An oxygen content prediction method and device applied to a single crystal silicon pulling process

Abstract

This application provides a method and apparatus for predicting oxygen content in the monocrystalline silicon pulling process. It can continuously collect time-series data of target process parameters that are temporally correlated with oxygen content during the monocrystalline silicon pulling process. Based on the collected time-series data, it constructs the input data for an oxygen content prediction model, and then outputs the predicted oxygen content in real time through the pre-trained oxygen content prediction model. This not only overcomes the technical deficiency of existing technologies that can only passively detect oxygen content after the monocrystalline silicon pulling process is completed, but also effectively reduces the reliance on manual operation in offline detection methods, thus reducing labor costs and improving oxygen content detection efficiency.

Description

Technical Field

[0001] This application relates to the field of monocrystalline silicon preparation technology, and more specifically, to a method and apparatus for predicting oxygen content in the monocrystalline silicon pulling process. Background Technology

[0002] In the production of monocrystalline silicon, oxygen content is a key indicator that determines the core performance of silicon wafers, such as resistivity and minority carrier lifetime, and directly affects the quality of downstream photovoltaic and semiconductor devices. Currently, existing technologies mainly employ offline detection methods. After the entire monocrystalline silicon pulling process is completed, samples of the produced monocrystalline silicon are sliced, and the oxygen content is measured from the slices using methods such as infrared spectroscopy analysis.

[0003] However, on the one hand, the offline testing process is cumbersome, requiring dedicated personnel to complete the entire process of sampling, testing, and data analysis. Therefore, the existing technology suffers from low oxygen content detection efficiency and high labor costs. On the other hand, since offline testing measures oxygen content from the pulled monocrystalline silicon, the existing technology also suffers from strong detection lag and cannot determine whether the oxygen content is abnormal during the pulling process. As a result, it is difficult to recover the monocrystalline silicon production losses caused by abnormal oxygen content by adjusting process parameters during the pulling process. Summary of the Invention

[0004] In view of this, this application provides a method and apparatus for predicting oxygen content in the monocrystalline silicon pulling process. This method can continuously collect time-series data of target process parameters that are temporally correlated with oxygen content during the monocrystalline silicon pulling process. Based on the collected time-series data, it constructs the input data for an oxygen content prediction model, thereby outputting the predicted oxygen content in real time through a pre-trained oxygen content prediction model. This not only overcomes the technical deficiency of existing technologies that can only passively detect oxygen content after the monocrystalline silicon pulling process is completed, but also effectively reduces the reliance on manual operation in offline detection methods, thus reducing labor costs and improving oxygen content detection efficiency.

[0005] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings.

[0006] In a first aspect, embodiments of this application provide a method for predicting oxygen content in a single-crystal silicon pulling process, the oxygen content prediction method comprising: From the monocrystalline silicon pulling process, the parameter values ​​of the target process parameters are collected in real time at each sampling moment to obtain the original time-series data of the target process parameters; wherein, the target process parameters are process parameters in the monocrystalline silicon pulling process that have a time-series correlation with oxygen content; The original timing data is segmented according to the multiple process segments in the monocrystalline silicon pulling process to obtain the timing data corresponding to each of the multiple process segments. For each process segment, the timing data corresponding to that process segment is divided into multiple sub-timing data corresponding to each time window according to a preset step size time window; The sub-time series data corresponding to the multiple time windows are input into the pre-trained oxygen content prediction model. The oxygen content prediction model predicts the oxygen content corresponding to each sub-time series data and outputs the oxygen content prediction result of the process segment under each time window.

[0007] Secondly, embodiments of this application provide an oxygen content prediction device applied in the single-crystal silicon pulling process, the oxygen content prediction device comprising: The parameter acquisition module is used to acquire the parameter values ​​of the target process parameters at each sampling time in real time during the monocrystalline silicon pulling process, and obtain the raw time-series data of the target process parameters; wherein, the target process parameters are process parameters in the monocrystalline silicon pulling process that have a time-series correlation with oxygen content; The first processing module is used to segment the original timing data according to multiple process segments in the single crystal silicon pulling process, so as to obtain the timing data corresponding to each of the multiple process segments. The second processing module is used to split the timing data corresponding to each process segment into sub-timing data corresponding to multiple time windows according to a preset step size time window for each process segment. The prediction module is used to input the sub-time series data corresponding to the multiple time windows into a pre-trained oxygen content prediction model, and to predict the oxygen content corresponding to each sub-time series data through the oxygen content prediction model, and output the oxygen content prediction result corresponding to the process segment under each time window.

[0008] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the oxygen content prediction method described above.

[0009] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the oxygen content prediction method described above.

[0010] The technical solutions provided by the embodiments of this application may include the following beneficial effects: This application provides a method and apparatus for predicting oxygen content in the monocrystalline silicon pulling process. It can continuously collect time-series data of target process parameters that are temporally correlated with oxygen content during the monocrystalline silicon pulling process. Based on the collected time-series data, it constructs the input data for an oxygen content prediction model, and then outputs the predicted oxygen content in real time through the pre-trained oxygen content prediction model. This not only overcomes the technical deficiency of existing technologies that can only passively detect oxygen content after the monocrystalline silicon pulling process is completed, but also effectively reduces the reliance on manual operation in offline detection methods, thus reducing labor costs and improving oxygen content detection efficiency. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A schematic flowchart of an oxygen content prediction method applied in the single-crystal silicon pulling process provided in an embodiment of this application is shown. Figure 2 This illustration shows a schematic diagram of an oxygen content prediction device applied in the single-crystal silicon pulling process, provided by an embodiment of this application. Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0014] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0015] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0016] One embodiment of this application provides a method for predicting oxygen content in a monocrystalline silicon pulling process. This method can be run in an oxygen content prediction device, which establishes a communication connection with various sensors installed on the monocrystalline silicon pulling equipment. This allows for the real-time and continuous acquisition of time-series data of various process parameters during the monocrystalline silicon pulling process.

[0017] It should be noted that the oxygen content prediction device can be an electronic device with data processing capabilities, such as a terminal device. This application does not limit the specific type of device or equipment for oxygen content prediction.

[0018] To facilitate understanding of the embodiments of this application, a method and apparatus for predicting oxygen content in the single-crystal silicon pulling process provided by the embodiments of this application will be described in detail below.

[0019] Reference Figure 1 As shown, Figure 1 The diagram illustrates a flow chart of an oxygen content prediction method applied in the single-crystal silicon pulling process according to an embodiment of this application. The oxygen content prediction method includes steps S101-S104; specifically: S101, from the single crystal silicon pulling process, the parameter values ​​of the target process parameters at each sampling time are collected in real time to obtain the original time series data of the target process parameters.

[0020] Here, the target process parameters are the process parameters in the monocrystalline silicon pulling process that have a time-series relationship with oxygen content; among them, the target process parameters may include, but are not limited to, the heating power and pulling rate of the monocrystalline silicon pulling equipment (such as a monocrystalline furnace).

[0021] Specifically, step S101 runs through the entire monocrystalline silicon pulling process. That is, in the entire monocrystalline silicon pulling process, the oxygen content prediction device can continuously collect the parameter values ​​of each target process parameter at each sampling time from multiple sensors installed on the monocrystalline silicon pulling equipment (such as a monocrystalline furnace) in real time according to a preset sampling frequency (e.g., the sampling frequency can be 5 Hz / second), thereby obtaining the original time series data of each target process parameter (equivalent to a sequence composed of the target process parameters at each sampling time).

[0022] It should be noted that the number of specific parameter values ​​of the target process parameters included in the above-mentioned raw time series data is related to the preset acquisition frequency. For example, if the preset acquisition frequency is 5 Hz / second, then 5 parameter values ​​of the target process parameters can be acquired within 1 second.

[0023] S102, the original timing data is segmented according to the multiple process segments in the single crystal silicon pulling process to obtain the timing data corresponding to the multiple process segments respectively.

[0024] Here, in the monocrystalline silicon pulling process, many core process parameters will run through multiple process stages. That is, for the original time series data of a target process parameter, if the monocrystalline silicon pulling process has gone through multiple process stages up to the current acquisition time, the original time series data of the target process parameter will include the parameter values ​​corresponding to the target process parameter in the above-mentioned multiple process stages.

[0025] Therefore, considering that the impact of the same target process parameter on oxygen content may differ in different process stages (i.e., there is a time-series correlation between the target process parameter and oxygen content), in order to ensure that the subsequent oxygen content prediction model can output accurate oxygen content prediction results, it is necessary to segment the original time-series data of each target process parameter according to the above-mentioned multiple process stages before inputting the original time-series data into the oxygen content prediction model, so as to ensure that the target process parameter in the same segment of time-series data comes from the same process stage.

[0026] It should be noted that the above-mentioned multiple process stages may include, but are not limited to: the crystal pulling process stage, the shoulder forming process stage, the equal diameter process stage, and the finishing process stage in the monocrystalline silicon pulling process.

[0027] For example, taking a sampling frequency of 5 Hz / second as an example, when the target process parameter a is continuously sampled for n seconds in real time, the original timing data consisting of 5n parameter values ​​can be obtained up to the current sampling time. If m seconds of the n seconds belong to the chip-leading process segment, then the timing data corresponding to the chip-leading process segment can be determined as the timing data consisting of the 5m parameter values ​​sampled first in the original timing data.

[0028] Furthermore, considering that the original time-series data may contain noise, abnormal parameter values, and other interference data, as an optional embodiment, before executing step S102, the oxygen content prediction device can also preprocess the above-mentioned original time-series data according to the method shown in steps a1-a3 below, and use the preprocessed original time-series data as the original time-series data for segmented processing in step S102, so as to ensure that the obtained time-series data format meets the time-series input requirements of the subsequent oxygen content prediction model. Specifically: Step a1, using 3 The criteria are used to remove outliers from the original time series data, and the proportion of the removed outliers in the original time series data is calculated to obtain the data missing rate corresponding to the original time series data.

[0029] Here, 3 The criterion is a statistical outlier detection method, in which, during step a1, the oxygen content prediction device can be based on 3 The criteria are to calculate the mean and standard deviation of each parameter value in the original time series data, and identify parameter values ​​that deviate from the above mean by more than 3 times the standard deviation as outliers that need to be removed.

[0030] Specifically, since outliers are parameter values ​​that need to be removed from the original time series data, the proportion of outliers in the original time series data is also equivalent to the proportion of missing parameter values ​​in the original time series data after removing outliers (i.e., the data missing rate).

[0031] For example, if the original time series data contains 100 parameter values, and there are 3 outliers that need to be removed, then the data missing rate of the original time series data can be determined to be 3%.

[0032] Step a2: In response to the data missing rate being less than or equal to a preset missing rate threshold, the missing values ​​in the original time series data are interpolated and filled in using a linear interpolation method to obtain supplementary time series data.

[0033] Here, the preset missing rate threshold can be flexibly adjusted according to the actual preprocessing requirements. For example, the preset missing rate threshold can be 3% or 4%, etc. The specific value of the preset missing rate threshold is not limited in this application embodiment.

[0034] It should be noted that when interpolating and filling in missing values ​​in the original time series data using linear interpolation, the position of the missing value in the original time series data is determined based on the position of the outlier in the original time series data. For example, if the outlier to be removed from the original time series data is the parameter value corresponding to timestamp t1 (i.e., the sampling time is t1), then when executing step a2, it is necessary to interpolate and fill in the parameter value corresponding to timestamp t1 using linear interpolation to ensure that the supplemented time series data is still time series data with continuous and complete timestamps.

[0035] Step a3: Normalize the supplementary timing data to obtain normalized supplementary timing data, which will then be used as the original timing data for subsequent segmentation according to the multiple process segments.

[0036] Here, when performing step a3, the supplementary time series data can be normalized using the Min-Max normalization algorithm to map all parameter values ​​in the supplementary time series data to the [0,1] interval, thereby eliminating the dimensional differences between different parameters.

[0037] Specifically, taking the parameter value x in the supplementary time series data as an example, it can be normalized according to the following formula 1 to obtain the normalized parameter value x': x' = (x - x_min) / (x_max - x_min) Formula 1; Where x is the parameter value of the target process parameter in the supplementary time series data; x' is the result of normalizing the parameter value x; x_max is the maximum value of the target process parameter in the historically acquired time-series data; x_min is the minimum value of the target process parameter in the historically collected time-series data.

[0038] S103, for each process segment, according to a preset step size time window, the timing data corresponding to the process segment is split into multiple sub-timing data corresponding to each time window.

[0039] Here, in order to obtain time series data whose data format meets the time series input requirements of the subsequent oxygen content prediction model, after performing the above step S102 (that is, ensuring that the target process parameters in the same segment of time series data after segmentation come from the same process segment), it is also necessary to split the time series data corresponding to the same process segment into multiple sub-time series data with equal time steps (i.e., preset step lengths), so that the subsequent oxygen content prediction model can use the time step corresponding to a time window as a unit, and predict the oxygen content within the time step based on the sub-time series data collected within the time step.

[0040] It should be noted that the specific value of the preset step size corresponding to the time window can be flexibly adjusted according to the actual model prediction requirements. For example, the preset step size can be 10 seconds or 15 seconds, etc. This application embodiment does not limit the specific value of the preset step size.

[0041] S104, input the sub-time series data corresponding to the multiple time windows into the pre-trained oxygen content prediction model, and use the oxygen content prediction model to predict the oxygen content corresponding to each sub-time series data, and output the oxygen content prediction result corresponding to the process segment under each time window.

[0042] Here, since the sub-time series data corresponds one-to-one with the time window, the oxygen content prediction result corresponding to a sub-time series data is also equivalent to the oxygen content prediction result corresponding to the process segment in a time window (i.e., the time window corresponding to the sub-time series data). Thus, in the monocrystalline silicon pulling process, the oxygen content prediction result of each process segment in each time window can be output in real time through the above oxygen content prediction model.

[0043] It should be noted that the specific model structure of the above-mentioned oxygen content prediction model can be an LSTM (Long Short-Term Memory) model or other prediction models capable of processing time series data. The embodiments of this application do not impose mandatory limitations on the specific model structure of the above-mentioned oxygen content prediction model.

[0044] Specifically, during the model training phase, the oxygen content prediction device can be trained to obtain the aforementioned oxygen content prediction model using the methods shown in steps b1-b7: Step b1: Collect the parameter values ​​of the target process parameters at each historical sampling time from the historical single-crystal silicon pulling process to obtain the historical time-series data of the target process parameters.

[0045] Here, the method of collecting historical time series data in step b1 is the same as the method of collecting raw time series data in step S101 above, and the repetition will not be repeated here.

[0046] It should be noted that, considering that the number of time series samples representing abnormal oxygen content in historical time series data may be relatively small, in order to solve the problem of data imbalance in model training, oversampling can be used for data augmentation on the relatively small number of time series samples (such as historical time series data under abnormal oxygen content). At the same time, a slight shift on the time axis of the model training data is performed, which is beneficial to improving the generalization ability of the original prediction model (i.e., the oxygen content prediction model that has not yet completed training).

[0047] Step b2: Divide the historical time series data into segments according to the multiple process segments to obtain sample time series data corresponding to the multiple process segments respectively.

[0048] Here, the segmentation method for historical time series data in step b2 is the same as the specific segmentation method in step S102 above, and the repetition will not be repeated here.

[0049] Step b3: Using offline detection, determine the oxygen content detection results of the historical single crystal silicon under the various process stages from the historical single crystal silicon that has been pulled.

[0050] Here, after the historical single-crystal silicon is pulled, a sample test can be performed on the historical single-crystal silicon. The growth time of the historical single-crystal silicon is different for different process stages. Therefore, for each process stage, a test sample matching the process stage needs to be extracted from the historical single-crystal silicon according to the sampling point location that matches the process stage.

[0051] Specifically, after extracting a test sample that matches the process stage, the intensity of infrared light passing through the test sample can be measured using a Fourier transform infrared spectrometer to obtain the infrared absorption spectrum. Based on the relationship between the peak value representing the absorption intensity in the infrared absorption spectrum and the oxygen content, the oxygen content detection results corresponding to the historical single-crystal silicon in this process stage can be obtained.

[0052] Step b4: For each process segment, according to the time window, the sample time series data corresponding to the process segment is split into multiple sub-sample time series data corresponding to each time window.

[0053] Here, the method of splitting the sample time series data in step b4 is the same as the specific splitting method in step S103 above, and the repetition will not be repeated here.

[0054] Step b5: Based on the oxygen content detection results corresponding to the process segment, determine the actual oxygen content corresponding to the sub-sample time series data.

[0055] Here, for each process segment, since the oxygen content detection result for that process segment is obtained through offline detection, the oxygen content detection result can be used as the actual oxygen content corresponding to the model training data (i.e., the time series data of multiple sub-samples split out under that process segment) for that process segment.

[0056] Step b6: Input the subsample time series data into the original prediction model, predict the oxygen content corresponding to the subsample time series data through the original prediction model, and output the oxygen content prediction result corresponding to the subsample time series data.

[0057] Here, the original prediction model is the oxygen content prediction model that has not yet been trained. The specific model structure of the original prediction model can be an LSTM (Long Short-Term Memory) model or other prediction models that can process time series data. This application does not impose a mandatory limitation on the specific model structure of the original prediction model.

[0058] Step b7: Based on the prediction loss between the oxygen content prediction result corresponding to the subsample time series data and the actual oxygen content corresponding to the subsample time series data, train the original prediction model until the original prediction model converges, and use the converged original prediction model as the trained oxygen content prediction model.

[0059] Here, when calculating the prediction loss between the actual oxygen content (i.e. the actual oxygen content mentioned above) of the same set of subsample time series data and the oxygen content prediction result output by the model, the mean squared error loss function can be used, or other types of loss functions can be used. The specific function of the loss function can be flexibly adjusted according to the actual model training requirements.

[0060] It should be noted that after the model training is completed, the trained oxygen content prediction model can be tested by selecting historical time series data from the database that is different from the model training data (i.e., the historical time series data in step b1 above) in the same way as the model training method shown in steps b1-b7 above. The repeated parts will not be described again here.

[0061] Specifically, by inputting the model test set into the trained oxygen content prediction model, the oxygen content prediction results output by the model for the model test set can be obtained. At this time, based on the oxygen content prediction results corresponding to the model test set and the actual oxygen content in the model test set, multiple evaluation indicators can be calculated to evaluate the model performance of the oxygen content prediction model, thereby evaluating the model performance of the oxygen content prediction model.

[0062] It should be noted that the above evaluation indicators may include, but are not limited to, mean squared error, mean absolute error, etc. The specific selection of the above evaluation indicators is not limited in the embodiments of this application.

[0063] After performing step S104 above, the oxygen content prediction device can also determine the specific oxygen content level to which each oxygen content prediction result belongs from a set of preset oxygen content levels, as shown in steps c1-c2 below. This allows relevant control personnel to determine whether there is an abnormality in the oxygen content during the current drawing process based on the oxygen content level to which the oxygen content prediction result belongs. Specifically: Step c1: For each oxygen content prediction result, determine the target preset value range to which the oxygen content prediction result belongs from the preset value ranges corresponding to multiple oxygen content levels.

[0064] Specifically, three oxygen content levels can be preset: high oxygen content, normal oxygen content, and low oxygen content. The preset value range for each oxygen content level can be determined based on the actual oxygen content control requirements.

[0065] For example, if the oxygen content is >1.0× atoms / cm³ indicates a high oxygen content. Therefore, the preset value range for this oxygen content level, "high oxygen content," can be set to "oxygen content > 1.0 × [missing value]". atoms / cm³.

[0066] Step c2: Determine the oxygen content level corresponding to the preset value range of the target as the level judgment result corresponding to the oxygen content prediction result.

[0067] For example, the preset value range corresponding to the oxygen content level of "high oxygen content" is "oxygen content > 1.0 × Taking "atoms / cm³" as an example, if the oxygen content prediction model outputs an oxygen content prediction result of 1.1× If the oxygen content is calculated as atoms / cm³, then the target preset value range for the predicted oxygen content is determined to be "oxygen content > 1.0 × The oxygen content prediction result of "atoms / cm³" corresponds to the grade judgment of "oxygen content is too high".

[0068] After performing step S104 above, based on the oxygen content prediction results output by the model, the oxygen content prediction device can also automatically adjust the target process parameters in the monocrystalline silicon pulling process through the automatic adjustment methods shown in steps d1-d3 below, to ensure that the oxygen content in the pulling process is always within the normal range. Specifically: Step d1: Obtain the parameter prompt information associated with the target process parameter within the process segment from the preset process parameter adjustment rule library.

[0069] Here, a process parameter adjustment rule library can be built into the oxygen content prediction device. The process parameter adjustment rule library records multiple parameter prompts associated with the target process parameter in each process segment.

[0070] Specifically, the aforementioned parameter prompts may include first parameter prompts and second parameter prompts; wherein, the first parameter prompt indicates the parameter adjustment prompt when the target process parameter value is too high (i.e., exceeds the normal range of the target process parameter value in that process segment) within a process segment, and the second parameter prompt indicates the parameter adjustment prompt when the target process parameter value is too low (i.e., below the normal range of the target process parameter value in that process segment) within a process segment.

[0071] Step d2: For each oxygen content prediction result, generate a parameter adjustment instruction for the target process parameter based on the grade determination result corresponding to the oxygen content prediction result and the parameter prompt information.

[0072] Here, when executing step d2, it can be determined whether the above parameter adjustment instruction needs to be generated based on the level determination result corresponding to the oxygen content prediction result; wherein, when the level determination result indicates that the oxygen content prediction result is abnormal (e.g., the level determination result indicates that the oxygen content prediction result belongs to the oxygen content high level or the oxygen content low level), it is determined that the above parameter adjustment instruction needs to be generated; when the level determination result indicates that the oxygen content prediction result is not abnormal (e.g., the level determination result indicates that the oxygen content prediction result belongs to the oxygen content normal level), it is determined that the above parameter adjustment instruction does not need to be generated.

[0073] Specifically, when it is determined that the above parameter adjustment instructions need to be generated, the specific instruction content of the generated parameter adjustment instructions can be determined based on the above parameter prompt information associated with the target process parameters within the process segment.

[0074] Step d3: Adjust the parameter values ​​of the target process parameters within this process segment according to the parameter adjustment instructions.

[0075] For example, if the target process parameter is argon flow rate, and the oxygen content prediction result is in the high oxygen content category and the above parameter prompt information is the corresponding parameter adjustment prompt information "increase argon flow rate by 5-8 L / min" when argon flow rate is low, then the parameter adjustment instruction "increase argon flow rate by 5-8 L / min" can be generated, and the parameter value of argon flow rate can be adjusted according to this parameter adjustment instruction within this process segment.

[0076] For example, if the target process parameter is the lifting rate, and the oxygen content prediction result is in the high oxygen content category, and the above parameter prompt information is the parameter adjustment prompt information "reduce the lifting rate by 0.05-0.1 mm / min" when the lifting rate is high, then the parameter adjustment instruction "reduce the lifting rate by 0.05-0.1 mm / min" can be generated, and the parameter value of the lifting rate can be adjusted according to the parameter adjustment instruction within this process segment.

[0077] After performing step S104 above, the oxygen content prediction device can further display the time-series variation trend of the target process parameters and the oxygen content prediction results output by the model on the graphical user interface (i.e., the display screen on one side of the oxygen content prediction device) using the methods shown in steps e1-e2 below. This allows relevant control personnel to refer to the specific information displayed on the graphical user interface and manually adjust the target process parameters according to the actual production conditions. Specifically: Step e1: Based on the original time series data of the target process parameters, determine the parameter time series change trend of the target process parameters over time.

[0078] Here, referring to the relevant explanation of the original time series data in step S101 above, it can be seen that since the original time series data is essentially time series data composed of the parameter values ​​of the target process parameters at each sampling time, by using the sampling time as the horizontal axis representing the time dimension and the parameter values ​​of the target process parameters as the vertical axis, a curve can be drawn to obtain the visualization data corresponding to the time series change trend of the above parameters.

[0079] Step e2: Display the time-series change trend of the parameters and each oxygen content prediction result output by the oxygen content prediction model on the graphical user interface.

[0080] Specifically, the graphical user interface can display the curves corresponding to the time-series changes of the above parameters, and based on the sampling time on the curve, display the oxygen content prediction results that match the sampling time at each sampling time (i.e., the oxygen content prediction results corresponding to the time window to which the sampling time belongs).

[0081] Here, regarding the aforementioned oxygen content prediction model, as an optional embodiment, the oxygen content prediction device can also monitor the online performance of the oxygen content prediction model through the methods shown in steps f1-f4 below. This allows for timely fine-tuning of the model parameters when the online performance of the oxygen content prediction model falls short of expectations, ensuring the accuracy of the oxygen content prediction results output by the model. Specifically: Step f1: In response to the end of the single-crystal silicon pulling process, the actual oxygen content of the single-crystal silicon at the target growth moment in the pulling process is determined from the pulled single-crystal silicon through offline detection.

[0082] Here, the specific implementation method of step f1 is the same as the offline detection method in step b3 above, and the repetitions will not be repeated here.

[0083] It should be noted that the target growth time is used to represent the growth time corresponding to any process segment in the drawing process.

[0084] Step f2: From all the oxygen content prediction results output by the oxygen content prediction model, determine the target oxygen content prediction result whose timestamp corresponding to the time window matches the target growth time.

[0085] Specifically, when executing step f2, since each oxygen content prediction result corresponds to a time interval (i.e., a time window) in the monocrystalline silicon pulling process, the target oxygen content prediction result whose timestamp matches the target growth time can be determined from all the oxygen content prediction results output by the oxygen content prediction model based on the target growth time mentioned above.

[0086] Step f3: Determine whether the error between the actual oxygen content and the predicted target oxygen content exceeds a preset error threshold.

[0087] Here, the specific value of the error threshold can be flexibly adjusted according to the actual model evaluation requirements. This application does not impose a mandatory limitation on the specific value of the error threshold.

[0088] Specifically, when the error between the actual oxygen content and the predicted target oxygen content output by the model is greater than the error threshold, it can be determined that the online performance of the oxygen content prediction model is not up to expectations, and the model parameters of the oxygen content prediction model need to be fine-tuned.

[0089] Step f4: If it is determined that the error exceeds the error threshold, then new model training data is constructed based on the original time series data and the actual oxygen content, and the model parameters of the oxygen content prediction model are fine-tuned based on the new model training data to obtain a new version of the oxygen content prediction model with updated model parameters.

[0090] Here, the specific method for fine-tuning the model parameters of the oxygen content prediction model is similar to the method for adjusting the model parameters in the aforementioned model training phase, and the repetition will not be repeated here.

[0091] Based on the oxygen content prediction method provided in the embodiments of this application, the time-series data of the target process parameters that are temporally correlated with oxygen content can be continuously collected in real time during the monocrystalline silicon pulling process. The model input data of the oxygen content prediction model can be constructed based on the collected time-series data, and the oxygen content prediction result can be output in real time through the pre-trained oxygen content prediction model. This not only overcomes the technical defect of the prior art that oxygen content can only be passively detected after the monocrystalline silicon pulling process is completed, but also effectively reduces the dependence on manual operation in offline detection methods, which is conducive to reducing manual operation costs and improving oxygen content detection efficiency.

[0092] Based on the same inventive concept, this application also provides an oxygen content prediction device corresponding to the above-mentioned oxygen content prediction method. Since the principle of solving the problem by the oxygen content prediction device in the embodiments of this application is similar to that of the above-mentioned oxygen content prediction method in the embodiments of this application, the implementation of the oxygen content prediction device can refer to the implementation of the above-mentioned oxygen content prediction method, and the repeated parts will not be described again.

[0093] Reference Figure 2 As shown, Figure 2 This illustration shows a schematic diagram of an oxygen content prediction device applied in the single-crystal silicon pulling process, provided by an embodiment of this application. The oxygen content prediction device includes: The parameter acquisition module 201 is used to acquire the parameter values ​​of the target process parameters at each sampling time in real time during the monocrystalline silicon pulling process, and obtain the original time-series data of the target process parameters; wherein, the target process parameters are process parameters in the monocrystalline silicon pulling process that have a time-series correlation with oxygen content; The first processing module 202 is used to segment the original timing data according to multiple process segments in the single crystal silicon pulling process, so as to obtain timing data corresponding to the multiple process segments respectively. The second processing module 203 is used to split the timing data corresponding to each process segment into sub-timing data corresponding to multiple time windows according to a preset step size time window for each process segment. The prediction module 204 is used to input the sub-time series data corresponding to the multiple time windows into the pre-trained oxygen content prediction model, and predict the oxygen content corresponding to each sub-time series data through the oxygen content prediction model, and output the oxygen content prediction result corresponding to the process segment under each time window.

[0094] In an optional implementation, before segmenting the original timing data according to multiple process stages in the monocrystalline silicon pulling process, the first processing module 202 is further configured to: Use 3 The criteria are used to remove outliers from the original time series data, and the proportion of the removed outliers in the original time series data is calculated to obtain the data missing rate corresponding to the original time series data. In response to the data missing rate being less than or equal to a preset missing rate threshold, missing values ​​in the original time series data are interpolated and imputed using linear interpolation to obtain supplementary time series data; wherein, the position of the missing value in the original time series data is determined based on the position of the outlier in the original time series data; The supplementary timing data is normalized to obtain normalized supplementary timing data, which is then used as the original timing data for subsequent segmentation according to the multiple process segments.

[0095] In an optional implementation, the prediction module 204 is used to train the oxygen content prediction model using the following method: From the historical single-crystal silicon pulling process, the parameter values ​​of the target process parameters at each historical sampling time are collected to obtain the historical time-series data of the target process parameters; The historical time series data is segmented according to the multiple process segments to obtain sample time series data corresponding to the multiple process segments respectively; By using offline detection, the oxygen content detection results of the historical single crystal silicon under the various process stages were determined from the historical single crystal silicon that had been pulled. For each process segment, according to the time window, the sample time series data corresponding to that process segment is split into multiple sub-sample time series data corresponding to each time window; Based on the oxygen content detection results corresponding to the process segment, the actual oxygen content corresponding to the sub-sample time series data is determined; The subsample time series data is input into the original prediction model, and the oxygen content corresponding to the subsample time series data is predicted by the original prediction model, and the oxygen content prediction result corresponding to the subsample time series data is output. Based on the prediction loss between the predicted oxygen content corresponding to the subsample time series data and the actual oxygen content corresponding to the subsample time series data, the original prediction model is trained until the original prediction model converges, and the converged original prediction model is used as the trained oxygen content prediction model.

[0096] In an optional embodiment, the oxygen content prediction device further includes: a level determination module, wherein the level determination module is used for: For each oxygen content prediction result, the target preset value range to which the oxygen content prediction result belongs is determined from the preset value ranges corresponding to multiple oxygen content levels. The oxygen content level corresponding to the preset value range of the target is determined as the level judgment result corresponding to the oxygen content prediction result.

[0097] In an optional embodiment, the oxygen content prediction device further includes: a first adjustment module, wherein the first adjustment module is used for: Obtain parameter prompt information associated with the target process parameter within the process segment from the preset process parameter adjustment rule library; For each oxygen content prediction result, a parameter adjustment instruction for the target process parameter is generated based on the grade determination result corresponding to the oxygen content prediction result and the parameter prompt information. According to the parameter adjustment instructions, the parameter values ​​of the target process parameters are adjusted within this process segment.

[0098] In an optional embodiment, the oxygen content prediction device further includes: a second adjustment module, wherein the second adjustment module is used for: Based on the original time-series data of the target process parameters, determine the time-series change trend of the target process parameters over time; The time-series variation trend of the parameters and each oxygen content prediction result output by the oxygen content prediction model are displayed on the graphical user interface.

[0099] In an optional embodiment, the oxygen content prediction device further includes: a model optimization module, wherein the model optimization module is used for: In response to the end of the monocrystalline silicon pulling process, the actual oxygen content of the monocrystalline silicon at the target growth moment in the pulling process is determined from the pulled monocrystalline silicon through offline detection. From all the oxygen content prediction results output by the oxygen content prediction model, determine the target oxygen content prediction result that matches the timestamp of the time window and the target growth time. Determine whether the error between the actual oxygen content and the predicted target oxygen content exceeds a preset error threshold; If it is determined that the error exceeds the error threshold, new model training data is constructed based on the original time-series data and the actual oxygen content, and the model parameters of the oxygen content prediction model are fine-tuned based on the new model training data to obtain a new version of the oxygen content prediction model with updated model parameters.

[0100] like Figure 3 As shown, this application embodiment also provides an electronic device 300 for executing the oxygen content prediction method in this application (the electronic device 300 is also equivalent to the aforementioned oxygen content prediction device). The electronic device includes a memory 301, a processor 302, and a computer program stored in the memory 301 and executable on the processor 302. The memory 301 and the processor 302 are connected via a bus for communication. When the processor 302 executes the computer program, it implements the steps of the aforementioned oxygen content prediction method.

[0101] Specifically, the memory 301 and processor 302 mentioned above can be general-purpose memory and processor, without any specific limitations. When the processor 302 runs the computer program stored in the memory 301, it can execute the oxygen content prediction method mentioned above.

[0102] Corresponding to the oxygen content prediction method in this application, this application embodiment also provides a computer-readable storage medium storing a computer program, which is executed by a processor to perform the steps of the oxygen content prediction method described above.

[0103] Specifically, the storage medium can be a general-purpose storage medium, such as a removable disk or hard disk, and the computer program on the storage medium, when run, can execute the oxygen content prediction method described above.

[0104] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0105] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0106] In addition, the functional units in the embodiments provided in this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0107] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0109] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application. All should be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A method for predicting oxygen content in the single-crystal silicon pulling process, characterized in that, The oxygen content prediction method includes: From the monocrystalline silicon pulling process, the parameter values ​​of the target process parameters are collected in real time at each sampling moment to obtain the original time-series data of the target process parameters; wherein, the target process parameters are process parameters in the monocrystalline silicon pulling process that have a time-series correlation with oxygen content; The original timing data is segmented according to the multiple process segments in the monocrystalline silicon pulling process to obtain the timing data corresponding to each of the multiple process segments. For each process segment, the timing data corresponding to that process segment is divided into multiple sub-timing data corresponding to each time window according to a preset step size time window; The sub-time series data corresponding to the multiple time windows are input into the pre-trained oxygen content prediction model. The oxygen content prediction model predicts the oxygen content corresponding to each sub-time series data and outputs the oxygen content prediction result of the process segment under each time window.

2. The oxygen content prediction method according to claim 1, characterized in that, Before segmenting the original time-series data according to multiple process stages in the monocrystalline silicon pulling process, the oxygen content prediction method further includes: Use 3 The criteria are used to remove outliers from the original time series data, and the proportion of the removed outliers in the original time series data is calculated to obtain the data missing rate corresponding to the original time series data. In response to the data missing rate being less than or equal to a preset missing rate threshold, missing values ​​in the original time series data are interpolated and imputed using linear interpolation to obtain supplementary time series data; wherein, the position of the missing value in the original time series data is determined based on the position of the outlier in the original time series data; The supplementary timing data is normalized to obtain normalized supplementary timing data, which is then used as the original timing data for subsequent segmentation according to the multiple process segments.

3. The oxygen content prediction method according to claim 1, characterized in that, The oxygen content prediction model was trained using the following method: From the historical single-crystal silicon pulling process, the parameter values ​​of the target process parameters at each historical sampling time are collected to obtain the historical time-series data of the target process parameters; The historical time series data is segmented according to the multiple process segments to obtain sample time series data corresponding to the multiple process segments respectively; By using offline detection, the oxygen content detection results of the historical single crystal silicon under the various process stages were determined from the historical single crystal silicon that had been pulled. For each process segment, according to the time window, the sample time series data corresponding to that process segment is split into multiple sub-sample time series data corresponding to each time window; Based on the oxygen content detection results corresponding to the process segment, the actual oxygen content corresponding to the sub-sample time series data is determined; The subsample time series data is input into the original prediction model, and the oxygen content corresponding to the subsample time series data is predicted by the original prediction model, and the oxygen content prediction result corresponding to the subsample time series data is output. Based on the prediction loss between the predicted oxygen content corresponding to the subsample time series data and the actual oxygen content corresponding to the subsample time series data, the original prediction model is trained until the original prediction model converges, and the converged original prediction model is used as the trained oxygen content prediction model.

4. The oxygen content prediction method according to claim 1, characterized in that, The oxygen content prediction method also includes: For each oxygen content prediction result, the target preset value range to which the oxygen content prediction result belongs is determined from the preset value ranges corresponding to multiple oxygen content levels. The oxygen content level corresponding to the preset value range of the target is determined as the level judgment result corresponding to the oxygen content prediction result.

5. The oxygen content prediction method according to claim 4, characterized in that, The oxygen content prediction method also includes: Obtain parameter prompt information associated with the target process parameter within the process segment from the preset process parameter adjustment rule library; For each oxygen content prediction result, a parameter adjustment instruction for the target process parameter is generated based on the grade determination result corresponding to the oxygen content prediction result and the parameter prompt information. According to the parameter adjustment instructions, the parameter values ​​of the target process parameters are adjusted within this process segment.

6. The oxygen content prediction method according to claim 1, characterized in that, The oxygen content prediction method also includes: Based on the original time-series data of the target process parameters, determine the time-series change trend of the target process parameters over time; The time-series variation trend of the parameters and each oxygen content prediction result output by the oxygen content prediction model are displayed on the graphical user interface.

7. The oxygen content prediction method according to claim 1, characterized in that, The oxygen content prediction method also includes: In response to the end of the monocrystalline silicon pulling process, the actual oxygen content of the monocrystalline silicon at the target growth moment in the pulling process is determined from the pulled monocrystalline silicon through offline detection. From all the oxygen content prediction results output by the oxygen content prediction model, determine the target oxygen content prediction result that matches the timestamp of the time window and the target growth time. Determine whether the error between the actual oxygen content and the predicted target oxygen content exceeds a preset error threshold; If it is determined that the error exceeds the error threshold, new model training data is constructed based on the original time-series data and the actual oxygen content, and the model parameters of the oxygen content prediction model are fine-tuned based on the new model training data to obtain a new version of the oxygen content prediction model with updated model parameters.

8. An oxygen content prediction device applied in the single-crystal silicon pulling process, characterized in that, The oxygen content prediction device includes: The parameter acquisition module is used to acquire the parameter values ​​of the target process parameters at each sampling time in real time during the monocrystalline silicon pulling process, and obtain the raw time-series data of the target process parameters; wherein, the target process parameters are process parameters in the monocrystalline silicon pulling process that have a time-series correlation with oxygen content; The first processing module is used to segment the original timing data according to multiple process segments in the single crystal silicon pulling process, so as to obtain the timing data corresponding to each of the multiple process segments. The second processing module is used to split the timing data corresponding to each process segment into sub-timing data corresponding to multiple time windows according to a preset step size time window for each process segment. The prediction module is used to input the sub-time series data corresponding to the multiple time windows into a pre-trained oxygen content prediction model, and to predict the oxygen content corresponding to each sub-time series data through the oxygen content prediction model, and output the oxygen content prediction result corresponding to the process segment under each time window.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of the oxygen content prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the oxygen content prediction method as described in any one of claims 1 to 7.

Images