A method and system for controlling the temperature of a thermal field based on crystal growth

Abstract

The application relates to the technical field of crystal growth control, and discloses a crystal growth thermal field temperature control method and system. The method comprises the following steps: acquiring thermal field temperature distribution data in a crystal growth process, combining a crystal growth rate and material phase change parameters to determine key influence factors of thermal field temperature control, analyzing a mapping relationship between the thermal field temperature and the crystal growth quality according to the thermal field temperature distribution data and the key influence factors, collecting historical thermal field temperature adjustment records and crystal defect data, combining a thermal field temperature dynamic adjustment model to generate a temperature adjustment strategy set, simulating thermal field temperature variation trends under different adjustment strategies based on the temperature adjustment strategy set, extracting stability features and response speed features in the temperature variation trends, screening an optimal temperature adjustment strategy according to the stability features and the response speed features, and dynamically adjusting temperature control parameters of a crystal growth thermal field in combination with real-time thermal field temperature monitoring data.

Description

Technical Field

[0001] This invention relates to the field of crystal growth control technology, specifically to a method and system for controlling the temperature of a crystal growth thermal field. Background Technology

[0002] Crystal growth is a critical process in materials science and the semiconductor industry, and its quality directly affects the performance of the final product. Thermal field temperature is the most crucial control parameter in crystal growth; the uniformity, stability, and dynamic response characteristics of the temperature distribution play a decisive role in crystal defect formation, growth rate, and final crystal quality. Traditional crystal growth thermal field control mainly relies on methods based on fixed temperature profiles or simple proportional-integral-derivative (PDI-D) control. These methods are ill-suited to the complex nonlinear and time-varying characteristics of crystal growth. At different stages of crystal growth, such as seeding, shoulder formation, and constant-diameter growth, the requirements for thermal field temperature vary significantly, making it impossible to achieve full-process optimization with fixed control strategies.

[0003] Existing thermal field temperature control methods primarily focus on maintaining the stability of the set temperature, lacking effective regulation of the spatial distribution of the temperature field. The spatial distribution characteristics of the temperature gradient and thermal field symmetry near the crystal growth interface are crucial for suppressing defect formation, but traditional single-point temperature control cannot fully reflect the spatial state of the thermal field. Furthermore, adjustments to control parameters largely rely on operator experience, lacking systematic theoretical guidance. When abnormalities occur during the growth process, such as interface flipping or compositional supercooling, traditional control methods exhibit a lag in response, often intervening only after crystal defects have formed, impacting yield and material properties.

[0004] With the development of sensor technology, an increasing number of temperature monitoring points are being deployed in crystal growth equipment, generating massive amounts of historical temperature data. However, this data has not been fully utilized to optimize control strategies. There is a lack of effective correlation analysis between historical temperature regulation records and corresponding crystal quality data, failing to form a knowledge loop from process data to quality control. The complex physicochemical changes in the crystal growth process make it difficult to accurately establish temperature control models. Data-driven intelligent control methods offer a new approach to solving this problem, but their application in crystal growth thermal field control still requires further exploration. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for controlling the temperature of a crystal growth thermal field, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a method for controlling the temperature of a crystal growth thermal field, the method comprising:

[0007] Acquire thermal field temperature distribution data during crystal growth, and combine crystal growth rate and material phase transformation parameters to determine the key influencing factors of thermal field temperature control;

[0008] Based on the thermal field temperature distribution data and key influencing factors, a dynamic adjustment model for thermal field temperature is established to analyze the mapping relationship between thermal field temperature and crystal growth quality.

[0009] Historical thermal field temperature regulation records and crystal defect data are collected and combined with a dynamic thermal field temperature regulation model to generate a set of temperature regulation strategies.

[0010] Based on a set of temperature regulation strategies, the temperature change trend of the thermal field under different regulation strategies is simulated, and the stability and response speed characteristics in the temperature change trend are extracted.

[0011] Based on stability and response speed characteristics, the optimal temperature regulation strategy is selected, and the temperature control parameters of the crystal growth thermal field are dynamically adjusted by combining real-time thermal field temperature monitoring data.

[0012] Preferably, the key influencing factors for determining the thermal field temperature control include:

[0013] The thermal field temperature distribution data is divided into regions, and the temperature gradient data of each region is extracted.

[0014] Analyze the correlation between temperature gradient data and crystal growth rate to identify temperature fluctuation-sensitive regions;

[0015] By combining the material phase transformation parameters, the deviation of heat conduction efficiency in the temperature fluctuation sensitive area is calculated;

[0016] Based on the deviation value of heat conduction efficiency, the key influencing factors and their priorities for thermal field temperature control are determined.

[0017] Preferably, the establishment of the dynamic temperature regulation model for the thermal field includes:

[0018] Extract periodic variation patterns from thermal field temperature distribution data to construct temporal features of temperature changes;

[0019] By combining crystal growth quality inspection data, a correlation rule between the temporal characteristics of temperature changes and crystal defect types is established.

[0020] Based on the association rules, generate the constraints for the dynamic temperature adjustment model of the thermal field;

[0021] Based on constraints, the parameter configuration of the dynamic temperature regulation model of the thermal field is optimized.

[0022] Preferably, the set of temperature regulation strategies includes:

[0023] Cluster analysis was performed on historical thermal field temperature regulation records to extract typical temperature regulation patterns;

[0024] The effectiveness of typical temperature regulation modes is evaluated by combining crystal defect data;

[0025] Based on the effectiveness evaluation results, a set of temperature regulation strategies is generated, and the applicable conditions for each strategy are marked.

[0026] Preferably, the simulation of the thermal field temperature change trend under different adjustment strategies includes:

[0027] A thermal field temperature simulation environment is constructed based on a set of temperature regulation strategies.

[0028] In the thermal field temperature simulation environment, real-time thermal field temperature monitoring data is injected as the initial condition;

[0029] Run the simulation environment and record the temperature change curves under each adjustment strategy;

[0030] Stability and response speed characteristics are extracted from the temperature change curve.

[0031] Preferably, the extraction of stability and response speed characteristics from the temperature change trend includes:

[0032] Calculate the standard deviation of the temperature change curve as a quantitative indicator of stability characteristics;

[0033] The time required for the temperature change curve to reach a set threshold is used as a quantitative indicator of the response speed characteristic.

[0034] By combining quantitative indicators, a performance evaluation report on the temperature regulation strategy is generated.

[0035] Preferably, the selection of the optimal temperature regulation strategy includes:

[0036] Based on the performance evaluation report, exclude adjustment strategies that fail to meet the standards for stability or response speed characteristics;

[0037] Among the remaining regulation strategies, the regulation strategy with the best overall performance is selected as the optimal temperature regulation strategy;

[0038] Record the parameter configuration and execution conditions of the optimal temperature regulation strategy.

[0039] Preferably, the temperature control parameters for dynamically adjusting the crystal growth thermal field include:

[0040] Real-time monitoring of thermal field temperature distribution data to identify areas of abnormal temperature;

[0041] Based on the optimal temperature regulation strategy, generate correction parameters for the temperature anomaly region;

[0042] The correction parameters are injected into the thermal field temperature control system to complete the dynamic adjustment of the temperature control parameters.

[0043] Preferably, the dynamic adjustment of the temperature control parameters includes:

[0044] Verify whether the thermal field temperature distribution data after the parameter correction is executed meets expectations;

[0045] If it does not meet expectations, the temperature regulation strategy screening process will be retried.

[0046] If it meets expectations, record the current temperature control parameters as the baseline configuration.

[0047] Preferably, the present invention further includes a crystal growth thermal field temperature control system, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the crystal growth thermal field temperature control method described above.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] This invention, through comprehensive analysis of thermal field temperature distribution, growth rate, and material phase transformation parameters, identifies key influencing factors for temperature control, making the control strategy more targeted to the essential needs of crystal growth. Traditional methods often focus only on the absolute value of temperature, while this method considers multiple factors affecting crystal quality, making temperature control more scientific and systematic.

[0050] The established dynamic temperature regulation model for the thermal field reveals the intrinsic relationship between temperature and crystal quality, providing a theoretical basis for the formulation of control strategies. By analyzing patterns in historical data, this model helps to understand the formation mechanism of crystal defects under different temperature conditions, thereby proactively avoiding undesirable temperature patterns in control.

[0051] By generating a set of temperature regulation strategies based on historical data and selecting the optimal strategy through simulation analysis, the scientific nature and reliability of control decisions are greatly improved. Traditional experience-based adjustments are replaced by systematic strategy generation and evaluation, reducing the uncertainty of human factors and improving the consistency and repeatability of control.

[0052] Extracting and optimizing the stability and response speed characteristics of temperature change trends ensures the dynamic performance of the control system. The crystal growth process requires both stable temperature and rapid response to process changes; this method achieves this balance through multi-objective optimization, meeting the control needs of different stages of crystal growth.

[0053] The real-time monitoring and dynamic adjustment mechanism enables the control system to be adaptive, continuously optimizing control parameters based on the actual growth state. This closed-loop control method effectively addresses the time-varying characteristics and various disturbances in the crystal growth process, improving control accuracy and robustness, and providing a reliable guarantee for obtaining high-quality crystals. This method promotes the development of crystal growth control towards intelligence and precision. Attached Figure Description

[0054] Figure 1 A graph showing the relationship between temperature regulation and growth rate;

[0055] Figure 2 A sub-flowchart for identifying key influencing factors in thermal field temperature control;

[0056] Figure 3 A sub-flowchart for establishing a dynamic temperature regulation model for the thermal field;

[0057] Figure 4 This is a multi-dimensional performance evaluation chart for the strategy. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] This invention provides a method and system for controlling the temperature of a crystal growth thermal field. The method includes: during the operation of the crystal growth equipment, acquiring real-time temperature distribution data of the thermal field using a temperature sensor array. This data covers the temperature values ​​at different locations in the thermal field and their changes over time. Simultaneously, monitoring crystal growth rate parameters, such as crystal pulling speed or deposition rate, and integrating material phase transformation parameters, such as melting point and crystallization temperature. Based on this data, using correlation analysis and statistical methods, identifying the factors most sensitive to temperature fluctuations as key influencing factors, these factors may include temperature gradients or thermal conductivity characteristics in specific regions of the thermal field. Based on the thermal field temperature distribution data and key influencing factors, a dynamic temperature adjustment model for the thermal field is constructed. This model employs time-series analysis techniques, such as autoregression or machine learning algorithms, to characterize the mathematical relationship between temperature changes and crystal growth quality indicators. Through model simulation, the impact of temperature adjustment on growth quality can be quantified. Historical thermal field temperature adjustment records and corresponding crystal defect data are collected from a historical database. These records include temperature setpoint adjustment sequences and actual response curves. Combining the dynamic temperature adjustment model, pattern mining is performed on the historical data to generate a set of temperature adjustment strategies, each strategy containing specific temperature control instructions and expected effects. Using a computer simulation platform, the temperature variation trend of the thermal field under different temperature regulation strategies was simulated based on a set of temperature regulation strategies. During the simulation, real-time monitoring data was injected as initial conditions, and multi-scenario simulations were run. Key features, such as stability features and response speed features, were extracted from the generated temperature change curves. Based on the evaluation results of stability and response speed features, a multi-objective optimization algorithm was used to select the optimal temperature regulation strategy. Combined with real-time thermal field temperature monitoring data, the temperature control parameters of the crystal growth thermal field, such as heater power or cooling rate, were dynamically adjusted through a feedback control loop to ensure that the thermal field temperature is always within the optimal range.

[0060] Example 1: The acquired crystal growth thermal field temperature distribution data is a multidimensional dataset containing spatial coordinate information and timestamps. The data originates from a high-temperature resistant thermocouple sensor array and an infrared thermal imager uniformly arranged around the thermal field. Data preprocessing includes signal filtering to eliminate random noise interference and time synchronization calibration of readings from different sensors to ensure data consistency in the spatiotemporal dimensions. Region partitioning is performed based on the geometry of the thermal field and the physical characteristics of crystal growth, dividing the entire thermal field into several functional sub-regions, such as the core heating zone, crystal growth interface zone, edge insulation zone, and bottom heat dissipation zone. The boundary coordinates of each sub-region are predefined and stored in the system configuration file, based on the distribution of isotherms within the thermal field and the physical layout of the heating elements. The temperature gradient data extraction process employs a spatial difference calculation method. For multiple temperature measurement points arranged within each sub-region, the temperature difference between adjacent measurement points is calculated and compared with the distance between points to obtain discrete temperature gradient values. These discrete temperature gradient values ​​are used to generate a continuous gradient distribution cloud map using the Kriging interpolation algorithm. This allows for the calculation of characteristic parameters such as the average temperature gradient, maximum temperature gradient, and gradient direction distribution within each sub-region. The sampling frequency of the temperature gradient data is consistent with the crystal growth rate monitoring frequency, typically set to once per second, thus enabling the capture of rapid temperature changes during the growth process.

[0061] When analyzing the correlation between temperature gradient data and crystal growth rate, a time series alignment method is used to match the two types of data. Crystal growth rate is measured in real time using a laser rangefinder or image processing technology, recording the change in crystal length per unit time. Correlation analysis uses a sliding window Pearson correlation coefficient, with the window size dynamically adjusted according to the crystal growth stage. A larger time window is used during the constant-diameter growth stage to smooth short-term fluctuations, while a smaller window is used during the shoulder formation or finishing stages to capture rapid changes. By calculating the absolute value of the correlation coefficient between the temperature gradient sequence and the growth rate sequence for each sub-region, sub-regions with correlation coefficients consistently higher than a set threshold are identified and marked as temperature fluctuation sensitive areas. A thermophysical property database of the material is introduced, incorporating the calculation process of material phase transition parameters. This database contains parameters such as thermal conductivity, specific heat capacity, and thermal diffusivity for different crystalline materials in the solid, liquid, and near phase transition points. The calculation of the thermal conductivity deviation value is based on the discrete form of the unsteady-state heat conduction equation. The actual temperature field data of the sensitive area is used as the boundary condition, and the theoretical thermal conductivity of the material is used as the reference value. The heat flux density distribution within the region is solved using the finite difference method. Actual thermal conductivity is obtained by measuring the ratio of heat flux density to temperature gradient, while theoretical thermal conductivity is calculated based on the material's standard thermal conductivity. The deviation value of thermal conductivity is the percentage error between the actual and theoretical values. The calculation process takes into account the effect of temperature on the material's thermal conductivity and uses temperature-dependent variable property parameters.

[0062] The process of identifying key influencing factors and their priorities involves establishing a multi-index evaluation system. Evaluation indicators include the magnitude of the thermal conductivity deviation, the fluctuation range of the deviation, the geometric dimensions of sensitive areas, and their positional importance within the thermal field. Each sensitive area is assigned a comprehensive score, calculated using a formula that combines the absolute value of the deviation, the standard deviation of the deviation, and a regional weighting coefficient. The regional weighting coefficient is determined by the area's impact on crystal quality; for example, the weighting coefficient for the crystal growth interface area is higher than that for the edge insulation area. Sensitive areas with comprehensive scores exceeding a threshold are identified as key influencing factors and prioritized according to their scores from highest to lowest. The ranking results are stored in a list, where each key influencing factor corresponds to an area number, deviation value, and priority level. The entire implementation process is automated through a dedicated data acquisition and processing module. This module employs a multi-threaded architecture: one thread is responsible for real-time temperature data acquisition and region division; another thread calculates the temperature gradient and correlation coefficient in parallel; and a third thread performs thermal conductivity deviation calculation and priority ranking. Data is transferred between different processing stages via shared memory and includes timestamps to ensure data synchronization. All algorithm parameters and thresholds are managed through configuration files, allowing adjustments based on different crystal growth process requirements. Processing results are updated in real time and displayed on the monitoring interface, while also being written to a historical database for subsequent model building. The system also includes an anomaly detection mechanism; when the thermal conductivity deviation in a certain area suddenly increases beyond the safe range, an alarm is triggered, prompting operators to intervene and investigate.

[0063] Example 2: Establishing a dynamic temperature regulation model for the thermal field begins with in-depth time-series analysis of continuously acquired thermal field temperature distribution data. The temperature data is a sequence with high-precision timestamps, and the sampling points cover every preset sub-region of the thermal field. The method used to extract periodic variation patterns combines spectral analysis and pattern recognition. A Fast Fourier Transform algorithm is applied to the long-term series data of each temperature measurement point to identify the fundamental and harmonic components hidden behind random fluctuations. These components correspond to periodic factors such as heater power adjustment cycles, cooling system operating cycles, and diurnal variations in ambient temperature. Constructing the time-series characteristics of temperature changes requires integrating multi-dimensional indicators in both the time and frequency domains. Time-domain features include the mean, variance, skewness, and kurtosis within the sliding window, while frequency-domain features extract the amplitude, phase, and energy percentage of the dominant frequency. All these features are organized into feature vectors at fixed time intervals and labeled with the corresponding crystal growth stages. Establishing association rules based on crystal growth quality inspection data requires simultaneously obtaining crystal defect inspection reports. Defect data comes from online optical inspection systems, X-ray topology analyzers, and subsequent metallographic inspection results. Defect types are classified and coded according to international standards, including quantitative indicators such as dislocation density, stacking fault probability, and impurity concentration level. Association rule mining algorithms are used to establish the correlation rules between temperature change time-series characteristics and crystal defect types, searching for frequent itemsets where temperature feature values ​​and defect types co-occur from a large amount of historical data. Each association rule includes antecedent and consequent terms, and three statistical indicators—support, confidence, and lift—are calculated to measure the rule strength. Minimum support and minimum confidence thresholds are set during rule generation to filter out random associations. Constraints for the dynamic temperature regulation model of the thermal field are generated based on the association rules, transforming strong association rules into mathematical constraint forms. For example, when a set of temperature feature values ​​is strongly correlated with high dislocation density, a corresponding upper limit constraint on temperature fluctuation is generated, expressed as a set of inequalities or boundary conditions. The parameter configuration of the dynamic temperature regulation model of the thermal field is optimized based on the constraints. The model uses partial differential equations with time delay terms to describe the evolution of the thermal field temperature. The undetermined parameters in the equations include the thermal diffusivity, boundary heat transfer coefficient, and internal heat source intensity. The optimization process is transformed into a constrained least squares problem. The objective function is the sum of squared residuals between the model output and the actual measured value. The constraints are derived from the transformation results of the aforementioned association rules. The solution algorithm uses the sequential quadratic programming method, iteratively adjusting the model parameters until the convergence criterion is met.

[0064] The process of generating a set of temperature regulation strategies begins with cluster analysis of historical thermal field temperature regulation records. These records are stored in a relational database, with each record containing information such as a timestamp, regulation type, operation parameters, and temperature distribution changes before and after execution. An improved K-means clustering algorithm is used to map multidimensional regulation record vectors to a feature space, with Mahalanobis distance used to eliminate the influence of dimensions. The number of clusters is determined using the silhouette coefficient method, and each cluster center represents a typical temperature regulation pattern. Pattern characteristics include regulation amplitude, effective area, and duration sequence. The extracted typical temperature regulation patterns are matched with contemporaneous crystal defect data to evaluate the effectiveness of each pattern. A comparative experimental design is used to evaluate the effectiveness of typical temperature regulation patterns, selecting time intervals from historical data where specific regulation patterns were applied, and control intervals where the pattern was not applied. Effectiveness evaluation indicators include the percentage reduction in defect rate, the degree of improvement in crystal quality parameters, and the magnitude of improvement in growth process stability. Paired t-tests are used for statistical testing, calculating p-values ​​to determine the significance of the improvement. The significance level is set at 0.05; only regulation patterns that pass the significance test are considered effective. A set of temperature regulation strategies is generated based on the effectiveness evaluation results. Each strategy is described in a structured data format, including fields such as regulation mode number, applicable growth stage, target defect type, expected effect index, and operating parameter range. The strategy set is stored in a database table with multi-level indexes to support fast retrieval. Each strategy is annotated with detailed applicable conditions, including limiting parameters such as crystal material type, thermal field structure specifications, and growth rate range, ensuring the strategy's targeted application. The strategy set update mechanism is designed as an incremental learning mode. When new regulation records and defect data accumulate to a certain scale, cluster analysis and effectiveness re-evaluation are automatically triggered, achieving self-optimization of the strategy library.

[0065] A dynamic temperature control model and strategy were established during the silicon carbide single crystal growth process. The crystal growth system was equipped with multiple independently controlled graphite heating elements surrounding the crucible, and thirty-six type B thermocouples distributed on different planes of the thermal field. Temperature distribution data was acquired five times per second, and the data stream was transmitted to the data processing unit in real time. The data included the absolute temperature value and the temperature difference relative to the reference point at each temperature measurement point. A time series decomposition algorithm was used to extract the periodic variation pattern, decomposing the continuous 72-hour growth data into trend, periodic, and residual terms. The periodic term was identified through spectral analysis, with the main periodic components including a 120-minute main heating cycle and a 15-minute auxiliary heating cycle. A multi-scale feature extraction method was used to construct the temporal features of temperature change. The mean rate of temperature change was calculated on a 30-second short-scale, the temperature fluctuation amplitude was calculated on a 10-minute medium-scale, and the approximate entropy of the temperature trajectory was calculated on a 1-hour long-scale. The temporal feature vector contained 128 dimensions, covering time domain, frequency domain, and nonlinear features. Crystal growth quality inspection data comes from full-section scanning tests of the ingot after growth, including X-ray bicrystalline diffraction full width at half maximum (FWHM), dislocation density distribution maps, and resistivity uniformity data. Defect type coding uses a three-digit code system, with the first digit indicating the major defect category and the last two digits indicating the specific subtype. An improved Apriori algorithm is used to establish the association rules between temperature change time-series characteristics and crystal defect types, with a minimum support of 5% and a minimum confidence of 75%. The rule mining process discovered several strong association patterns; for example, when the temperature fluctuation amplitude in region 3 exceeds eight degrees Celsius and the fluctuation frequency is within a specific frequency band, there is a significant correlation with the occurrence of screw dislocation defects. Association rules are stored in production form, with the antecedent being a combination of temperature characteristic conditions and the consequent being the probability distribution of defect types. Constraints for the dynamic temperature adjustment model of the thermal field are generated based on the association rules, transforming qualitative rules into mathematical constraint expressions. For example, the association rule for screw dislocation defects is transformed into a constraint condition for the temperature fluctuation amplitude in region 3, expressed as a system of inequalities, limiting the temperature fluctuation amplitude in this region to no more than five degrees Celsius. Model parameter configuration includes the diffusion term coefficients in the heat conduction equation, boundary condition update frequency, and feedback gain matrix element values. The optimization process employs a constrained particle swarm optimization algorithm with a population size of fifty particles and five hundred iterations. The objective function is the root mean square error between the model's predicted temperature field and the actual measured field. When generating the temperature regulation strategy set, cluster analysis was performed on temperature regulation records from two hundred growth experiments over the past three months. The cluster feature vector includes eight dimensions: regulation timing, effective area, regulation amplitude, and duration. A density clustering algorithm automatically determined the number of clusters to be seven. Each cluster center corresponds to a typical temperature regulation mode, with characteristics including a slow heating mode in the preheating stage, a micro-oscillation mode in the constant diameter stage, and a stepped cooling mode in the final stage.The effectiveness evaluation of typical temperature regulation modes employs a counterfactual reasoning framework. A control group is constructed from historical data, and the average treatment effect of each mode on the defect rate is calculated using causal inference methods. The effectiveness evaluation results generate a strategy effectiveness scoring table, which includes fields such as regulation mode number, applicable defect type, effect significance index, and upper and lower bounds of the confidence interval. A set of temperature regulation strategies is generated based on the scoring table. Strategies are stored using standardized templates, and each strategy includes structured information such as trigger conditions, operation sequence, expected effect, and risk level. The applicable conditions for each strategy are detailed, specifying process parameter boundaries such as crystal orientation, doping concentration, and gas pressure range. A semantic-based search index is established in the strategy library, supporting multi-condition combined queries. A strategy update mechanism includes a periodic review process; after every ten growth experiments, the effectiveness of all strategies is re-evaluated, old strategies with insignificant effects are eliminated, and newly discovered effective modes are added. During implementation, a strategy effectiveness tracking system is established, recording actual effect data after each strategy application, including indicators such as temperature response curves, crystal quality parameters, and growth yield. The strategy optimization loop is implemented using a reinforcement learning algorithm, dynamically adjusting strategy parameters based on application effects to form a continuously improving closed-loop system. The strategy version management system records the historical changes to the strategy library, supporting longitudinal comparative analysis of strategy effects. The strategy generation and verification platform is integrated into the crystal growth control system, enabling automated pipeline operations from data acquisition to strategy application. The system includes a strategy security verification module to perform simulation testing and security checks on newly generated strategies, ensuring they do not trigger risks such as thermal instability. A strategy knowledge graph constructs the relationships between strategies, forming a strategy recommendation network to improve the efficiency of strategy retrieval and application.

[0066] Example 3: The thermal field temperature simulation environment is constructed based on the principles of computational fluid dynamics and heat transfer. A numerical model is established, and the simulation environment uses a three-dimensional unsteady-state heat conduction equation coupled with fluid motion equations to describe the heat transfer process within the thermal field. The model is discretized using the finite volume method, dividing the thermal field space into millions of hexahedral mesh elements. Each element stores physical quantities such as temperature, velocity, and pressure. Boundary conditions include parameters such as heater power density distribution, cooling water jacket convective heat transfer coefficient, and ambient radiative heat dissipation coefficient. Material property parameters are based on the actual structural configuration of the crystal growth device, including the thermophysical parameters of multiple layers such as graphite heating elements, carbon fiber insulation materials, and quartz crucibles. The simulation environment runs on a parallel computing cluster, using an explicit time-progression algorithm to solve the partial differential equations. Each time step is automatically adjusted according to the Courant number condition. Real-time thermal field temperature monitoring data is injected into the thermal field temperature simulation environment as initial conditions. The injection process is achieved through data assimilation technology. The real-time monitoring data comes from thermocouple measurements distributed across various layers of the thermal field. These measurements are spatially interpolated to generate the initial distribution of the entire three-dimensional temperature field. The data assimilation algorithm employs an ensemble Kalman filter to optimally fuse measured data with simulation predictions, correcting the initial simulation field to reduce model errors. The injection operation is performed every five minutes to ensure the simulation environment remains synchronized with the actual thermal field. During simulation operation, each strategy in the temperature control strategy set is sequentially loaded into the simulation parameter library. Strategy parameters include operable variables such as heater power adjustment curves, gas flow control sequences, and crucible lifting / lowering speed variations.

[0067] Recording temperature change curves under various control strategies requires setting up multiple virtual monitoring points, whose locations correspond exactly to the actual thermocouple placement points. The temperature change curve recording frequency is set to ten sampling points per second, recording the temperature time series for each monitoring point, the maximum temperature difference across the entire thermal field, and derived parameters such as the temperature distribution uniformity index. The simulation runtime covers a complete control cycle of crystal growth, typically ranging from thirty minutes to two hours, with the specific duration dynamically adjusted according to the growth stage. When extracting stability characteristics from the temperature change curves, the statistical properties of the curve data are calculated, with the sliding window standard deviation as the core indicator. The window width is set to sixty seconds, and the window sliding step size is ten seconds. Finally, the average of all window standard deviations is taken as the quantitative indicator of the curve's stability characteristics. The step response analysis method is used to extract response speed characteristics. After the control strategy is executed, the time required for the temperature curve to reach the target temperature stability band from its initial value is measured. The stability band is defined as the positive and negative fluctuation range of the target temperature value, with the fluctuation threshold set according to the thermal sensitivity of the crystal material. The response time is counted from the moment the control command is issued until the temperature value first enters the stability band and remains within the threshold for ten seconds. This time value serves as a quantitative indicator of response speed characteristics, while also recording auxiliary indicators such as overshoot and number of oscillations during the adjustment process.

[0068] A performance evaluation report for temperature regulation strategies is generated by combining quantitative indicators. The report uses a standardized template and includes evaluation dimensions such as strategy number, simulation time range, stability characteristic value, response speed characteristic value, degree of improvement in temperature uniformity, and energy consumption increment. Each dimension has a scoring standard; the smaller the stability characteristic value, the higher the score, and the smaller the response speed characteristic value, the higher the score. The comprehensive score is calculated using a weighted summation method, with weight coefficients configured according to crystal growth process requirements. Different weight allocation schemes can be set for different growth stages. The performance evaluation report is output as a structured data file, and visual charts are generated to compare the performance of each strategy, including radar charts, trend curves, and heat maps. The entire simulation and evaluation process is fully automated and pipelined. The simulation task scheduler manages the parallel simulation of multiple strategies, the performance evaluation module automatically analyzes the simulation results and generates reports, and the evaluation results are directly transmitted to the strategy selection system as a decision-making basis. The simulation environment also has a verification mechanism that periodically compares the simulation predictions with the actual measured values. When the error exceeds the allowable range, the model parameter correction process is automatically triggered to ensure that the simulation accuracy meets the requirements of engineering applications. During feature extraction, the preprocessing of the temperature change curve includes trend removal and outlier elimination. Detrending is performed using a multinomial fitting method, decomposing the original sequence into trend and fluctuation components, with stability analysis only applied to the fluctuation component. Outlier detection uses the Laida criterion, considering data points exceeding three standard deviations as outliers and replacing them with linear interpolation of adjacent points. The quantification index calculation program has undergone multiple iterations of optimization, employing parallel algorithms to accelerate the processing speed of large-scale temperature data. The performance evaluation report generation module integrates a business rule engine, enabling dynamic adjustment of evaluation criteria based on the specific requirements of different crystal materials. The entire system establishes a complete version management mechanism; all simulation cases, temperature curve data, and evaluation reports are labeled with version identifiers and timestamps, supporting historical data traceability and result reproduction. A secure data exchange interface is established between the simulation environment and the real-time monitoring system, employing a checksum mechanism to ensure data transmission integrity. Simulation results are automatically converted and standardized in unit format when output to the strategy library.

[0069] Example 4: The performance evaluation report includes quantitative indicators of stability characteristics, quantitative indicators of response speed characteristics, and other auxiliary evaluation parameters for each candidate strategy. A dual threshold filtering mechanism is used to exclude substandard adjustment strategies. The stability characteristic threshold is determined based on the thermal sensitivity of the crystal material and the requirements of the growth process, while the response speed characteristic threshold depends on the dynamic response requirements of the growth process. The strategy exclusion condition is designed as a logical OR relation, meaning that if any characteristic indicator exceeds the allowable range, the strategy is marked as substandard. The exclusion operation generates a strategy elimination list, recording the excluded strategy's number, the type of substandard characteristic, and the degree of deviation between the actual value and the threshold. Among the remaining adjustment strategies, the strategy with the best overall performance is selected using a multi-objective decision analysis method, establishing a comprehensive evaluation system that includes quantitative calculation and qualitative assessment. The quantitative calculation part uses a standardized scoring conversion formula to convert characteristic values ​​of different dimensions into dimensionless percentage scores. The stability characteristic score calculation formula is: 100 points are awarded when the actual standard deviation is less than or equal to the optimal standard deviation; points are deducted proportionally when it exceeds the optimal standard deviation. A similar rule is used to calculate the response speed characteristic score; the shorter the actual response time, the higher the score. The qualitative assessment considers engineering factors such as the complexity of strategy implementation, energy consumption levels, and equipment wear and tear risks, and is conducted manually by process experts according to the scoring criteria. The overall performance score is derived by a weighted sum of the quantitative and qualitative scores, with the weighting dynamically adjusted according to the growth stage. Response speed has a higher weight during the high-speed growth stage, while stability has a higher weight during the precision control stage.

[0070] The decision-making process for selecting the optimal temperature control strategy introduces the concept of Pareto optimal solution. When the comprehensive performance scores of multiple strategies are similar, further analysis is conducted to determine whether the characteristic distributions of these strategies are at the Pareto front. The Pareto front identification method employs a non-dominated sorting algorithm, marking strategies that are superior to any other strategy in all characteristics as the front solution set. When determining the final optimal strategy from the front solution set, the Euclidean distance between each strategy and the ideal solution is calculated using the ideal point method, and the strategy with the smallest distance is selected as the optimal temperature control strategy. The stability and response speed characteristics of the ideal solution are taken as the theoretically achievable best values, derived from historical best records or theoretical calculations. Recording the parameter configuration and execution conditions of the optimal temperature control strategy requires establishing a complete strategy archive, which includes basic information such as strategy number, generation time, applicable material type, and thermal field configuration version. The parameter configuration section records in detail all adjustable parameters of the temperature control loop, including PID parameters such as proportional gain coefficient, integral time constant, and derivative time constant, as well as advanced parameters such as feedforward compensation coefficient and filter time constant. The execution conditions clearly define the scope of application of the strategy, including boundary conditions such as crystal diameter range, growth rate range, ambient temperature range, and cooling water flow rate range. The strategy file is stored in the form of structured database records, and a readable configuration file is generated for direct use by the temperature control system.

[0071] The anomaly handling mechanism in the strategy selection process incorporates a multi-level fallback scheme. When the number of remaining strategies is insufficient, the threshold standard is automatically relaxed; when none of the strategies meet the requirements, a strategy refactoring process is triggered. The strategy comparison and selection algorithm has undergone robustness testing and can handle situations where measurement errors exist in the feature data. After the optimal strategy is determined, the system automatically generates a strategy verification plan, which includes monitoring points and evaluation indicators for implementing the strategy in the next growth cycle. The strategy version management system records the time, user, and decision basis of each selection operation, supporting long-term tracking and optimization of strategy effectiveness. The entire selection process achieves automated closed-loop management, with the average processing time from performance evaluation report input to optimal strategy output controlled within five minutes, meeting the real-time control requirements of the crystal growth process. See Table 1, which shows the feature thresholds used in the strategy selection process.

[0072] Table 1: Standard Thresholds for Temperature Regulation Strategies

[0073] Feature type Threshold conditions critical value unit Applicable growth stage Stability characteristics Upper limit of standard deviation 2.5 °C Equal diameter growth stage Stability characteristics Upper limit of standard deviation 4.0 °C Shoulder relaxation stage Response speed characteristics Maximum response time 180 Second Equal diameter growth stage Response speed characteristics Maximum response time 300 Second Shoulder relaxation stage Temperature uniformity Minimum improvement 15 % All stages Energy consumption Maximum increment 8 % All stages

[0074] The screening algorithm displays processing progress and intermediate results in real time during execution. The user interface provides a strategy comparison visualization tool, supporting parallel coordinate graphs to display the feature distribution of multiple strategies. The decision support system provides strategy recommendation descriptions, detailing the advantages and potential risks of each candidate strategy. Once the optimal strategy is determined, an implementation plan schedule is automatically generated, clearly defining the strategy activation time, parameter transition curves, and expected effect monitoring points. Pre-implementation verification testing is performed in a simulation environment to ensure no significant risks before deployment to the actual growth system. The entire screening system establishes a robust logging mechanism, recording all operation steps, judgment logic, and decision results in the audit log, meeting process traceability and quality control requirements. The strategy effect feedback mechanism compares the actual application effect with the expected target; deviation data is used to optimize the parameter settings of the screening algorithm, enabling the system to self-improve.

[0075] See Figure 4The upper part of the chart selects the optimal and worst strategies, showcasing their performance distribution across five core dimensions: stability, response speed, temperature uniformity, energy consumption, and defect improvement. The upper subplot clearly shows that Strategy 1 scores highly across all dimensions, demonstrating balanced and excellent performance; while Strategy 7 scores significantly lower across all dimensions, clearly illustrating the multi-dimensional performance differences between the strategies. The lower part quantifies the overall performance of all strategies by weighted summation of multi-dimensional indicators: the scores of each strategy are sorted in descending order, clearly showing the overall performance hierarchy of different temperature regulation strategies. By intuitively comparing the performance differences between strategies across multiple dimensions and quantifying the merits of strategies using bar charts, this provides an intuitive and quantitative basis for eliminating substandard strategies and selecting the overall optimal strategy. It supports the technical process from generating a set of temperature regulation strategies to selecting the optimal strategy, reflecting the evaluation logic for multi-objective optimization of temperature regulation strategies.

[0076] Example 5: Real-time monitoring of thermal field temperature distribution data is achieved through a distributed sensor network. The network contains multiple temperature measurement units, each integrating a thermocouple and signal conditioning circuitry. The data acquisition frequency is set to ten times per second. The temperature anomaly area identification algorithm is based on the sliding window statistical principle, with a window width set to sixty seconds. The moving average and moving standard deviation of the temperature data within the window are calculated. The anomaly determination condition adopts a dynamic threshold method, and the threshold calculation formula is:

[0077]

[0078] in: Represents the dynamic anomaly threshold. This represents the moving average of the temperature data within the window. The coefficient represents the moving standard deviation of the temperature data within the window. These are adjustable parameters set based on the thermal sensitivity of the crystalline material. When the real-time temperature value of any temperature measurement point exceeds the dynamic anomaly threshold for three consecutive sampling cycles, the area where that point is located is marked as a temperature anomaly region. The boundary of the anomaly region is determined by a region growing algorithm, expanding outwards from the anomaly point until a temperature measurement point with a normal temperature is encountered.

[0079] Correction parameters for the temperature anomaly region are generated based on the optimal temperature regulation strategy, employing a fuzzy inference mechanism. Input variables include the temperature deviation value, deviation change rate, and anomaly duration in the anomaly region. Output variables are the heater power correction or cooling system flow regulation. The fuzzy inference rule base contains multiple IF-THEN rules, in the form of "if the temperature deviation is negative and the deviation change rate is positive, then the power correction is positive." The fuzzification process converts precise input values ​​into membership degrees of fuzzy sets. The inference engine performs fuzzy logic operations based on the rule base. The defuzzification process converts the fuzzy output into precise correction parameter values. The correction parameters include two dimensions: amplitude and direction. The amplitude value is normalized to convert it into the actual control signal range. The correction parameters are injected into the thermal field temperature control system via a digital communication interface using the Modbus TCP standard. The correction parameters are encapsulated into a specific data structure containing the target device address, parameter type, numerical value, and timestamp fields. The injection operation uses a transaction mechanism to ensure data integrity; after sending the correction parameters, it waits for a device confirmation frame, and automatically retransmits if no confirmation is received within a timeout period. After receiving the correction parameters, the temperature control system gradually adjusts the actuator state according to the preset smooth transition curve to avoid sudden temperature changes. Heater power adjustment is achieved by changing the thyristor firing angle, and cooling system flow regulation is accomplished by adjusting the opening of the electric valve. A multi-index evaluation system is established to verify whether the thermal field temperature distribution data after the correction parameters are executed meets expectations. The expected standards include four dimensions: temperature recovery time, overshoot, steady-state error, and fluctuation amplitude. Temperature recovery time refers to the time from the start of correction to the temperature returning to the normal range; overshoot measures the maximum deviation of the temperature from the target value during the correction process; steady-state error represents the difference between the stable temperature and the target value; and fluctuation amplitude assesses the temperature fluctuation after stabilization. Meeting expectations requires simultaneously satisfying the following conditions: temperature recovery time is less than the set threshold, overshoot does not exceed the allowable range, steady-state error is within the accuracy requirements, and fluctuation amplitude is lower than the standard value.

[0080] When the verification results do not meet expectations, the temperature regulation strategy selection process is retried, employing an incremental learning mechanism. All data from the current anomaly handling process is added to the historical database, including anomaly characteristics, corrective parameters taken, and actual effect records. The strategy selection algorithm recalculates the performance indicators of each strategy based on the original strategy set and the new data, updating the strategy ranking. The optimal strategy selected after re-selection may be the same as the original strategy, or a more suitable strategy may be chosen based on newly discovered temperature characteristics. A maximum retry limit is set for the re-selection process to prevent the system from entering an infinite loop. When the verification results meet expectations, the current temperature control parameters are recorded as the baseline configuration. The recording operation includes two stages: parameter archiving and knowledge updating. Parameter archiving saves successful control parameter combinations to the baseline parameter library, marking applicable conditions and effect evaluations. The knowledge updating stage analyzes the experience of this successful adjustment, extracts the association rules between anomaly characteristics and effective corrective parameters, and updates the fuzzy inference rule library. The baseline configuration record is in a structured data format, containing information from multiple dimensions such as environmental parameters, equipment status parameters, control parameters, and effect indicators. A version management mechanism is established in the baseline configuration library; each successful adjustment generates a new configuration version, while historical versions are retained for use under different operating conditions. The system periodically evaluates the effectiveness of baseline configurations, eliminates older versions that have deteriorated in performance, and optimizes the quality of the configuration library.

[0081] The dynamic adjustment process establishes a complete closed-loop control logic, forming an automated pipeline from anomaly detection to parameter correction and effect verification. The real-time monitoring interface displays the current temperature distribution map, highlighted abnormal areas, parameter correction progress, and verification results. All operation records generate detailed logs, including timestamps, operation types, parameter values, and execution results, supporting post-event analysis and process optimization. The system employs a multi-level alarm mechanism, triggering different alarm levels based on the severity of the anomaly; minor anomalies are handled automatically, while major anomalies require manual confirmation. The temperature control parameter dynamic adjustment system establishes standard data interfaces with the upstream process management system and the downstream equipment control system, achieving full-process automated integration. The system periodically performs self-diagnostic checks to verify sensor accuracy, communication link status, and control logic integrity, ensuring long-term operational reliability.

[0082] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0083] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for controlling the temperature of a crystal growth thermal field, characterized in that, Includes the following steps: Acquire thermal field temperature distribution data during crystal growth, and combine crystal growth rate and material phase transformation parameters to determine the key influencing factors of thermal field temperature control; Based on the thermal field temperature distribution data and key influencing factors, a dynamic adjustment model for thermal field temperature is established to analyze the mapping relationship between thermal field temperature and crystal growth quality. Historical thermal field temperature regulation records and crystal defect data are collected and combined with a dynamic thermal field temperature regulation model to generate a set of temperature regulation strategies. Based on a set of temperature regulation strategies, the temperature change trend of the thermal field under different regulation strategies is simulated, and the stability and response speed characteristics in the temperature change trend are extracted. Based on stability and response speed characteristics, the optimal temperature regulation strategy is selected, and the temperature control parameters of the crystal growth thermal field are dynamically adjusted in combination with real-time thermal field temperature monitoring data. The key influencing factors for determining the temperature control of the thermal field include: The thermal field temperature distribution data is divided into regions, and the temperature gradient data of each region is extracted. Analyze the correlation between temperature gradient data and crystal growth rate to identify temperature fluctuation-sensitive regions; By combining the material phase transformation parameters, the deviation of heat conduction efficiency in the temperature fluctuation sensitive area is calculated; Based on the deviation value of heat conduction efficiency, the key influencing factors and their priorities for thermal field temperature control are determined.

2. The method for controlling the temperature of a crystal growth thermal field according to claim 1, characterized in that, The establishment of the dynamic temperature regulation model for the thermal field includes: Extract periodic variation patterns from thermal field temperature distribution data to construct temporal features of temperature changes; By combining crystal growth quality inspection data, a correlation rule between the temporal characteristics of temperature changes and crystal defect types is established. Based on the association rules, generate the constraints for the dynamic temperature adjustment model of the thermal field; Based on constraints, the parameter configuration of the dynamic temperature regulation model of the thermal field is optimized.

3. The method for controlling the temperature of a crystal growth thermal field according to claim 1, characterized in that, The set of temperature regulation strategies includes: Cluster analysis was performed on historical thermal field temperature regulation records to extract typical temperature regulation patterns; The effectiveness of typical temperature regulation modes is evaluated by combining crystal defect data; Based on the effectiveness evaluation results, a set of temperature regulation strategies is generated, and the applicable conditions for each strategy are marked.

4. The method for controlling the temperature of a crystal growth thermal field according to claim 1, characterized in that, The simulated temperature change trends of the thermal field under different adjustment strategies include: A thermal field temperature simulation environment is constructed based on a set of temperature regulation strategies. In the thermal field temperature simulation environment, real-time thermal field temperature monitoring data is injected as the initial condition; Run the simulation environment and record the temperature change curves under each adjustment strategy; Stability and response speed characteristics are extracted from the temperature change curve.

5. The method for controlling the temperature of a crystal growth thermal field according to claim 4, characterized in that, The stability and response speed characteristics extracted from the temperature change trend include: Calculate the standard deviation of the temperature change curve as a quantitative indicator of stability characteristics; The time required for the temperature change curve to reach a set threshold is used as a quantitative indicator of the response speed characteristic. By combining quantitative indicators, a performance evaluation report on the temperature regulation strategy is generated.

6. The method for controlling the temperature of a crystal growth thermal field according to claim 1, characterized in that, The optimal temperature regulation strategy for selection includes: Based on the performance evaluation report, exclude adjustment strategies that fail to meet the standards for stability or response speed characteristics; Among the remaining regulation strategies, the regulation strategy with the best overall performance is selected as the optimal temperature regulation strategy; Record the parameter configuration and execution conditions of the optimal temperature regulation strategy.

7. The method for controlling the temperature of a crystal growth thermal field according to claim 1, characterized in that, The temperature control parameters for dynamically adjusting the crystal growth thermal field include: Real-time monitoring of thermal field temperature distribution data to identify areas of abnormal temperature; Based on the optimal temperature regulation strategy, generate correction parameters for the temperature anomaly region; The correction parameters are injected into the thermal field temperature control system to complete the dynamic adjustment of the temperature control parameters.

8. The method for controlling the temperature of a crystal growth thermal field according to claim 7, characterized in that, The dynamic adjustment of the temperature control parameters includes: Verify whether the thermal field temperature distribution data after the parameter correction is executed meets expectations; If it does not meet expectations, the temperature regulation strategy screening process will be retried. If it meets expectations, record the current temperature control parameters as the baseline configuration.

9. A temperature control system based on a crystal growth thermal field, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the crystal growth thermal field temperature control method according to any one of claims 1 to 8.

Images