algorithm to obtain the optimized power allocation parameters and generate an initial power compensation scheme, resulting in a preliminarily adjusted scheme. Power allocation features are extracted from the preliminarily adjusted scheme and classified using a support vector machine algorithm to determine the stability of the scheme. If the stability of the scheme meets the preset conditions, the preliminarily adjusted scheme is fine-tuned using a power allocation optimization method to obtain the final power allocation scheme. A region calibration command is generated based on the final power allocation scheme, and the calibrated heating power allocation result is output.

[0079] Furthermore, assuming the current detected temperature of a heating area is 85℃, which is lower than the preset threshold of 90℃, temperature offset calibration analysis and power compensation optimization are required. First, the system collects area temperature data through sensors and calculates the temperature deviation: A temperature deviation of -5℃ indicates performance degradation.

[0080] Next, temperature offset calibration analysis was performed to obtain the offset analysis results. Specifically, a regression model for temperature and power was established based on historical data, and its formula is as follows: ,in, For power, The temperature is obtained by least squares fitting. The value is used to calculate the theoretical power corresponding to the current temperature of 85℃, and then the error value is calculated using the actual power output value. Based on the error compensation algorithm, the compensation power is calculated using proportional and integral control. ,in With a sampling interval of 0.1s, the cumulative error (Assuming 10 samples), scaling factor Integral coefficient The final compensated power is then obtained. Finally, the heating power allocation parameters are optimized, and the adjusted power is the sum of the actual power and the compensated power. This power value is allocated to the heating units, and the adjustment effect is recorded.

[0081] If the temperature detected later is still below the set threshold of 90℃, further iterations and optimizations will be performed. The model parameters will be adjusted in conjunction with ambient temperature and humidity data (such as humidity of 50% which may affect heat conduction) to ensure that the temperature is stable within the range of 90℃±1℃. This process is automatically executed by the embedded controller, and the closed-loop logic ensures accurate power distribution.

[0082] In one embodiment, component assembly error data in the power compensation scheme is acquired, and an error data set is generated through a data extraction process. The temperature asymmetry point distribution in the error data set is extracted to obtain temperature distribution characteristics. Based on the temperature distribution characteristics, the overlapping area between the temperature asymmetry points and the performance degradation area is determined. The degree of overlap is calculated based on the overlapping area determination result to obtain the quantized value of the overlapping area. A deviation correction coefficient is generated based on the quantized value of the overlapping area. The power compensation scheme is adjusted using the deviation correction coefficient to obtain optimized compensation parameters. The optimized compensation parameters are used to update the component assembly error data and generate a new error data set.

[0083] In this embodiment, component assembly error data in the power compensation scheme is acquired, and an error data set is generated through a data extraction process. The temperature asymmetry point distribution is extracted from the error data set to obtain temperature distribution characteristics. A statistical threshold analysis method is used to process the temperature distribution characteristics to determine the overlapping areas between temperature asymmetry points and performance degradation regions. The degree of overlap is calculated based on the overlap region determination results, yielding a quantified value for the overlapping area. A linear regression algorithm is used to calculate the deviation correction coefficient using the quantified value of the overlapping area; the power compensation scheme is adjusted using the deviation correction coefficient to obtain optimized compensation parameters. The component assembly error data is updated based on the optimized compensation parameters, generating a new error data set.

[0084] In this scheme, based on the obtained power compensation scheme, the temperature distribution data corresponding to the component assembly error is first collected by a thermal imaging sensor array. When the assembly gap is 0.2mm, a local temperature anomaly point of 85.3℃ is detected, which deviates from the normal operating temperature of 78.5℃ by 6.8℃. Next, a Gaussian mixture model algorithm is used to identify the temperature asymmetry point. By calculating the temperature gradient matrix, the coordinate position is found to be... An abnormal temperature gradient was detected at a location with a gradient value of 2.4℃ / mm, exceeding the normal threshold of 1.8℃ / mm by 33.3%. Subsequently, the system used a statistical threshold judgment algorithm to set a temperature anomaly detection threshold, and after traversing all temperature sampling points, it was found that a total of 12 points exceeded the threshold.

[0085] Furthermore, a regional overlap calculation method is used to spatially match and analyze the temperature anomaly region with the preset performance degradation region. If the overlap area is too high compared to the total anomaly region, the temperature deviation within the overlap region exhibits a linear distribution characteristic. Based on the temperature deviation data of the overlap region, a deviation correction coefficient K=0.847 is calculated using a least squares fitting algorithm. The deviation correction coefficient is obtained using the formula: The calculated parameters are used for precise adjustment of subsequent power compensation parameters to ensure that the system can maintain a stable thermal balance under the influence of assembly errors.

[0086] In one embodiment, the parameters of the thermal field temperature model are adjusted by the deviation correction coefficient to obtain an optimized temperature model; the optimized temperature model is then used to reconstruct the temperature distribution to generate initial temperature distribution data.

[0087] Ambient temperature data is acquired from a real-time data acquisition system to determine the validity of the real-time data. If the deviation between the real-time data and the initial temperature distribution data exceeds a preset threshold, the temperature distribution is adjusted through data fusion analysis to obtain a corrected temperature distribution. The corrected temperature distribution is used to predict the temperature at future time points to generate a temperature prediction result. The accuracy of the prediction result is judged by comparing the temperature prediction result with the real-time data acquisition result. If the accuracy of the prediction result is lower than the preset standard, the model update mechanism is updated to regenerate the optimized temperature model.

[0088] In this embodiment, the parameters of the thermal field temperature model are adjusted by a deviation correction coefficient to obtain an optimized temperature model. The optimized temperature model is then used to reconstruct the temperature distribution and generate initial temperature distribution data. Ambient temperature data is acquired from a real-time data acquisition system to determine the validity of the real-time data. If the deviation between the real-time data and the initial temperature distribution data exceeds a preset threshold, the temperature distribution is adjusted through data fusion analysis to obtain a corrected temperature distribution. The corrected temperature distribution is then used to predict future temperatures using a support vector regression algorithm, generating a temperature prediction result. The accuracy of the prediction result is judged by comparing it with the real-time data acquisition result. If the accuracy of the prediction result is lower than a preset standard, the model update mechanism is activated, and a new optimized temperature model is generated.

[0089] In this scheme, during the thermal field temperature simulation, the deviation correction coefficient is first calculated based on historical monitoring data. For example, 100 sets of experimental data are collected, and the average deviation between the actual temperature and the initial model prediction is 2.5℃. The least squares algorithm is used to fit the correction coefficient. Then, the model parameters are adjusted through the correction coefficient to ensure that the error rate is reduced to within 0.8% after correction, thus laying an accurate foundation for subsequent steps.

[0090] Next, the correction coefficient is used to remove data points in the high deviation area of ​​the model. That is, 15% of the abnormal node data with deviations exceeding 1.5℃ are removed by threshold screening method, and the thermal field model is reconstructed by Gaussian process regression algorithm. After the finite element analysis is integrated to solve the heat conduction equation, an updated simulation framework is generated. The robustness of the simulation framework is verified by Monte Carlo simulation with a confidence interval of 95%.

[0091] Then, the model is validated by combining real-time sensor data. Specifically, the collected temperature sequence data is fused with the predicted and observed values ​​using the Kalman filter algorithm. The residual analysis after filtering shows the root mean square error, which is used to confirm the reliability of the optimized model.

[0092] The predictions and observations are obtained through the state equation and the observation equation:

[0093] The state equation is , For state At any moment The value; The state transition coefficient is 1.02 in this scheme. This refers to process noise during the state evolution process; For the state in the previous moment The value;

[0094] The observation equation is in, For the measured value at time The value; The observation coefficient is set to 1 in this embodiment; For state At any moment The value; This refers to the measurement noise generated during the sensor's operation.

[0095] Finally, based on the verification results, the temperature distribution prediction is determined. For example, the peak temperature in the target area is predicted to be 85.7℃. The distribution map is inverted by using a backpropagation neural network (3 hidden layers, learning rate 0.01). The analysis process includes calculating gradient descent to update weights. After 50 iterations, the prediction accuracy is improved by 12%, forming a uniform thermal field distribution map. This map can guide the adjustment of process parameters and avoid the risk of overheating. The entire process is optimized in a closed loop through automated scripts.

[0096] In one embodiment, the collected temperature distribution data is used to generate an initial temperature distribution prediction result; deviation data is extracted from the prediction result to generate a deviation map; based on the deviation map, a preset threshold is used to determine whether there is a residual deviation; if there is a residual deviation, the deviation location and magnitude are recorded to obtain deviation distribution information; the parameters of the compensation scheme are iteratively adjusted based on the deviation distribution information to generate an updated compensation scheme; the updated compensation scheme is used to regenerate the temperature distribution prediction result to obtain new temperature distribution data.

[0097] The deviation map is generated and the residual deviation is judged from the new temperature distribution data. If the residual deviation is lower than the preset threshold, the final stable temperature control parameter is determined. The temperature distribution prediction model is updated based on the final stable temperature control parameter to obtain the optimized prediction result.

[0098] In this embodiment, a statistical analysis method is used to statistically analyze the collected temperature distribution data to generate an initial temperature distribution prediction result. Deviation data is extracted from the prediction result, and a deviation map is generated using a deviation analysis algorithm. Based on a preset threshold, it is determined whether there is a residual deviation in the deviation map. If a residual deviation exists, the deviation location and magnitude are recorded to obtain deviation distribution information. The gradient descent algorithm is used to calculate the gradient of the deviation distribution information, and the compensation scheme parameters are iteratively adjusted to generate an updated compensation scheme. The updated compensation scheme is used to regenerate the temperature distribution prediction result, resulting in new temperature distribution data. The deviation map generation and residual deviation determination are repeated from the new temperature distribution data. If the residual deviation is lower than a preset threshold, the final stable temperature control parameters are determined. The final stable temperature control parameters are used to update the temperature distribution prediction model, resulting in an optimized prediction result.

[0099] In this scheme, during the temperature distribution prediction stage, the temperature field of the crystal pulling furnace is first simulated and calculated using the finite element analysis software ANSYS. After determining the initial boundary conditions, the predicted temperature distribution is obtained by solving the heat conduction equation. Subsequently, the predicted temperature is compared with the actual measurement data through deviation map analysis to confirm the existence of residual deviation. Based on the significant deviation, an iterative optimization process is initiated. The compensation scheme is adjusted using a genetic algorithm to optimize the compensation cooling wind speed and add the thickness of the local insulation layer. Finally, the temperature control parameters are stabilized as the target center temperature and surface temperature, and the deviation is reduced to within the preset threshold. Verification simulation shows that the volume deviation rate is less than 2%, thereby achieving precise control. The entire process is executed in the MATLAB environment through automated scripts to ensure that the parameters are stable and suitable for mass production.

[0100] In one embodiment, temperature distribution data within the silicon crystal growth cavity is collected in real time using a sensor array to obtain a multi-point temperature matrix; if there are local high-temperature or low-temperature regions in the temperature matrix, the heater power is adjusted to obtain an optimized power allocation scheme; the heater is driven to perform power adjustment according to the optimized power allocation scheme to obtain an updated thermal field distribution state;

[0101] The thermal field distribution state is compared with the preset equilibrium threshold. If the deviation exceeds the threshold, the thermal field correction parameters are calculated to obtain the corrected thermal field equilibrium model. The temperature control parameters are updated according to the corrected thermal field equilibrium model to obtain the dynamic temperature equilibrium state of the micro-defect induced zone.

[0102] The temperature equilibrium state of the micro-defect induced zone is monitored in real time, and the potential defect formation trend is predicted to obtain the defect risk assessment result. The parameters of the iterative control algorithm are adjusted according to the defect risk assessment result to obtain the optimized temperature control strategy.

[0103] In this embodiment, a multi-point temperature matrix is ​​obtained by real-time acquisition of temperature distribution data within the silicon crystal growth cavity using a sensor array. If local high-temperature or low-temperature regions exist in the temperature matrix, a preset gradient descent algorithm is used to adjust the heater power to obtain an optimized power allocation scheme. The heater is driven to perform power adjustment according to the optimized power allocation scheme, resulting in an updated thermal field distribution state. The thermal field distribution state is compared with a preset equilibrium threshold. If the deviation exceeds the threshold, finite element analysis is used to calculate thermal field correction parameters, resulting in a corrected thermal field equilibrium model. The temperature control parameters are updated using the corrected thermal field equilibrium model to obtain the dynamic temperature equilibrium state of the micro-defect induced region. Based on the support vector machine algorithm combined with real-time monitoring of the temperature equilibrium state of the micro-defect induced region, the potential defect formation trend is predicted, resulting in a defect risk assessment result. The parameters of the iterative control algorithm are adjusted according to the defect risk assessment result to obtain an optimized temperature control strategy.

[0104] In this embodiment, more specifically, during the silicon crystal growth process, the initial temperature distribution is first calculated using thermal field simulation software. The crystal growth furnace is divided into 10,000 grid elements using the finite element method. The input parameters include a heater power of 5kW, a crystal diameter of 200mm, and a growth rate of 0.8mm / min. The simulation results show that the crystal center temperature is 1420.0℃, the edge temperature is 1415.5℃, and the temperature gradient is 4.5℃. The finite element analysis also includes the final stable temperature control parameters obtained by solving the heat conduction equation.

[0105] Next, the temperature control parameters are input into the real-time control system. The gradient descent algorithm is used to optimize the heater's power. Combined with the moving average filtering with a historical data window width of a preset value, the power is dynamically adjusted from the initial power parameter to the target power parameter, so as to achieve the purpose of controlling the temperature deviation within the preset range.

[0106] Subsequently, a thermal field equalization correction module is introduced. Fourier transform is used to perform frequency domain analysis on the temperature field image, identifying high-frequency noise components and filtering out thresholds. Then, Gaussian kernel convolution is applied to correct non-uniform regions. The corrected equalization model is as follows: Among them, the equilibrium increment Statistical variance of 1000 paths simulated in Monte Carlo simulation Calculated at ℃, ; This is the reference temperature.

[0107] The local temperature is dynamically adjusted for the micro-defect-induced region, such as the oxygen precipitation-sensitive region in the middle of the crystal, to 1418.0℃, ensuring that the defect density is reduced to below 10^4 / cm³. Through the continuous feedback loop of the aforementioned parameter optimization and power adjustment, a closed-loop thinking chain is formed. Among them, the final stable temperature control parameter directly affects the power iteration accuracy, while the equalization correction is based on the temperature field data after power output to further refine the process, avoiding the accumulation of thermal stress induced by defects such as vacancy clusters. Ultimately, a dynamic temperature equilibrium state is achieved throughout the growth process, improving the crystal quality by at least 15%.

[0108] This invention provides a thermal zone temperature optimization system for a crystal pulling furnace, which uses the thermal zone temperature optimization method to control the thermal zone temperature of the crystal pulling furnace, comprising:

[0109] The data acquisition and processing module is used to collect thermal field temperature and heater power data, and generate temperature deviation maps and power reduction indicators.

[0110] An anomaly diagnosis module is used to identify anomalies based on the deviation map and power index, and to determine areas of heater performance degradation and temperature asymmetry points caused by assembly errors.

[0111] The calibration compensation module is used to calculate the temperature offset and generate a power compensation scheme when the temperature of the performance degradation region is too low.

[0112] The deviation correction module is used to determine whether the temperature asymmetry point overlaps with the performance degradation region, and to generate deviation correction coefficients based on the quantization value of the overlapping region to update the thermal field temperature model.

[0113] The iterative control module is used to perform temperature prediction and residual deviation analysis using the updated model. Through iterative optimization, it determines the final stable temperature control parameters and drives the heater to adjust its power to achieve dynamic temperature equalization control in the silicon crystal micro-defect induced region.

[0114] The present invention also provides a crystal pulling furnace, wherein the thermal field temperature optimization method of the crystal pulling furnace is applied to control the thermal field temperature of the crystal pulling furnace.

[0115] Inside the crystal pulling furnace, a sensor array is used to collect real-time data on the thermal field temperature distribution and heater power output in the silicon crystal micro-defect induced zone. For example, the temperature sensor array records a core temperature of 1450.2℃ and an edge temperature of 1428.5℃, with a power output of 85.3kW. Subsequently, a Kalman filter is used as a noise filtering mechanism to process the signal. The filter gain is calculated through state equations and observation equations to obtain the smoothed current temperature deviation. For example, the map shows a core deviation of -2.1℃ and an edge deviation of +1.8℃. Based on historical averages, the power decrease index is 3.2%, thus providing a reliable basis for subsequent analysis.

[0116] Next, based on this deviation map and power degradation index, the isolated forest algorithm is used to detect outliers in the visualized temperature data; for example, a contamination rate of 0.05 is set for training the model, calculating the outlier score for each point, and identifying multiple outliers; such as coordinates. If a temperature jump occurs, these abnormal points are filled and repaired using radial basis function interpolation. Further, the performance degradation area of ​​the heater is determined to be where the output of the lower heater decays, and the temperature asymmetry points caused by assembly errors are directly related to the degradation compensation requirements. If the temperature in the performance degradation area is lower than a preset threshold, these areas are analyzed through temperature offset calibration. The least squares error compensation method is used to optimize the heating power distribution parameters, and the weight coefficients in the linear model are solved, i.e., the contribution weight of each heating power to temperature change. This yields the adjusted power compensation scheme for the upper and lower heaters, which ensures energy balance based on a physical heat conduction model.

[0117] Furthermore, temperature asymmetry point data corresponding to component assembly errors are obtained, as shown in the point set on the left. Statistical thresholds are used to determine whether these points overlap with performance degradation areas. The mean and standard deviation are calculated. If the deviation exceeds a preset value, the points overlap, and a deviation correction coefficient for the overlapping area is obtained. This deviation correction coefficient is used to bridge the logical correction between asymmetry and degradation. Anomalies are eliminated using the deviation correction coefficient, and the finite element method is used to reconstruct and update the thermal field temperature simulation model. Real-time data is used to verify the correlation coefficients and determine the core and edge temperatures of the optimized temperature distribution prediction results. This verification process enhances the model's robustness.

[0118] Next, based on the optimized temperature distribution prediction results and deviation map analysis, the remaining deviation is determined, for example, by calculating the root mean square and determining whether it exceeds a preset threshold. If it exceeds the preset threshold, gradient descent is used for iterative optimization to adjust the compensation scheme. After multiple iterations with the learning rate until convergence, the final number of power allocation parameters, including upper, lower, and asymmetric correction parameters for stable temperature control, is obtained. The above iterations ensure that the deviation is minimized. The power allocation parameters of these stable temperature control parameters are then applied in real time to drive the heater to perform power adjustment through scheme iteration. Combined with thermal field equalization correction such as Fourier transform filtering of low-frequency drift, the dynamic temperature equalization state of the micro-defect induced region during silicon crystal growth is obtained, maintaining fluctuations below a preset value to promote uniform crystal growth. The entire process forms a closed-loop adaptive control chain.

[0119] Furthermore, this invention collects thermal field temperature distribution data using a sensor array with a sampling frequency of 10Hz; calculates a temperature deviation map with a deviation threshold of ±2℃; detects outliers using the isolated forest algorithm with a contamination rate of 0.05; repairs outliers using Kriging interpolation; if the temperature in the performance degradation area is below 1400℃, a power compensation scheme is generated using least squares temperature offset calibration; it obtains assembly error temperature asymmetry points and calculates the deviation correction coefficient for overlapping areas; it updates the finite element thermal field model based on the correction coefficient; it verifies the prediction results using real-time data; it analyzes the remaining deviation through the deviation map, and if the remaining deviation is >1℃, it uses gradient descent iterative optimization of the compensation scheme with a learning rate of 0.05; it obtains stable temperature control parameters; and it drives the heater to adjust its power to achieve a temperature fluctuation range of ±5℃.

[0120] In this invention, the residual deviation is used to quantify the overall degree of deviation between the temperature distribution prediction result and the real-time acquired data. The calculation formula is as follows:

[0121] ; in, It represents the number of data points. Let i be the predicted temperature value; For the i-th measured temperature This invention acquires real-time data on thermal field temperature and heater power, uses a noise filtering mechanism to generate a temperature deviation map and power drop index, and combines outlier detection and data interpolation to repair temperature asymmetry points, accurately identifying performance degradation areas and assembly error points. For areas with temperatures below a threshold, this invention optimizes heating power allocation through error compensation, generates a power compensation scheme, and updates the thermal field temperature simulation model using a deviation correction coefficient. Based on real-time data verification and iterative optimization, this invention adjusts the compensation scheme to eliminate residual deviations, obtains stable temperature control parameters, and ultimately drives the heater to achieve dynamic power adjustment, achieving thermal field equilibrium. This invention significantly improves the temperature uniformity of silicon crystal growth, effectively reduces the incidence of micro-defects, and enhances crystal quality and production stability.

[0122] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for optimizing the thermal field temperature of a crystal pulling furnace, characterized in that, include: Based on the collected thermal field temperature distribution data and heater power output data of the silicon crystal micro-defect induced zone in the crystal pulling furnace, the current temperature deviation map and power reduction index are obtained. Based on outlier detection, anomalies are identified in the image-based temperature data of temperature deviation map and power reduction index. The anomalies are filled and repaired, and the areas of heater performance degradation and temperature asymmetry caused by assembly errors are identified based on the anomalies. If the temperature of the identified performance degradation region is lower than the preset threshold, the performance degradation region is analyzed by temperature offset calibration and the heating power distribution parameters are optimized to obtain an adjusted power compensation scheme. Based on the power compensation scheme, the temperature asymmetry point data corresponding to the component assembly error is obtained, and it is determined whether the asymmetry point data overlaps with the performance degradation area to obtain the deviation correction coefficient of the overlapping area. The thermal field temperature simulation model was reconstructed and updated after removing the deviation correction coefficient, and the optimized temperature distribution prediction results were verified and confirmed by combining real-time data. By combining the deviation map with the temperature distribution prediction results, it is determined whether there is a residual deviation in the prediction results. If there is a residual deviation, the compensation scheme is adjusted by iterative optimization process to obtain the final stable temperature control parameters. The heater is driven to perform power regulation in real time by using the final stable temperature control parameters, and the dynamic temperature equilibrium state of the micro-defect induced region is obtained by combining thermal field equalization correction.

2. The method for optimizing the thermal field temperature of a crystal pulling furnace as described in claim 1, characterized in that, Real-time acquisition of temperature distribution data and heater power output data within the crystal pulling furnace is performed and stored as a raw dataset. Noise filtering is applied to the acquired signals in the raw dataset to generate a denoised dataset. A temperature deviation dataset is generated by calculating the difference between the thermal field temperature and the preset standard temperature based on the denoised dataset. If the deviation value in the temperature deviation dataset exceeds a preset threshold, spatial interpolation is performed on the deviation value to generate a temperature deviation map. The trend of heater power output data is extracted from the denoised dataset and the power drop index is calculated. The power drop index is compared with the preset power threshold to determine whether there are any power anomalies. If power anomalies exist, the temperature deviation map area corresponding to the anomaly point is marked to generate a defect-induced zone distribution.

3. The method for optimizing the thermal field temperature of a crystal pulling furnace as described in claim 2, characterized in that, Based on the temperature deviation map and power reduction index, outliers are identified by setting a preset threshold and the distribution data of outliers is obtained; outlier detection is performed on the imaged temperature data to determine the location of outliers; and the outliers are repaired to generate repaired temperature distribution data. Spatial correlation is calculated using the repaired temperature distribution data to determine the areas of heater performance degradation; if the spatial correlation is lower than a preset threshold, local weighted regression analysis is used to identify temperature asymmetry points and determine the areas affected by assembly errors. The correlation between power reduction index and temperature distribution data is obtained, and the degree of heater performance degradation is determined based on a linear regression model. Temperature distribution data before and after repair are compared, and the accuracy of data repair is verified before generating the final degradation area and asymmetric point distribution.

4. The method for optimizing the thermal field temperature of a crystal pulling furnace as described in claim 3, characterized in that, If the temperature of the performance degradation area is detected to be lower than the preset threshold, the temperature value of the abnormal area is obtained from the temperature detection data, the temperature offset is calculated, and the temperature offset analysis result is obtained. Based on the temperature offset analysis results, predict the impact of temperature offset on power distribution and determine the error compensation parameters; if the error compensation parameters exceed the preset range, adjust the heating power distribution through an iterative optimization algorithm to obtain the optimized power distribution parameters; An initial power compensation scheme is generated using the optimized power allocation parameters, resulting in a preliminarily adjusted scheme. Power allocation features are extracted from the initially adjusted scheme and classified to determine the stability of the scheme. If the stability of the scheme meets the preset conditions, the initially adjusted scheme is fine-tuned through the power allocation optimization method to obtain the final power allocation scheme. Generate regional calibration instructions based on the final power allocation scheme and output the calibrated heating power allocation results.

5. The method for optimizing the thermal field temperature of a crystal pulling furnace as described in claim 4, characterized in that, The component assembly error data in the power compensation scheme is obtained, and an error data set is generated through the data extraction process; the temperature asymmetry point distribution in the error data set is extracted to obtain the temperature distribution characteristics; Based on the temperature distribution characteristics, the overlapping area between the temperature asymmetry point and the performance degradation area is determined. The degree of overlap is calculated based on the overlapping area determination results to obtain the quantified value of the overlapping area. A deviation correction coefficient is generated based on the quantified value of the overlapping area. The power compensation scheme is adjusted using the deviation correction coefficient to obtain optimized compensation parameters; the optimized compensation parameters are then used to update the component assembly error data and generate a new error data set.

6. The method for optimizing the thermal field temperature of a crystal pulling furnace as described in claim 5, characterized in that, The parameters of the thermal field temperature model are adjusted by the deviation correction coefficient to obtain the optimized temperature model; the optimized temperature model is then used to reconstruct the temperature distribution and generate the initial temperature distribution data. Ambient temperature data is acquired from a real-time data acquisition system to determine the validity of the real-time data. If the deviation between the real-time data and the initial temperature distribution data exceeds a preset threshold, the temperature distribution is adjusted through data fusion analysis to obtain a corrected temperature distribution. The corrected temperature distribution is used to predict the temperature at future time points to generate a temperature prediction result. The accuracy of the prediction result is judged by comparing the temperature prediction result with the real-time data acquisition result. If the accuracy of the prediction result is lower than the preset standard, the model update mechanism is updated to regenerate the optimized temperature model.

7. The method for optimizing the thermal field temperature of a crystal pulling furnace as described in claim 6, characterized in that, The collected temperature distribution data is used to generate an initial temperature distribution prediction result; deviation data is extracted from the prediction result to generate a deviation map; based on the deviation map, a preset threshold is used to determine whether there is a residual deviation. If there is a residual deviation, the location and magnitude of the deviation are recorded to obtain the deviation distribution information; the parameters of the compensation scheme are iteratively adjusted based on the deviation distribution information to generate an updated compensation scheme; the updated compensation scheme is used to regenerate the temperature distribution prediction result to obtain new temperature distribution data. The deviation map is generated and the residual deviation is judged from the new temperature distribution data. If the residual deviation is lower than the preset threshold, the final stable temperature control parameter is determined. The temperature distribution prediction model is updated based on the final stable temperature control parameter to obtain the optimized prediction result.

8. The method for optimizing the thermal field temperature of a crystal pulling furnace as described in claim 7, characterized in that, Temperature distribution data within the silicon crystal growth cavity is collected in real time using a sensor array to obtain a multi-point temperature matrix. If local high or low temperature regions exist in the temperature matrix, the heater power is adjusted to obtain an optimized power allocation scheme. The heater is then driven to perform power regulation according to the optimized power allocation scheme to obtain an updated thermal field distribution state. The thermal field distribution state is compared with the preset equilibrium threshold. If the deviation exceeds the threshold, the thermal field correction parameters are calculated to obtain the corrected thermal field equilibrium model. The temperature control parameters are updated according to the corrected thermal field equilibrium model to obtain the dynamic temperature equilibrium state of the micro-defect induced zone. Real-time monitoring of the temperature equilibrium state in the micro-defect induction zone and prediction of potential defect formation trends yield defect risk assessment results; The parameters of the iterative control algorithm are adjusted based on the defect risk assessment results to obtain an optimized temperature control strategy.

9. A thermal zone temperature optimization system for a crystal pulling furnace, comprising using the thermal zone temperature optimization method for a crystal pulling furnace as described in any one of claims 1-8 for controlling the thermal zone temperature of the crystal pulling furnace, characterized in that, The data acquisition and processing module is used to collect thermal field temperature and heater power data, and generate temperature deviation maps and power reduction indicators. An anomaly diagnosis module is used to identify anomalies based on the deviation map and power index, and to determine areas of heater performance degradation and temperature asymmetry points caused by assembly errors. The calibration compensation module is used to calculate the temperature offset and generate a power compensation scheme when the temperature of the performance degradation region is too low. The deviation correction module is used to determine whether the temperature asymmetry point overlaps with the performance degradation region, and to generate deviation correction coefficients based on the quantization value of the overlapping region to update the thermal field temperature model. The iterative control module is used to perform temperature prediction and residual deviation analysis using the updated model, determine the final stable temperature control parameters through iterative optimization, and drive the heater to adjust the power to achieve balanced control of the dynamic temperature of the silicon crystal micro-defect induced region.

10. A crystal pulling furnace, characterized in that, The thermal field temperature of the crystal pulling furnace is controlled by applying the thermal field temperature optimization method as described in any one of claims 1-8.

Images