An integrated circuit deposition film thickness prediction method based on HHO-stacking ensemble learning

Abstract

The present invention provides an integrated circuit deposition film thickness prediction method based on HHO-stacking ensemble learning, including: the first step: extracting an integrated circuit deposition film feature data set; the second step: performing normalization processing on the data; the third step: using the Harris hawk optimization (HHO) algorithm to perform feature screening on the normalized data set and selecting the optimal feature set; the fourth step: establishing a stacking ensemble learning model; the fifth step: using grid search to optimize the hyperparameters of each learner in the stacking ensemble learning; the sixth step: using the optimized stacking ensemble learning model to predict the integrated circuit deposition film thickness. The present invention helps to predict the quality conditions of each key stage, thereby ensuring the product yield.

Description

Technical Field

[0001] The present invention belongs to the field of semiconductor integrated circuit manufacturing, and relates to a method for predicting the deposition film thickness of an integrated circuit based on HHO-stacking ensemble learning. Background Art

[0002] A chip is composed of several layers of thin films stacked together. Thin film deposition is an essential step in the integrated circuit manufacturing process. The thickness of the integrated circuit thin film refers to the thickness of a layer of thin film used to cover circuit elements, and its thickness has a crucial impact on the performance and stability of the integrated circuit. Therefore, in the integrated circuit manufacturing process, it is particularly important to add a measurement link after key processes to inspect the quality situation.

[0003] Currently, in the prediction in the field of integrated circuit manufacturing, it is relatively common to use virtual metrology to judge the results of the manufacturing process or device performance. Feature screening can eliminate useless or interfering features. Currently, the filtering method is mostly used for feature screening, but this method cannot guarantee finding the best feature set in the feature selection problem and will consume too much time, and the prediction effect cannot meet the target requirements.

[0004] With the development of the AI industry, using machine learning methods to perform regression prediction on the results of each process saves time and effort and thus has been widely applied. At the same time, compared with traditional feature screening methods, swarm intelligence optimization algorithms simulate the habits of animal populations, which can not only obtain better results but also not consume too much computing time. The Harris Hawks Optimization (HHO) algorithm is a novel swarm intelligence optimization algorithm, and research has proven its excellent performance. At the same time, stacking ensemble learning is a novel ensemble learning method that can integrate multiple machine learning methods, enhancing the diversity of the methods used and thus improving the prediction effect. Summary of the Invention

[0005] (1) Technical Problems to be Solved

[0006] The technical problem to be solved by the present invention is to provide a method for predicting the deposition film thickness of an integrated circuit based on HHO-stacking ensemble learning with high prediction accuracy in view of the above-mentioned defects in the prior art.

[0007] (2) Technical Solutions

[0008] In order to achieve the above technical objectives, the main technical solutions adopted by the present invention include:

[0009] In the first aspect, the present invention provides a method for predicting the deposition film thickness of an integrated circuit based on HHO-stacking ensemble learning, including:

[0010] The first step: Extract the integrated circuit deposition film feature data set;

[0011] Second step: Normalize each data in the integrated circuit deposition film feature dataset to obtain the normalized integrated circuit deposition film feature dataset;

[0012] Third step: Use the Harris Hawks Optimization (HHO) algorithm to perform feature screening on the normalized dataset and select the optimal feature set;

[0013] Fourth step: Randomly divide all the data in the obtained integrated circuit deposition film feature dataset into a training set and a test set according to a preset ratio, and establish a stacked ensemble learning model;

[0014] Fifth step: Use grid search to optimize the hyperparameters of each learner in the stacked ensemble learning to obtain the integrated circuit deposition film thickness prediction model with optimized hyperparameters;

[0015] Sixth step: Use the optimized stacked ensemble learning model to predict the integrated circuit deposition film thickness.

[0016] A method for predicting the integrated circuit deposition film thickness based on HHO-stacked ensemble learning proposed by the present invention, according to all process parameters of the integrated circuit thin film deposition process, through the Harris Hawks Optimization (HHO) algorithm and the stacked ensemble learning model, performs regression prediction on the process results of each key stage in the integrated circuit thin film deposition, so as to ensure the quality of each stage and ultimately ensure the performance and stability of the integrated circuit. Among them, the normalization process maps the extracted integrated circuit deposition film feature data to the same scale, standardizes the data, thereby reducing the calculation time and ensuring the prediction accuracy; taking advantage of the strong optimization ability, simple principle, few parameters and high accuracy of the Harris Hawks Optimization (HHO) algorithm, perform feature screening on the normalized integrated circuit deposition film thickness feature dataset, filter out redundant feature variables from it, and select the optimal feature set; the model hyperparameters are the external configurations of the model, and their values cannot be estimated from the data. Use grid search to optimize the hyperparameters of each learner in the stacked ensemble learning to find the values that can make the model converge as soon as possible and obtain the best accuracy.

[0017] Preferably, the training set is used for model training, the test set is used to evaluate the prediction performance of the model, and the training set and the test set can be preset with corresponding ratios.

[0018] Among them, before constructing the stacked ensemble learning model, the normalized data is randomly divided into a training set and a test set according to a preset ratio. Use the data in the training set to train the model and use the test set to evaluate the prediction performance of the model to prevent the data model from being too complex and having too many parameters, thereby reducing the error.

[0019] Preferably, the stacked ensemble learning model has a two-layer structure, including a base learner and a meta-learner. The base learner includes RF, XGBoost, SVM, and LightGBM, and the meta-learner includes a CatBoost model. The first layer selects the base learner, and the output results of these base learners are used to form a new training set, which is input into the meta-learner of the second layer and selected as the meta-learner of the second layer.

[0020] Among them, stacked ensemble learning draws on the advantages of each model used, synthesizes a model with better model performance and prediction ability, and obtains the final prediction result, thereby improving the accuracy of regression prediction.

[0021] Preferably, in the stacked ensemble learning model, the evaluation metrics for hyperparameter optimization include selecting the mean squared error MSE or / and the R2 score.

[0022] The formula for the mean squared error MSE is:

[0023]

[0024] where y i is the true value of the i-th sample, is the predicted value of the i-th sample, and n is the total number of samples;

[0025] R 2 The formula for the score is:

[0026]

[0027] where y i is the true value of the i-th sample, is the predicted value of the i-th sample, is the average of n true values.

[0028] Among them, the mean squared error MSE measures the matching degree between the predicted value and the true value y i Generally, the smaller the mean squared error MSE, the better; the R 2 score is the coefficient of determination, which is used to evaluate the performance of the regression model. Generally, the higher the R 2 score, the better the model's explanatory results.

[0029] (III) Beneficial Effects

[0030] The beneficial effects of the present invention are as follows: The present invention uses the novel and performance-excellent HHO swarm intelligence optimization algorithm for feature screening to filter out redundant feature variables, select the optimal feature set, and proposes a stacking ensemble learning method with RF, XGBoost, SVM, and LightGBM as base learners and CatBoost as the meta-learner for regression prediction. Compared with traditional single models, the multiple high-performance models used in the present invention improve the prediction accuracy, thereby ensuring the quality of each key stage and the product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 FIG. schematically shows a flowchart of an integrated circuit deposition film thickness prediction method based on HHO-stacking ensemble learning according to the present invention;

[0032] Figure 2 FIG. is a stacking ensemble learning structure diagram. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] In order to better explain the present invention for easy understanding, the present invention will be described in detail below with reference to the accompanying drawings through specific embodiments.

[0034] In order to better understand the above technical solutions, the exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more clear and thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0035] Embodiment 1

[0036] Referring to Figure 1 , the present invention provides an integrated circuit deposition film thickness prediction method based on HHO-stacking ensemble learning, including:

[0037] The first step: Extract the integrated circuit deposition film feature data set;

[0038] The second step: Normalize each data in the integrated circuit deposition film feature data set to obtain a normalized integrated circuit deposition film feature data set;

[0039] The third step: Use the HHO Harris hawk optimization algorithm to perform feature screening on the normalized data set and select the optimal feature set;

[0040] Fourth step: Randomly divide all the data in the obtained pooled integrated circuit deposition film feature dataset into a training set and a test set according to a preset ratio; among them, the training set is used for model training, and the test set is only used to evaluate the prediction performance of the model; for example, the preset ratio corresponding to the training set and the test set is 3:1;

[0041] And establish a stacked ensemble learning model. The stacked ensemble learning model has a two-layer structure, including a base learner and a meta-learner. The base learner includes RF, XGBoost, SVM, and LightGBM, and the meta-learner includes the CatBoost model. The first layer is selected as the base learner, and the output results of these base learners are used to form a new training set, which is input into the meta-learner in the second layer, and the CatBoost model is selected as the meta-learner in the second layer.

[0042] Refer to Figure 2 , this step uses a stacking strategy. The output of the first-layer base learner is used as a feature for training through the second-layer meta-learner, and then the results of all the first-layer base learners are integrated and output to obtain the final prediction result, which greatly improves the model prediction effect.

[0043] Among them, RF is the random forest algorithm, which is an improved method proposed for the overfitting problem of decision trees. Its training can be highly parallelized, the variance of the trained model is small, and the generalization ability is strong; XGBoost is the extreme gradient boosting tree, and it has very good prediction effects for classification or regression problems; SVM is the support vector machine, which is a discriminant method and is a supervised learning model in the field of machine learning, used for pattern recognition, classification, and regression analysis; LightGBM is a gradient boosting framework, which has the characteristics of faster training speed, lower memory usage, and the ability to handle large-scale data; CatBoost is a gradient boosting tree algorithm, which has an adaptive learning rate and can handle categorical features, can handle classification and regression problems, and performs well on multiple datasets.

[0044] Fifth step: Use grid search to optimize the hyperparameters of each learner in the stacked ensemble learning to obtain an integrated circuit deposition film thickness prediction model with optimized hyperparameters; Select the mean squared error MSE or / and R2 score as the evaluation index for hyperparameter optimization, where:

[0045] The mean squared error MSE is as follows:

[0046]

[0047] Among them, y i is the true value of the i-th sample, is the predicted value of the i-th sample, and n is the total number of samples;

[0048] The formula for the R2 score is as follows:

[0049]

[0050] where y i is the true value of the i-th sample, is the predicted value of the i-th sample, is the average value of n true values;

[0051] Sixth step: Use the optimized stacked ensemble learning model to predict the deposition film thickness of the integrated circuit.

[0052] The present invention proposes a method for predicting the deposition film thickness of an integrated circuit based on HHO-stacked ensemble learning. Using this method, the process results of each stage of the integrated circuit thin film deposition can be regressively predicted, thereby ensuring the product qualification rate. When using the HHO algorithm for feature screening, this method is superior to the traditional method in terms of the speed of obtaining the optimal set, and when finally using the optimized stacked ensemble learning model for prediction, the prediction accuracy is improved.

[0053] Those skilled in the art should understand that those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention should also include these modifications and variations.

[0054] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality" means two or more unless otherwise specifically defined.

[0055] In the present invention, unless otherwise clearly defined and limited, the terms "installation", "connection", "connection", "fixation", etc. should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or integrated; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium; it can be the internal communication of two components or the interaction relationship between two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0056] In the present invention, unless otherwise clearly specified or limited, a first feature being "on" or "under" a second feature may mean that the first and second features are in direct contact, or that the first and second features are indirectly in contact via an intermediate medium. Further, a first feature being "above", "over" or "on top of" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature has a higher level of height than the second feature. A first feature being "under", "below" or "beneath" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature has a lower level of height than the second feature.

[0057] In the description of this specification, the descriptions of terms such as "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples", etc., mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0058] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art may make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An integrated circuit deposition film thickness prediction method based on HHO-stacked ensemble learning, characterized in that Including: The first step: Extract the integrated circuit deposition film feature dataset; The second step: Normalize each data in the integrated circuit deposition film feature dataset to obtain the normalized integrated circuit deposition film feature dataset; The third step: Use the Harris Hawks Optimization (HHO) algorithm to perform feature screening on the normalized dataset and select the optimal feature set; The fourth step: Randomly divide all the data in the obtained integrated circuit deposition film feature dataset into a training set and a test set according to a preset ratio, and establish a stacked ensemble learning model; The fifth step: Use grid search to optimize the hyperparameters of each learner in the stacked ensemble learning to obtain the integrated circuit deposition film thickness prediction model with optimized hyperparameters; The sixth step: Use the optimized stacked ensemble learning model to predict the integrated circuit deposition film thickness; The stacked ensemble learning model has a two-layer structure, including a base learner and a meta-learner. The base learner includes RF, XGBoost, SVM, and LightGBM, and the meta-learner includes a CatBoost model. The first layer selects the base learners, and the output results of these base learners form a new training set, which is input into the meta-learner of the second layer, and the CatBoost model is selected as the meta-learner of the second layer.

2. The method for predicting the deposition film thickness of an integrated circuit based on HHO-stacked ensemble learning according to claim 1, wherein: The training set is used for model training, and the test set is used to evaluate the prediction performance of the model. The training set and the test set can be preset with corresponding ratios.

3. The method for predicting the deposition film thickness of an integrated circuit based on HHO-stacked ensemble learning according to claim 1, characterized in that: In the stacked ensemble learning model, the evaluation metrics for hyperparameter optimization include selecting the mean squared error (MSE) or / and the R2 score. The formula for the mean squared error (MSE) is: Where, is the true value of the i-th sample, is the predicted value of the i-th sample, n is the total number of samples; R 2 The formula for the fraction is: Where, is the true value of the i-th sample, is the predicted value for the i-th sample, is the average of n true values.

Images