A temperature control method and related apparatus
By using a temperature control method based on a predictive model during crystal growth, the problem of insufficient temperature control accuracy was solved, thereby improving the consistency of liquid temperature and increasing production efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BAODING JING XIN SHI CHUANG ELECTRIC
- Filing Date
- 2026-03-26
- Publication Date
- 2026-07-03
AI Technical Summary
The accuracy of temperature control during crystal growth in existing technologies is insufficient, resulting in poor consistency of liquid temperature after temperature control is completed, which affects crystal quality and production efficiency.
A temperature control method based on three predictive models is adopted, which combines current working data to predict the timing of shutting down the low heater and the power correction amount of the main heater, thereby achieving standardized control of the crystal furnace temperature.
It improves the consistency of liquid temperature after temperature control, reduces the limitations of subjective judgment and fixed parameter adjustment, and enhances the accuracy of temperature control and production efficiency.
Smart Images
Figure CN122331670A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of crystal growth technology, specifically to a temperature control method and related apparatus. Background Technology
[0002] Temperature is a core control parameter in the crystal growth process, directly affecting crystal quality, growth rate, and defect density.
[0003] In related technologies, the melting process of solid silicon material can be observed in real time, and the power of the heater can be manually adjusted, or the temperature of the crystal furnace can be controlled based on a set of fixed parameters.
[0004] However, the accuracy of temperature control using the above methods is insufficient. Summary of the Invention
[0005] In view of this, this application provides a temperature control method and related apparatus. Based on three prediction models combined with current working data, it can predict the timing of shutting down the low heater and the power correction value of the main heater after shutting down the bottom heater, with the goal of achieving better liquid temperature consistency after temperature control. Then, it controls the temperature of the crystal furnace based on the prediction results. Compared with related technologies, it avoids the limitations of manual subjective judgment or fixed parameter adjustment, and can achieve better liquid temperature consistency after temperature control, realize standardized control of the crystal furnace temperature, and improve the accuracy of temperature control.
[0006] To solve the above problems, the technical solution provided in this application is as follows: On one hand, embodiments of this application provide a temperature control method, which is applied in the material preparation stage, and the method includes: Obtain the current operating data of the crystal furnace; Based on the current working data and the first prediction model, the timing for shutting down the bottom heater is determined. Shutting down the bottom heater at the specified timing ensures that the expected consistency of the liquid temperature after temperature control ends reaches the first target threshold. The bottom heater is turned off based on the aforementioned timing; Based on the current working data, the second prediction model is used to predict the first time duration for the solution percentage in the crystal furnace to reach the second preset threshold. Based on the first duration and the current working data, the power correction amount of the main heater is predicted by the third prediction model. The power correction amount is used to make the expected liquid temperature consistency reach the second target threshold. The main heater is controlled based on the power correction amount.
[0007] As one possible implementation, the method further includes: If the solution percentage does not reach the second preset threshold after the first time period, the bottom heater is turned on. If the solution percentage reaches the second preset threshold, the bottom heater is turned off; Based on current operating data and the third prediction model, the power correction amount for the main heater is predicted.
[0008] As one possible implementation, turning off the bottom heater if the solution percentage reaches the second preset threshold includes: If the solution percentage reaches the second preset threshold or the rate of change of the solution percentage reaches a preset rate of change, the bottom heater is turned off.
[0009] As one possible implementation, the method further includes: Acquire historical working data, including liquid temperature consistency data after temperature control; Filter the historical working data to obtain valid data if the liquid temperature consistency data falls within a preset range. Based on the effective data, the initial first prediction model, the initial second prediction model, and the initial third prediction model are trained respectively to obtain the first prediction model, the second prediction model, and the third prediction model.
[0010] As one possible implementation, the current working data includes solid silicon feed data, power adjustment data, and temperature data.
[0011] As one possible implementation, the first prediction model is built on a Long Short-Term Memory (LSTM) network.
[0012] In another aspect, embodiments of this application provide a temperature control device, the device comprising: The acquisition unit is used to acquire the current working data of the crystal furnace; The determining unit is used to determine the timing for shutting down the bottom heater based on the current working data and the first prediction model. Shutting down the bottom heater based on the timing can ensure that the expected consistency of the liquid temperature after the temperature control ends reaches a first target threshold. Control unit, configured to shut off the bottom heater based on the stated timing; The prediction unit is used to predict, based on the current working data and through a second prediction model, the first time duration for the solution percentage in the crystal furnace to reach a second preset threshold. The prediction unit is further configured to predict the power correction amount of the main heater based on the first duration and the current working data through a third prediction model. The power correction amount is used to make the expected liquid temperature consistency reach the second target threshold. The control unit is also used to control the main heater based on the power correction amount.
[0013] As one possible implementation, the control unit is further configured to: If the solution percentage does not reach the second preset threshold after the first time period, the bottom heater is turned on. If the solution percentage reaches the second preset threshold, the bottom heater is turned off; The prediction unit is also used to predict the power correction amount of the main heater based on the current working data and the third prediction model.
[0014] As one possible implementation, the control unit is used for: If the solution percentage reaches the second preset threshold or the rate of change of the solution percentage reaches a preset rate of change, the bottom heater is turned off.
[0015] As one possible implementation, the device further includes a screening unit and a training unit; The acquisition unit is also used to acquire historical working data, which includes liquid temperature consistency data after temperature control. The filtering unit is used to filter the liquid temperature consistency data in the historical working data within a preset range to obtain valid data. The training unit is used to train the initial first prediction model, the initial second prediction model, and the initial third prediction model based on the effective data, respectively, to obtain the first prediction model, the second prediction model, and the third prediction model.
[0016] As one possible implementation, the control unit is further configured to: The current working data includes solid silicon material feeding data, power adjustment data, and temperature data.
[0017] As one possible implementation, the control unit is further configured to: The first prediction model is built on a Long Short-Term Memory (LSTM) network.
[0018] In another aspect, this application provides a computer device, which includes a processor and a memory: The memory is used to store computer programs; The processor is configured to execute the method described in any of the above-described embodiments according to the computer program.
[0019] In another aspect, this application provides a computer-readable storage medium for storing a computer program that, when executed by a computer device, implements the method described in any of the above-mentioned embodiments.
[0020] In another aspect, this application provides a computer program product including a computer program, which, when run on a computer device, causes the computer device to perform any of the methods described above.
[0021] As can be seen from the above technical solution, when applied to the material preparation stage, the following steps are taken: First, the current operating data of the crystal furnace is acquired. Then, based on the current operating data and a first prediction model, the timing for shutting down the bottom heater is determined. Shutting down the bottom heater at this timing ensures that the expected liquid temperature consistency reaches the first target threshold after temperature control. Furthermore, by predicting the timing for shutting down the bottom heater, the bottom heater can be controlled, resulting in better temperature control. Next, based on the current operating data, a second prediction model predicts the first time interval for the solution percentage in the crystal furnace to reach the second preset threshold. Then, based on the first time interval and the current operating data, the prediction model predicts the power correction amount for the main heater. This power correction amount is used to ensure that the expected liquid temperature consistency reaches the second target threshold. Finally, the main heater is controlled based on the power correction amount to guarantee the liquid temperature consistency effect after temperature control. Therefore, based on three prediction models and current working data, the timing of shutting down the low heater and the power correction value of the main heater after shutting down the bottom heater can be predicted with the goal of achieving better liquid temperature consistency after temperature control. Then, the temperature of the crystal furnace can be controlled based on the prediction results. Compared with related technologies, this avoids the limitations of subjective judgment or fixed parameter adjustment, and can achieve better liquid temperature consistency after temperature control. This realizes standardized control of the crystal furnace temperature and improves the accuracy of temperature control. Attached Figure Description
[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0023] Figure 1 A schematic flowchart of a temperature control method provided in an embodiment of this application; Figure 2 A flowchart for model building provided in this application embodiment; Figure 3 A logic flowchart for temperature control provided in an embodiment of this application; Figure 4 This is a schematic diagram of a temperature control device provided in an embodiment of this application. Detailed Implementation
[0024] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.
[0025] As described in the background section, by observing the crystal's melting process in real time, the heater power can be adjusted based on human experience, or the temperature can be automatically controlled by a machine using a set of fixed parameters. However, this method suffers from a small observation window (e.g., limited field of view of a CCD camera) and numerous factors such as inconsistent solid material thickness, varying refill weights, inconsistent furnace insulation, and attenuation of thermal insulation make it difficult to standardize power adjustment strategies and guarantee consistent liquid temperature after temperature control, resulting in insufficient accuracy in temperature control. This not only leads to additional consumption of labor and energy but also seriously affects the stability and product quality of subsequent crystal processing. Furthermore, this model requires a large number of skilled workers to continuously monitor and intervene, resulting in high labor costs and hindering further improvements in production efficiency and capacity, becoming a significant bottleneck restricting cost control and large-scale production.
[0026] To address the aforementioned issues, this application provides a temperature control method. Based on three prediction models combined with current working data, it aims to improve the consistency of liquid temperature after temperature control. The method predicts the timing of shutting down the low-level heater and the power correction value for the main heater after shutting down the bottom heater. Based on the prediction results, the temperature of the crystal furnace is controlled. Compared to related technologies, this method avoids the limitations of subjective human judgment or fixed parameter adjustments, achieving better consistency of liquid temperature after temperature control. It realizes standardized temperature control of the crystal furnace and improves the accuracy of temperature control.
[0027] The solutions provided in this application relate to the field of crystal growth technology, and are specifically illustrated through the following embodiments.
[0028] See Figure 1 The diagram shown is a schematic flow chart of a temperature control method provided in an embodiment of this application, including steps S101-S106. This method is applied to the material preparation stage of crystal growth, that is, the stage of melting solid silicon material into a liquid.
[0029] S101: Obtain the current operating data of the crystal furnace.
[0030] Current working data refers to working data between the current time and the present time, and this application embodiment does not impose specific limitations on it.
[0031] As one possible approach, current working data includes solid silicon feed data, power adjustment data, and temperature data.
[0032] The solid silicon material feeding data refers to the data during the solid silicon material feeding stage. This application embodiment does not impose specific limitations on this, such as including the weight of solid silicon material fed each time, the number of buckets of solid silicon material fed each time, the time interval between each bucket, etc.
[0033] The greater the weight of solid silicon material, the greater the total heat required for melting, or the worse the temperature uniformity. Therefore, the solid silicon material addition data is one of the important factors affecting temperature control.
[0034] Power adjustment data refers to the data used to adjust the power of the heaters in the crystal furnace. This includes, for example, the initial power, adjusted power, and adjustment timing of the bottom heater, as well as the initial power, adjusted power, and adjustment timing of the main heater. Power adjustment data directly affects temperature consistency and is one of the important considerations for temperature control.
[0035] Temperature data refers to the temperature inside the crystal furnace, accumulated heat, etc. Temperature data can be directly used to characterize or evaluate temperature consistency.
[0036] Therefore, based on these three types of data, the working state of the crystal furnace can be characterized, thereby achieving more precise temperature control and better temperature consistency.
[0037] Furthermore, the current working data includes not only directly obtainable data, but also data further constructed based on the obtained data. For example, the current working data includes main heater power, auxiliary heater power, solid silicon material feeding weight, weight per cylinder, cylinder-to-cylinder time interval, crucible rotation, argon flow rate, feeding time, running time, crystal pulling power, hot zone size, power utilization efficiency, main / auxiliary power ratio, power efficiency, time efficiency, unit energy consumption, cooling efficiency, crucible rotation solution ratio, crucible rotation temperature effect, thin film solution ratio, material turning solution ratio, solution-raw material balance, thin film material proportion fluctuation, material turning proportion fluctuation, hot screen position fluctuation, crucible rotation fluctuation, and crucible rotation change rate, etc.
[0038] S102: Based on the current working data and the first prediction model, determine when to shut down the bottom heater.
[0039] The bottom heater primarily heats the bottom of the crucible to ensure that the silicon material at the bottom melts.
[0040] The first prediction model is used to predict the probability of shutting down the bottom heater; a higher probability indicates better consistency of the expected liquid temperature after temperature control ends. This application does not impose specific limitations on the first prediction model. For example, the first prediction model may be constructed based on a fully connected neural network, supporting gradient optimization.
[0041] As one possible implementation, the first prediction model is built on a Long Short-Term Memory (LSTM) network.
[0042] LSTM possesses a "memory" capability, enabling it to learn and retain important information from long-sequence data while filtering out irrelevant information, making it highly adept at processing and predicting time-series data. Therefore, a first prediction model built upon LSTM can predict the probability of the bottom heater being shut down in the near future.
[0043] Based on the probability prediction results, the optimal time to shut down the bottom heater can be determined. For example, the time with the highest probability can be used as the time to shut down the bottom heater, thereby making the expected liquid temperature consistency better.
[0044] In this embodiment of the application, turning off the bottom heater at this time can ensure that the expected consistency of the liquid temperature after the temperature control ends reaches the first target threshold.
[0045] Among them, liquid temperature uniformity refers to the degree of uniform distribution of liquid temperature, and the first target threshold refers to a pre-set threshold. This application embodiment does not impose specific restrictions on this, such as the liquid temperature distribution error being within 1 degree Celsius.
[0046] As one possible implementation, a first preset threshold can be set, and the bottom heater can be turned off when the probability of reaching the first preset threshold is closest. The first preset threshold refers to a pre-set threshold, which is not specifically limited in this embodiment; for example, it could be 0.5. When the probability of turning off the bottom heater is greater than the first preset threshold, turning off the bottom heater can result in better consistency of the liquid temperature after temperature control.
[0047] S103: Turn off the bottom heater based on timing.
[0048] For example, a shutdown signal is generated after the timing for shutting down the bottom heater is determined, and the bottom heater is shut down at the time corresponding to that timing based on the shutdown signal.
[0049] S104: Based on the current working data, predict the first time duration for the solution ratio in the crystal furnace to reach the second preset threshold using the second prediction model.
[0050] The second prediction model is used to predict the time required for the solution percentage to reach the second preset threshold after the bottom heater is turned off. This application embodiment does not impose specific limitations on the second prediction model.
[0051] As one possible approach, historical work data is acquired and filtered based on the absence of human intervention. The time taken for the solution percentage to reach a second preset threshold is used as a label to construct an effective dataset. The second prediction model is then trained under supervision using the Lightweight Gradient Boosting Decision Tree (GBDT) algorithm to obtain the trained second prediction model.
[0052] The second preset threshold is a pre-set threshold, and this application embodiment does not impose specific restrictions on it, such as 80%, 95%, etc.
[0053] After the bottom heater is shut down, the main heater provides heat independently to even out the temperature distribution and prevent localized overheating. The current operating data is input into the second prediction model to predict the first time it will take for the solution percentage to reach the second preset threshold.
[0054] Furthermore, it can be determined whether the timing of shutting down the bottom heater is appropriate.
[0055] S105: Based on the first duration and current operating data, predict the power correction amount of the main heater through the third prediction model.
[0056] The third prediction model is used to predict the power correction amount of the main heater. This application does not impose specific limitations on the third prediction model.
[0057] As one possible approach, historical working data is acquired, and a valid dataset is constructed by filtering data based on the consistency of liquid temperature within a preset range. By analyzing the correspondence between the timing of multiple opening and closing of the bottom heater and the power correction amount, a lightweight GBDT algorithm is used to supervise the training of the third prediction model, resulting in the trained third prediction model.
[0058] The power correction amount refers to the amount of data used to adjust the power of the main heater to ensure that the expected liquid temperature consistency reaches the second target threshold. This application does not impose specific limitations on the second target threshold.
[0059] The first duration and current working data are input into the third prediction model to predict the power correction amount of the main heater. Based on the power correction amount, the power of the main heater can be adjusted to achieve better liquid temperature consistency.
[0060] S106: Control the main heater based on the power correction amount.
[0061] For example, after obtaining the power correction amount, a correction signal is generated, and the main heater is controlled based on the correction signal.
[0062] Therefore, this technical solution is applied to the material preparation stage. First, the current operating data of the crystal furnace is acquired. Then, based on the current operating data and a first prediction model, the timing for shutting down the bottom heater is determined. Shutting down the bottom heater at this timing ensures that the expected liquid temperature consistency reaches the first target threshold after temperature control. Furthermore, by predicting the timing for shutting down the bottom heater, the bottom heater can be controlled, resulting in better temperature control. Next, based on the current operating data, a second prediction model predicts the first time interval for the solution percentage in the crystal furnace to reach the second preset threshold. Then, based on the first time interval and the current operating data, the prediction model predicts the power correction amount for the main heater. This power correction amount is used to ensure that the expected liquid temperature consistency reaches the second target threshold. Finally, the main heater is controlled based on the power correction amount to guarantee the liquid temperature consistency effect after temperature control. Therefore, based on three prediction models and current working data, the timing of shutting down the low heater and the power correction value of the main heater after shutting down the bottom heater can be predicted with the goal of achieving better liquid temperature consistency after temperature control. Then, the temperature of the crystal furnace can be controlled based on the prediction results. Compared with related technologies, this avoids the limitations of subjective judgment or fixed parameter adjustment, and can achieve better liquid temperature consistency after temperature control. This realizes standardized control of the crystal furnace temperature and improves the accuracy of temperature control.
[0063] As one possible implementation, the method also includes: A1: If the solution percentage does not reach the second preset threshold after the first time period, turn on the bottom heater.
[0064] A2: If the solution percentage reaches the second preset threshold, turn off the bottom heater.
[0065] A3: Based on the current working data and the third prediction model, predict the power correction amount for the main heater.
[0066] The current operating data of the crystal furnace is continuously monitored. After the first time period, if the solution ratio does not reach the second preset threshold, it indicates that the prediction result of the second prediction model has an error. Then, the bottom heater is restarted to increase the temperature inside the crystal furnace and ensure that the solid silicon material at the bottom of the crucible melts. When the solution ratio reaches the second preset threshold, the bottom heater is turned off. Then, the current operating data, such as the timing of multiple times the bottom heater is turned on and off, the duration of the bottom heater being turned off, etc., are input into the third prediction model to analyze the correspondence between time and power correction and obtain the power correction amount of the main heater for this time.
[0067] Therefore, if the solution percentage fails to reach the second preset threshold after the first time, the bottom heater is turned on to ensure that the solid silicon material is completely melted. Then, based on the real-time current working data, the power correction value of the main heating power can be accurately predicted, thereby ensuring the consistency of the liquid temperature after temperature control.
[0068] As one possible implementation, A2 includes: If the solution percentage reaches the second preset threshold or the rate of change of the solution percentage reaches the preset rate of change, the bottom heater is turned off.
[0069] The rate of change of the solution percentage indicates whether the temperature inside the crystal furnace is suitable. An excessively rapid rate of change indicates overheating, which may lead to localized overheating.
[0070] The preset rate of change refers to a rate of change that is set in advance, and this application does not impose specific limitations on this.
[0071] For example, when the solution percentage reaches a second preset threshold, a shutdown signal is generated to shut down the bottom heater, or when the rate of change of the solution percentage is detected to reach a preset rate of change, a shutdown signal is generated to shut down the bottom heater.
[0072] Therefore, when the conditions are met, only the main heater is retained for temperature control to balance the thermal field, prevent local overheating, and achieve better temperature uniformity.
[0073] As one possible implementation, the method also includes: B1: Obtain historical work data.
[0074] B2: Filter historical working data to obtain valid data based on liquid temperature consistency within a preset range; B3: Based on valid data, train the initial first prediction model, the initial second prediction model, and the initial third prediction model respectively to obtain the first prediction model, the second prediction model, and the third prediction model.
[0075] Among them, historical working data includes liquid temperature consistency data after temperature control, which is used to characterize the uniformity of liquid temperature distribution.
[0076] The preset range refers to a pre-set threshold range. This application does not impose specific restrictions on this. For example, the target range refers to a temperature difference of liquid within 1 degree Celsius.
[0077] For example, after obtaining historical working data, data is filtered based on liquid temperature consistency data. Data with liquid temperature consistency within a preset range is taken as valid data. Then, the initial first prediction model is trained based on the valid data to obtain the first prediction model. The initial second prediction model is trained to obtain the second prediction model. The initial third model is trained to obtain the third prediction model.
[0078] Therefore, data based on liquid temperature consistency data can be removed from abnormal situations such as human intervention or unreasonable power adjustments, resulting in effective data. This data can then be used to train the prediction model, improving model convergence efficiency and thus enhancing the model's prediction accuracy.
[0079] In addition, after the solid silicon material is fully melted, the actual liquid temperature can be obtained. At this time, dynamic linear adjustment is made according to the liquid temperature change trend to ensure that the liquid temperature can reach the standard after the temperature control is completed.
[0080] To more clearly describe this temperature control method, the following explanation will be provided in conjunction with a specific implementation scenario, including S1-S5.
[0081] S1: Building the model.
[0082] See Figure 2 The diagram shown is a flowchart of a model building process provided in an embodiment of this application. First, batch historical production data (i.e., historical work data) is acquired. Then, feature engineering is constructed to extract features from the historical work data, including: main power at the start of each temperature control cycle, secondary power, refill weight, weight per cylinder, cylinder-to-cylinder time interval, crucible rotation, argon flow rate, feeding time, running time, crystal pulling power, hot zone size, power utilization efficiency, main / secondary power ratio, power efficiency, time efficiency, unit energy consumption, cooling efficiency, crucible rotation solution ratio, crucible rotation temperature influence, thin film solution ratio, material turning solution ratio, solution raw material balance, thin film material ratio fluctuation, material turning ratio fluctuation, hot screen position fluctuation, crucible rotation fluctuation, and crucible rotation change rate. Records of abnormal human intervention, unreasonable power adjustment timing, and ineffective normal intervention are cleaned to obtain a dataset. Then, models are constructed based on the dataset according to different objectives and different algorithms.
[0083] First prediction model: Use LSTM to build a probabilistic model of the bottom heater turning off.
[0084] The second prediction model uses a Light Gradient Boosting Machine (lightGBM) to construct a time range from low to high until a threshold is reached.
[0085] The third prediction model uses lightGBM to construct the actual time taken for the main power to reach the threshold and infers the main power correction amount.
[0086] After model training is complete, the temperature control process begins, which involves real-time access to the model to control the temperature. See also Figure 3 The diagram shown is a logic flowchart of a temperature control system provided in an embodiment of this application. At the start of temperature control, the power of the main heater is 100KW, and the power of the bottom heater is 80KW. Then, steps S2-S5 are executed.
[0087] S2: Determine whether to turn off the bottom heater.
[0088] After the temperature control begins, the current operating data is acquired and input into the first prediction model. The model outputs the inference probability value for shutting down the bottom heater, and checks whether it is greater than a probability threshold (e.g., 0.5). If it is, the bottom heater is shut down. Otherwise, the current operating data is continuously monitored, and the first prediction model is invoked to make predictions until the inference probability value is greater than the probability threshold, at which point the bottom heater is shut down.
[0089] S3: Determine the solution ratio threshold after turning off the heater.
[0090] The second prediction model is invoked to predict the inference time (i.e., the first duration) from when the bottom heater is turned off until the solution percentage reaches a (second preset) threshold. After the bottom heater is turned off, the change in solution percentage is continuously monitored. If the solution percentage still has not reached the (second preset) threshold after the duration reaches the inference time of the second prediction model, the bottom heater is turned back on. After reaching the threshold, the bottom heater is turned off. (This addresses abnormal situations such as thick solid blocks, changes in shape, and slower melting speed).
[0091] S4: After the bottom heater is turned off, adjust the power of the main heater.
[0092] After the bottom heater is turned off, the current working data is obtained and input into the third prediction model to obtain the power correction amount and adjust the power of the main heater.
[0093] S5: Temperature control ends. After the solid silicon material has fully melted, further temperature adjustments are made.
[0094] After the silicon material is fully melted and the crucible is positioned, the actual liquid temperature is obtained. The deviation from the target liquid temperature of 1450±4°C is dynamically and linearly adjusted based on the temperature change trend over the previous 10 cycles. This ensures that the liquid temperature meets the standard after temperature control is completed.
[0095] Therefore, this application introduces a data analysis method to select key features in the temperature control process from historical work data and expert experience. An effective dataset is constructed from these key features and used as input for model training, resulting in a data-driven temperature control model. This improves the automated decision-making capability in the Czochralski single crystal silicon process, reduces the possibility of manual intervention and misjudgment, and significantly improves process stability and crystal quality.
[0096] Based on the above embodiments, this application provides a temperature control device, with reference to... Figure 4 The diagram shown is a schematic of a temperature control device provided in an embodiment of this application. The device 400 includes: Acquisition unit 401 is used to acquire the current working data of the crystal furnace; The determining unit 402 is used to determine the timing for shutting down the bottom heater based on the current working data and the first prediction model. Shutting down the bottom heater based on the timing can ensure that the expected consistency of the liquid temperature after the temperature control ends reaches a first target threshold. Control unit 403 is configured to shut off the bottom heater based on the stated timing; Prediction unit 404 is used to predict, based on current working data and through a second prediction model, the first time duration for the solution percentage in the crystal furnace to reach a second preset threshold. The prediction unit 404 is further configured to predict the power correction amount of the main heater based on the first duration and the current working data through a third prediction model. The power correction amount is used to make the expected liquid temperature consistency reach the second target threshold. The control unit 403 is also used to control the main heater based on the power correction amount.
[0097] As one possible implementation, the control unit is further configured to: If the solution percentage does not reach the second preset threshold after the first time period, the bottom heater is turned on. If the solution percentage reaches the second preset threshold, the bottom heater is turned off; The prediction unit is also used to predict the power correction amount of the main heater based on the current working data and the third prediction model.
[0098] As one possible implementation, the control unit is used for: If the solution percentage reaches the second preset threshold or the rate of change of the solution percentage reaches a preset rate of change, the bottom heater is turned off.
[0099] As one possible implementation, the device further includes a screening unit and a training unit; The acquisition unit is also used to acquire historical working data, which includes liquid temperature consistency data after temperature control. The filtering unit is used to filter the liquid temperature consistency data in the historical working data within a preset range to obtain valid data. The training unit is used to train the initial first prediction model, the initial second prediction model, and the initial third prediction model based on the effective data, respectively, to obtain the first prediction model, the second prediction model, and the third prediction model.
[0100] As one possible implementation, the control unit is further configured to: The current working data includes solid silicon material feeding data, power adjustment data, and temperature data.
[0101] As one possible implementation, the control unit is further configured to: The first prediction model is built on a Long Short-Term Memory (LSTM) network.
[0102] Based on the above embodiments, this application provides a computer device, which includes a processor and a memory: The memory is used to store computer programs; The processor is used to execute the temperature control method described above according to the computer program.
[0103] Based on the above embodiments, this application provides a computer-readable storage medium for storing a computer program, which, when executed by a computer device, implements the above-described temperature control method.
[0104] Based on the above embodiments, this application provides a computer program product including a computer program, which, when run on a computer device, causes the computer device to execute the above-described temperature control method.
[0105] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.
[0106] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A temperature control method characterized by, The method is applied in the material preparation stage, and the method includes: Obtain the current operating data of the crystal furnace; Based on the current working data and the first prediction model, the timing for shutting down the bottom heater is determined. Shutting down the bottom heater at the specified timing ensures that the expected consistency of the liquid temperature after temperature control ends reaches the first target threshold. The bottom heater is turned off based on the aforementioned timing; Based on the current working data, the second prediction model is used to predict the first time duration for the solution percentage in the crystal furnace to reach the second preset threshold. Based on the first duration and the current working data, the power correction amount of the main heater is predicted by the third prediction model. The power correction amount is used to make the expected liquid temperature consistency reach the second target threshold. The main heater is controlled based on the power correction amount.
2. The method of claim 1, wherein, The method further includes: If the solution percentage does not reach the second preset threshold after the first time period, the bottom heater is turned on. If the solution percentage reaches the second preset threshold, the bottom heater is turned off; Based on current operating data and the third prediction model, the power correction amount for the main heater is predicted.
3. The method of claim 2, wherein, The step of turning off the bottom heater if the solution percentage reaches the second preset threshold includes: If the solution percentage reaches the second preset threshold or the rate of change of the solution percentage reaches a preset rate of change, the bottom heater is turned off.
4. The method of claim 1, wherein, The method further includes: Acquire historical working data, including liquid temperature consistency data after temperature control; Filter the historical working data to obtain valid data if the liquid temperature consistency data falls within a preset range. Based on the effective data, the initial first prediction model, the initial second prediction model, and the initial third prediction model are trained respectively to obtain the first prediction model, the second prediction model, and the third prediction model.
5. The method according to claim 1, characterized in that, The current working data includes solid silicon material feeding data, power adjustment data, and temperature data.
6. The method according to claim 1, characterized in that, The first prediction model is built on a Long Short-Term Memory (LSTM) network.
7. A temperature control device, characterized in that, The device includes: The acquisition unit is used to acquire the current working data of the crystal furnace; The determining unit is used to determine the timing for shutting down the bottom heater based on the current working data and the first prediction model. Shutting down the bottom heater based on the timing can ensure that the expected consistency of the liquid temperature after the temperature control ends reaches a first target threshold. Control unit, configured to shut off the bottom heater based on the stated timing; The prediction unit is used to predict, based on the current working data and through a second prediction model, the first time duration for the solution percentage in the crystal furnace to reach a second preset threshold. The prediction unit is further configured to predict the power correction amount of the main heater based on the first duration and the current working data through a third prediction model. The power correction amount is used to make the expected liquid temperature consistency reach the second target threshold. The control unit is also used to control the main heater based on the power correction amount.
8. A computer device, characterized in that, The computer device includes a processor and memory: The memory is used to store computer programs; The processor is configured to perform the method according to any one of claims 1-6 according to the computer program.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program that, when executed by a computer device, performs the method described in any one of claims 1-6.
10. A computer program product comprising a computer program, characterized in that, When it is run on a computer device, it causes the computer device to perform the method described in any one of claims 1-6.