Training Method for Prediction Model of Wafer Grinding Processing Removal Rate

Abstract

The present application provides a training method for a prediction model of the removal rate in wafer grinding processing, which relates to the field of wafer grinding processing. The method obtains the parameter characteristics of multiple batches of wafer grinding processing; obtains the service life of the auxiliary materials for multiple batches of wafer grinding processing, and the auxiliary materials at least include a grinding pad; the grinding pad is divided into multiple life cycle stages based on the service life, and a performance degradation model is established for each life cycle stage; for each batch of wafer grinding processing, based on the service life of the grinding pad and the performance degradation model in this batch of wafer grinding processing, determine the performance degradation characteristics of this batch of wafer grinding processing, and each batch of wafer grinding processing includes at least one performance degradation characteristic; merge the performance degradation characteristics with the parameter characteristics to construct an enhanced data set; train a machine learning model through the enhanced data set to obtain a removal rate prediction model. The prediction model obtained by this training method can more accurately predict the removal rate in wafer grinding processing.

Description

Technical Field

[0001] The present application relates to the field of wafer grinding and processing, and particularly to a method for training a prediction model of the removal rate in wafer grinding and processing. Background Art

[0002] In the process of wafer production in the semiconductor and photovoltaic industries, the grinding process is a key link. The main purpose is to remove the saw marks and processed damaged layers generated by cutting, so that the wafer surface reaches the required flatness. With the increase in wafer size and the improvement of precision requirements, it is crucial to accurately control the removal rate during the grinding process.

[0003] At present, there are certain limitations in the removal rate prediction methods in the related art. On the one hand, it is difficult for the mechanism-based modeling method to comprehensively consider the complex interactions of various factors during the grinding process, resulting in low model prediction accuracy. On the other hand, although the data-driven statistical modeling method can use historical data for learning and prediction, most of the existing methods are for the polishing process, and there are significant differences in parameters and auxiliary material characteristics from the grinding process, and the prediction accuracy is not ideal when directly applied to the grinding process. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a method for training a prediction model of the removal rate in wafer grinding and processing, and the prediction model obtained by this training method can accurately predict the removal rate in wafer grinding and processing.

[0005] In a first aspect, the present application provides a method for training a prediction model of the removal rate in wafer grinding and processing, and the method includes:

[0006] Obtain the parameter features of multiple batches of wafer grinding and processing; wherein, the parameter features of each batch are extracted from the process parameters during the wafer grinding and processing of this batch, and the parameter features of each batch are multiple and at least include the removal rate;

[0007] Obtain the service life of the auxiliary materials for multiple batches of wafer grinding and processing, and the auxiliary materials at least include the grinding pad; wherein, the grinding pad includes multiple life cycle stages divided based on its own service life, and a performance degradation model is established for each life cycle stage; wherein, the performance degradation model is a functional relationship between the removal rate and the service life;

[0008] For each batch of wafer grinding and processing, determine the performance degradation features of this batch of wafer grinding and processing based on the service life of the grinding pad and the performance degradation model in this batch of wafer grinding and processing, and each batch of wafer grinding and processing includes at least one performance degradation feature;

[0009] Combine the performance degradation features of multiple batches with the parameter features of multiple batches to construct an enhanced data set;

[0010] By strengthening the training of the machine learning model with a dataset, a removal rate prediction model is obtained.

[0011] In one embodiment, multiple life cycle stages are defined for the polishing pad, including:

[0012] Based on the service life of the polishing pad, three life cycle stages are defined for the polishing pad, namely the first cycle stage, the second cycle stage, and the third cycle stage. The polishing pad in the first cycle stage has not been worn, while the polishing pads in the second and third cycle stages have been worn.

[0013] Among them, the first cycle stage is associated with a first service life range, the second cycle stage is associated with a second cumulative service life range, and the third cycle stage is associated with a third cumulative service life range; the maximum value of the first cumulative service life range is less than or equal to the minimum value of the second cumulative service life range, and the maximum value of the second cumulative service life range is less than or equal to the minimum value of the third cumulative service life range.

[0014] In one embodiment, a performance degradation model is established for each life cycle stage, including:

[0015] A first degradation model is established for the first cycle stage, and the relationship between the removal rate and the first cumulative service life of the polishing pad in the first degradation model is non-linear.

[0016] A second degradation model is established for the second cycle stage, and the exponential decay function is used to express the relationship between the removal rate and the second cumulative service life of the polishing pad in the second degradation model.

[0017] A third degradation model is established for the third cycle stage, and the linear function relationship is used to express the relationship that the removal rate decreases as the third cumulative service life of the polishing pad increases.

[0018] In one embodiment, for each batch of wafer polishing processes, based on the service life of the polishing pad and the performance degradation model in this batch of wafer polishing processes, the performance degradation characteristics of this batch of wafer polishing processes are determined, including:

[0019] For each batch of wafer polishing processes, based on the service life of the polishing pad in this batch of wafer polishing processes, the life cycle stage passed by the polishing pad is determined, which is defined as the life cycle stage set, and the life cycle stage set includes at least one life cycle stage.

[0020] For any life cycle stage in the life cycle stage set, calculate the integral of the removal rate of the life cycle stage with respect to the service life.

[0021] After adding up the integrals of each life cycle stage in the life cycle stage set and dividing by the cumulative service life, the performance degradation characteristics of this batch of wafer polishing processes are obtained.

[0022] In one embodiment, the method further includes:

[0023] Obtaining equipment marks and processing technologies for grinding and processing wafers in multiple batches;

[0024] For the grinding and processing of wafers in each batch, determining first-category features based on the equipment marks of this batch, and determining second-category features based on the processing technology of this batch;

[0025] Combining the first-category features and second-category features of multiple batches into an enhanced data set.

[0026] In one embodiment, for the grinding and processing of wafers in each batch, determining first-category features based on the equipment marks of this batch, and determining second-category features based on the processing technology of this batch, includes:

[0027] Using label encoding to map equipment identifiers to unique integers to obtain first-category features, and using one-hot encoding to convert the processing technology into a binary vector to obtain first-category features. The binary vector can cover different wafer types and sizes.

[0028] In one embodiment, the machine learning model is an XGBoost regression model, and the hyperparameters of the XGBoost regression model are optimized through a combination of grid search algorithm and genetic algorithm; wherein, optimizing the hyperparameters of the XGBoost regression model through a combination of grid search and genetic algorithm includes:

[0029] Searching the pre-defined hyperparameter space based on the grid search algorithm to obtain a benchmark hyperparameter combination;

[0030] Taking the benchmark hyperparameter combination as the center point, determining a preselected hyperparameter range, and determining an initial parameter set based on the preselected hyperparameter range;

[0031] Selecting the optimal hyperparameter combination from the initial parameter set through the selection, crossover, and mutation operations of the genetic algorithm.

[0032] In a second aspect, an embodiment of the present application provides a method for predicting the removal rate of wafer grinding and processing. The prediction method includes: obtaining the instantaneous process parameters during the current batch of wafer grinding and processing and the instantaneous service life of the grinding pad, inputting them into the removal rate prediction model, and obtaining the predicted removal rate of wafer grinding and processing; wherein, the removal rate prediction model is obtained through the training method of the prediction model for the removal rate of wafer grinding and processing.

[0033] In a third aspect, the present application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the training method of the prediction model for the wafer grinding processing removal rate in the first aspect or the method for predicting the wafer grinding processing removal rate in the second aspect.

[0034] In a fourth aspect, the present application provides a training device for a prediction model of wafer grinding processing removal rate, and the device includes:

[0035] A first acquisition module, configured to acquire parameter features of multiple batches of wafer grinding processing; wherein, the parameter features of each batch are extracted from the process parameters during the wafer grinding processing of this batch, and there are multiple parameter features for each batch and at least include the removal rate;

[0036] A second acquisition module, configured to acquire the service life of auxiliary materials for multiple batches of wafer grinding processing, and the auxiliary materials at least include a grinding pad; wherein, the grinding pad includes multiple life cycle stages divided based on its own service life, and a performance degradation model is established for each life cycle stage; wherein, the performance degradation model is a functional relationship between the removal rate and the service life;

[0037] A feature extraction module, configured to, for each batch of wafer grinding processing, determine the performance degradation features of this batch of wafer grinding processing based on the service life of the grinding pad and the performance degradation model in this batch of wafer grinding processing, and each batch of wafer grinding processing includes at least one performance degradation feature;

[0038] A training module, configured to combine the performance degradation features of multiple batches with the parameter features of multiple batches to construct an enhanced data set; the training module is further configured to train a machine learning model through the enhanced data set to obtain a removal rate prediction model.

[0039] The above-mentioned training method of the prediction model for the wafer grinding processing removal rate obtains the parameter features and the service life of auxiliary materials of multiple batches of wafer grinding processing, and focuses on multiple life cycle stages of the grinding pad and its performance degradation model. For each batch, the performance degradation features are determined by combining the service life of the grinding pad and the performance degradation model. These features are combined with the parameter features to construct an enhanced data set, and a machine learning model is trained through this data set to obtain a removal rate prediction model. The prediction model obtained by this training method can more accurately predict the removal rate in wafer grinding processing. Description of the Drawings

[0040] Figure 1 It is a flowchart of the training method of the prediction model for the wafer grinding processing removal rate in an embodiment;

[0041] Figure 2Flow chart for establishing a performance degradation model for each life cycle stage in an embodiment;

[0042] Figure 3 Flow chart for determining the performance degradation characteristics of wafer grinding and processing for this batch in an embodiment;

[0043] Figure 4 Flow chart for determining the first category of features and the second category of features and merging them into an enhanced data set in an embodiment;

[0044] Figure 5 Flow chart for optimizing the hyperparameters of the XGBoost regression model through a combination of grid search algorithm and genetic algorithm in an embodiment;

[0045] Figure 6 Structural diagram of an electronic device in an embodiment;

[0046] Figure 7 Diagram of a training device for a prediction model of the removal rate of wafer grinding and processing in another embodiment. Detailed implementation

[0047] To make the objectives, technical solutions and advantages of this application clearer and more understandable, the following further details this application in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0048] In one embodiment, as Figure 1 shown, a method for training a prediction model of the removal rate of wafer grinding and processing is provided, and the method includes the following steps:

[0049] Step 101: Obtain the parameter features of multiple batches of wafer grinding and processing; among them, the parameter features of each batch are extracted from the process parameters during the wafer grinding and processing of this batch, and the parameter features of each batch are multiple and at least include the removal rate.

[0050] The parameter features of each batch at least include the removal rate. Among them, the removal rate of each batch can be obtained by measuring the thickness difference of the wafer before and after grinding, obtaining the average removal amount, and dividing it by the grinding time, so as to obtain the average removal rate of the wafer for each batch.

[0051] Exemplarily, assuming that the number of wafers in a batch is m, the thickness of the wafer before grinding is:

[0052] .

[0053] The thickness of the wafer after grinding is:

[0054] .

[0055] The average removal amount of this batch is . Further, the average removal rate of this batch is obtained as . Where t is the grinding time.

[0056] Where m is the number of wafers in a batch. thk_pre is the thickness of the wafer before grinding, which is an array containing the thickness of each wafer, in the form of . thk_after is the thickness of the wafer after grinding, which is also an array corresponding to the thickness of each wafer after grinding, in the form of . MR is the average removal amount of this batch. MRR is the average removal rate of this batch.

[0057] During the wafer grinding process for each batch, there can also be multiple process parameters, such as pressure, grinding fluid flow rate, platen temperature, upper platen current, and lower platen current, etc., and statistical characteristics such as the mean and standard deviation of each process parameter are calculated.

[0058] Step 102: Obtain the service life of the auxiliary materials for the wafer grinding process of multiple batches. The auxiliary materials at least include a grinding pad; wherein, the grinding pad includes multiple life cycle stages divided based on its own service life, and a performance degradation model is established for each life cycle stage; wherein, the performance degradation model is a functional relationship between the removal rate and the service life.

[0059] During the wafer grinding process, in order to accurately predict the removal rate and optimize the processing technology, it is necessary to obtain the service life of the auxiliary materials for the wafer grinding process of multiple batches, where the auxiliary materials at least include a grinding pad.

[0060] It should be noted that a corresponding performance degradation model can be established for each life cycle stage. Different performance degradation models can describe the functional relationship between the removal rate and the service life. Exemplarily, in the initial use stage of the grinding pad, the performance of the grinding pad is relatively new and the removal rate is relatively high. As the use time increases, the grinding pad enters the use stage, the performance gradually stabilizes, and the change in the removal rate is relatively slow. When the grinding pad enters the stage of excessive wear, the performance significantly decreases, and the removal rate can gradually decrease over time.

[0061] Step 103: For the wafer grinding process of each batch, determine the performance degradation characteristics of the wafer grinding process of this batch based on the service life of the grinding pad and the performance degradation model in the wafer grinding process of this batch. The wafer grinding process of each batch includes at least one performance degradation characteristic.

[0062] The service life of a polishing pad is divided into multiple life cycle stages, and each stage corresponds to a performance degradation model. Different performance degradation models can describe the functional relationship between the removal rate and the service life. By combining the actual service life of the polishing pad with the corresponding performance degradation model, the performance degradation characteristics of the wafer polishing process for this batch can be determined.

[0063] The wafer polishing process for each batch includes at least one performance degradation characteristic, which is used to reflect the performance change of the polishing pad during the processing of this batch, thereby providing an important basis for predicting the removal rate and optimizing the polishing process.

[0064] Step 104: Combine the performance degradation characteristics of multiple batches with the parameter characteristics of multiple batches to construct an enhanced dataset.

[0065] The performance degradation characteristics are determined based on the service life of the polishing pad in each batch and its corresponding performance degradation model. The performance degradation characteristics can reflect the performance changes of the polishing pad at different life cycle stages. The parameter characteristics include the statistical characteristics of the key parameters during the polishing process of each batch, such as pressure, polishing liquid flow rate, platen temperature, upper platen current, and lower platen current, as well as the classification characteristics of the equipment and process.

[0066] By combining the performance degradation characteristics with the parameter characteristics corresponding to each batch, a comprehensive dataset can be formed, which can be named an enhanced dataset. The enhanced dataset can contain detailed information for each batch, providing more comprehensive and rich data support for subsequent model training and prediction, and helping to improve the accuracy of prediction and the generalization ability of the model.

[0067] Exemplarily, assume there are three batches of data. For Batch 1, the parameter characteristics are P 11 , P 12 ,..., P 1n , and the performance degradation characteristic is F 1. For Batch 2, the parameter characteristics are P 21 , P 22 ,..., P 2n , and the performance degradation characteristic is F 2. For Batch 3, the parameter characteristics are P 31 , P 32 ,..., P 3n , and the performance degradation characteristic is F 3.

[0068] Further, the merged enhanced dataset contains all features of each batch, that is, the enhanced dataset can be P 11 , P 12 ,..., P 1n , F 1], P 21 , P 22 ,..., P 2 n , F 2], P 31 , P 32 ,..., P 3n , F 3]

[0069] Step 105: Train a machine learning model with the enhanced dataset to obtain a removal rate prediction model.

[0070] Use the previously constructed enhanced dataset containing performance degradation features and parameter features of multiple batches as input, and apply machine learning algorithms (such as XGBoost, random forest, etc.) for model training.

[0071] During the model training process, the model will learn the patterns and relationships in the data, especially the complex associations between the removal rate and various features. The model parameters are optimized by minimizing the prediction error (such as mean squared error) so that it can predict the removal rate as accurately as possible based on the input feature data. After sufficient training and validation, the finally obtained removal rate prediction model will be able to effectively predict the removal rate in new grinding process batches based on real-time or historical performance degradation features and parameter features, thus providing strong support for optimizing the grinding process.

[0072] In this embodiment, by obtaining the parameter features of multiple batches of wafer grinding processes, including key indicators such as removal rate and the service life of auxiliary materials, and combining the life cycle stage and performance degradation model of the grinding pad, the performance degradation features of each batch are determined. These performance degradation features are combined with the parameter features to construct an enhanced dataset for training a machine learning model to obtain a removal rate prediction model. The prediction model obtained by this training method can more accurately predict the removal rate in wafer grinding processes.

[0073] In one embodiment, multiple life cycle stages are defined for the polishing pad, including: dividing the polishing pad into three life cycle stages based on its service life, namely the first cycle stage, the second cycle stage, and the third cycle stage. The polishing pad in the first cycle stage has not been worn, while the polishing pads in the second and third cycle stages have been worn. Among them, the first cycle stage is associated with a first service life range, the second cycle stage is associated with a second cumulative service life range, and the third cycle stage is associated with a third cumulative service life range. The maximum value of the first cumulative service life range is less than or equal to the minimum value of the second cumulative service life range, and the maximum value of the second cumulative service life range is less than or equal to the minimum value of the third cumulative service life range.

[0074] The service life of the polishing pad is divided into three life cycle stages, namely the first cycle stage, the second cycle stage, and the third cycle stage. In the first cycle stage, the polishing pad has not experienced wear and is in a brand-new state. The service life range of the first cycle stage is defined as the first service life range. As the polishing pad is used, it gradually enters the second cycle stage. In the second cycle stage, the polishing pad begins to show wear, and its service life range is the second cumulative service life range. Finally, the polishing pad enters the third cycle stage, where the degree of wear is more severe, and the corresponding service life range is the third cumulative service life range.

[0075] To ensure that the service life ranges of different cycle stages are ordered and non-overlapping, the maximum value of the first cumulative service life range is less than or equal to the minimum value of the second cumulative service life range, indicating that the service life of the first cycle stage is completely included in the service life range of the second cycle stage. Similarly, the maximum value of the second cumulative service life range is less than or equal to the minimum value of the third cumulative service life range, ensuring that the service life of the second cycle stage is completely included in the service life range of the third cycle stage.

[0076] In this embodiment, by dividing the polishing pad into different cycle stages, the usage status of the polishing pad can be understood more accurately, and thus the performance degradation of the polishing pad can be tracked more precisely.

[0077] In one embodiment, as Figure 2 shown, a performance degradation model is established for each life cycle stage, including the following steps:

[0078] Step 201: Establish a first degradation model for the first cycle stage. In the first degradation model, the removal rate has a non-linear relationship with the first cumulative service life of the polishing pad.

[0079] Establish a first degradation model for the first cycle stage. Exemplarily, assume that the first degradation model adopts a quadratic function form Among them R (T ) represents the removal rate, T represents the first cumulative service life of the polishing pad, a , b , c are model parameters obtained by fitting historical data.

[0080] The quadratic regression model includes terms, such that the relationship between the removal rate R ( T ) and the usage time T is no longer a simple linear relationship, but can show a trend of first decreasing and then increasing, or first increasing and then decreasing. The non-linear relationship can more realistically reflect the change of the removal rate when the polishing pad starts to be used.

[0081] Step 202: Establish a second degradation model for the second cycle stage. In the second degradation model, an exponential decay function is used to express the relationship between the removal rate and the second cumulative service life of the polishing pad;

[0082] For the second cycle stage, establish a second degradation model, and use an exponential decay function to describe the relationship between the removal rate and the second cumulative service life of the polishing pad. The specific formula of the second degradation model is:

[0083] .

[0084] Where, R ( T ) represents the removal rate at the usage time T , r is the removal rate of the polishing pad at the start of the second cycle stage, k is the decay constant, representing the rate of decrease of the removal rate, e is the base of the natural logarithm.

[0085] The second degradation model selects an exponential decay function, which can describe the gradually slowing down trend of the removal rate over time.

[0086] Step 203: Establish a third degradation model for the third cycle stage. In the third degradation model, a linear function relationship is used to express the relationship that the removal rate decreases as the third cumulative service life of the polishing pad increases.

[0087] In the third cycle stage, the polishing pad has been used for some time, and its performance has significantly declined. The removal rate gradually decreases as the cumulative service life increases. A linear function can be used to simply and clearly express the linear decreasing trend of the removal rate with the usage time.

[0088] The third degradation model can be expressed as: R ( T ) = a -bT . Among them, R ( T ) represents the removal rate at time T , a is the removal rate at the beginning of the third cycle stage, b is the decline rate of the removal rate over time T . The negative sign indicates that as T increases, R ( T ) gradually decreases.

[0089] In this embodiment, through the performance degradation model of the polishing pad in different life cycles, the performance changes of the polishing pad in different life cycle stages can be more accurately described, thereby providing an important basis for predicting the removal rate and optimizing the polishing process.

[0090] In one embodiment, as Figure 3 shown, for the polishing process of each batch of wafers, based on the service life and performance degradation model of the polishing pad in the polishing process of this batch of wafers, the performance degradation characteristics of the polishing process of this batch of wafers are determined, including the following steps:

[0091] Step 301: For the polishing process of each batch of wafers, based on the service life of the polishing pad in the polishing process of this batch of wafers, determine the life cycle stage passed by the polishing pad, which is defined as the life cycle stage set, and the life cycle stage set includes at least one life cycle stage.

[0092] For the polishing process of each batch of wafers, determine the life cycle stage passed by it according to the service life of the polishing pad in this batch, and define these stages as a life cycle stage set. For example, assume that the total service life of the polishing pad is 100 hours, and it is divided into three life cycle stages: the first cycle stage (0 - 20 hours), the second cycle stage (20 - 60 hours), and the third cycle stage (60 - 100 hours). If the polishing pad has been used for 45 hours during a certain batch of processing, then the life cycle stage set where the polishing pad is located is the second cycle stage. If the polishing pad of a certain batch has been used for 75 hours, then its life cycle stage set is the third cycle stage.

[0093] Step 302: For any life cycle stage in the life cycle stage set, calculate the integral of the removal rate of the life cycle stage with respect to the service life.

[0094] For the first cycle stage in the life cycle stage set, the removal rate of the first cycle stage , calculate the integral from T = 0 to T = T 1:

[0095] .

[0096] The second cycle stage in the set of life cycle stages, and the removal rate function of the second cycle stage is , calculate from T = T 1 to T = T 2 integral:

[0097] .

[0098] The third cycle stage in the set of life cycle stages, and the removal rate function of the third cycle stage is R ( T )= a - bT , calculate from T = T 2 to T = T 3 integral:

[0099] .

[0100] Step 303: Add up the integrals of each life cycle stage in the set of life cycle stages and divide by the cumulative service life to obtain the performance degradation characteristics of the wafer grinding process for this batch.

[0101] Calculate the integral of the first cycle stage (from T =0 to T = T 1) in the set of life cycle stages ; Calculate the integral of the second cycle stage (from T = T 1 to T = T 2) in the set of life cycle stages ; Calculate the integral of the third cycle stage (from T = T 2 to T = T 3) in the set of life cycle stages .

[0102] Add up the three integrals to get the total integral = + + , divide by the cumulative service life , to obtain the performance degradation characteristics of the wafer grinding process for this batch, that is . Among them, the performance degradation characteristics D of the wafer grinding process can reflect the average impact of the removal rate of the grinding pad at different target life cycle stages on the total service life.

[0103] In this embodiment, the life cycle stage of the polishing pad in this batch of processing is determined, that is, the life cycle stage set; the integral of the life cycle stage removal rate in each life cycle stage set with respect to the service life is calculated respectively, the integrals of the life cycle stages in all life cycle stage sets are added together, and then divided by the cumulative service life to obtain the performance degradation characteristics of this batch of wafer polishing processing. The performance degradation characteristics comprehensively reflect the performance changes of the polishing pad in different life cycle stages, providing an important basis for evaluating the polishing effect and optimizing the process.

[0104] In one embodiment, as Figure 4 shown, the method further includes:

[0105] Step 401: Obtain the equipment marks and processing technologies of multiple batches of wafer polishing processing.

[0106] The equipment mark can be a unique identifier used to distinguish different polishing equipment, such as an equipment number or name, and the equipment number or name information can be obtained through the equipment management system or production records. The processing technology can be the specific process parameters adopted during the polishing process, including the type of polishing liquid, polishing pressure, polishing speed, polishing time, etc., and the process parameter information is usually recorded in process documents or production instructions.

[0107] Step 402: For each batch of wafer polishing processing, determine the first category of features based on the equipment mark of this batch, and determine the second category of features based on the processing technology of this batch.

[0108] For each batch of wafer polishing processing, determine the first category of features based on the equipment mark of this batch. By encoding the equipment mark, such as using integer encoding or label encoding, the equipment information can be converted into features recognizable by the model. For example, if there are three different polishing equipment in the factory, the different polishing equipment can be marked as Equipment 1, Equipment 2, and Equipment 3 respectively, and these marks are converted into integer category features.

[0109] Determine the second category of features based on the processing technology of this batch. The processing technology refers to the specific process parameters and processes adopted during the polishing process. By classifying and encoding the process parameters, the processing technology information can be converted into features recognizable by the model. For example, if there are four different processing technologies in the factory, one-hot encoding can be used to represent each process as a binary vector.

[0110] Step 403: Combine the first category of features and the second category of features of multiple batches into an enhanced data set.

[0111] Align the first category features (equipment markings) and the second category features (processing techniques) of each batch with other relevant features (such as removal rate, grinding time, etc.) according to the batch number, and integrate them into a dataset. Among them, this dataset is an enhanced dataset R n×z , where n is the number of processing batches, z is the number of data features for each batch.

[0112] In this embodiment, by obtaining the equipment markings and processing techniques of multiple batches of wafer grinding, and respectively converting them into the first category features and the second category features, and then merging them into the enhanced dataset, the diversity and representativeness of the dataset can be significantly enhanced. This process helps the prediction model of the wafer grinding removal rate to more comprehensively learn the influence of different equipment and processes on the grinding effect, thereby improving the generalization ability and prediction accuracy of the prediction model, and enabling it to better adapt to various actual production scenarios.

[0113] In one embodiment, for each batch of wafer grinding, determine the first category features based on the equipment markings of this batch, and determine the second category features based on the processing technique of this batch, including:

[0114] Use label encoding to map the equipment identifier to a unique integer to obtain the first category features, and use one-hot encoding to convert the processing technique into a binary vector to obtain the first category features. The binary vector can cover different wafer types and sizes.

[0115] Specifically, for each batch of wafer grinding, determine the first category features through the equipment markings of this batch. Specifically, use label encoding to map the equipment identifier to a unique integer. For example, equipment A is mapped to 1, equipment B is mapped to 2, etc., so as to obtain the first category features.

[0116] Determine the second category features through the processing technique of this batch. Use one-hot encoding to convert the processing technique into a binary vector. For example, 6-inch standard wafer grinding can be represented as [1, 0, 0, 0], 8-inch standard wafer grinding can be represented as [0, 1, 0, 0], 6-inch seed crystal grinding can be represented as [0, 0, 1, 0], and 8-inch seed crystal grinding can be represented as [0, 0, 0, 1], so as to obtain the second category features. The binary vector can cover different wafer types and sizes.

[0117] In this embodiment, in this way, the information of the equipment and the process can be effectively encoded into numerical features that the model can process, so as to better reflect the influence of the equipment and the process on the grinding effect.

[0118] In one embodiment, such as Figure 5As shown, the machine learning model is an XGBoost regression model, and the hyperparameters of the XGBoost regression model are optimized through a combination of grid search algorithm and genetic algorithm; among them, optimizing the hyperparameters of the XGBoost regression model through a combination of grid search and genetic algorithm includes the following steps:

[0119] Step 501: Search the predefined hyperparameter space based on the grid search algorithm to obtain a benchmark hyperparameter combination;

[0120] Execute a relatively extensive grid search algorithm to traverse the predefined hyperparameter space, obtain preliminary results, and calculate the performance of each hyperparameter combination. The following grid search algorithm formula can be used:

[0121] 。

[0122] Among them, represents the benchmark hyperparameter combination obtained by the grid search algorithm; represents the predefined hyperparameter space, including all possible hyperparameter combinations; represents the loss function calculated through cross-validation, usually the mean squared error (MSE) or root mean squared error (RMSE). It should be noted that the benchmark hyperparameter combination can refer to the optimal hyperparameter combination.

[0123] Step 502: Use the benchmark hyperparameter combination as the center point, determine the preselected hyperparameter range, and based on the preselected hyperparameter range, determine the initial parameter set;

[0124] By comparing the average validation losses of all possible hyperparameter combinations, select the hyperparameter combination with the minimum average validation loss as the center point of the preselected hyperparameter range. Set a smaller range near the center point of the preselected hyperparameter range as the preselected hyperparameter range.

[0125] Suppose the benchmark hyperparameter combination obtained by the grid search algorithm is:

[0126] {learning_rate = 0.1, max_depth = 6, n_estimators = 100, gamma = 0.1}.

[0127] Based on the optimal hyperparameter combination, narrow the range of each hyperparameter:

[0128] Narrow the range of learning_rate to [0.05, 0.1, 0.15];

[0129] Narrow the range of max_depth to [5, 6, 7];

[0130] Narrow the range of n_estimators to [80, 100, 120];

[0131] Narrow the range of gamma to [0.09, 0.1].

[0132] Using each narrowed hyperparameter range, a preselected hyperparameter range can be obtained. Further, within the preselected hyperparameter range, the grid search algorithm is used again to generate all possible hyperparameter combinations within the preselected hyperparameter range, and cross-validation is performed on each hyperparameter combination to calculate the average validation loss. The hyperparameter combination with the minimum average validation loss is selected as the initial parameter set.

[0133] Step 503: Screen out the optimal hyperparameter combination from the initial parameter set through the selection, crossover, and mutation operations of the genetic algorithm.

[0134] In the genetic algorithm, an individual is usually represented by a chromosome, and a chromosome is a candidate solution in the solution space. Each chromosome can be represented as a vector or a string, here as a hyperparameter combination θ. The quality of each hyperparameter combination θ is evaluated based on the fitness function, and the combinations with high fitness are selected to enter the next generation. The fitness function is defined as the negative root mean square error, and the fitness function is:

[0135] .

[0136] Select the value that maximizes the fitness function. Among them, the fitness function is the negative root mean square error, and the parameter combinations with high fitness are selected through the selection operation.

[0137] Calculate the selection probability P(x i ) as the probability that the hyperparameter combination x i is selected, and the calculation formula is:

[0138] .

[0139] Among them, is the probability that the hyperparameter combination x i is selected, is the fitness value of the j-th hyperparameter combination, and N is the number of hyperparameter combinations.

[0140] Perform the crossover operation on the selected hyperparameter combinations. Exemplarily, select two hyperparameter combinations, exchange some parameter values, and generate new hyperparameter combinations. For example, assume the current hyperparameter combination is θ t , and a new combination θ t+1 is generated through the crossover operation. The calculation formula is:

[0141]

[0142] Perform mutation operations to randomly change some parameter values in a certain hyperparameter combination, increasing the diversity of the population. For example, make a small adjustment to one parameter in a certain hyperparameter combination to generate a new hyperparameter combination. Mutation operations help avoid getting stuck in local optimal solutions and improve the global search ability. Through multi-generation iterative optimization, the hyperparameter combinations in the population will be continuously improved and eventually approach the optimal solution. In each generation, repeat the selection, crossover, and mutation operations to gradually screen out hyperparameter combinations with better performance. After multi-generation evolution, the optimal hyperparameter combination is obtained. Use the optimal hyperparameter combination to train the XGBoost regression model. Through the selection, crossover, and mutation operations of the genetic algorithm, screen out the optimal hyperparameter combination from the initial parameter set to improve the prediction performance of the model and finally obtain an optimal model under the goal of minimizing the root mean square error.

[0143] In this embodiment, the hyperparameters of the XGBoost regression model are optimized by combining grid search and genetic algorithm. The grid search is used to narrow the hyperparameter range to obtain a preselected range, and an initial parameter set is obtained within this range. The selection, crossover, and mutation operations of the genetic algorithm are used to screen out the optimal hyperparameter combination. This method effectively improves the efficiency and accuracy of hyperparameter optimization, enables the model to achieve better performance under the goal of minimizing the root mean square error, and enhances its generalization ability.

[0144] In one embodiment, obtain the instant process parameters during the current batch of wafer grinding process and the instant service life of the grinding pad, and input them into the removal rate prediction model to obtain the predicted removal rate of the wafer grinding process; wherein, the removal rate prediction model is obtained through the training method of the prediction model for the removal rate of the wafer grinding process.

[0145] Obtain the instant process parameters during this batch of processing. The instant process parameters include grinding pressure, grinding fluid flow rate, platen temperature, upper platen current, and lower platen current, etc. Obtain the instant service life of the grinding pad, that is, the time the grinding pad has been used. Take the instant process parameters and the instant service life of the grinding pad as inputs and input them into the pre-trained removal rate prediction model. The removal rate prediction model can predict the removal rate of the current batch of wafer grinding process according to the input data. The predicted removal rate can help optimize the grinding process and improve processing efficiency and quality control.

[0146] In this embodiment, by obtaining the instant process parameters and the instant service life of the grinding pad during the current batch of wafer grinding process and inputting them into the removal rate prediction model, the predicted removal rate of this batch of wafer grinding process can be obtained, thereby realizing the real-time evaluation and prediction of the current grinding processing effect and providing a timely reference basis for process adjustment and optimization.

[0147] Based on the same concept, the present application further provides an electronic device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it can implement the training method of the prediction model for the removal rate of wafer grinding processing or the method for predicting the removal rate of wafer grinding processing.

[0148] In one embodiment, an electronic device is provided, including a memory and a processor. The electronic device may be a terminal, and its internal structure diagram may be as Figure 6 shown. The electronic device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the electronic device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements the training method of the prediction model for the removal rate of wafer grinding processing or the method for predicting the removal rate of wafer grinding processing. The display screen of the electronic device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic device may be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the electronic device, or an external keyboard, a touchpad, or a mouse, etc.

[0149] Based on the same concept, as Figure 7 shown, the present application further provides a training device for the prediction model of the removal rate of wafer grinding processing. The device includes:

[0150] A first acquisition module 701, configured to acquire parameter features of multiple batches of wafer grinding processing; wherein, the parameter features of each batch are extracted from the process parameters during the wafer grinding processing of this batch, and the parameter features of each batch are multiple and at least include the removal rate;

[0151] A second acquisition module 702, configured to acquire the service life of auxiliary materials for multiple batches of wafer grinding processing. The auxiliary materials at least include a grinding pad; wherein, the grinding pad includes multiple life cycle stages divided based on its own service life, and a performance degradation model is established for each life cycle stage; wherein, the performance degradation model is a functional relationship between the removal rate and the service life;

[0152] A feature extraction module 703, configured to, for each batch of wafer grinding processing, determine the performance degradation feature of this batch of wafer grinding processing based on the service life of the grinding pad and the performance degradation model in this batch of wafer grinding processing. Each batch of wafer grinding processing includes at least one performance degradation feature;

[0153] A training module 704 is configured to combine performance degradation features of multiple batches with parameter features of multiple batches to construct an enhanced dataset. The training module is further configured to train a machine learning model with the enhanced dataset to obtain a removal rate prediction model.

[0154] In one embodiment, the second acquisition module 702 divides a plurality of life cycle stages for the polishing pad, specifically: based on the service life of the polishing pad, the polishing pad is divided into three life cycle stages, namely a first cycle stage, a second cycle stage, and a third cycle stage. The polishing pad in the first cycle stage is not worn, and the polishing pads in the second cycle stage and the third cycle stage are worn; wherein, the first cycle stage is associated with a first service life range, the second cycle stage is associated with a second cumulative service life range, and the third cycle stage is associated with a third cumulative service life range; the maximum value of the first cumulative service life range is less than or equal to the minimum value of the second cumulative service life range, and the maximum value of the second cumulative service life range is less than or equal to the minimum value of the third cumulative service life range.

[0155] In one embodiment, the second acquisition module 702 establishes a performance degradation model for each life cycle stage, specifically: establishing a first degradation model for the first cycle stage, where the removal rate and the first cumulative service life of the polishing pad in the first degradation model are in a non-linear relationship; establishing a second degradation model for the second cycle stage, where an exponential decay function is used to express the relationship between the removal rate and the second cumulative service life of the polishing pad in the second degradation model; establishing a third degradation model for the third cycle stage, where a linear function relationship is used to express the relationship that the removal rate decreases as the third cumulative service life of the polishing pad increases.

[0156] In one embodiment, for each batch of wafer polishing processes, the feature extraction module 703 determines the performance degradation features of the batch of wafer polishing processes based on the service life of the polishing pad and the performance degradation model in the batch of wafer polishing processes, specifically: for each batch of wafer polishing processes, based on the service life of the polishing pad in the batch of wafer polishing processes, determine the life cycle stage that the polishing pad has passed through so far, which is defined as the target life cycle stage; calculate the integral of the removal rate with respect to the service life for each target life cycle stage respectively; divide the sum of the integrals of each target life cycle stage by the cumulative service life to obtain the performance degradation features of the batch of wafer polishing processes.

[0157] In one embodiment, the first acquisition module 701 acquires the equipment marks and processing technologies of multiple batches of wafer polishing processes; for each batch of wafer polishing processes, determine the first category features based on the equipment marks of the batch, and determine the second category features based on the processing technologies of the batch; combine the first category features and the second category features of multiple batches into the enhanced dataset.

[0158] In one embodiment, for each batch of wafer grinding process, the first acquisition module 701 determines the first category of features based on the device markings of this batch, and determines the second category of features based on the processing technology of this batch. Specifically, it is used to: map the device identification to a unique integer using label encoding to obtain the first category of features, and convert the processing technology into a binary vector using one-hot encoding to obtain the second category of features. The binary vector can cover different wafer types and sizes.

[0159] In one embodiment, the machine learning model is an XGBoost regression model, and the hyperparameters of the XGBoost regression model are optimized through a combination of grid search and genetic algorithm; wherein, the training module 704 optimizes the hyperparameters of the XGBoost regression model through a combination of grid search and genetic algorithm. Specifically, it includes: searching the predefined hyperparameter space based on the grid search algorithm to obtain a benchmark hyperparameter combination; using the benchmark hyperparameter combination as the center point, determining the preselected hyperparameter range, and determining the initial parameter set based on the preselected hyperparameter range; screening out the optimal hyperparameter combination from the initial parameter set through the selection, crossover, and mutation operations of the genetic algorithm.

[0160] The embodiment of the present application also provides a device for predicting the removal rate of wafer grinding process. The prediction device includes a prediction module. The prediction module is used to obtain the instant process parameters during the current batch of wafer grinding process and the instant service life of the grinding pad, and input them into the removal rate prediction model to obtain the predicted removal rate of the wafer grinding process.

[0161] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0162] The above embodiments only represent several implementation manners of the present application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several deformations and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A training method for a prediction model of the removal rate in wafer grinding processing, characterized in that, The method includes: Obtaining parameter characteristics of wafer grinding and processing for multiple batches; the parameter characteristics of each batch are extracted from the process parameters during the wafer grinding and processing of that batch, and there are multiple parameter characteristics for each batch and at least include the removal rate; Obtaining the service life of auxiliary materials for the wafer grinding and processing of the multiple batches, the auxiliary materials at least including a grinding pad; the grinding pad includes multiple life cycle stages divided based on its own service life, and a performance degradation model is established for each life cycle stage; the performance degradation model is a functional relationship between the removal rate and the service life; For the wafer grinding and processing of each batch, determining the performance degradation characteristics of the wafer grinding and processing of that batch based on the service life of the grinding pad and the performance degradation model in the wafer grinding and processing of that batch, and there is at least one performance degradation characteristic for the wafer grinding and processing of each batch; Combining the performance degradation characteristics of multiple batches with the parameter characteristics of multiple batches to construct an enhanced data set; Training a machine learning model through the enhanced data set to obtain a removal rate prediction model; For the wafer grinding and processing of each batch, determining the performance degradation characteristics of the wafer grinding and processing of that batch based on the service life of the grinding pad and the performance degradation model in the wafer grinding and processing of that batch, including: for the wafer grinding and processing of each batch, determining the life cycle stage passed by the grinding pad based on the service life of the grinding pad in the wafer grinding and processing of that batch, defined as the life cycle stage set, and the life cycle stage set at least includes one life cycle stage; for any one life cycle stage in the life cycle stage set, calculating the integral of the removal rate of the life cycle stage with respect to the service life; adding up the integrals of each life cycle stage in the life cycle stage set and then dividing by the cumulative service life to obtain the performance degradation characteristics of the wafer grinding and processing of that batch.

2. The training method of the prediction model for the removal rate of wafer grinding and processing, characterized in that Dividing multiple life cycle stages for the grinding pad, including: Based on the service life of the grinding pad, dividing the grinding pad into three life cycle stages, namely the first cycle stage, the second cycle stage, and the third cycle stage. The grinding pad in the first cycle stage is not worn, and the grinding pads in the second cycle stage and the third cycle stage are worn; Wherein, the first cycle stage is associated with a first cumulative service life range, the second cycle stage is associated with a second cumulative service life range, and the third cycle stage is associated with a third cumulative service life range; the maximum value of the first cumulative service life range is less than or equal to the minimum value of the second cumulative service life range, and the maximum value of the second cumulative service life range is less than or equal to the minimum value of the third cumulative service life range.

3. The training method of the prediction model for the removal rate of wafer grinding and processing according to claim 2, characterized in that Establishing a performance degradation model for each life cycle stage, including: Establishing a first degradation model for the first cycle stage, and the relationship between the removal rate and the first cumulative service life of the grinding pad in the first degradation model is non-linear; Establishing a second degradation model for the second cycle stage, and using an exponential decay function to express the change relationship of the removal rate with the second cumulative service life of the grinding pad in the second degradation model; A third degradation model is established for the third cycle stage, and in the third degradation model, a linear function relationship is used to express the relationship that the removal rate decreases as the third cumulative service life of the polishing pad increases.

4. The training method of the prediction model for the removal rate of wafer grinding and processing according to any one of claims 1 to 3, characterized in that, The method further includes: Obtaining equipment marks and processing technologies for grinding and processing wafers in multiple batches; For the grinding and processing of wafers in each batch, determining first category features based on the equipment marks of this batch, and determining second category features based on the processing technology of this batch; Merging the first category features and second category features of multiple batches into the enhanced data set.

5. The training method of the prediction model for the removal rate of wafer grinding and processing, characterized in that, For the grinding and processing of wafers in each batch, determining first category features based on the equipment marks of this batch, and determining second category features based on the processing technology of this batch, including: Using label encoding to map the equipment identification to a unique integer to obtain the first category features, and using one-hot encoding to convert the processing technology into a binary vector to obtain the first category features, and the binary vector can cover different wafer types and sizes.

6. The training method of the prediction model for the removal rate of wafer grinding and processing according to claim 1, wherein, The machine learning model is an XGBoost regression model, and the hyperparameters of the XGBoost regression model are optimized by combining grid search and genetic algorithm; wherein, optimizing the hyperparameters of the XGBoost regression model by combining grid search and genetic algorithm includes: Searching the predefined hyperparameter space based on the grid search algorithm to obtain a benchmark hyperparameter combination; Taking the benchmark hyperparameter combination as the center point, determining a preselected hyperparameter range, and determining an initial parameter set based on the preselected hyperparameter range; Selecting the optimal hyperparameter combination from the initial parameter set through the selection, crossover and mutation operations of the genetic algorithm.

7. A method for predicting the removal rate of wafer grinding and processing, characterized in that Obtaining the instant process parameters and the instant service life of the polishing pad during the grinding and processing of the current batch of wafers, inputting them into the removal rate prediction model, and obtaining the predicted removal rate of the wafer grinding and processing; wherein, the removal rate prediction model is obtained by the training method of the prediction model of the removal rate of wafer grinding and processing according to any one of the above claims 1 to 6.

8. An electronic device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the training method of the prediction model of the removal rate of wafer grinding and processing according to any one of claims 1 to 6 or the method for predicting the removal rate of wafer grinding and processing according to claim 7.

9. A training device for a prediction model of the removal rate in wafer grinding processing, characterized in that, The device includes: A first acquisition module, configured to acquire parameter features of grinding and processing wafers in multiple batches; wherein, the parameter features of each batch are extracted from the process parameters during the grinding and processing of wafers in this batch, and the parameter features of each batch are multiple and at least include the removal rate; A second acquisition module, configured to acquire the service life of auxiliary materials for the grinding and processing of the multiple batches of wafers, and the auxiliary materials at least include a polishing pad; wherein, the polishing pad includes multiple life cycle stages divided based on its own service life, and a performance degradation model is established for each life cycle stage; wherein, the performance degradation model is a functional relationship between the removal rate and the service life; A feature extraction module, which is used for each batch of wafer grinding processes, to determine the performance degradation features of the wafer grinding process of this batch based on the service life of the grinding pad in the wafer grinding process of this batch and the performance degradation model. Each batch of wafer grinding processes includes at least one performance degradation feature; A training module, which is used to combine the performance degradation features of multiple batches with the parameter features of multiple batches to construct an enhanced data set; the training module is also used to train a machine learning model through the enhanced data set to obtain a removal rate prediction model; For each batch of wafer grinding processes, determining the performance degradation features of the wafer grinding process of this batch based on the service life of the grinding pad in the wafer grinding process of this batch and the performance degradation model includes: for each batch of wafer grinding processes, determining the life cycle stage passed by the grinding pad based on the service life of the grinding pad in the wafer grinding process of this batch, which is defined as a life cycle stage set, and the life cycle stage set includes at least one life cycle stage; for any one life cycle stage in the life cycle stage set, calculating the integral of the removal rate of this life cycle stage with respect to the service life; adding up the integrals of each life cycle stage in the life cycle stage set and then dividing by the cumulative service life to obtain the performance degradation features of the wafer grinding process of this batch.

Images