Training method, prediction method, storage medium and terminal for predicting the distribution of polishing material removal rate

Abstract

The present invention discloses a polishing material removal rate distribution prediction model training method, prediction method, storage medium and terminal, wherein the training method comprises: obtaining multiple polishing material removal rate distribution data, performing dimensionality reduction processing on each polishing material removal rate distribution data respectively through a reversible dimensionality reduction method to obtain distribution key features; performing correlation analysis on each parameter in a polishing process and polishing pad physical property parameter set respectively with the distribution key features of the polishing material removal rate distribution data to obtain high-correlation process and physical property parameters of the polishing material removal rate distribution data; integrating the high-correlation process and physical property parameters of the polishing material removal rate distribution data with the distribution key features into a training data set, and training a preset network model based on the training data set to obtain a polishing material removal rate distribution prediction model. The present invention adopts a machine learning model, performs learning and prediction based on a large amount of actual data, and improves the scientificity and accuracy of decision-making.

Description

Technical Field

[0001] The present invention relates to the technical field of semiconductor integrated circuit manufacturing, and in particular to a polishing material removal rate distribution prediction model training method, a prediction method, a storage medium and a terminal. Background Art

[0002] The flatness of silicon wafers is a key quality standard in the field of wafer manufacturing. It is directly related to the consistency of subsequent process steps and the stability of chip performance. Since the initial shape of silicon wafers is difficult to adjust, their flatness is mainly determined by the material removal rate (MRR) distribution during the polishing stage.

[0003] In this process, the MRR distribution is mainly affected by the polishing process parameters, polishing pad and polishing liquid. Therefore, it is very important to select the appropriate polishing pad and polishing liquid according to the initial shape of the silicon wafer and design a complete set of process parameters for it. The traditional parameter optimization method relies on manually designed experiments (Design of Experiments, DOE), manually analyzing experimental data, summarizing rules, and then adjusting parameters. However, due to the large selection space of process, polishing pad and polishing liquid parameters and the complex influencing mechanism, the manual method can only provide preliminary guidance, and there are obvious challenges in efficiency and quantitative control, and the optimization effect on MRR distribution is limited. Summary of the invention

[0004] The purpose of this application is to provide a polishing material removal rate distribution prediction model training method, prediction method, storage medium and terminal, which are used to solve the problem that the traditional polishing material removal rate distribution can only be optimized by manually adjusting parameters, which has limitations and has limited optimization effect on MRR distribution.

[0005] In a first aspect, the present application provides a polishing material removal rate distribution prediction model training method, comprising:

[0006] Acquire multiple pieces of polishing material removal rate distribution data, and perform dimensionality reduction processing on each piece of polishing material removal rate distribution data by a reversible dimensionality reduction method to obtain key distribution features of the polishing material removal rate distribution data;

[0007] Perform correlation analysis on each parameter in the polishing process and polishing pad physical property parameter set and the key distribution characteristics of the polishing material removal rate distribution data to obtain highly correlated process and physical property parameters of the polishing material removal rate distribution data;

[0008] The highly correlated process and physical property parameters of the polishing material removal rate distribution data and the key distribution features are integrated into a training data set, and a preset network model is trained based on the training data set to obtain a polishing material removal rate distribution prediction model.

[0009] In one embodiment of the present application, each polishing material removal rate distribution data is cleaned and standardized, and the cleaning and standardization of the polishing material removal rate distribution data includes:

[0010] The polishing material removal rate distribution data includes a plurality of data points arranged in ascending order of radius. The data point with a radius of zero in the polishing material removal rate distribution data is subtracted from each data point in the polishing material removal rate distribution data, and the data excluding the data point with a radius of zero in the polishing material removal rate distribution data is formed into a data matrix.

[0011] In one embodiment of the present application, performing dimensionality reduction processing on each polishing material removal rate distribution data by a reversible dimensionality reduction method includes:

[0012] A bidirectional converter is constructed by using a principal component analysis method, and a dimension reduction process is performed on each polishing material removal rate distribution data by using the bidirectional converter;

[0013] The bidirectional converter is also used to convert the distribution key features predicted by the polishing material removal rate distribution prediction model into polishing material removal rate distribution data.

[0014] In one embodiment of the present application, during the dimensionality reduction processing of each polishing material removal rate distribution data by the bidirectional converter, when greater than or equal to 90% of the variance in all polishing material removal rate distribution data is explained, the corresponding number of principal components selected is the dimensionality reduction dimension of the polishing material removal rate distribution data.

[0015] In one embodiment of the present application, a Pearson correlation coefficient analysis method is used to perform correlation analysis between each parameter in the polishing process and polishing pad physical property parameter set and the distribution key characteristics of the polishing material removal rate distribution data.

[0016] In one embodiment of the present application, the correlation analysis of the distribution key features of the target parameters and the polishing material removal rate distribution data in the polishing process and polishing pad physical property parameter set is performed by the Pearson correlation coefficient analysis method, including:

[0017] Calculating the Pearson correlation coefficient between the target parameter and each key feature in the distribution key features of the polishing material removal rate distribution data to obtain the correlation coefficient value corresponding to each key feature, multiplying each correlation coefficient value by the explained variance ratio corresponding to the key feature and summing the results to obtain the correlation value between the target parameter and the distribution key features of the polishing material removal rate distribution data;

[0018] Determine whether the correlation value is greater than a preset component threshold value, if so, determine that the target parameter is a highly correlated process and physical property parameter of the polishing material removal rate distribution data, otherwise determine that the target parameter is not a highly correlated process and physical property parameter of the polishing material removal rate distribution data;

[0019] The target parameter is any parameter in the set of polishing process and polishing pad physical property parameters; the explained variance ratio corresponding to the correlation coefficient value is the explained variance ratio of the key feature corresponding to the correlation coefficient value in the polishing material removal rate distribution data.

[0020] In one embodiment of the present application, the preset network model is a multi-task ridge regression model.

[0021] In a second aspect, the present application provides a polishing material removal rate distribution prediction method, comprising:

[0022] Obtaining the polishing process to be predicted and the physical property parameters of the polishing pad, and inputting the polishing process to be predicted and the physical property parameters of the polishing pad into the polishing material removal rate distribution prediction model obtained by the polishing material removal rate distribution prediction model training method to obtain the predicted distribution key features;

[0023] The predicted distribution key features are converted into predicted polishing material removal rate distribution data.

[0024] In a third aspect, the present application provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the polishing material removal rate distribution prediction model training method or the polishing material removal rate distribution prediction method.

[0025] In a fourth aspect, the present application provides a terminal, comprising: a processor and a memory, wherein the memory is communicatively connected to the processor;

[0026] The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory, so that the terminal performs the polishing material removal rate distribution prediction model training method or the polishing material removal rate distribution prediction method.

[0027] Compared with the prior art, one or more embodiments of the above scheme may have the following advantages or beneficial effects:

[0028] The polishing material removal rate distribution prediction model training method provided by the embodiment of the present invention is applied to perform dimensionality reduction processing on the polishing material removal rate distribution data by a reversible dimensionality reduction method to obtain the distribution key features of the polishing material removal rate distribution data, and to perform correlation analysis on the parameters and distribution key features in the polishing process and polishing pad physical property parameter set to obtain the high correlation process and physical property parameters of the polishing material removal rate distribution data; the training model is trained based on the high correlation process and physical property parameters and distribution key features to obtain a polishing material removal rate distribution prediction model that can predict the distribution key features based on the parameters, and the distribution key features can be further converted into polishing material removal rate distribution data. A high-precision prediction of the MRR distribution after silicon wafer polishing is achieved, with an accuracy of nanometers. The method can also achieve control and quantitative optimization of process parameters and polishing pads at the production line end, thereby meeting the high flatness requirements on the production line and optimizing production efficiency and product quality. The present invention adopts a machine learning model to learn and predict based on a large amount of actual data, thereby improving the scientificity and accuracy of decision-making. This data-driven method can more accurately understand the complex mechanisms in the material removal process, thereby achieving more precise control and optimization.

[0029] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0031] Figure 1 Shown is a flow chart of the polishing material removal rate distribution prediction model training method described in an embodiment of the present application.

[0032] Figure 2 Shown is a process schematic diagram of the polishing material removal rate distribution prediction model training method described in an embodiment of the present application.

[0033] Figure 3 Shown is a schematic diagram of the process of a polishing material removal rate distribution prediction model training method described in an embodiment of the present application.

[0034] Figure 4 Shown is a schematic flow chart of a polishing material removal rate distribution prediction method according to an embodiment of the present application.

[0035] Figure 5 Shown is a schematic diagram of the structure of the terminal described in an embodiment of the present application. DETAILED DESCRIPTION

[0036] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0037] It should be noted that the illustrations provided in the following embodiments are only used to illustrate the basic concept of the present application in a schematic manner. Therefore, the drawings only show components related to the present application rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0038] The following embodiments of the present application provide a polishing material removal rate distribution prediction model training method, prediction method, storage medium and terminal, which are used to solve the problem that the traditional polishing material removal rate distribution can only be optimized by manually adjusting parameters, which has limitations and has limited optimization effect on MRR distribution.

[0039] like Figure 1-3 As shown, this embodiment provides a polishing material removal rate distribution prediction model training method, including the following steps.

[0040] Step S101 , obtaining a plurality of polishing material removal rate distribution data, and performing dimensionality reduction processing on each polishing material removal rate distribution data by a reversible dimensionality reduction method to obtain distribution key features of the polishing material removal rate distribution data.

[0041] Specifically, a plurality of polishing material removal rate distribution data are obtained, each of which includes a plurality of data points, and the plurality of data points of each polishing material removal rate distribution data are arranged in order from small to large radius. For example, there are n polishing material removal rate distribution data, each of which includes 750 data points, that is, X∈R n×750 , and r=0 represents the central data point with a radius of zero in each polishing material removal rate distribution data, then r=1, 2, …, 749 represent the data sorted from the center to the edge position.

[0042] In order to eliminate the baseline offset of data at different radii, this embodiment also cleans and standardizes each piece of polishing material removal rate distribution data obtained. Specifically, cleaning and standardizing each piece of polishing material removal rate distribution data includes: subtracting the data point with a radius of zero in the polishing material removal rate distribution data from each data point in the polishing material removal rate distribution data, and forming a data matrix with the data from which the data point with a radius of zero in the polishing material removal rate distribution data is subtracted. Specifically, it can be expressed as:

[0043] x i ' j =x ij -x i0 for all i=1,2,…,n andj=0,1,2,…,749 (1)

[0044] where x ij is the MRR value of the i-th data at the radius position j, x i0 is the MRR value of the i-th data at a radius of 0, that is, the MRR value of the center point.

[0045] This embodiment uses principal component analysis to construct a bidirectional converter, which can extract key features from the polishing material removal rate distribution data, that is, convert the high-dimensional polishing material removal rate distribution data into a low-dimensional representation containing the main information; at the same time, these features can also be converted back to the polishing material removal rate distribution data. This is the basis for achieving accurate prediction and optimization. This process not only simplifies the data structure, but also retains the most critical information for the prediction task.

[0046] Further, the bidirectional converter constructed by the principal component analysis method performs dimensionality reduction processing on each polishing material removal rate distribution data to obtain the key distribution characteristics of the polishing material removal rate distribution data. Specifically, the bidirectional converter constructed by the principal component analysis method performs dimensionality reduction processing on each polishing material removal rate distribution data to obtain the key distribution characteristics of the polishing material removal rate distribution data. Specifically, it is assumed that all the polishing material removal rate distribution data after cleaning and standardization form a data matrix X, and then the covariance matrix of the data matrix X is calculated. The covariance matrix calculation formula is as follows:

[0047]

[0048] Where X' T is the transpose of X'.

[0049] Then perform eigenvalue decomposition on the covariance matrix S to obtain eigenvalues ​​λ1,λ2,…λ 750 and the corresponding eigenvectors v1,v2,…v 750 :

[0050] Svk =λ k v k for k=1,2,…,750 (3)

[0051] Eigenvector v k The kth principal component direction is taken as the kth principal component direction. The explained variance ratio ratiok of the kth principal component is the kth eigenvalue divided by the sum of all eigenvalues, and the calculation formula is as follows:

[0052]

[0053] The larger the eigenvalue, the higher the explained variance ratio. Then according to the cumulative explained variance ratio Select the most important p principal components and calculate the projection of each data on these principal components (that is, the data matrix and principal component score matrix after reducing to P dimensions). The calculation formula is as follows:

[0054] TX′V p (5)

[0055] Where V p is the matrix V containing the first p eigenvectors p ∈R 750×p ,T is the principal component score matrix, that is, the data matrix after reducing to p dimensions, T∈R n×p .

[0056] When the selected polishing material removal rate distribution data is greater than or equal to 90% of the variance is explained, that is, The corresponding number of selected principal components p is the dimension reduction dimension of the polishing material removal rate distribution data. For example, when p = 6, more than 90% of the variance in the polishing material removal rate distribution data can be explained. Therefore, we believe that the original 750-dimensional data can be effectively reduced to 6 dimensions, while retaining the most critical information in the data and simplifying the subsequent analysis and calculation process. After that, all the polishing material removal rate distribution data are reduced to 6-dimensional data to obtain the key distribution characteristics of the polishing material removal rate distribution data.

[0057] It should be noted that when selecting the dimension reduction of the polishing material removal rate distribution data, it can also be set that when the variance in the polishing material removal rate distribution data is greater than or equal to other appropriate percentage values, the corresponding number of principal components p selected is the dimension reduction dimension of the polishing material removal rate distribution data.

[0058] Step S102 , performing correlation analysis on each parameter in the polishing process and polishing pad physical property parameter set and the distribution key characteristics of the polishing material removal rate distribution data to obtain highly correlated process and physical property parameters of the polishing material removal rate distribution data.

[0059] This embodiment uses the Pearson correlation coefficient analysis method to determine which parameters in the polishing process and polishing pad physical property parameter set have a significant correlation with the distribution key characteristics of the polishing material removal rate distribution data.

[0060] Assume that MRR is defined actual ∈R n×750 is the MRR set after n polishing material removal rate distribution data are processed by formula (1). The feature vector obtained by the bidirectional transformer constructed by principal component analysis is reduced in dimension to obtain the predicted target Y∈R n×6 , where n is the number of data items, T k represents the kth principal component, k = 1, 2, ..., 6, C is the set of all possible polishing process and polishing pad physical property parameters, and the total number of parameter sets is M, C j represents the jth parameter, j = 1, 2, ..., M. In order to screen the principal component score T k Significantly correlated parameters, calculate each parameter C j With T k Pearson correlation coefficient r kj :

[0061]

[0062] Among them, T ki represents the kth principal component of the i-th MRR distribution data, C ji is the jth polishing parameter corresponding to the i-th MRR distribution data, It is T ki The mean of It is C ji The mean of .

[0063] Assuming that the target parameter is any parameter in the set of polishing process and polishing pad physical property parameters, the target parameter in the set of polishing process and polishing pad physical property parameters is subjected to correlation analysis with the distribution key features of the polishing material removal rate distribution data by the Pearson correlation coefficient analysis method. Specifically, the Pearson correlation coefficient of each key feature in the distribution key features of the target parameter and the polishing material removal rate distribution data is calculated by formula (6) to obtain the correlation coefficient value corresponding to each key feature. Then, each correlation coefficient value is multiplied by the explained variance ratio corresponding to the key feature, and then all the digits after the multiplication are summed to obtain the correlation value of the target parameter and the distribution key feature of the polishing material removal rate distribution data. The explained variance ratio corresponding to the key feature is the value calculated by formula (4).

[0064] Finally, it is determined whether the correlation value is greater than the preset component threshold. If so, the target parameter is determined to be a highly correlated process and physical property parameter of the polishing material removal rate distribution data. Otherwise, it is determined that the target parameter is not a highly correlated process and physical property parameter of the polishing material removal rate distribution data. In this way, all parameters in the polishing process and polishing pad physical property parameter set with highly correlated process and physical property parameters of the polishing material removal rate distribution data can be determined.

[0065] Step S103, integrating the highly correlated process and physical property parameters of the polishing material removal rate distribution data with the key distribution features into a training data set, and training a preset network model based on the training data set to obtain a polishing material removal rate distribution prediction model.

[0066] After obtaining the highly correlated process and physical property parameters and key distribution characteristics of the polishing material removal rate distribution data, the highly correlated process and physical property parameters and key distribution characteristics of the polishing material removal rate distribution data can be combined into a training data set, and the preset network model can be trained using the training data set to obtain a polishing material removal rate distribution prediction model.

[0067] In this embodiment, the preset network model adopts a multi-task ridge regression model. is the prediction target, D is the selected key parameter subset, and the ridge regression model is expressed as:

[0068]

[0069] Among them, B represents the coefficient matrix, each column corresponds to the coefficient of a key feature after the MRR distribution is reduced in dimension, and E is the error matrix, each column corresponds to the error of a key feature after the MRR distribution is reduced in dimension.

[0070] Calculate the loss function of ridge regression, the formula is as follows:

[0071] L(B)=|Y-DB| 2 +λ|B| 2 (8)

[0072] Among them, λ is the regularization parameter, which controls the strength of the regularization term, |B 2 is the L2 norm of B.

[0073] Solve B by minimizing the loss function, using the analytical solution method, that is:

[0074] B=(D T D+λI) -1 D T Y (9)

[0075] Where I is the identity matrix, D T is the transpose of D.

[0076] It should be noted that the obtained polishing material removal rate distribution prediction model also has the function of converting the key distribution features predicted by the polishing material removal rate distribution prediction model into polishing material removal rate distribution data using a bidirectional converter constructed by principal component analysis.

[0077] The upcoming forecast The formula for restoring the original MRR dimension through the inverse PCA process is:

[0078]

[0079] The root mean square error RSME of the prediction is expressed as:

[0080]

[0081] The polishing material removal rate distribution prediction model training method provided by the embodiment of the present invention performs dimensionality reduction processing on the polishing material removal rate distribution data by a reversible dimensionality reduction method to obtain the distribution key features of the polishing material removal rate distribution data, and performs correlation analysis on the parameters and distribution key features in the polishing process and polishing pad physical property parameter set to obtain the high-correlation process and physical property parameters of the polishing material removal rate distribution data; the training model is trained based on the high-correlation process and physical property parameters and distribution key features to obtain a polishing material removal rate distribution prediction model that can predict the distribution key features based on the parameters, and the distribution key features can be further converted into polishing material removal rate distribution data. A high-precision prediction of the MRR distribution after silicon wafer polishing is achieved, with an accuracy of nanometers. The method can also achieve control and quantitative optimization of process parameters and polishing pads at the production line end, thereby meeting the high flatness requirements on the production line and optimizing production efficiency and product quality. The present invention adopts a machine learning model to learn and predict based on a large amount of actual data, thereby improving the scientificity and accuracy of decision-making. This data-driven method can more accurately understand the complex mechanisms in the material removal process, thereby achieving more precise control and optimization.

[0082] like Figure 4 As shown, this embodiment also provides a polishing material removal rate distribution prediction method, which includes the following steps.

[0083] Step S201 , obtaining the polishing process to be predicted and the physical property parameters of the polishing pad, and inputting the polishing process to be predicted and the physical property parameters of the polishing pad into a polishing material removal rate distribution prediction model to obtain predicted distribution key features.

[0084] Specifically, the polishing process and the polishing pad physical property parameters to be predicted are obtained as the polishing process and the polishing pad physical property parameters to be predicted, and then the polishing process and the polishing pad physical property parameters to be predicted are input into the polishing material removal rate distribution prediction model, which is obtained by the polishing material removal rate distribution prediction model training method. The output of the polishing material removal rate distribution prediction model is used as the predicted distribution key feature.

[0085] Step S202: converting the predicted distribution key features into predicted polishing material removal rate distribution data.

[0086] Specifically, a bidirectional converter constructed by principal component analysis in the polishing material removal rate distribution prediction model is used to convert the predicted distribution key features into predicted polishing material removal rate distribution data.

[0087] The method of the present invention can more accurately understand the complex mechanisms in the material removal process, thereby achieving more precise control and optimization.

[0088] The present application embodiment also provides a computer-readable storage medium. A person of ordinary skill in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the processor through a program, and the program can be stored in a computer-readable storage medium, and the storage medium is a non-transitory medium, such as a random access memory, a read-only memory, a flash memory, a hard disk, a solid-state hard disk, a magnetic tape, a floppy disk, an optical disc, and any combination thereof. The above-mentioned storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more available media integrated. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a digital video disc (DVD)), or a semiconductor medium (for example, a solid-state disk (SSD)), etc.

[0089] like Figure 5 As shown, an embodiment of the present application provides a terminal.

[0090] The terminal of this embodiment includes a processor and a memory connected to each other; the memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory, so that all or part of the steps in the method of the above embodiment can be implemented when the terminal executes.

[0091] The beneficial effects of all or part of the steps of the method in the above embodiment are the same as the beneficial effects obtained by applying the terminal provided by the embodiment of the present invention, and will not be described in detail here.

[0092] It should be noted that the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage. Similarly, the processor may also be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0093] Although the embodiments disclosed in the present invention are as above, the contents described are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the protection scope of the present invention shall still be subject to the scope defined in the attached claims.

Claims

1. A polishing material removal rate distribution prediction model training method, comprising: Acquire multiple pieces of polishing material removal rate distribution data, and perform dimensionality reduction processing on each piece of polishing material removal rate distribution data by a reversible dimensionality reduction method to obtain key distribution features of the polishing material removal rate distribution data; Perform correlation analysis on each parameter in the polishing process and polishing pad physical property parameter set and the key distribution characteristics of the polishing material removal rate distribution data to obtain highly correlated process and physical property parameters of the polishing material removal rate distribution data; Integrate the highly correlated process and physical property parameters of the polishing material removal rate distribution data with the key distribution features into a training data set, and train a preset network model based on the training data set to obtain a polishing material removal rate distribution prediction model; Among them, through the Pearson correlation coefficient analysis method, each parameter in the polishing process and polishing pad physical property parameter set is respectively analyzed with the distribution key characteristics of the polishing material removal rate distribution data, including: Calculate the Pearson correlation coefficient between the target parameter and each key feature in the distribution key feature of the polishing material removal rate distribution data to obtain the correlation coefficient value corresponding to each key feature, multiply each correlation coefficient value by the explained variance ratio corresponding to the key feature and sum them to obtain the correlation value between the target parameter and the distribution key feature of the polishing material removal rate distribution data; Determine whether the correlation value is greater than a preset component threshold value, if so, determine that the target parameter is a highly correlated process and physical property parameter of the polishing material removal rate distribution data, otherwise determine that the target parameter is not a highly correlated process and physical property parameter of the polishing material removal rate distribution data; The target parameter is any parameter in the polishing process and polishing pad physical property parameter set; the explained variance ratio corresponding to the correlation coefficient value is the explained variance ratio of the key feature corresponding to the correlation coefficient value in the polishing material removal rate distribution data.

2. The training method according to claim 1, characterized in that: Each polishing material removal rate distribution data is cleaned and standardized, and the cleaning and standardization of the polishing material removal rate distribution data includes: The polishing material removal rate distribution data includes a plurality of data points arranged in ascending order of radius. The data point with a radius of zero in the polishing material removal rate distribution data is subtracted from each data point in the polishing material removal rate distribution data, and the data excluding the data point with a radius of zero in the polishing material removal rate distribution data is formed into a data matrix.

3. The training method according to claim 1, characterized in that: The dimensionality reduction processing of each polishing material removal rate distribution data by the reversible dimensionality reduction method includes: A bidirectional converter is constructed by using a principal component analysis method, and a dimension reduction process is performed on each polishing material removal rate distribution data by using the bidirectional converter; The bidirectional converter is also used to convert the distribution key features predicted by the polishing material removal rate distribution prediction model into polishing material removal rate distribution data.

4. The training method according to claim 3, characterized in that: During the dimension reduction process of each polishing material removal rate distribution data through the bidirectional converter, the number of principal components corresponding to when greater than or equal to 90% of the variance in all polishing material removal rate distribution data is explained is selected as the dimension reduction dimension of the polishing material removal rate distribution data.

5. The training method according to claim 1, characterized in that: The preset network model is a multi-task ridge regression model.

6. A method for predicting polishing material removal rate distribution, characterized in that: include: Obtaining the polishing process to be predicted and the physical property parameters of the polishing pad, and inputting the polishing process to be predicted and the physical property parameters of the polishing pad into the polishing material removal rate distribution prediction model obtained by the polishing material removal rate distribution prediction model training method according to any one of claims 1 to 5, so as to obtain the predicted distribution key features; The predicted distribution key features are converted into predicted polishing material removal rate distribution data.

7. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it implements the polishing material removal rate distribution prediction model training method described in any one of claims 1 to 5 or the polishing material removal rate distribution prediction method described in claim 6.

8. A terminal, characterized in that: include: A processor and a memory, wherein the memory is communicatively connected to the processor; The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory, so that the terminal performs the polishing material removal rate distribution prediction model training method as described in any one of claims 1 to 5 or the polishing material removal rate distribution prediction method as described in claim 6.

Images