Integrated circuit deposition film thickness prediction method and system

Abstract

The present invention discloses a method and system for predicting the thickness of deposited films of integrated circuits based on RFE-CV-XGBoost, which belongs to the field of semiconductor manufacturing. The method comprises: step S1, pre-processing the deposited film data of a database; step S2, screening the characteristic values ​​contained in the deposited film data pre-processed in step S1 according to a predetermined feature screening rule; step S3, processing the features of step S2 by a cross-validated feature recursive elimination method; step S4, establishing a regression prediction model based on XGBoost, and training and testing using the features of step S3; step S5, optimizing the hyperparameters of the regression prediction model using a genetic algorithm, finding the optimal hyperparameters, and predicting the thickness of deposited films of integrated circuits using the final regression prediction model. The present invention saves the film thickness measurement step after the thin film deposition process through data modeling, shortens the production cycle of semiconductor products, and saves measurement cost expenditure.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor manufacturing measurement, and in particular to a method and system for predicting the thickness of deposited films of integrated circuits. Background Art

[0002] Semiconductor chip manufacturing technology is one of the core technologies of the information industry and is widely used in new energy, information and communication equipment, and high-end communications fields.

[0003] Metrology refers to measuring various target parameters during chip manufacturing, such as thin film deposition thickness, chemical mechanical polishing depth, plasma implantation depth, and ion implantation concentration. Virtual metrology (VM) methods use mathematical models or simulations to predict the results of various target parameters involved in each wafer processing step without the cost and time of metrology equipment. These methods are based on statistical models that link readily observable process data (e.g., process tool sensor readings, wafer context data) with relevant metrology data. This data-driven modeling of the process reduces the loss of metrology equipment and materials, shortens production cycles, and ultimately improves economic efficiency. In recent years, domestic and international scholars have conducted in-depth research on the application of machine learning in semiconductor virtual metrology. However, these methods have limitations. For example, incomplete data processing can retain many useless features, reducing prediction accuracy and increasing the model training burden. Furthermore, processed features can still contain features that negatively impact model results, meaning that the data information contained in the features contains a large proportion of noise. At the same time, conventional machine learning-based prediction models require a lot of manual intervention and a lot of effort to select models and optimize model hyperparameters, which makes the establishment of prediction models in the semiconductor field difficult and complex.

[0004] In view of this, the present invention is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for predicting the thickness of deposited films of integrated circuits, which can solve the problems of inaccurate prediction of film thickness by existing machine learning prediction models in the field of thin film deposition in integrated circuit manufacturing and time-consuming optimization of model hyperparameters.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A method for predicting the thickness of an integrated circuit deposited film comprises the following steps:

[0008] Step S1, preprocessing the deposited thin film data in the database to obtain preprocessed deposited thin film data;

[0009] Step S2, screening the characteristic values ​​contained in the deposited thin film data pre-processed in step S1 according to characteristic screening rules to obtain screened characteristics;

[0010] Step S3, processing the filtered features obtained in step S2 by cross-validation feature recursive elimination to delete features that have a negative impact on the performance of the XGBoost model;

[0011] Step S4, establishing a regression prediction model based on the XGBoost model, and using the features obtained in step S3 as a training data set to train and test the regression prediction model;

[0012] In step S5, a genetic algorithm is used to optimize the hyperparameters of the regression prediction model trained and tested in step S4, and a set of hyperparameters that makes the regression prediction model perform best is found from all the hyperparameters of the regression prediction model. The set of hyperparameters is used as the hyperparameters of the final regression prediction model, and the thickness of the deposited film of the integrated circuit is predicted using the optimized final regression prediction model.

[0013] A system for predicting the thickness of deposited films of integrated circuits, comprising:

[0014] A data preprocessing module can preprocess the deposited film data obtained from the database to obtain preprocessed deposited film data;

[0015] a feature screening module, which is in communication with the data preprocessing module and can screen the feature values ​​contained in the deposited thin film data preprocessed by the data preprocessing module according to a predetermined feature screening rule to obtain screened features;

[0016] A cross-validated feature recursive elimination module is communicatively connected to the feature screening module and is capable of processing the filtered features output by the feature screening module through a cross-validated feature recursive elimination method to delete features that have a side effect on the performance of the XGBoost model;

[0017] A prediction model construction module is respectively connected to the cross-validation-based feature recursive elimination module and the hyperparameter optimization module, capable of establishing a regression prediction model based on the XGBoost model, and using the features output by the cross-validation-based feature recursive elimination module as a training data set to train and test the regression prediction model;

[0018] A hyperparameter optimization module is communicatively connected to the prediction model construction module and can use a genetic algorithm to perform hyperparameter optimization on the trained and tested regression prediction model established by the prediction model construction module, and find a set of hyperparameters that makes the regression prediction model perform best from all the hyperparameters of the regression prediction model, and use them as the hyperparameters of the final regression prediction model. The final regression prediction model is used to predict the thickness of the deposited film of the integrated circuit.

[0019] Compared with the prior art, the integrated circuit deposition film thickness prediction method and system provided by the present invention have the following beneficial effects:

[0020] By introducing feature screening rules, the features with little significance in the deposited thin film data after pre-processing are screened out. At the same time, a feature recursive elimination method based on cross-validation is introduced to eliminate features that have a negative effect on the performance of the regression prediction model; and the genetic algorithm is used to optimize the hyperparameters of the RFE-CV-XGBoost algorithm, which greatly improves the parameter adjustment efficiency of the regression prediction model hyperparameters and reduces the heavy burden on semiconductor measurement personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 This is a flowchart of the steps of a method for predicting the thickness of deposited films of integrated circuits provided by an embodiment of the present invention.

[0023] Figure 2 2 is a box plot diagram of the variance standard in the feature screening rule provided by an embodiment of the present invention.

[0024] Figure 3 Schematic diagram of a box plot provided by an embodiment of the present invention.

[0025] Figure 4 It is a heat map diagram of the Pearson coefficient standard in the feature screening rule provided by the embodiment of the present invention.

[0026] Figure 5 This is a flowchart of using a genetic algorithm to optimize hyperparameters according to an embodiment of the present invention.

[0027] Figure 6 This is a block diagram of a system for predicting the thickness of deposited films of integrated circuits provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the specific content of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments, and do not constitute a limitation of the present invention. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0029] First, the following terms may be used in this article:

[0030] The term “and / or” means that either or both of them can be realized at the same time. For example, X and / or Y includes both “X” or “Y” and “X and Y”.

[0031] The terms "include," "comprises," "contains," "has," or other similar expressions should be interpreted as non-exclusive. For example, "including certain technical features (such as raw materials, components, ingredients, carriers, dosage forms, materials, dimensions, parts, components, mechanisms, devices, steps, procedures, methods, reaction conditions, processing conditions, parameters, algorithms, signals, data, products, or manufactured articles, etc.) should be interpreted as including not only the technical features explicitly listed, but also other technical features known in the art that are not explicitly listed.

[0032] Unless otherwise specified or limited, the terms "mounted," "connected," "connect," and "fixed" should be interpreted broadly. For example, they can refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediary; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in this document based on specific circumstances.

[0033] When concentration, temperature, pressure, size or other parameters are expressed in the form of a numerical range, the numerical range should be understood to specifically disclose all ranges formed by the pairing of any upper limit, lower limit, or preferred value within the numerical range, regardless of whether the range is explicitly stated. For example, if a numerical range of "2 to 8" is stated, the numerical range should be interpreted as including ranges of "2 to 7," "2 to 6," "5 to 7," "3 to 4 and 6 to 7," "3 to 5 and 7," "2 and 5 to 7," etc. Unless otherwise specified, the numerical ranges stated herein include both their endpoints and all integers and fractions within the numerical range.

[0034] The terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings and are only for the convenience and simplification of description, and do not explicitly or implicitly indicate that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore should not be understood as a limitation to this document.

[0035] The integrated circuit deposition film thickness prediction method and system provided by the present invention are described in detail below. Any information not described in detail in the examples of the present invention is prior art known to those skilled in the art. Where specific conditions are not specified in the examples of the present invention, the results were performed in accordance with conventional conditions in the art or the conditions recommended by the manufacturer. Reagents or instruments used in the examples of the present invention, where the manufacturer is not specified, are all commercially available conventional products.

[0036] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting the thickness of deposited films of integrated circuits, comprising the following steps:

[0037] Step S1, preprocessing the deposited thin film data in the database to obtain preprocessed deposited thin film data;

[0038] Step S2, screening the characteristic values ​​contained in the deposited thin film data pre-processed in step S1 according to characteristic screening rules to obtain screened characteristics;

[0039] Step S3, processing the filtered features obtained in step S2 by cross-validation feature recursive elimination to delete features that have a negative impact on the performance of the XGBoost model;

[0040] Step S4, establishing a regression prediction model based on the XGBoost model, and using the features obtained in step S3 as a training data set to train and test the regression prediction model;

[0041] In step S5, a genetic algorithm is used to optimize the hyperparameters of the regression prediction model trained and tested in step S4, and a set of hyperparameters that makes the regression prediction model perform best is found from all the hyperparameters of the regression prediction model. The set of hyperparameters is used as the hyperparameters of the final regression prediction model, and the thickness of the deposited film of the integrated circuit is predicted using the optimized final regression prediction model.

[0042] In step S1 of the above method, the deposited thin film data in the database is preprocessed in the following manner, including:

[0043] Stored in the database are process parameters involved in the integrated circuit thin film deposition process as deposited film data;

[0044] The deposited film data with missing values ​​and missing processes in the database are deleted, and the data set consisting of the deposited film data obtained after deletion is normalized. The formula of the conversion function is:

[0045]

[0046] Where, X * represents the normalized result; X represents the sample data of the deposited thin film data, Max represents the maximum value of the sample data of the deposited thin film data, and Min represents the minimum value of the sample data of the deposited thin film data.

[0047] By normalizing the data set, the impact of magnitude differences between features can be reduced.

[0048] In step S2 of the above method, the characteristic values ​​contained in the deposited thin film data pre-processed in step S1 are screened according to a predetermined characteristic screening rule in the following manner, including:

[0049] Delete features with variance less than 0.015 and features with Pearson correlation coefficient greater than 0.9.

[0050] The above feature screening rules comprehensively consider the variance and the linear correlation between features to delete features. Figure 2 In the analysis of the box plot shown, features with a variance less than 0.015 are deleted, the Pearson correlation coefficient between the features is calculated, and a heat map is drawn to find features with a large degree of linear correlation, that is, features with a Pearson coefficient greater than 0.9.

[0051] In step S3 of the above method, the filtered features obtained in step S2 are processed by cross-validated feature recursive elimination in the following manner to remove features that have a side effect on the performance of the XGBoost model, including:

[0052] The filtered features obtained in step S2 are processed by a feature recursive elimination algorithm based on cross-validation. The performance of the XGBoost model with 7-fold cross-validation is used as the evaluation index. The features are deleted one by one. If the mean square error (MSE) of the XGBoost model decreases after deleting the features and the coefficient of determination (R) decreases, the feature is deleted. 2 If it rises, the feature is deleted, otherwise it is not deleted.

[0053] In step S4 of the above method, a regression prediction model is established based on the XGBoost model in the following manner, and the features obtained in step S3 are used as a training data set to train and test the regression prediction model, including:

[0054] The data set formed by the features obtained in step S3 is randomly divided into a training set, a test set, and a validation set according to a ratio of 7:2:1; the regression prediction model is trained using the training set; and the prediction performance of the trained regression prediction model is evaluated using the test set;

[0055] After RFE-CV eliminates features with too much noise information, the remaining features are used as parameter inputs of the RFE-CV-XGBoost regression prediction model. At this time, all features contain more information that can reflect the changes in the results, which has a positive effect on improving the accuracy of the prediction results.

[0056] In step S5 of the above method, the hyperparameters of the regression prediction model trained and tested in step S4 are optimized using a genetic algorithm in the following manner, and a set of hyperparameters that makes the regression prediction model perform best is found from all the hyperparameters of the regression prediction model, and the hyperparameters are used as the hyperparameters of the final regression prediction model, including:

[0057] The genetic optimization algorithm is used to optimize the number and depth of trees formed by all hyperparameters of the regression prediction model after training and testing through the validation set, and a set of hyperparameters that makes the regression prediction model perform best is found from all the hyperparameters of the regression prediction model, which are used as the hyperparameters of the final regression prediction model.

[0058] In step S4 of the above method, in the data set formed by the features obtained in step S3, the features related to the deposition, ventilation, and heating steps in the deposited film data are used as input parameters of the regression prediction model;

[0059] The output parameter of the regression prediction model is the predicted result of the deposited film thickness.

[0060] See also Figure 6 The embodiment of the present invention further provides a system for predicting the thickness of deposited films of integrated circuits, comprising:

[0061] A data preprocessing module can preprocess the deposited film data obtained from the database to obtain preprocessed deposited film data;

[0062] a feature screening module, which is in communication with the data preprocessing module and can screen the feature values ​​contained in the deposited thin film data preprocessed by the data preprocessing module according to a predetermined feature screening rule to obtain screened features;

[0063] A cross-validated feature recursive elimination module is in communication with the feature screening module and is capable of processing the filtered features output by the feature screening module through a cross-validated feature recursive elimination method to delete features that have a negative impact on the performance of the XGBoost model;

[0064] A prediction model construction module is respectively connected to the cross-validation-based feature recursive elimination module and the hyperparameter optimization module, capable of establishing a regression prediction model based on the XGBoost model, and using the features output by the cross-validation-based feature recursive elimination module as a training data set to train and test the regression prediction model;

[0065] A hyperparameter optimization module is communicatively connected to the prediction model construction module and can use a genetic algorithm to perform hyperparameter optimization on the trained and tested regression prediction model established by the prediction model construction module, and find a set of hyperparameters that makes the regression prediction model perform best from all the hyperparameters of the regression prediction model, and use them as the hyperparameters of the final regression prediction model. The final regression prediction model is used to predict the thickness of the deposited film of the integrated circuit.

[0066] The above system also includes:

[0067] a parameter acquisition module, communicatively connected to the thin film deposition equipment, capable of acquiring process parameters involved in the integrated circuit thin film deposition process as deposited film data, the process parameters including setting parameters of the thin film deposition equipment used in the integrated circuit thin film deposition process and external material supply parameters for the integrated circuit deposition process;

[0068] The database construction module is respectively connected to the parameter acquisition module and the data preprocessing module, and can establish a database based on the process parameters involved in the integrated circuit thin film deposition process obtained by the parameter acquisition module as deposited thin film data, and provide the deposited thin film data to the data preprocessing module.

[0069] In summary, the method and system of the embodiment of the present invention, by introducing feature screening rules, screens out features of little significance in the deposited thin film data after pretreatment, and at the same time introduces a feature recursive elimination method based on cross-validation to eliminate features that have a negative effect on the performance of the regression prediction model; and uses a genetic algorithm to optimize the hyperparameters of the RFE-CV-XGBoost algorithm, which greatly improves the tuning efficiency of the hyperparameters of the regression prediction model and reduces the heavy burden on semiconductor measurement personnel.

[0070] In order to more clearly demonstrate the technical solution and technical effects provided by the present invention, the integrated circuit deposition film thickness prediction method and system based on RFE-CV-XGBoost provided by the embodiments of the present invention are described in detail below with specific examples.

[0071] Example 1

[0072] like Figure 1 As shown, this embodiment provides a method for predicting the thickness of deposited films of integrated circuits, comprising the following steps:

[0073] Step S1, preprocessing the deposited thin film data derived from the thin film deposition equipment, such as deleting missing values, screening data of missing processes, and normalizing the data;

[0074] In step S1, the data with missing values ​​and missing processes are deleted, and the data set is normalized due to the influence of the magnitude difference between features. The formula of the conversion function is as follows:

[0075]

[0076] Where, X * Represents the normalized result; X represents the sample data, Max represents the maximum value of the sample data, and Min represents the minimum value of the sample data.

[0077] Step S2: Establishing a feature screening rule for eigenvalues, i.e., deleting variables by comprehensively considering the variance and the linearity between features;

[0078] In step S2, see Figure 2 、 Figure 3 and Figure 4 , the feature screening rule when processing features, that is, to delete variables by comprehensively considering the variance and the linearity between features, Figure 2 The box plot analysis was used to delete features with a variance less than 0.015. The Pearson correlation coefficient was used to measure the linear correlation between independent variables, and a heat map was drawn to find features with a Pearson coefficient greater than 0.9, that is, to find variables with a high degree of linear correlation and delete them.

[0079] In order to compare the feature screening results, the random forest algorithm with n_estimators = 500 was used as the measurement basis, and the determination coefficient R2 and mean square error MSE were used as the evaluation criteria. The R2 values ​​of the following five cases were compared: before screening, only deleting features with too small variance, only deleting features with too large a proportion of outliers, deleting features with too small variance and too large a proportion of outliers, and deleting features with too small variance, too large a proportion of outliers, and Pearson coefficients greater than 0.9. 2 and MSE, see Figure 5 , we found that the best results can be obtained by simultaneously removing features with too small variance, too large proportion of outliers, and Pearson coefficient greater than 0.9.

[0080] The dataset was randomly divided into training, test, and validation sets in a ratio of 7:2:1. The training set was used for model training, the validation set was used for model hyperparameter optimization, and the test set was not involved in the training process and was only used to evaluate the model's predictive performance.

[0081] Step S3: Use the cross-validation-based feature recursive elimination method to further screen the features and eliminate the features that have a negative effect on the performance of the XGBoost model.

[0082] Taking the XGBoost model effect with n_estimators = 500 (number of trees) as the evaluation index, try to delete features one by one. After each feature is deleted, evaluate the film prediction effect of the XGBoost model after training. If R 2 If the value of the feature increases and the MSE decreases, it means that deleting the feature improves the model performance, so it is deleted. Otherwise, it is not deleted. All the features in the final feature set have a positive effect on the performance of the XGBoost model and contain more useful information.

[0083] Step S4: establishing a regression prediction model based on the RFE-CV-XGBoost model, and using the trained regression prediction model to predict the deposited film thickness.

[0084] This example focuses on the application of the RFE-CV-XGBoost model in semiconductor virtual metrology. XGBoost, short for eXtremeGradientBoosting, offers numerous improvements over traditional gradient boosting algorithms for machine learning. It is faster than other ensemble algorithms using gradient boosting and boasts exceptional performance in both classification and regression. RFE-CV-XGBoost follows RFE-CV with further feature screening before performing XGBoost modeling. The features used in modeling are those that positively impact model performance.

[0085] Step S5: Optimize the hyperparameters of the established regression prediction model using a genetic algorithm. Genetic algorithms include three basic genetic operators: selection, crossover, and mutation. Selection and crossover operations can complete most search tasks within the feasible domain, while compilation enables the genetic algorithm to find the global optimal solution.

[0086] This example uses XGBoost based on RFE-CV for training. The actual operation is as follows:

[0087] 1) Data preprocessing, including deletion of missing values, screening and deletion of data with missing processes, data normalization, etc.;

[0088] 2) Feature selection: Variables were deleted by comprehensively considering the variance and the degree of linear correlation between features. By analyzing the box plot, features with a variance less than 0.015 were deleted. The linear correlation between independent variables was measured using the Pearson correlation coefficient, and a heat map was drawn to identify features with a Pearson coefficient greater than 0.9. These were considered to be highly linearly correlated variables and were subsequently deleted.

[0089] 3) Use the recursive feature elimination method based on cross-validation to further screen the features and eliminate the features that have a negative impact on the model performance. Take the XGBoost model effect with n_estimators = 500 (the number of trees) as the evaluation index, try to delete features one by one, and evaluate the film prediction effect after each feature is deleted. If R 2 If the value increases and the MSE decreases, it means that deleting the feature improves the model performance, and it is deleted. Otherwise, it is not deleted. All the features in the final feature set have a positive effect on the performance of the model and contain more useful information.

[0090] After feature screening, we found that the six features that negatively impacted the results were SID_MEAN_SIH4_ON, SIH4_TOP_TIME_SIH4_ON, GSH_MEAN_HEAT_UP, SID_MEAN_DEP, SID_MEAN_DEP, and BUF_PRESSURE_DEP. These represent the reaction dome edge temperature during the silane flow step, the silane concentration at the top of the reaction chamber during the silane flow step, the temperature of the heater surrounding the reaction chamber during the heating step, the dome edge temperature during the deposition step, and the buffer pressure during the deposition step. For pure data modeling, these features negatively impacted model performance.

[0091] 4) Build the XGBoost code. Assume that there are k trees in the ensemble algorithm. The ensemble result of k trees is the accumulation of leaf weights of all previous trees. So let the ensemble result of k trees be Add the leaf weight f on the newly created tree k+1 (x i ), we can get the prediction results of a total of k+1 trees after the k+1th iteration So, we have:

[0092]

[0093] Let this process continue until we find the solution that minimizes the loss function. this It is the prediction result of the model;

[0094] 5) Use genetic algorithm to optimize the hyperparameters of the established regression prediction model, bring in the global optimal hyperparameters obtained by genetic algorithm, and finally obtain the MSE and R of the optimized model. 2 They are significantly better than other optimized traditional machine learning models. The results are shown in the table below, which proves the superiority of this method.

[0095]

[0096] In summary, the method of this embodiment has the following advantages and beneficial effects compared to the prior art:

[0097] By introducing feature screening rules and comprehensively considering variance and linearity between features, we screen out variables with little significance. At the same time, we introduce a feature recursive elimination method based on cross-validation to eliminate variables that have a negative impact on model performance. We also use genetic algorithms to optimize the hyperparameters of the RFE-CV-XGBoost algorithm and find a set of hyperparameters that gives the best model performance. This also significantly improves the efficiency of adjusting the model hyperparameters and reduces the heavy burden on semiconductor measurement staff. The final regression prediction model results are superior to other optimized traditional machine learning algorithms, both in R 2 They will perform better on MSE.

[0098] Example 2

[0099] This embodiment further provides an integrated circuit deposition film thickness prediction system, comprising:

[0100] a parameter acquisition module, in communication with the integrated circuit thin film deposition apparatus, for determining process parameters involved in the integrated circuit thin film deposition process, the process parameters including setting parameters of the apparatus used in the integrated circuit thin film deposition process and external material supply parameters for the integrated circuit deposition process;

[0101] a database construction module, in communication with the parameter acquisition module, for establishing a database of various process parameters involved in the integrated circuit thin film deposition process based on the process parameters acquired by the parameter acquisition module;

[0102] A data preprocessing module, in communication with the database construction module, is used to remove missing values, filter and delete data of missing processes, and normalize the data;

[0103] A feature screening module, in communication with the data preprocessing module, is used to delete redundant features, i.e., to streamline features to improve prediction accuracy;

[0104] A cross-validation-based feature recursive elimination module is in communication with the feature screening module, i.e., the RFE-CV module is used to remove features that have a negative effect on the prediction results;

[0105] A prediction model building module is communicatively connected to the cross-validation-based feature recursive elimination module, and is used to establish a regression prediction model based on RFE-CV-XGBoost, and use the trained regression prediction model to predict the deposited film thickness of the integrated circuit;

[0106] The genetic algorithm hyperparameter optimization module is in communication with the prediction model building module and is used to optimize various hyperparameters related to model performance in RFE-CV-XGBoost, find a set of hyperparameters that makes the final model prediction result perform best, introduce them into the model, and obtain the final RFE-CV-XGBoost model with the best prediction effect.

[0107] The integrated circuit deposition film thickness prediction system provided in this embodiment can execute an integrated circuit deposition film thickness prediction method based on RFE-CV-XGBoost provided in the method embodiment of the present invention, can execute any combination of implementation steps of the method embodiment, and has the corresponding functions and beneficial effects of the method.

[0108] The present invention combines genetic algorithms and cross-validation methods to realize the screening of hyperparameters of the RFE-CV-XGBoost prediction model, quickly finds a set of hyperparameters that makes the model reach the optimal solution, and reduces the workload of researchers in manually adjusting parameters. The prediction method and system proposed in the present invention can be widely used in multiple processes of semiconductor thin film deposition, such as plasma enhanced chemical vapor deposition, low pressure chemical vapor deposition and ultra-high vacuum chemical vapor deposition. By inputting the process data in the production process into the thin film prediction system, the thickness of the deposited film can be automatically predicted, avoiding complex physical measurements. The film thickness prediction results can also be used as a reference for the subsequent chemical mechanical polishing grinding parameter setting. Through data modeling, the present invention saves the film thickness measurement step after the thin film deposition process, shortens the production cycle of semiconductor products, saves measurement cost expenditure, and provides theoretical guidance for technical parameter adjustment of the thin film deposition process.

[0109] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0110] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims. The information disclosed in the background technology section of this article is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as an admission or any form of implication that the information constitutes prior art already known to those skilled in the art.

Claims

1. A method for predicting the thickness of deposited films of integrated circuits, characterized in that: The following steps are involved: Step S1, preprocessing the deposited thin film data in the database to obtain preprocessed deposited thin film data; Step S2, screening the characteristic values ​​contained in the deposited thin film data pre-processed in step S1 according to characteristic screening rules to obtain screened characteristics; Step S3, processing the filtered features obtained in step S2 by cross-validation feature recursive elimination to delete features that have a negative impact on the performance of the XGBoost model; Step S4, establishing a regression prediction model based on the XGBoost model, and using the features obtained in step S3 as a training data set to train and test the regression prediction model; Step S5, using a genetic algorithm to optimize the hyperparameters of the regression prediction model trained and tested in step S4, finding a set of hyperparameters that makes the regression prediction model perform best from all the hyperparameters of the regression prediction model, and using the optimized final regression prediction model to predict the thickness of the deposited film of the integrated circuit, including: using a genetic optimization algorithm to optimize the number and depth of trees formed by all the hyperparameters of the regression prediction model after training and testing through a verification set, and finding a set of hyperparameters that makes the regression prediction model perform best from all the hyperparameters of the regression prediction model, and using the hyperparameters of the final regression prediction model.

2. The method for predicting the thickness of deposited films of integrated circuits according to claim 1, wherein: In step S1, the deposited thin film data in the database is preprocessed in the following manner, including: Stored in the database are process parameters involved in the integrated circuit thin film deposition process as deposited film data; The deposited film data with missing values ​​and missing processes in the database are deleted, and the data set consisting of the deposited film data obtained after deletion is normalized. The formula of the conversion function is: ； Where, represents the normalized result; X represents the sample data of the deposited thin film data, Max represents the maximum value of the sample data of the deposited thin film data, and Min represents the minimum value of the sample data of the deposited thin film data.

3. The method for predicting the thickness of deposited films of integrated circuits according to claim 1 or 2, wherein: In step S2, the characteristic values ​​contained in the deposited thin film data pre-processed in step S1 are screened according to a predetermined characteristic screening rule in the following manner, including: Remove features with variance less than 0.015 and remove features with Pearson correlation coefficient greater than 0.

9.

4. The method for predicting the thickness of deposited films of integrated circuits according to claim 1 or 2, wherein: In step S3, the filtered features obtained in step S2 are processed by cross-validated feature recursive elimination in the following manner to delete features that have a side effect on model performance, including: The filtered features obtained in step S2 are processed by a feature recursive elimination algorithm based on cross-validation. The performance of the XGBoost model with 7-fold cross-validation is used as the evaluation index. The features are deleted one by one. If the mean square error (MSE) of the XGBoost model decreases after deleting the features and the coefficient of determination decreases, the feature is deleted. If it rises, the feature is deleted, otherwise it is not deleted.

5. The method for predicting the thickness of deposited films of integrated circuits according to claim 1 or 2, wherein: In step S4, a regression prediction model is established based on the XGBoost model in the following manner, and the features obtained in step S3 are used as a training data set to train and test the regression prediction model, including: The data set formed by the features obtained in step S3 is randomly divided into a training set, a test set and a validation set in a ratio of 7:2:1; the regression prediction model is trained using the training set; and the prediction performance of the trained regression prediction model is evaluated using the test set.

6. The method for predicting the thickness of deposited films of integrated circuits according to claim 5, wherein: In step S4, in the data set formed by the features obtained in step S3, the features related to the deposition, ventilation, and heating steps in the deposited film data are used as input parameters of the regression prediction model; The output parameter of the regression prediction model is the predicted result of the deposited film thickness.

7. A system for predicting the thickness of deposited films of integrated circuits, for implementing the method according to any one of claims 1 to 6, characterized in that: include: A data preprocessing module can preprocess the deposited film data obtained from the database to obtain preprocessed deposited film data; a feature screening module, which is in communication with the data preprocessing module and can screen the feature values ​​contained in the deposited thin film data preprocessed by the data preprocessing module according to a predetermined feature screening rule to obtain screened features; A cross-validated feature recursive elimination module is in communication with the feature screening module and is capable of processing the filtered features output by the feature screening module through a cross-validated feature recursive elimination method to delete features that have a negative impact on the performance of the XGBoost model; A prediction model construction module is respectively connected to the cross-validation-based feature recursive elimination module and the hyperparameter optimization module, capable of establishing a regression prediction model based on the XGBoost model, and using the features output by the cross-validation-based feature recursive elimination module as a training data set to train and test the regression prediction model; A hyperparameter optimization module is communicatively connected to the prediction model construction module and can use a genetic algorithm to perform hyperparameter optimization on the trained and tested regression prediction model established by the prediction model construction module, and find a set of hyperparameters that makes the regression prediction model perform best from all the hyperparameters of the regression prediction model, and use them as the hyperparameters of the final regression prediction model. The final regression prediction model is used to predict the thickness of the deposited film of the integrated circuit.

8. The system for predicting the thickness of deposited films of integrated circuits according to claim 7, wherein: Also includes: a parameter acquisition module, communicatively connected to the thin film deposition equipment, capable of acquiring process parameters involved in the integrated circuit thin film deposition process as deposited film data, the process parameters including setting parameters of the thin film deposition equipment used in the integrated circuit thin film deposition process and external material supply parameters for the integrated circuit deposition process; The database construction module is respectively communicated with the parameter acquisition module and the data preprocessing module, and can establish a database based on the process parameters involved in the integrated circuit thin film deposition process obtained by the parameter acquisition module as deposited thin film data, and provide the deposited thin film data to the data preprocessing module.

Images