Dynamic control method, system and equipment for single crystal silicon production process

Abstract

The present invention relates to the field of dynamic control technology, and provides a dynamic control method, system and equipment for a single crystal silicon production process. The method comprises: installing a multi-mode sensor unit in a pulling furnace, acquiring multi-mode monitoring data, analyzing and extracting finishing features, inputting a risk prediction model, optimizing a pulling speed adjustment strategy according to the prediction results, and correcting the deviation between the prediction and the actual pulling speed, thereby solving the technical problems of potential fracture risks in the finishing stage and lack of an effective mechanism to timely identify and correct potential production risks, comprehensively and real-timely monitoring the growth process of single crystal silicon, ensuring that environmental changes and crystal growth states in production are captured in time, and timely discovering anomalies through analysis of characteristic data of the crystal finishing stage to avoid production accidents caused by fractures or unstable states, ensuring the accuracy and real-time nature of the adjustment process, and achieving the technical effects of high efficiency, stability and safety in the single crystal silicon production process.

Description

Technical Field

[0001] The present invention relates to the technical field related to dynamic control, and in particular to a dynamic control method, system and equipment for a single crystal silicon production process. Background Art

[0002] With the advancement of science and technology, the demand for the semiconductor industry continues to grow. As the basis of semiconductor materials, single crystal silicon is widely used in electronics, photovoltaics, integrated circuits and other fields. The production process of single crystal silicon, especially the control during the crystal growth process, directly determines the quality and yield of the crystal. With the expansion of production scale, the manufacturing process of single crystal silicon has become more and more complicated. In particular, the influence of multiple environmental variables such as temperature, humidity, strain, etc. during the pulling process on the growth state and quality of the crystal has attracted more and more attention.

[0003] At present, the production process of single crystal silicon mainly relies on empirical function control methods, such as process control through simple sensor detection. However, conventional control methods cannot capture the various dynamic changes in the production process in real time, especially in the pulling process. The effects of factors such as temperature, humidity and strain on crystal growth require real-time feedback and precise control. In addition, due to the many variables in the crystal growth process, it is often difficult to achieve comprehensive monitoring and optimal control of multiple factors, which can easily lead to potential breakage or substandard quality in the production process.

[0004] In summary, the existing technology has potential risks of breakage in the finishing stage and lacks an effective mechanism to promptly identify and correct technical problems of potential production risks. Summary of the invention

[0005] The present application aims to solve the technical problems in the prior art of potential fracture hazards in the final stage and the lack of effective mechanisms to timely identify and correct potential production risks by providing a dynamic control method, system and equipment for the single crystal silicon production process.

[0006] In view of the above problems, the embodiments of the present application provide a dynamic control method, system and equipment for a single crystal silicon production process.

[0007] The first aspect disclosed in the present application provides a dynamic control method for a single crystal silicon production process, wherein the method comprises: installing a multimodal sensing unit in a pulling furnace, the multimodal sensing unit comprising a temperature sensor, a humidity sensor, a strain sensor, and a visual monitoring element; using the multimodal sensing unit to obtain multimodal monitoring data, the multimodal monitoring data comprising crystal growth status, interface stability, and furnace environment parameters; performing feature analysis on the multimodal monitoring data to extract finishing feature data corresponding to the crystal finishing stage; synchronizing the key feature data into a constructed risk prediction model to identify crystal growth status and potential fracture risks; based on the risk prediction results, starting a preset optimization space to optimize the pulling speed adjustment strategy, outputting the optimal pulling speed control strategy, and comparing the deviation between the predicted pulling speed and the actual pulling speed for correction control.

[0008] The second aspect disclosed in the present application provides a dynamic control system for a single crystal silicon production process, wherein the system includes: an installation module for installing a multimodal sensing unit in a pulling furnace, wherein the multimodal sensing unit includes a temperature sensor, a strain sensor and a visual monitoring element; a data acquisition module for using the multimodal sensing unit to acquire multimodal monitoring data, wherein the multimodal monitoring data includes crystal growth status, interface stability and furnace environment parameters; a data extraction module for performing feature analysis on the multimodal monitoring data and extracting finishing feature data corresponding to the crystal finishing stage; a risk identification module for synchronizing the key feature data into a constructed risk prediction model to identify crystal growth status and potential fracture risks; and a correction control module for starting a preset optimization space to optimize a pulling speed adjustment strategy based on risk prediction results, outputting an optimal pulling speed control strategy, and comparing the deviation between the predicted pulling speed and the actual pulling speed for correction control.

[0009] The third aspect disclosed in the present application provides a dynamic control device for a single crystal silicon production process, wherein the device includes: a memory for storing executable instructions; and a processor for executing the executable instructions stored in the memory to implement the dynamic control method for a single crystal silicon production process described in any one of the first aspects.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] By installing a multi-modal sensing unit in the pulling furnace, multi-mode monitoring data is obtained, the finishing features are analyzed and extracted, and input into the risk prediction model. The pulling speed adjustment strategy is optimized according to the prediction results, and the deviation between the predicted and actual pulling speeds is corrected. The growth process of single crystal silicon is monitored comprehensively and in real time, ensuring that environmental changes and crystal growth states in production are captured in time. By analyzing the characteristic data of the crystal finishing stage, anomalies are discovered in time to avoid production accidents caused by breakage or unstable states, ensuring the accuracy and real-time nature of the adjustment process, and achieving the technical effects of high efficiency, stability and safety in the single crystal silicon production process.

[0012] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A schematic flow chart of a dynamic control method for a single crystal silicon production process is provided for an embodiment of the present application;

[0014] Figure 2 A schematic structural diagram of a dynamic control system for a single crystal silicon production process is provided for an embodiment of the present application;

[0015] Figure 3 A schematic structural diagram of a dynamic control device for a single crystal silicon production process is provided for this application.

[0016] Explanation of the reference numerals: installation module 11 , data acquisition module 12 , data extraction module 13 , risk identification module 14 , correction control module 15 , processor 21 , memory 22 , input device 23 , output device 24 . DETAILED DESCRIPTION

[0017] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0018] Embodiment 1

[0019] like Figure 1 As shown, the embodiment of the present application provides a dynamic control method for a single crystal silicon production process, wherein the method comprises:

[0020] S100: installing a multi-modal sensing unit in the drawing furnace, wherein the multi-modal sensing unit includes a temperature sensor, a humidity sensor, a strain sensor, and a visual monitoring element; S200: using the multi-modal sensing unit to obtain multi-mode monitoring data, wherein the multi-mode monitoring data includes crystal growth status, interface stability, and furnace environment parameters; S300: performing feature analysis on the multi-mode monitoring data to extract finishing feature data corresponding to the crystal finishing stage.

[0021] Specifically, the multimodal sensing unit is capable of simultaneously monitoring multiple different physical quantities including temperature sensors, humidity sensors, strain sensors and visual monitoring elements. These sensors work together to provide comprehensive data on the growth status of single crystal silicon; the temperature sensor is used to measure the temperature in the pulling furnace, and the temperature has a direct impact on the crystal growth rate and quality; the humidity sensor is used to monitor the humidity in the furnace, and changes in humidity may affect the growth environment and quality of the crystal; the strain sensor is used to measure the stress and strain during the crystal growth process; the visual monitoring element is used to monitor the physical state of crystal growth, such as shape changes and defects, through images or videos; the multimodal monitoring data refers to the data collected by the above sensors, including the crystal growth state, interface stability and furnace environmental parameters; the crystal growth state refers to the physical and chemical state of the crystal during the growth process, such as the growth rate, crystal integrity, etc.; the interface stability refers to the stability of the contact surface between the crystal and other materials in the furnace; the furnace environmental parameters include environmental factors such as temperature, humidity, and pressure that affect crystal growth; feature analysis is used to extract useful information and patterns from the original data; the ending feature data specifically refers to the key data near the end of the crystal growth stage.

[0022] Multimodal sensing units are installed in the pulling furnace. These sensors are responsible for real-time monitoring of the crystal growth status, interface stability and furnace environmental parameters; temperature sensors provide accurate readings of the furnace temperature, and humidity sensors monitor the humidity in the furnace, because changes in humidity may affect the growth environment and quality of the crystal; strain sensors measure stress and strain during crystal growth, which is crucial for predicting and preventing crystal breakage; visual monitoring elements monitor the physical state of crystal growth, such as shape changes and defects, through images or videos.

[0023] By using these multimodal sensing units, multimodal monitoring data can be obtained, including crystal growth status, interface stability and furnace environmental parameters; by performing feature analysis on these multidimensional data, key feature data related to the crystal finishing stage can be extracted, and key parameters in the crystal growth process can be identified; in this way, the growth process of single crystal silicon can be monitored comprehensively and in real time, ensuring that environmental changes and crystal growth status in production are captured in a timely manner, thereby achieving efficient, stable and safe control of the crystal growth process.

[0024] S400: Synchronize the key feature data to the constructed risk prediction model to identify the crystal growth status and potential fracture risks; S500: Based on the risk prediction results, start the preset optimization space to optimize the pulling speed adjustment strategy, output the optimal pulling speed control strategy, and compare the deviation between the predicted pulling speed and the actual pulling speed for correction control.

[0025] Specifically, key feature data refers to data closely related to the crystal growth state and potential fracture risk extracted from the data collected by the multimodal sensing unit through feature analysis; the risk prediction model is a model built based on machine learning or other data analysis technologies, which is used to predict the crystal growth state and potential fracture risk based on the input feature data; the crystal growth state refers to the physical and chemical state of the crystal during the growth process, including the growth rate, crystal integrity, etc.; potential fracture risk identification refers to identifying the risk factors that may cause crystal fracture by analyzing the data during the crystal growth process; the preset optimization space refers to the parameter range pre-set in the risk prediction model. The optimization of the pulling speed adjustment strategy refers to finding the best pulling speed adjustment strategy within the preset optimization space to achieve optimal control of crystal growth; the optimal pulling speed regulation strategy refers to the pulling speed adjustment strategy calculated based on the risk prediction results and the preset optimization space, which can minimize the risk of fracture and improve the crystal quality; the deviation between the predicted pulling speed and the actual pulling speed refers to the difference between the pulling speed predicted by the optimal pulling speed regulation strategy and the actual pulling speed; the correction control refers to adjusting the pulling speed according to the deviation between the predicted pulling speed and the actual pulling speed to ensure the accuracy and stability of the crystal growth process.

[0026] During the production process of single crystal silicon, the key feature data obtained from the multimodal sensing unit are synchronized to the constructed risk prediction model. The risk prediction model can analyze these data and identify possible problems in the crystal growth process, such as abnormal growth rate or potential fracture risks; based on the results of the risk prediction model, the preset optimization space is started to optimize the pulling speed adjustment strategy; within the parameter range preset by the risk prediction model, the best pulling speed adjustment strategy is found to achieve optimal control of the crystal growth process; the optimal pulling speed regulation strategy is output; the deviation between the predicted pulling speed and the actual pulling speed is compared, and correction control is performed to ensure the accuracy and real-time nature of the crystal growth process, and the pulling speed is monitored in real time through a closed-loop feedback control system, and adjusted according to the difference between the actual pulling speed and the predicted pulling speed to ensure the stability and safety of the crystal growth process, improve the crystal quality, and also improve the production efficiency, so as to achieve the high efficiency, stability and safety of the single crystal silicon production process.

[0027] Furthermore, a multi-modal sensing unit is installed in the drawing furnace, and the method of the present application includes:

[0028] Based on the multimodal sensing unit, a data transmission network is determined, and the data transmission network is used to transmit the multimodal monitoring data; an encrypted communication protocol is used to perform data security protection on the data transmission network.

[0029] Specifically, the data transmission network refers to the communication network used to transmit the multimodal monitoring data collected by the multimodal sensing unit from the sensor to the data processing center or the risk prediction model; the multimodal monitoring data includes the data collected from the temperature sensor, the humidity sensor, the strain sensor and the visual monitoring element, which reflects the crystal growth state, the interface stability and the furnace environment parameters; the encrypted communication protocol is used to protect the data from unauthorized access or tampering during the data transmission process, and ensure the confidentiality, integrity and availability of the data through encryption technology; data security protection refers to taking measures to protect the security of the data during the transmission process, prevent data leakage, tampering or loss, and ensure the security and reliability of the data.

[0030] In the process of monocrystalline silicon production, the multimodal monitoring data collected by the multimodal sensing unit needs to be transmitted in real time to the data processing center or the risk prediction model for further analysis, and a stable and reliable data transmission network needs to be established. The network is responsible for quickly and accurately transmitting the data collected by the sensors. The establishment of the data transmission network includes selecting appropriate communication media (such as optical fiber, wireless network, etc.), designing the network topology, configuring network equipment (such as routers, switches, etc.), and ensuring the reliability and stability of the network, so as to ensure that the data can be transmitted from the sensor to the data processing center in a timely and secure manner.

[0031] In order to protect the security of these data during transmission, encrypted communication protocols are used to perform data security protection on the data transmission network, which usually involves the use of secure communication protocols (such as TLS / SSL, IPSec, etc.) to encrypt data transmission to ensure that the data cannot be interpreted even if it is intercepted during transmission. In addition, other security measures can be taken, such as access control, intrusion detection systems, etc., to further improve the security of data transmission; by adopting encrypted communication protocols and data security protection measures, sensitive data can be effectively prevented from being accessed or tampered with without authorization, the company's technical secrets and production data can be protected, and the safety and reliability of the monocrystalline silicon production process can be ensured.

[0032] Furthermore, the key feature data is synchronized into the constructed risk prediction model, and the method of the present application includes:

[0033] Connecting to a distributed database to obtain a pulling speed control record; performing data mining based on the pulling speed control record to analyze a mapping relationship between a historical pulling speed change pattern and crystal quality; and configuring parameters of the risk prediction model through the mapping relationship between the historical pulling speed change pattern and crystal quality.

[0034] Specifically, a distributed database allows data to be stored and processed across multiple physical locations while appearing externally as a single, unified data storage system; pulling speed control records refer to the historical records of adjustments to the pulling speed of the pulling furnace during the production of single crystal silicon. These records include pulling speed values ​​at different time points and the corresponding crystal growth conditions; data mining is a process of analyzing large amounts of data to discover patterns, trends and correlations, with the aim of extracting useful information from the data; historical pulling speed change patterns refer to the changing trends and patterns of pulling speeds during past production processes, which are related to the growth quality and efficiency of crystals; the mapping relationship of crystal quality refers to the correlation between pulling speed changes and crystal quality, and analysis can determine which pulling speed changes have a positive or negative impact on crystal quality; a risk prediction model is a model used to predict possible risks (such as fracture risks) during crystal growth, which can be predicted based on input parameters; parameter configuration refers to adjusting the input parameters of the model based on specific data and analysis results to optimize the performance of the model and prediction accuracy.

[0035] During the single crystal silicon production process, the distributed database is connected to obtain the pulling speed control records, which contain detailed information on the adjustment of the pulling speed under different production conditions and the impact of these adjustments on the crystal quality. Based on these pulling speed control records, data mining can be performed to analyze the mapping relationship between the historical pulling speed change pattern and the crystal quality. The data mining process involves statistical analysis, machine learning algorithms and other technologies to identify the specific impact of pulling speed changes on crystal quality.

[0036] By analyzing the mapping relationship obtained from historical data, the risk prediction model is parameterized, which means adjusting the model's weights, biases or other parameters based on the patterns and trends found in the historical data to improve the model's accuracy in predicting crystal growth status and potential fracture risks. This helps the model better reflect the complexity of the actual production process, thereby improving the control accuracy of the production process and the quality of crystals. Historical data is used to optimize the risk prediction model so that it can more accurately predict and adapt to changes in the production process, thereby improving the efficiency and safety of the monocrystalline silicon production process. In this way, production accidents can be reduced, the yield rate can be increased, and more efficient and stable monocrystalline silicon production can be achieved.

[0037] Furthermore, the risk prediction model is configured with parameters, and the method of the present application includes:

[0038] Identify a first key factor affecting crystal quality and a second key factor affecting growth efficiency; and adjust the weight bias of the risk prediction model based on the first key factor and the second key factor.

[0039] Specifically, the first key factor and the second key factor refer to the two factors that have the greatest impact on crystal quality and growth efficiency in the production process of single crystal silicon. Among them, the first key factor refers to the parameters that directly affect the crystal quality, such as the accuracy of temperature control, and the second key factor refers to the parameters that affect the growth efficiency, such as pulling speed; the risk prediction model is a model used to predict the risks that may occur during the crystal growth process (such as fracture risk), which can be predicted based on the input parameters; in machine learning and statistical models, weights and biases are key parameters that affect the model output. The weights determine the degree of influence of the input features on the model output, and the bias is the baseline value of the model output. The weight bias is adjusted to optimize the prediction performance of the model.

[0040] In the single crystal silicon production process, identifying the first key factor affecting crystal quality and the second key factor affecting growth efficiency usually involves in-depth analysis of historical data to determine which parameters have the greatest impact on crystal quality and growth efficiency. For example, the first key factor is the accuracy of temperature control, which directly affects the lattice structure and defect rate of the crystal; the second key factor is the pulling speed, which determines the speed and efficiency of crystal growth.

[0041] By identifying these key factors, the risk prediction model can be more finely tuned. Specifically, the model can be optimized by adjusting the weights and biases associated with these key factors in the model; for example, if the accuracy of temperature control has the greatest impact on crystal quality, then the weight of temperature-related features in the model may be increased to reflect its importance. Similarly, if pulling speed has the greatest impact on growth efficiency, then the weight of pulling speed-related features will be adjusted accordingly.

[0042] Adjusting the weight bias enables the risk prediction model to more accurately predict the risks that may occur during crystal growth and provide more effective control strategies. In this way, the crystal quality is improved, the growth efficiency is optimized, and ultimately the economic benefits of monocrystalline silicon production are improved. The model adjustment based on key factors makes the risk prediction model more in line with actual production conditions, improving the practicality of the model and the accuracy of prediction.

[0043] Further, the first key factor affecting crystal quality and the second key factor affecting growth efficiency are identified, and the method of the present application includes:

[0044] Using signal processing technology, initialization information is separated; based on the initialization information, data dimension reduction processing is performed to obtain dimension reduction processing information; based on the dimension reduction processing information, various factors affecting crystal quality are ranked in importance to determine the first key factor affecting crystal quality.

[0045] Specifically, signal processing technology refers to a series of methods used to analyze and process signals, which can extract key information from the signal or improve the quality of the signal for better analysis; initialization information refers to the basic information or features extracted from the original data before any further processing, which provides a starting point for subsequent analysis; data dimensionality reduction refers to reducing the number of variables in the data set while retaining as much important information as possible from the original data. Dimensionality reduction helps to simplify the model, improve computational efficiency, and reduce the risk of overfitting; dimensionality reduction information refers to the data after dimensionality reduction, which contains the most important features of the original data set but is expressed in a lower dimension; importance ranking is a process of evaluating the degree of influence of different factors on the results, and ranking can determine which factors are the most important; the first key factor refers to the most influential factor among the multiple factors that affect the quality of the crystal, determined by importance ranking.

[0046] During the production of single crystal silicon, signal processing technology is used to analyze the raw data collected from the multimodal sensing unit in order to separate the initialization information, including the basic characteristics of the crystal growth state, such as temperature change, strain degree, etc. Based on this initialization information, data dimensionality reduction is performed to simplify the data, remove redundant information, and extract the most influential features. For example, principal component analysis (PCA) or other machine learning algorithms are used to identify and retain the variables that have the greatest impact on crystal quality.

[0047] After dimensionality reduction, dimensionality reduction information will be obtained, which contains the most important features of the original data set in a lower dimension; based on this dimensionality reduction information, the various factors affecting crystal quality are ranked by importance, involving statistical analysis or machine learning models to determine which factors have the greatest impact on crystal quality; the first key factor affecting crystal quality is determined by importance ranking, which can be any one of temperature control, pulling speed, strain, etc., depending on the ranking results. Determine the first key factor and focus resources and attention on the most critical control parameters, thereby achieving more precise process control and higher production efficiency.

[0048] Furthermore, the deviation between the predicted casting speed and the actual casting speed is compared to perform correction control, and the method of the present application further includes:

[0049] A safety threshold for pulling speed adjustment is set; during each pulling speed adjustment process, a judgment is made based on the safety threshold for pulling speed adjustment. If the speed adjustment is exceeded, the current pulling speed adjustment operation is paused and a safety reminder is issued.

[0050] Specifically, the pulling speed adjustment safety threshold refers to a safety range or limit value set for the pulling speed adjustment during the single crystal silicon production process, which is used to ensure that the pulling speed adjustment does not exceed the limit that may affect the crystal quality and production safety; pulling speed adjustment refers to the adjustment of the pulling speed of the pulling furnace according to the actual crystal growth conditions during the single crystal silicon growth process, so as to optimize the crystal growth conditions; safety reminder refers to when the operation exceeds the safety threshold, it is triggered to remind the operator to take immediate action to avoid possible risks or damages.

[0051] In the process of single crystal silicon production, setting the safety threshold for pulling speed adjustment is risk management. The setting of the safety threshold is based on the analysis of historical data, production experience and theoretical knowledge to determine the maximum and minimum safety values ​​of the pulling speed, and define the range in which the pulling speed can vary without compromising the crystal quality or causing damage to the equipment.

[0052] During each pulling speed adjustment process, a judgment is made based on the preset pulling speed adjustment safety threshold; if the pulling speed adjustment value exceeds this safety range, the current pulling speed adjustment operation will be automatically paused to prevent operations that may adversely affect the crystal growth process, and at the same time, a safety reminder will be issued; the purpose of the safety reminder is to ensure that the operator immediately notices the abnormal situation and takes necessary measures, such as checking sensor readings, re-evaluating growth conditions or adjusting growth strategies, thereby enhancing the safety and reliability of the production process and reducing production accidents caused by operational errors or out-of-control conditions. In this way, the pulling speed adjustment safety threshold and the safety reminder mechanism work together to ensure the stability and safety of the single crystal silicon growth process, while also improving production efficiency and crystal quality.

[0053] Furthermore, the present application method also includes:

[0054] In the case where the pulling speed adjustment safety threshold is exceeded, a feedback control strategy is triggered; based on the feedback control strategy, an adverse impact supplementary factor is determined, and the risk prediction model is updated.

[0055] Specifically, the feedback control strategy refers to the use of the system's output feedback to adjust the system's input in order to keep the system's performance at a desired level or range; adverse impact supplementary factors refer to those factors that may have a negative impact on the crystal growth process but are not fully considered in normal operations. These factors may include environmental changes, equipment aging, raw material fluctuations, etc.; risk prediction models are used to predict the risks that may occur during the crystal growth process (such as fracture risks), which can be predicted based on the input parameters and continuously updated and optimized based on new data.

[0056] In the process of single crystal silicon production, once the pulling speed adjustment exceeds the preset safety threshold, the feedback control strategy will be automatically triggered, and then through real-time monitoring and adjustment, it will ensure that the crystal growth process remains within a safe and effective range. The feedback control strategy includes reducing the pulling speed, adjusting the environmental parameters in the furnace, or taking other necessary measures to correct the deviation; based on this feedback control strategy, further analyze and determine the adverse impact supplementary factors, which are the reasons that cause the pulling speed adjustment to exceed the safety threshold, and may also be other potential risks that affect crystal growth; through the identification and analysis of these supplementary factors, a more comprehensive understanding of the various influencing factors in the crystal growth process can be achieved.

[0057] After determining the supplementary factors of adverse impact, the risk prediction model will be updated, including adding new parameters, adjusting the weights of existing parameters or optimizing the model's algorithm. In this way, the risk prediction model can more accurately reflect the complexity of the actual production process and improve the accuracy and reliability of the prediction; the risk prediction model is updated to better understand the various risks in the crystal growth process and take more effective control measures to improve the crystal quality and achieve more efficient and stable monocrystalline silicon production. Through this continuous learning and optimization, the risk prediction model can adapt to changes in the production process and improve the adaptability and flexibility of the production system.

[0058] Furthermore, and updating the risk prediction model, the method of the present application includes:

[0059] Using a machine learning framework, an online learning mechanism is deployed; based on the online learning mechanism, performance optimization iterations are performed during the risk prediction model update process.

[0060] Specifically, the machine learning framework is used to design, train, and deploy machine learning models, providing the algorithms, data processing tools, and automated processes required to build and optimize models; the online learning mechanism is a machine learning strategy in which the model is updated instantly when new data is received, rather than periodically in batch mode. Online learning allows the model to quickly adapt to new data and changes; performance optimization iteration refers to an ongoing process in which the performance of the model is constantly evaluated and improved based on new data and feedback, involving the adjustment of model parameters, optimization of algorithms, and improvement of prediction accuracy.

[0061] During the production of monocrystalline silicon, the machine learning framework provides the necessary tools and algorithms to enable the risk prediction model to converge quickly and update its parameters and structure in real time; the machine learning framework may include deep learning libraries, statistical model tools or other specialized machine learning technologies; the online learning mechanism allows the risk prediction model to receive new multimodal sensor data in real time during the production process and adjust the model parameters instantly based on this data, which means that the model can quickly adapt to changes in production conditions, such as temperature fluctuations, changes in pulling speed or other environmental factors, thereby improving the accuracy of the prediction and the speed of response.

[0062] According to the online learning mechanism, the risk prediction model performs performance optimization iterations during the update process, which means that the model will continuously evaluate the difference between its predicted results and actual results, and adjust the model parameters based on these differences to reduce the prediction error. This iterative process can be performed manually or through automated algorithms to ensure that the model is always in the best state; performance optimization iterations enable the risk prediction model to be continuously improved and adapt to dynamic changes in the production process; in this way, the risks in the crystal growth process can be more accurately predicted, and more effective control strategies can be provided, thereby improving crystal quality and production efficiency. This continuous learning and optimization method makes the production process more intelligent, flexible and efficient.

[0063] In summary, the dynamic control method, system and device for the single crystal silicon production process provided by the embodiments of the present application have the following technical effects:

[0064] The invention adopts the method of installing a multi-modal sensing unit in the pulling furnace; adopting the multi-modal sensing unit to obtain multi-modal monitoring data; performing feature analysis on the multi-modal monitoring data to extract the finishing feature data corresponding to the finishing stage of the crystal; synchronizing the key feature data into the constructed risk prediction model to identify the growth status of the crystal and the potential fracture risk; based on the risk prediction result, starting the preset optimization space to optimize the pulling speed adjustment strategy, outputting the optimal pulling speed control strategy, and comparing the deviation between the predicted pulling speed and the actual pulling speed for correction control. The present application provides a dynamic control method, system and equipment for the single crystal silicon production process, comprehensively and real-timely monitors the growth process of the single crystal silicon, ensures that the environmental changes and the crystal growth status in the production are captured in time, and timely discovers the abnormality through the analysis of the feature data of the finishing stage of the crystal to avoid the production accidents caused by fracture or unstable state, ensures the accuracy and real-time of the adjustment process, and achieves the technical effect of high efficiency, stability and safety of the single crystal silicon production process.

[0065] Embodiment 2

[0066] Based on the same inventive concept as the dynamic control method for the single crystal silicon production process in the aforementioned embodiment, Figure 3As shown, the embodiment of the present application provides a dynamic control system for a single crystal silicon production process, wherein the system comprises:

[0067] An installation module 11 for installing a multimodal sensing unit in the drawing furnace, wherein the multimodal sensing unit includes a temperature sensor, a strain sensor and a visual monitoring element;

[0068] A data acquisition module 12, for acquiring multi-mode monitoring data using the multi-mode sensing unit, wherein the multi-mode monitoring data includes crystal growth state, interface stability and furnace environment parameters;

[0069] A data extraction module 13 is used to perform feature analysis on the multi-mode monitoring data and extract closing feature data corresponding to the closing stage of the crystal;

[0070] A risk identification module 14 is used to synchronize the key feature data to the constructed risk prediction model to identify the crystal growth state and potential fracture risk;

[0071] The correction control module 15 is used to start the preset optimization space to optimize the casting speed adjustment strategy based on the risk prediction result, output the optimal casting speed control strategy, and compare the deviation between the predicted casting speed and the actual casting speed to perform correction control.

[0072] Furthermore, the installation module 11 includes:

[0073] Based on the multimodal sensing unit, determining a data transmission network, wherein the data transmission network is used to transmit the multimodal monitoring data;

[0074] An encrypted communication protocol is used to provide data security protection for the data transmission network.

[0075] Furthermore, the risk identification module 14 includes:

[0076] Connect to the distributed database to obtain the speed control records;

[0077] Performing data mining based on the pulling speed control records to analyze the mapping relationship between the historical pulling speed change pattern and the crystal quality;

[0078] The risk prediction model is configured with parameters through the mapping relationship between the historical pulling speed variation pattern and the crystal quality.

[0079] Furthermore, the risk identification module 14 also includes:

[0080] Identify the first key factor affecting crystal quality and the second key factor affecting growth efficiency;

[0081] The weight bias of the risk prediction model is adjusted through the first key factor and the second key factor.

[0082] Furthermore, the risk identification module 14 also includes:

[0083] Using signal processing technology, the initialization information is separated;

[0084] Based on the initialization information, perform data dimensionality reduction processing to obtain dimensionality reduction processing information;

[0085] Based on the dimensionality reduction processing information, various factors affecting the crystal quality are ranked in order of importance, and the first key factor affecting the crystal quality is determined.

[0086] Furthermore, the correction control module 15 also includes:

[0087] Set the pull speed adjustment safety threshold;

[0088] During each pulling speed adjustment process, a judgment is made based on the pulling speed adjustment safety threshold. If it is exceeded, the current pulling speed adjustment operation is suspended and a safety reminder is issued.

[0089] Furthermore, the risk identification module 14 also includes:

[0090] When the pulling speed adjustment safety threshold is exceeded, a feedback control strategy is triggered;

[0091] Based on the feedback control strategy, adverse impact supplementary factors are determined and the risk prediction model is updated.

[0092] Furthermore, the risk identification module 14 also includes:

[0093] Use machine learning frameworks to deploy online learning mechanisms;

[0094] According to the online learning mechanism, performance optimization iterations are performed during the risk prediction model update process.

[0095] Embodiment 3

[0096] like Figure 3 , which is a schematic diagram of the structure of a dynamic control device for a single crystal silicon production process provided in Example 3 of the present application, and is suitable for implementing a block diagram of an exemplary dynamic control device for a single crystal silicon production process according to an embodiment of the present invention. Figure 3 The dynamic control device for the single crystal silicon production process shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention. Figure 3 As shown, the dynamic control device for the single crystal silicon production process includes a processor 21, a memory 22, an input device 23 and an output device 24; the number of processors 21 in the dynamic control device for the single crystal silicon production process can be one or more. Figure 3Taking a processor 21 as an example, the processor 21, memory 22, input device 23 and output device 24 in the dynamic control device for the single crystal silicon production process can be connected through a bus or other means. Figure 3 The example of connecting through bus is taken in the following.

[0097] The memory 22, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as the program instructions / modules corresponding to the dynamic control method for the single crystal silicon production process in the embodiment of the present application. The processor 21 executes various functional applications and data processing of the computer device by running the software programs, instructions and modules stored in the memory 22, thereby realizing the above-mentioned dynamic control method for the single crystal silicon production process.

[0098] It should be noted that the above-mentioned sequence of the embodiments of the present application is for description only and does not represent the advantages or disadvantages of the embodiments. The above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-tasking and parallel processing are also possible or may be advantageous.

[0099] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0100] This specification and drawings are merely exemplary illustrations of the present application and are deemed to have covered any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art may make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these modifications and variations.

[0101] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0102] Furthermore, the first or second mentioned above may not only represent an order relationship, but may also represent a specific concept, and / or refer to multiple elements that can be selected individually or in full. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A dynamic control method for a single crystal silicon production process, characterized in that: The method comprises: Installing a multi-modal sensing unit in the drawing furnace, the multi-modal sensing unit comprising a temperature sensor, a humidity sensor, a strain sensor, and a visual monitoring element; The multi-mode sensing unit is used to obtain multi-mode monitoring data, wherein the multi-mode monitoring data includes crystal growth state, interface stability and furnace environment parameters; Performing feature analysis on the multi-mode monitoring data to extract closing feature data corresponding to the closing stage of the crystal; Synchronizing the key feature data into the constructed risk prediction model to identify the crystal growth status and potential fracture risk; Based on the risk prediction results, the preset optimization space is started to optimize the casting speed adjustment strategy, the optimal casting speed control strategy is output, and the deviation between the predicted casting speed and the actual casting speed is compared for correction control.

2. The method according to claim 1, characterized in that A multimodal sensing unit is installed in a drawing furnace, the method comprising: Based on the multimodal sensing unit, determining a data transmission network, wherein the data transmission network is used to transmit the multimodal monitoring data; An encrypted communication protocol is used to provide data security protection for the data transmission network.

3. The method according to claim 2, characterized in that Synchronizing the key feature data into the constructed risk prediction model, the method comprising: Connect to the distributed database to obtain the speed control records; Performing data mining based on the pulling speed control records to analyze the mapping relationship between the historical pulling speed change pattern and the crystal quality; The risk prediction model is configured with parameters through the mapping relationship between the historical pulling speed variation pattern and the crystal quality.

4. The method according to claim 3, characterized in that Parameter configuration is performed on the risk prediction model, and the method includes: Identify the first key factor affecting crystal quality and the second key factor affecting growth efficiency; The weight bias of the risk prediction model is adjusted through the first key factor and the second key factor.

5. The method according to claim 4, characterized in that Identifying a first key factor affecting crystal quality and a second key factor affecting growth efficiency, the method comprising: Using signal processing technology, the initialization information is separated; Based on the initialization information, perform data dimensionality reduction processing to obtain dimensionality reduction processing information; Based on the dimensionality reduction processing information, various factors affecting the crystal quality are ranked in order of importance, and the first key factor affecting the crystal quality is determined.

6. The method according to claim 5, characterized in that Comparing the deviation between the predicted casting speed and the actual casting speed to perform correction control, the method further includes: Set the pull speed adjustment safety threshold; During each pulling speed adjustment process, a judgment is made based on the pulling speed adjustment safety threshold. If it is exceeded, the current pulling speed adjustment operation is suspended and a safety reminder is issued.

7. The method according to claim 4, characterized in that The method further comprises: When the pulling speed adjustment safety threshold is exceeded, a feedback control strategy is triggered; Based on the feedback control strategy, adverse impact supplementary factors are determined and the risk prediction model is updated.

8. The method according to claim 7, characterized in that and updating the risk prediction model, the method comprising: Use machine learning frameworks to deploy online learning mechanisms; According to the online learning mechanism, performance optimization iterations are performed during the risk prediction model update process.

9. A dynamic control system for a single crystal silicon production process, characterized in that: A system for implementing the dynamic control method for a single crystal silicon production process according to any one of claims 1 to 8, the system comprising: an installation module for installing a multimodal sensing unit in the drawing furnace, the multimodal sensing unit including a temperature sensor, a strain sensor, and a visual monitoring element; A data acquisition module, used to acquire multi-mode monitoring data using the multi-mode sensing unit, wherein the multi-mode monitoring data includes crystal growth status, interface stability and furnace environment parameters; A data extraction module, used to perform feature analysis on the multi-mode monitoring data and extract closing feature data corresponding to the closing stage of the crystal; A risk identification module, used to synchronize the key feature data to the constructed risk prediction model to identify the crystal growth status and potential fracture risks; The correction control module is used to start the preset optimization space to optimize the casting speed adjustment strategy based on the risk prediction results, output the optimal casting speed control strategy, and compare the deviation between the predicted casting speed and the actual casting speed for correction control.

10. A dynamic control device for a single crystal silicon production process, characterized in that: include: A memory for storing executable instructions; The processor is used to implement the dynamic control method for a single crystal silicon production process as described in any one of claims 1 to 8 when executing the executable instructions stored in the memory.

Images