Silicon carbide epitaxial process recipe generation method, device, equipment and storage medium

Abstract

The present application relates to the technical field of silicon carbide, in particular to a silicon carbide epitaxy process formula generation method, device, equipment and storage medium; historical experimental process data is acquired; according to the preset data selection condition and the second component correlation coefficient, the historical experimental process data and the preset target process index are solved to obtain a group of candidate feasible solutions; according to the preset least square regression algorithm and the preset expected dependent variable, the group of candidate feasible solutions is screened to obtain process parameters; a silicon carbide epitaxy process formula and an evaluation report are generated according to the process parameters; the scheme avoids the blindness of experience and trial and error by positioning the key control factors of the silicon carbide epitaxy process; a group of candidate feasible solutions are obtained by the least square regression algorithm, and finally the process formula is generated, which provides an interpretable and reproducible scheme for process research and development, shortens the cycle and reduces the cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of silicon carbide, in particular to a silicon carbide epitaxy process formula generation method, device, equipment and storage medium. BACKGROUND

[0002] In industrial production and experimental research, accurate control of process parameters is crucial to product quality, production efficiency and experimental result reliability. Traditional process parameter optimization methods rely on manual experience trial and error or single variable experiments, which have problems of low efficiency and high cost, and are difficult to capture the complex coupling relationship between multiple parameters. Although existing data-driven optimization methods introduce statistical analysis or machine learning technology, the input and output data are not standardized, resulting in unbalanced influence weight of different magnitude parameters on the result. At the same time, in the feature extraction link, there is a lack of directional screening mechanism for process data characteristics, which is easily disturbed by redundant information and difficult to accurately extract key influencing factors. In addition, the single evaluation standard is used in the candidate solution screening stage, ignoring the multi-objective demand of process indicators, resulting in deviation of the optimization result from the actual production target, which cannot efficiently support the accurate decision of process parameters. SUMMARY

[0003] In order to overcome the deficiencies of the prior art, the purpose of the present application is to provide a silicon carbide epitaxy process formula generation method, device, equipment and storage medium.

[0004] To solve the above technical problems, the technical solutions adopted by the present application are as follows:

[0005] The present application provides a silicon carbide epitaxy process formula generation method, comprising: obtaining historical experimental process data; generating independent variables and dependent variables according to the historical experimental process data; standardizing the independent variables and dependent variables to obtain standardized independent variables and standardized dependent variables; selecting the first component correlation coefficient from the standardized dependent variables according to the preset maximum variance column condition; generating a weight vector according to the standardized independent variables and the first component correlation coefficient; generating a second component correlation coefficient according to the weight vector and the standardized independent variables; solving the historical experimental process data and the preset target process indicators according to the preset data selection condition and the second component correlation coefficient to obtain a group of candidate feasible solutions; screening the group of candidate feasible solutions according to the preset least squares regression algorithm and the preset expected dependent variable to obtain process parameters; generating a silicon carbide epitaxy process formula and an evaluation report according to the process parameters.

[0006] Further, the solving the historical experiment process data and the preset target process index according to the preset data selection condition and the second component correlation coefficient to obtain a group of candidate feasible solutions comprises: performing row number analysis on the historical experiment process data to obtain a historical data row number; performing calculation on the historical experiment process data according to the preset elimination condition, the historical data row number and the data selection condition to obtain a mean square error; performing data statistics on all the mean square errors to obtain a mean square error number; performing calculation on the standardized independent variable and the second component correlation coefficient to obtain a first load; judging whether the mean square error number satisfies a preset inner loop iteration stop condition; if the mean square error number does not satisfy the inner loop iteration stop condition, returning to perform the calculation on the historical experiment process data according to the elimination condition, the historical data row number and the data selection condition until the mean square error number satisfies the inner loop iteration stop condition; if the mean square error number satisfies the inner loop iteration stop condition, performing analysis on the mean square error and the second component correlation coefficient according to a preset outer loop iteration stop condition to obtain an optimal component number; obtaining all the first loads to obtain a first load set; and generating a group of candidate feasible solutions according to the optimal component number, the first load set and the target process index.

[0007] Further, the calculation on the historical experiment process data according to the preset elimination condition, the historical data row number and the data selection condition to obtain a mean square error comprises: screening the historical experiment process data according to the elimination condition and the historical data row number to obtain a training set and an elimination data set; training a preset partial least squares regression algorithm according to the training set to obtain a first partial least squares regression model; obtaining an independent variable substitution and an eliminated dependent variable from the elimination data set according to the data selection condition; predicting the independent variable substitution according to the first partial least squares regression model to obtain a dependent variable prediction value; and calculating the mean square error according to the dependent variable prediction value and the eliminated dependent variable.

[0008] Further, the analysis on the mean square error and the second component correlation coefficient according to the preset outer loop iteration stop condition to obtain an optimal component number comprises: calculating an average value of all the mean square errors to obtain a mean square error average value set; performing data statistics on the mean square error average value set to obtain a mean square error average value number; performing calculation on the standardized dependent variable and the second component correlation coefficient to obtain a second load; generating a residual matrix according to the second load, the standardized dependent variable and the second component correlation coefficient; judging whether the mean square error average value number or the residual matrix satisfies the outer loop iteration stop condition; when the second mean square error average value number or the residual matrix satisfies the outer loop iteration stop condition, obtaining a minimum mean square error average value from the mean square error average value set; obtaining a component number corresponding to the minimum mean square error average value according to the minimum mean square error average value to obtain the optimal component number.

[0009] Further, the generating a group of candidate feasible solutions according to the optimal composition number, the first load set and the target process index comprises: optimizing the first partial least squares regression model according to the optimal composition number to obtain an optimized least squares regression model; and solving the target process index according to the optimized least squares regression model and the first load set to obtain the group of candidate feasible solutions.

[0010] Further, the solving the target process index according to the optimized least squares regression model and the first load set to obtain the group of candidate feasible solutions comprises: obtaining all weight vectors to obtain a weight vector set; generating a first regression coefficient matrix and a generalized inverse matrix according to the weight vector set and the first load set; judging whether the first regression coefficient matrix is a column full-rank matrix; when the first regression coefficient matrix is the column full-rank matrix, solving the generalized inverse matrix, the dependent variable and the first regression coefficient matrix according to a preset solving formula to obtain a first accurate solution; solving the target process index according to the optimized least squares regression model and the first accurate solution to obtain the group of candidate feasible solutions; otherwise, iteratively solving the generalized inverse matrix, the dependent variable and the first regression coefficient matrix according to a numerical method to obtain a second accurate solution; and solving the target process index according to the optimized least squares regression model and the second accurate solution to obtain the group of candidate feasible solutions.

[0011] Further, the generating the first regression coefficient matrix and the generalized inverse matrix according to the weight vector set and the first load set comprises: generating a weight matrix according to the weight vector set; obtaining all second loads to obtain a second load set; generating a second load matrix according to the second load set; generating a first load matrix according to the first load set; generating the first regression coefficient matrix according to the weight matrix, the first load matrix and the second load matrix; and transforming the first regression coefficient matrix to obtain the generalized inverse matrix.

[0012] Further, the screening the group of candidate feasible solutions according to the preset least squares regression algorithm and the preset expected dependent variable to obtain the process parameter comprises: randomly extracting and returning the historical experimental process data according to a preset repetition number to obtain a plurality of Bootstrap samples; constructing a plurality of second partial least squares regression models and a plurality of second regression coefficient matrices according to the least squares regression algorithm and the plurality of Bootstrap samples; inversely solving the plurality of second regression coefficient matrices according to the preset expected dependent variable and the plurality of second partial least squares regression models to obtain a plurality of estimated values; generating a confidence interval according to the plurality of estimated values; and screening the group of candidate feasible solutions according to the confidence interval to obtain the process parameter.

[0013] Further, the silicon carbide epitaxy process formula generation device comprises a process data acquisition module configured to acquire historical experimental process data; a variable generation module configured to generate independent variables and dependent variables according to the historical experimental process data; a standardization module configured to standardize the independent variables and the dependent variables to obtain standardized independent variables and standardized dependent variables; a selection module configured to select a first component correlation coefficient from the standardized dependent variables according to a preset maximum variance column condition; a weight vector generation module configured to generate a weight vector according to the standardized independent variables and the first component correlation coefficient; a coefficient generation module configured to generate a second component correlation coefficient according to the weight vector and the standardized independent variables; a solving module configured to solve the historical experimental process data and a preset target process index according to a preset data selection condition and the second component correlation coefficient to obtain a group of candidate feasible solutions; a screening module configured to screen the group of candidate feasible solutions according to a preset least square regression algorithm and a preset expected dependent variable to obtain process parameters; and a formula and report generation module configured to generate a silicon carbide epitaxy process formula and an evaluation report according to the process parameters.

[0014] Further, the silicon carbide epitaxy process formula generation device comprises a memory and at least one processor, wherein the memory stores instructions; and the at least one processor invokes the instructions in the memory to enable the silicon carbide epitaxy process formula generation device to perform each step of the silicon carbide epitaxy process formula generation method according to any one of the above.

[0015] Further, a computer readable storage medium stores instructions, wherein the instructions are executed by a processor to implement each step of the silicon carbide epitaxy process formula generation method according to any one of the above.

[0016] In the technical solution of the present application, the core dependent variable is screened by the maximum variance principle, the independent variable influence strength is quantified by the weight vector, the key control factors are located from multiple groups of process parameters (such as hydrogen carrier gas flow, process temperature, etc.), the optimization efficiency is improved, and the blindness of traditional experience trial and error is avoided; the dimension and order of magnitude difference problem is solved by eliminating data interference and standardization processing, the independent variable and dependent variable association is dynamically optimized by the component iteration mechanism, the noise is stripped to focus on the core law, the objectivity and reliability of the analysis result are guaranteed, the complex process scene is adapted, the parameter synergy effect is captured by the second component correlation coefficient, the optimal solution is screened by the least square regression, the candidate formula conforming to the target index (such as epitaxial layer thickness, uniformity) is generated, the flexibility and practicability are taken into account, an interpretable and reproducible optimization scheme is provided for silicon carbide epitaxy process research and development, the research and development cycle is shortened, and the experimental cost is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0017] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description of the embodiments, taken in conjunction with the references to the following drawings, wherein:

[0018] Figure 1 The first flow chart of the method for generating a silicon carbide epitaxy process recipe provided by the embodiment of the present application;

[0019] Figure 2 The second flow chart of the method for generating a silicon carbide epitaxy process recipe provided by the embodiment of the present application;

[0020] Figure 3 The third flow chart of the method for generating a silicon carbide epitaxy process recipe provided by the embodiment of the present application;

[0021] Figure 4 The fourth flow chart of the method for generating a silicon carbide epitaxy process recipe provided by the embodiment of the present application;

[0022] Figure 5 The fifth flow chart of the method for generating a silicon carbide epitaxy process recipe provided by the embodiment of the present application;

[0023] Figure 6 The sixth flow chart of the method for generating a silicon carbide epitaxy process recipe provided by the embodiment of the present application;

[0024] Figure 7 The seventh flow chart of the method for generating a silicon carbide epitaxy process recipe provided by the embodiment of the present application;

[0025] Figure 8 The eighth flow chart of the method for generating a silicon carbide epitaxy process recipe provided by the embodiment of the present application;

[0026] Figure 9 The structure schematic diagram of the device for generating a silicon carbide epitaxy process recipe provided by the embodiment of the present application;

[0027] Figure 10 The structure schematic diagram of the device for generating a silicon carbide epitaxy process recipe provided by the embodiment of the present application. DETAILED DESCRIPTION

[0028] The terms "first", "second", "third", "fourth" and the like in the description and in the claims of the present application, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of such terms is interchangeable under appropriate circumstances such that the embodiments of the application described herein are, for example, capable of orderly or sequential execution, irrespective of the particular order or sequence used to describe the execution of the embodiments of the application. Moreover, the use of the term "including", "having", "having at least" and variations thereof, is intended to cover a non-exclusive inclusion, such that a process, method, system, product or apparatus that comprises, has, has at least or includes a list of steps or units can not necessarily be limited to those steps or units explicitly listed, but can include other steps or units not expressly listed or inherent to such process, method, product or apparatus.

[0029] For the purpose of facilitating understanding, the specific flow of the embodiments of the present application is described below, please refer to Figure 1 One embodiment of the method for generating a silicon carbide epitaxy process recipe in the embodiments of the present application includes:

[0030] 101, obtaining historical experimental process data;

[0031] In this embodiment, the historical experimental process data contains two types of core information, the "process operation parameters" (such as temperature, flow, etc.) that can be adjusted and the "performance indicators" (such as product purity, thickness, etc.) that reflect the experimental results;

[0032] 102, generating independent variables and dependent variables according to the historical experimental process data;

[0033] In this embodiment, the independent variables include the process parameters commonly used in the silicon carbide epitaxy process, hydrogen carrier gas flow, TCS flow, ethylene flow, flow, air floatation flow, process temperature; the dependent variables include the silicon carbide epitaxy process indicators: epitaxial layer thickness, thickness uniformity, doping concentration, doping concentration uniformity, surface roughness; the historical experimental process data also refers to these physical quantities; the number of columns of the independent variable matrix represents the number of process parameters (i.e. the number of independent variables), the number of columns of the dependent variable matrix represents the number of epitaxy process indicators (i.e. the number of dependent variables), the number of rows of the independent variable matrix and the dependent variable matrix is equal, one row represents one complete epitaxy process, and the number of rows represents the number of epitaxy processes performed in the historical experiments;

[0034] 103, standardizing the independent variables and the dependent variables to obtain standardized independent variables and standardized dependent variables;

[0035] In the embodiment, the independent variables and dependent variables are standardized (e.g., Z-score standardization, with a mean of 0 and a standard deviation of 1); the weight imbalance caused by the dimensional difference (e.g., temperature unit ℃, flow unit sccm) or order of magnitude difference (e.g., thickness and concentration) of different variables is eliminated, so that each variable can participate in the subsequent calculation "fairly", and the analysis result is not dominated by a variable with a large value;

[0036] 104. Selecting a first component correlation coefficient from the standardized dependent variables according to a preset maximum variance column condition;

[0037] In the embodiment, the column with the maximum variance is selected from the standardized dependent variables as the initial component u (i.e., the first component correlation coefficient);

[0038] 105. Generating a weight vector according to the standardized independent variables and the first component correlation coefficient;

[0039] In the embodiment, the weight vector wherein X is the standardized independent variable, is the device matrix of the standardized independent variable;

[0040] 106. Generating a second component correlation coefficient according to the weight vector and the standardized independent variable;

[0041] In the embodiment, t is the part of the standardized independent variable X that is most relevant to the current u (i.e., the first component correlation coefficient) (i.e., the second component correlation coefficient); the weight vector is calculated according to the standardized dependent variable Y and the second component correlation coefficient t, and then u is updated to reflect the u that is most relevant to t in Y; w is the weight vector of X and u , and the calculation target is to maximize the covariance of t and u; the covariance is a statistical quantity for measuring the degree of correlation, so according to the definition, the t obtained by multiplying X by the calculated w is the part that is most consistent with the direction of u and changes most synchronously;

[0042] 107. Solving the historical experimental process data and the preset target process index according to a preset data selection condition and the second component correlation coefficient to obtain a group of candidate feasible solutions;

[0043] In the embodiment, the historical experimental process data is known, a PLS model (least squares regression model) is generated according to the historical experimental process data, and then the model is solved in reverse (equivalent to obtaining its inverse); given the target process index, the inverse of the PLS model is obtained to obtain a group of candidate feasible solutions;

[0044] 108. Screening the group of candidate feasible solutions according to a preset least squares regression algorithm and a preset expected dependent variable to obtain the process parameters;

[0045] In the embodiment, the process parameters include hydrogen carrier gas flow, TCS flow, ethylene flow, N2 flow, process temperature;

[0046] 109. generating a silicon carbide epitaxy process formula and an evaluation report according to the process parameters;

[0047] In the embodiment, one or more candidate formulas can be generated by the output module, and the output module also allows the user to manually adjust the generated candidate formula;

[0048] In the embodiment, the core dependent variable is screened by the variance maximum principle, the influence strength of the independent variable is quantified in combination with the weight vector, the key control factors are located from multiple sets of process parameters (such as hydrogen carrier gas flow, process temperature, etc.), the optimization efficiency is improved, and the blindness of traditional experience trial and error is avoided; by eliminating data interference, the dimension and order of magnitude difference problem is solved by standardization processing, the correlation between the independent variable and the dependent variable is dynamically optimized by the component iteration mechanism, the core law is stripped and focused, the objectivity and reliability of the analysis result are guaranteed, the complex process scene is adapted, the parameter synergistic effect is captured by the second component correlation coefficient, the optimal solution is screened by the least square regression, the candidate formula meeting the target index (such as the thickness and uniformity of the epitaxial layer) is generated, the flexibility and practicality are considered, an interpretable and reproducible optimization scheme is provided for silicon carbide epitaxy process research and development, the research and development cycle is shortened, and the experimental cost is reduced.

[0049] Please refer to Figure 2 , the second embodiment of the silicon carbide epitaxy process formula generation method in the embodiment of the present application, step 107 comprises:

[0050] 201. performing row analysis on historical experimental process data to obtain a historical data row number;

[0051] In the embodiment, the historical experimental process data contains n data entries in total, the historical data row number represents the quantity scale of the historical experimental process data, and the historical data row number directly reflects the total number (n) of the historical experimental process data entry rows; this step provides a basis for subsequent sample division of cross-validation, and ensures data coverage and verification;

[0052] 202. calculating the historical experimental process data according to the preset rejection condition, the historical data row number and the data selection condition to obtain a mean square error;

[0053] In the embodiment, the mean square error (MSE) is calculated through cross-validation to evaluate the prediction ability of the model on the data not participating in the training, in combination with the preset elimination condition (such as “eliminate row by row”, 1 row of data is eliminated from the first row, and the remaining n-1 rows are used as the training set, and the eliminated 1 row is used as the validation set (including the independent variable and the dependent variable));

[0054] 203, data statistics are performed on all the mean square errors to obtain the number of mean square errors;

[0055] 204, the standardized independent variable and the second component correlation coefficient are calculated to obtain the first load;

[0056] In the embodiment, the load p of X (i.e. the first load ) is calculated,

[0057] 205, it is judged whether the number of mean square errors meets the preset inner loop iteration stop condition;

[0058] 206, if the number of mean square errors does not meet the inner loop iteration stop condition, the historical experimental process data is calculated according to the elimination condition, the number of historical data rows and the data selection condition, until the number of mean square errors meets the inner loop iteration stop condition;

[0059] In the embodiment, the inner loop iteration stop condition is that h mean square error average values can be calculated through all the mean square errors, and n mean square errors will be calculated in each inner loop; each row of data entry in the historical experimental process data is used as the validation set to participate in the evaluation, and each inner loop is stopped when n mean square errors are calculated;

[0060] 207, if the number of mean square errors meets the inner loop iteration stop condition, the mean square error and the second component correlation coefficient are analyzed according to the preset outer loop iteration stop condition to obtain the optimal component number;

[0061] 208, all the first loads are obtained to obtain a first load set;

[0062] In the embodiment, the first load p reflects the contribution degree of the independent variable to the correlation coefficient t of the second component: the greater the absolute value of the independent variable (such as the flow rate of a certain gas), the more significant the influence on the process result, which provides a basis for parameter priority ranking; the first load set is derived from full-sample cross-validation of n data entries in the historical experimental process data in the inner loop, and the first load is calculated repeatedly by removing data row by row, dynamically adjusting the input parameters (X and t), and repeatedly calculating the first load adapted to different training sets, and finally the first load set is obtained; the inner loop iteration stopping condition is that the outer loop passes through different component numbers, and the inner loop is repeatedly executed for multiple rounds until the average value of the h mean square errors can be calculated through all the mean square errors, and then the inner loop iteration stops;

[0063] 209、According to the optimal component number, the first load set and the target process index, a set of candidate feasible solutions is generated;

[0064] In the embodiment, the average value of the n MSE is calculated as the prediction effect index of the current component number; after the outer loop iterates different component numbers, the component number corresponding to the minimum MSE average value is selected from all the MSE average values, that is, the optimal component number A, which avoids the problems of "underfitting due to too few component numbers" or "overfitting due to too many component numbers", and ensures the model generalization ability;

[0065] In the embodiment, systematic cross-validation and load analysis are used to support process parameter optimization; the number of rows is analyzed to determine the data size, which lays a foundation for cross-validation; the "row-by-row removal" method is used to divide the training set, and the mean square error (MSE) is calculated through n inner loops to comprehensively evaluate the prediction accuracy of the model, maximize the use of historical data, and reduce the cost of additional experiments; the first load is calculated simultaneously to analyze the contribution of the independent variable, to determine the priority of the key parameters and to improve the explainability of the optimization; when the number of mean square errors is equal to the number of data rows, the optimal component number is selected through the outer loop, the component number corresponding to the minimum MSE is selected, the model complexity and generalization ability are balanced, and overfitting or underfitting is avoided; finally, the candidate solution is generated by combining the optimal component number, the load set and the target index, which not only meets the process requirements, but also is based on the data law, and provides a basis for efficiently screening reliable process parameters.

[0066] Please refer to Figure 3 In the third embodiment of the silicon carbide epitaxy process formula generation method in the embodiment of the present application, step 202 comprises:

[0067] 301、According to the removal condition and the number of historical data rows, the historical experimental process data is screened to obtain a training set and a removed data set;

[0068] In the embodiment, the number of historical data rows is n rows, the discard condition is to discard each row from the first row, the remaining n-1 rows form a training set, a verification that covers all samples is realized, the division method ensures that each sample can participate in evaluation as a verification set, and avoids verification deviation caused by fixed sample division; the discarded data set: the discarded 1 row is used as a verification sample, contains "substitute independent variable" (process parameters to be predicted) and "discard dependent variable" (corresponding actual result index), and is used for model precision test;

[0069] 302. Train the preset partial least squares regression algorithm according to the training set to obtain a first partial least squares regression model;

[0070] In the embodiment, the partial least squares regression algorithm has the advantages of effectively processing the multiple collinearity problem of independent variables, and extracting the core correlation between independent variables and dependent variables, and is especially suitable for process scenes (such as mutual influence of multiple gas flow and temperature in silicon carbide epitaxy) with multiple parameter coupling;

[0071] 303. Obtain the substitute independent variable and the discard dependent variable from the discarded data set according to the data selection condition;

[0072] In the embodiment, the data selection condition is to select the substitute independent variable of a row discarded in the inner loop and the discard dependent variable of the discarded row;

[0073] 304. Predict the independent variable according to the first partial least squares regression model to obtain a dependent variable prediction value;

[0074] 305. Calculate the mean square error according to the dependent variable prediction value and the discard dependent variable;

[0075] In the embodiment, the substitute independent variable of the row discarded in the inner loop is substituted into the temporary first partial least squares regression model to obtain a predicted dependent variable prediction value of the row, and the mean square error (MSE) between the predicted dependent variable prediction value and the discard dependent variable of the discarded row is calculated, which comprehensively reflects the prediction accuracy of the model on different samples;

[0076] In the embodiment, the cross-validation mechanism provides reliable support for process model precision evaluation; the "row-by-row elimination" method is used to divide samples, and n rows of historical data are dynamically divided into a training set (n-1 rows) and an eliminated data set (1 row), so that each sample participates in verification, deviation caused by fixed division is avoided, objective evaluation covering all samples is realized; the partial least squares regression model trained based on the training set can effectively process the multicollinearity problem of multiple parameters, capture the core correlation between independent variables and dependent variables, and adapt to multiple parameter coupling scenes such as silicon carbide epitaxy; the independent variables and dependent variables of the eliminated row are extracted through data selection conditions, the mean square error (MSE) is calculated after the predicted value is obtained by substituting the model, the prediction precision of the model on different samples is comprehensively reflected through n times of cycles; the value of historical data is maximized to provide a quantitative basis for subsequent model optimization and parameter selection, reduce research and development costs, and improve process reliability.

[0077] Referring to Figure 4 In the fourth embodiment of the silicon carbide epitaxy process formula generation method in the embodiment of the present application, step 207 comprises:

[0078] 401, calculate the average value of all mean square errors to obtain a mean square error average value set;

[0079] In the embodiment, the mean square error average value eliminates the influence of single sample fluctuation and more stably represents the overall prediction effect of the model under the current component number (the higher the component number, the higher the model complexity, and whether overfitting occurs needs to be evaluated by the mean value;

[0080] 402, perform data statistics on the mean square error average value set to obtain a mean square error average value number;

[0081] In the embodiment, n mean square errors will be calculated in the inner loop, and the average value of all n mean square errors represents the prediction effect of the component number;

[0082] 403, calculate the standardized dependent variable and the second component correlation coefficient to obtain a second load;

[0083] In the embodiment, the second load quantifies the contribution of the dependent variable to the second component correlation coefficient t and supplements the dependent variable correlation information not covered by the first load;

[0084] 404, generate a residual matrix according to the second load, the standardized dependent variable and the second component correlation coefficient;

[0085] In the embodiment, the residual matrix of the standardized dependent variable Y reflects the dependent variable information not explained by the current second component correlation coefficient t (the smaller the residual, the stronger the model's ability to explain the dependent variable);

[0086] 405, judge whether the mean square error average number or the residual matrix meets the outer loop iteration stop condition;

[0087] In the embodiment, the outer loop iteration stop condition is that the number of components selected by cross-validation is reached or the absolute value of the element of the residual matrix F is close to 0 (i.e., the model has sufficiently explained the information of the dependent variable), and the element of the residual matrix F close to 0 represents that the mathematical iteration will no longer be able to improve the accuracy of the model, and the general range is less than [1, 5];

[0088] 406, when the second mean square error average number or the residual matrix meets the outer loop iteration stop condition, all the mean square error average numbers are obtained to obtain a mean square error average number set;

[0089] In the embodiment, otherwise, the residual matrix F is generated according to the first load, the standardized independent variable and the second component correlation coefficient ; the dependent variable and the independent variable are updated according to the residual matrix F and the residual matrix E respectively, and the standardized independent variable and the dependent variable are returned to be executed, and the next group of t and u is extracted;

[0090] 407, the minimum mean square error average number is obtained from the mean square error average number set;

[0091] 408, the component number corresponding to the minimum mean square error average number is obtained to obtain the best component number;

[0092] In the embodiment, until the outer loop calculation is completed, h mean square error average numbers are obtained, and the component number with the minimum mean square error average number is selected as the best component number A, so that the model reaches the optimal balance between the "lowest complexity" and the "highest prediction accuracy";

[0093] In the embodiment, the best component number is accurately selected through the systematic iteration optimization mechanism, and reliable parameter support is provided for the process model; the n mean square errors in the inner loop are averaged to eliminate the influence of single sample fluctuation, stably reflect the model prediction effect of the current component number, and avoid the risk of overfitting; the contribution of the second load to the second component correlation coefficient is calculated to supplement the variable correlation information; the residual matrix is generated to intuitively present the information not explained by the model, and the smaller the residual is, the stronger the explanation ability is; the stop condition is dynamically judged through the outer loop, and if the preset component number or the residual approaches to 0, the iteration is terminated, otherwise, the data is updated and a new component is extracted; finally, the component number corresponding to the minimum mean square error average number is selected as the best value, and the optimal balance between the model complexity and the prediction accuracy is realized; the flow quantitatively evaluates the standardization, adapts to the complex process scene, improves the efficiency and reliability of the process optimization, and reduces the cost of manual parameter adjustment.

[0094] Please refer to Figure 5In the fifth embodiment of the method for generating a silicon carbide epitaxy process formula in the embodiments of the present application, step 209 comprises:

[0095] 501. Optimize the first partial least squares regression model according to the optimal component number to obtain an optimized least squares regression model;

[0096] In this embodiment, the optimized least squares regression model represents the mapping relationship between the independent variable matrix and the dependent variable matrix, and the optimized model clearly defines the mapping relationship between the independent variable matrix (process parameters such as hydrogen carrier gas flow and process temperature) and the dependent variable matrix (result indicators such as epitaxial layer thickness). This relationship is calibrated by the optimal component number, which not only retains the core correlation information but also avoids the high complexity of the model caused by redundant parameters.

[0097] 502. Solve the target process indicators according to the optimized least squares regression model and the first load set to obtain a group of candidate feasible solutions;

[0098] In this embodiment, in the silicon carbide epitaxy process, commonly used target process indicators include epitaxial layer thickness, thickness uniformity, doping concentration, doping concentration uniformity, and surface roughness. A group of process parameter combinations that meet the target indicator constraints, i.e., candidate feasible solutions, are finally generated, laying a foundation for subsequent precise screening of optimal process parameters.

[0099] In this embodiment, after the first partial least squares regression model is optimized by the optimal component number, the obtained optimized model clearly defines the mapping relationship between the process parameters (such as H2 hydrogen carrier gas flow and process temperature) and the result indicators (such as epitaxial layer thickness). After calibration by the optimal component number, the core correlation information is retained, and the high complexity of the model caused by redundant parameters is avoided, balancing between prediction accuracy and generalization ability. In combination with the optimized model and the first load set, the silicon carbide epitaxy core target indicators (thickness, uniformity, doping concentration, etc.) are solved, and process parameter combinations that meet the indicator constraints, i.e., candidate feasible solutions, are generated. This process focuses on key quality requirements, candidate solutions conform to optimization objectives, reduces invalid screening costs, and the parameter correlation is traceable, facilitating engineers to understand and adjust, efficiently connecting the processes before and after, and accelerating process research and development iteration.

[0100] Please refer to Figure 6 In the sixth embodiment of the method for generating a silicon carbide epitaxy process formula in the embodiments of the present application, step 502 comprises:

[0101] 601. Obtain all weight vectors to obtain a weight vector set;

[0102] In this embodiment, these vectors quantify the influence intensity of the independent variables on the second component correlation coefficient, which is a key basis for constructing a regression relationship.

[0103] 602、generating a first regression coefficient matrix and a generalized inverse matrix according to the weight vector set and the first load set;

[0104] In the embodiment, the first regression coefficient matrix is essentially a quantitative expression of the mapping relationship between the independent variable and the dependent variable, and is directly related to the process parameters and the result indicators; the generalized inverse matrix solves the problem of non-existent inverse matrix of non-square or singular matrix, and provides a mathematical tool for subsequent solution;

[0105] 603、judging whether the first regression coefficient matrix is a column full-rank matrix;

[0106] In the embodiment, whether the first regression coefficient matrix is a column full-rank matrix (column vectors are linearly independent) is judged, which determines the selection of the solution method and ensures the effectiveness of the solution;

[0107] 604、when the first regression coefficient matrix is a column full-rank matrix, the generalized inverse matrix, the dependent variable and the first regression coefficient matrix are solved according to a preset solution formula to obtain a first accurate solution;

[0108] In the embodiment, the first accurate solution has analyticality and can accurately meet the target indicator constraint;

[0109] In the embodiment, , that is is the first accurate solution, I is an identity matrix, the main diagonal elements in the matrix are 1, and the other elements are 0, is the generalized inverse matrix of the first regression coefficient matrix B, is the dependent variable matrix;

[0110] 605、solving the target process indicators according to the optimized least squares regression model and the first accurate solution to obtain a group of candidate feasible solutions;

[0111] 606、otherwise, the generalized inverse matrix, the dependent variable and the first regression coefficient matrix are iteratively solved according to a numerical method to obtain a second accurate solution;

[0112] In the embodiment, most of the time B is a non-square matrix and is not full-rank, so the numerical method is needed to iteratively solve, and the target is to minimize , which gradually approaches the optimal solution through iterative optimization and adapts to complex data structures;

[0113] 607、solving the target process indicators according to the optimized least squares regression model and the second accurate solution to obtain a group of candidate feasible solutions;

[0114] In the embodiment, the target process indicators include a target range of epitaxy indicators (such as a thickness of 10±0.5 um), a priority of the epitaxy indicators, whether to use some indicators according to the epitaxy target priority, definition of some constraint conditions of the process (such as a longest process time length, a specific gas flow range, etc.), a solving process taking into account the indicator priority, and ensuring that key quality parameters are preferentially met;

[0115] In the embodiment, the systematized matrix operation and the flexible solving mechanism are used to support the process parameter optimization; the weight vector set is used to quantify the influence of the independent variables on the second component correlation coefficient, the first load set is used to generate the first regression coefficient matrix, the mapping relationship between the process parameters and the result indicators is clearly presented, the generalized inverse matrix is used to solve the non-square matrix or the singular matrix, the matrix is used to judge whether the column is full rank, and the adaptive solution method is selected: the first accurate solution is solved by the analytic method when the column is full rank, the index constraint is met; the second accurate solution is solved by the numerical method when the column is not full rank, the optimal solution is approached by minimizing the error, and the complex data structure is adapted; the solving process integrates the target process indicator range, the priority and the process constraint (such as the flow limitation), and ensures that key quality parameters are preferentially met; the scheme takes into account the theoretical rigor and the actual applicability, the generation logic of the solution is traceable, the model and the application are efficiently connected, and the process formula landing is accelerated.

[0116] Please refer to Figure 7 In the seventh embodiment of the silicon carbide epitaxy process formula generation method in the embodiment of the application, step 602 comprises:

[0117] 701. A weight matrix is generated according to the weight vector set;

[0118] In the embodiment, the weight vector quantifies the influence strength of the independent variables on the second component correlation coefficient, and the weight matrix systematically integrates multiple sets of weight information, and fully reflects the influence law of the independent variables under different components;

[0119] 702. All second loads are obtained to obtain a second load set;

[0120] In the embodiment, the second load set is derived from the full-range test of multiple sets of second component correlation coefficients in the outer loop, the second load adapted to different second component correlation coefficients is repeatedly calculated by iterating different second component correlation coefficients, and the second load set is finally summarized;

[0121] 703. A second load matrix is generated according to the second load set;

[0122] In the embodiment, the second load matrix reflects the association between the dependent variable and the second component correlation coefficient, and the two together constitute a key quantitative indicator of variable association;

[0123] 704. A first load matrix is generated according to the first load set;

[0124] 705. generating a first regression coefficient matrix according to the weight matrix, the first loading matrix and the second loading matrix;

[0125] In the embodiment, the solution formula of the first regression coefficient matrix is:

[0126] , wherein, is the first regression coefficient matrix, is the weight matrix, is the first loading matrix, is the transpose matrix of the first loading matrix, is the transpose matrix of the second loading matrix, the first loading matrix reflects the contribution degree of the independent variable to the correlation coefficient of the second component, and the second loading matrix reflects the correlation between the dependent variable and the correlation coefficient of the second component, which together constitute the key quantitative indicators of variable correlation; the first regression coefficient matrix is a direct quantitative expression of the mapping relationship between the independent variable and the dependent variable, for example, the element value in the matrix directly reflects the influence coefficient of a certain process parameter (such as H2 flow) on the result index (such as the thickness of the epitaxial layer), which provides a mathematical correlation basis for parameter optimization;

[0127] 706. transforming the first regression coefficient matrix to obtain a generalized inverse matrix;

[0128] In the embodiment, is the generalized inverse matrix of the first regression coefficient matrix B, is the transpose matrix of the first regression coefficient matrix B; in the process data, the regression coefficient matrix is often a non-square matrix or a singular matrix (not full rank), and the traditional inverse matrix does not exist, and the generalized inverse matrix solves this problem through mathematical transformation, ensuring that effective solution can still be obtained in complex data scenarios, and providing necessary mathematical tools for subsequent accurate solution calculation;

[0129] In the embodiment, the systematized matrix construction and transformation provide mathematical support for process parameter optimization; the weight matrix is generated from the weight vector set, the multiple sets of weight information are systematically integrated, and the influence law of the independent variable under different components is completely reflected; the loading matrices generated by combining the first loading set and the second loading set reflect the contribution degree of the independent variable to the correlation coefficient of the second component and the correlation between the dependent variable and the correlation coefficient of the second component, which together constitute the key quantitative indicators; the first regression coefficient matrix generated by the specific formula directly quantifies the mapping relationship between the independent variable and the dependent variable, and clarifies the influence coefficient of the process parameter on the result index; the generalized inverse matrix obtained by transforming the regression coefficient matrix solves the problem of solving non-square or singular matrices, and adapts to complex data scenarios; the scheme systematically integrates the dispersed information, improves the model interpretability and solution rigor, lays a foundation for subsequent accurate solution calculation, and efficiently promotes process parameter optimization. ​

[0130] Referring to Figure 8 In the eighth embodiment of the method for generating a silicon carbide epitaxy process formula, step 108 comprises:

[0131] 801. Randomly sampling and returning the historical experimental process data according to a preset number of repetitions to obtain a plurality of Bootstrap samples;

[0132] In this embodiment, N samples are randomly sampled with replacement from the historical experimental process data, and the sampling is repeated M times, where M is usually in the range of 1000-10000, to obtain a plurality of Bootstrap samples. The sample set generated by sampling simulation data fluctuation not only retains the statistical characteristics of the original data, but also reflects the uncertainty (such as experimental error and sample bias) in the data, thereby providing a basis for subsequent model robustness analysis.

[0133] 802. A plurality of second partial least squares regression models and a plurality of second regression coefficient matrices are constructed according to the least squares regression algorithm and the plurality of Bootstrap samples;

[0134] In this embodiment, a partial least squares regression model is constructed for each Bootstrap sample to obtain a set of second partial least squares regression models and a set of second regression coefficient matrices This set of can reflect the uncertainty of B, and is particularly suitable for the common “parameter correlation instability” problem (such as the fluctuation of multiple gas flow coordination) in process data. Through multiple models, the possible correlation rules can be covered.

[0135] 803. A plurality of estimated values are obtained by inversely solving the plurality of second regression coefficient matrices according to a preset expected dependent variable and the plurality of second partial least squares regression models;

[0136] In this embodiment, for the expected dependent variable, each is inversely solved once to obtain M estimated values.

[0137] 804. A confidence interval is generated according to the plurality of estimated values;

[0138] In this embodiment, the confidence interval is taken as the 2.5% and 97.5% quantiles of the estimated values as the 95% confidence interval. This interval reflects the possible value range of the process parameters under 95% probability, and quantifies the uncertainty of the parameters.

[0139] 805. A set of candidate feasible solutions is screened according to the confidence interval to obtain process parameters;

[0140] In this embodiment, the previously generated candidate feasible solutions are screened according to the confidence interval, and the parameter combinations falling within the interval are retained, and finally the process parameters are obtained;

[0141] 806、evaluate the rule base to obtain the process recipe and the evaluation report;

[0142] In this embodiment, the screened parameters are evaluated in combination with the rule base (such as process safety constraints, equipment limitations) to generate a process recipe and an evaluation report containing parameter uncertainty and compliance;

[0143] In this embodiment, the process recipe and the evaluation report can be generated according to the process parameters, and the specific steps are as follows:

[0144] According to the target process indicators, the preset process steps and the process parameter definitions, the core entities are obtained;

[0145] According to the preset dependency relationships, the preset constraint relationships and the mapping relationship definitions, the core relationships are obtained;

[0146] Obtain the knowledge in the field of silicon carbide process, and extract the knowledge in the field of silicon carbide process to obtain structured data, semi-structured data and unstructured data;

[0147] According to the core entities, the core relationships, the structured data, the semi-structured data and the unstructured data, the knowledge graph is constructed;

[0148] The main function of the knowledge graph is to divide the process recipe into steps and instructions, and to check the safety and rationality. The main steps of construction are:

[0149] 1. Define the core entities (process targets, process steps, process parameters, equipment, materials, etc.);

[0150] 2. Define the core relationships: dependency relationships (such as process steps containing temperature parameters), constraint relationships (such as the value of parameter A being affected by parameter B), and mapping relationships (such as process target G corresponding to process step sequence S);

[0151] 3. Data acquisition and preprocessing: multidimensional acquisition of knowledge in the field of silicon carbide process, and the knowledge sources include but are not limited to literature, industry manual, safety specification, expert experience, etc. Data processing includes removing duplicate information, unifying terminology, etc.;

[0152] 4. Knowledge extraction: structured data (tables) can be directly mapped to entities (such as "reaction furnace A: maximum temperature 1800℃" in the equipment manual, which can be extracted as (reaction furnace A, maximum temperature, 1800℃)), semi-structured data (process documents) are matched or parsed using templates, and unstructured data (such as papers, expert interviews) are extracted using NLP technology and domain dictionaries;

[0153] 5. Knowledge storage: select graph database storage;

[0154] Rules are extracted from the knowledge graph;

[0155] Entity association analysis is performed on the rules and the knowledge graph to obtain entity association;

[0156] A rule library is constructed according to the preset rule type and entity association;

[0157] The specific construction steps of the rule library are as follows:

[0158] 1. Rule type: process step division rule, safety verification rule, and rationality verification rule

[0159] 2. Rule extraction: rules are extracted from the knowledge graph (for example, a safety rule is extracted from the knowledge graph "reaction furnace A-highest temperature-1800℃"; a rationality rule is extracted from expert experience "when the epitaxial growth time exceeds 60 min, the temperature needs to be reduced by 50℃ to avoid substrate damage")

[0160] 3. Association of rules and knowledge graph: association of entities in the rules and entities in the knowledge graph, parameters are obtained through the knowledge graph during reasoning, and process constraints are converted into executable logical rules based on the information of the knowledge graph;

[0161] In this embodiment, Bootstrap sampling and multi-model analysis are used to provide reliable guarantee for process parameter screening; N samples are extracted from historical data in a replacement manner, and multiple Bootstrap samples are generated by repeating 1000-10000 times, which not only retains the statistical characteristics of the original data, but also reflects the uncertainties such as experimental error; a partial least squares regression model and a regression coefficient matrix are constructed based on each sample, multiple models cover the parameter association fluctuation law, and adapt to the unstable scene of complex processes; multiple estimated values are obtained by inversely solving the expected dependent variable, a 95% confidence interval is constructed by taking the 2.5% and 97.5% quantiles, and the possible value range of the parameter is quantified; candidate solutions are screened according to the confidence interval, parameter combinations within the interval are retained, and process recipes and evaluation reports are generated by combining rule library verification; this scheme quantifies uncertainty, enhances parameter robustness, reduces invalid experiments, and outputs reports with compliance and interpretability, providing support for process optimization with high stability requirements.

[0162] The silicon carbide epitaxial process recipe generation method in the embodiment of the application is described above, and the silicon carbide epitaxial process recipe generation device in the embodiment of the application is described below, please refer to Figure 9 An embodiment of the silicon carbide epitaxial process recipe generation device in the embodiment of the application includes:

[0163] A process data acquisition module 1 is configured to acquire historical experimental process data;

[0164] A variable generation module 2 is configured to generate independent variables and dependent variables according to the historical experimental process data;

[0165] A standardization module 3 is configured to standardize the independent variables and the dependent variables to obtain standardized independent variables and standardized dependent variables;

[0166] A selection module 4 is configured to select a first component correlation coefficient from the standardized dependent variables according to a preset maximum variance column condition;

[0167] A weight vector generation module 5 is configured to generate a weight vector according to the standardized independent variables and the first component correlation coefficient;

[0168] A coefficient generation module 6 is configured to generate a second component correlation coefficient according to the weight vector and the standardized independent variables;

[0169] A solving module 7 is configured to solve the historical experimental process data and a preset target process index according to a preset data selection condition and the second component correlation coefficient to obtain a group of candidate feasible solutions;

[0170] A screening module 8 is configured to screen the group of candidate feasible solutions according to a preset least square regression algorithm and a preset expected dependent variable to obtain process parameters;

[0171] A formula and report generation module 9 is configured to generate a silicon carbide epitaxy process formula and an evaluation report according to the process parameters;

[0172] In this embodiment, the core dependent variable is screened by the maximum variance principle, the independent variable influence strength is quantified by combining the weight vector, the key control factors are located from a plurality of process parameters (such as hydrogen carrier gas flow, process temperature, etc.), the optimization efficiency is improved, and the blindness of traditional experience trial and error is avoided; by eliminating data interference, the dimension and order of magnitude difference problem is solved by standardization processing, the independent variable and dependent variable association is dynamically optimized by the component iteration mechanism, the noise is stripped to focus on the core law, the objectivity and reliability of the analysis result are guaranteed, the complex process scene is adapted, the parameter synergy effect is captured by the second component correlation coefficient, the optimal solution is screened by the least square regression, the candidate formula conforming to the target index (such as epitaxial layer thickness, uniformity) is generated, the flexibility and practicality are taken into account, an interpretable and reproducible optimization scheme is provided for silicon carbide epitaxy process research and development, the research and development cycle is shortened, and the experimental cost is reduced.

[0173] Figure 10is a structural schematic view of a silicon carbide epitaxy process recipe generation device provided by an embodiment of the present application. The silicon carbide epitaxy process recipe generation device 900 can have great differences due to different configurations or performances, and can include one or more central processing units (CPUs) 910 (for example, one or more processors) and a memory 920, one or more storage media 930 (for example, one or more mass storage devices) storing application programs 933 or data 932. The memory 920 and the storage media 930 can be temporary storage or persistent storage. The programs stored in the storage media 930 can include one or more modules (not shown in the figure), and each module can include a series of instruction operations in the silicon carbide epitaxy process recipe generation device 900. Further, the processor 910 can be configured to communicate with the storage media 930 and execute the series of instruction operations in the storage media 930 on the silicon carbide epitaxy process recipe generation device 900 to implement the steps of the silicon carbide epitaxy process recipe generation method provided by each of the above method embodiments.

[0174] The silicon carbide epitaxy process recipe generation device 900 can further include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input and output interfaces 960, and / or one or more operating systems 931, such as Windows Server, MacOS X, Unix, Linux, FreeBSD, and the like. Those skilled in the art can understand that the silicon carbide epitaxy process recipe generation device 900 can further include other components not shown in the figure, or some components can be combined, or different components can be arranged. Figure 10 The silicon carbide epitaxy process recipe generation device structure shown does not constitute a limitation on the silicon carbide epitaxy process recipe generation device, and can include more or fewer components than shown, or some components can be combined, or different components can be arranged.

[0175] The present application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium or a volatile computer readable storage medium. The computer readable storage medium stores instructions, and when the instructions are run on a computer, the computer executes the steps of the silicon carbide epitaxy process recipe generation method.

[0176] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system or device, unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0177] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0178] Finally, it should be noted that: the above only for the preferred examples of the present application, and not for limiting the present application, although the present application is described in detail with reference to the foregoing examples, for those skilled in the art, it still can be modified to the technical solutions recorded in the foregoing embodiments, or equivalent replacement of some technical features. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for generating a silicon carbide epitaxial process formulation, characterized in that, The method comprises the following steps: obtaining historical experimental process data; generating independent variables and dependent variables according to the historical experimental process data; standardizing the independent variables and the dependent variables to obtain standardized independent variables and standardized dependent variables; selecting first component correlation coefficients from the standardized dependent variables according to a preset maximum variance column condition; generating a weight vector according to the standardized independent variables and the first component correlation coefficients; generating second component correlation coefficients according to the weight vector and the standardized independent variables; solving the historical experimental process data and a preset target process index according to a preset data selection condition and the second component correlation coefficients to obtain a group of candidate feasible solutions; The step of solving the historical experimental process data and the preset target process index according to the preset data selection condition and the second component correlation coefficients to obtain a group of candidate feasible solutions comprises the following steps: performing row number analysis on the historical experimental process data to obtain a historical data row number; performing calculation on the historical experimental process data according to a preset elimination condition, the historical data row number and the data selection condition to obtain mean square errors; performing data statistics on all the mean square errors to obtain a mean square error number; performing calculation on the standardized independent variables and the second component correlation coefficients to obtain first loads; determining whether the mean square error number meets a preset inner loop iteration stop condition; if the mean square error number does not meet the inner loop iteration stop condition, returning to perform calculation on the historical experimental process data according to the elimination condition, the historical data row number and the data selection condition until the mean square error number meets the inner loop iteration stop condition; if the mean square error number meets the inner loop iteration stop condition, analyzing the mean square errors and the second component correlation coefficients according to a preset outer loop iteration stop condition to obtain an optimal component number; obtaining all the first loads to obtain a first load set; generating a group of candidate feasible solutions according to the optimal component number, the first load set and the target process index; screening the group of candidate feasible solutions according to a preset least square regression algorithm and a preset expected dependent variable to obtain process parameters; The step of screening the group of candidate feasible solutions according to the preset least square regression algorithm and the preset expected dependent variable to obtain process parameters comprises the following steps: randomly extracting and returning the historical experimental process data according to a preset repetition number to obtain a plurality of Bootstrap samples; constructing a plurality of second partial least square regression models and a plurality of second regression coefficient matrices according to the least square regression algorithm and the plurality of Bootstrap samples; performing inverse solution on the plurality of second regression coefficient matrices according to the preset expected dependent variable and the plurality of second partial least square regression models to obtain a plurality of estimated values; generating a confidence interval according to the plurality of estimated values; screening the group of candidate feasible solutions according to the confidence interval to obtain the process parameters; generating a silicon carbide epitaxial process formula and an evaluation report according to the process parameters.

2. The method of claim 1, wherein the process recipe is generated by: The step of performing calculation on the historical experimental process data according to the preset elimination condition, the historical data row number and the data selection condition to obtain mean square errors comprises the following steps: screening the historical experimental process data according to the elimination condition and the historical data row number to obtain a training set and a discarded data set; The preset partial least squares regression algorithm is trained according to the training set, so as to obtain a first partial least squares regression model; The substitute independent variables and the discarded dependent variables are obtained from the discarded data set according to the data selection condition; The dependent variable prediction value is obtained by predicting the substitute independent variables according to the first partial least squares regression model; The mean square error is calculated according to the dependent variable prediction value and the discarded dependent variable.

3. The method of claim 2, wherein the process recipe is generated by: The mean square error and the second component correlation coefficient are analyzed according to the preset outer loop iteration stop condition, so as to obtain the optimal component number, including: The average value of all mean square errors is calculated, so as to obtain a mean square error average value set; The data statistics of the mean square error average value set are performed, so as to obtain a mean square error average value number; The second load is calculated according to the standardized dependent variable and the second component correlation coefficient; The residual matrix is generated according to the second load, the standardized dependent variable and the second component correlation coefficient; It is judged whether the mean square error average value number or the residual matrix meets the outer loop iteration stop condition; When the second mean square error average value number or the residual matrix meets the outer loop iteration stop condition, the minimum mean square error average value is obtained from the mean square error average value set; The component number corresponding to the minimum mean square error average value is obtained according to the minimum mean square error average value, so as to obtain the optimal component number.

4. The method of claim 3, wherein the process recipe is generated by: A group of candidate feasible solutions is generated according to the optimal component number, the first load set and the target process index, including: The first partial least squares regression model is optimized according to the optimal component number, so as to obtain an optimized least squares regression model; The target process index is solved according to the optimized least squares regression model and the first load set, so as to obtain a group of candidate feasible solutions.

5. The method of claim 4, wherein the process recipe is generated by: The target process index is solved according to the optimized least squares regression model and the first load set, so as to obtain a group of candidate feasible solutions, including: All weight vectors are obtained, so as to obtain a weight vector set; The first regression coefficient matrix and the generalized inverse matrix are generated according to the weight vector set and the first load set; It is judged whether the first regression coefficient matrix is a column full rank matrix; When the first regression coefficient matrix is a column full rank matrix, the generalized inverse matrix, the dependent variable and the first regression coefficient matrix are solved according to a preset solving formula, so as to obtain a first exact solution; The target process index is solved according to the optimized least squares regression model and the first exact solution, so as to obtain a group of candidate feasible solutions; On the contrary, the generalized inverse matrix, the dependent variable and the first regression coefficient matrix are iteratively solved according to a numerical method, so as to obtain a second exact solution; The target process index is solved according to the optimized least squares regression model and the second exact solution, so as to obtain a group of candidate feasible solutions.

6. The method of claim 5, wherein the process recipe is generated by: The first regression coefficient matrix and the generalized inverse matrix are generated according to the weight vector set and the first load set, including: The weight matrix is generated according to the weight vector set; All second loads are obtained, so as to obtain a second load set; The second load matrix is generated according to the second load set; The first load matrix is generated according to the first load set; The first regression coefficient matrix is generated according to the weight matrix, the first load matrix and the second load matrix; The first regression coefficient matrix is transformed, so as to obtain the generalized inverse matrix.

7. A silicon carbide epitaxy process recipe generation apparatus, characterized by, Including: The process data acquisition module is configured to acquire historical experimental process data; The variable generation module is configured to generate independent variables and dependent variables according to historical experimental process data; The standardization module is configured to standardize the independent variables and the dependent variables to obtain standardized independent variables and standardized dependent variables; The selection module is configured to select a first component correlation coefficient from the standardized dependent variables according to a preset maximum variance column condition; The weight vector generation module is configured to generate a weight vector according to the standardized independent variables and the first component correlation coefficient; The coefficient generation module is configured to generate a second component correlation coefficient according to the weight vector and the standardized independent variables; The solving module is configured to solve the historical experimental process data and the preset target process index according to a preset data selection condition and the second component correlation coefficient to obtain a group of candidate feasible solutions, and specifically includes the following steps: Performing row analysis on the historical experimental process data to obtain a historical data row number; Performing calculation on the historical experimental process data according to a preset elimination condition, the historical data row number and the data selection condition to obtain a mean square error; Performing data statistics on all the mean square errors to obtain a mean square error number; Performing calculation on the standardized independent variables and the second component correlation coefficient to obtain a first load; Determining whether the mean square error number satisfies a preset inner loop iteration stop condition; If the mean square error number does not satisfy the inner loop iteration stop condition, returning to perform calculation on the historical experimental process data according to the elimination condition, the historical data row number and the data selection condition until the mean square error number satisfies the inner loop iteration stop condition; If the mean square error number satisfies the inner loop iteration stop condition, analyzing the mean square error and the second component correlation coefficient according to a preset outer loop iteration stop condition to obtain an optimal component number; Obtaining all the first loads to obtain a first load set; Generating a group of candidate feasible solutions according to the optimal component number, the first load set and the target process index; The screening module is configured to screen the group of candidate feasible solutions according to a preset least square regression algorithm and a preset expected dependent variable to obtain the process parameters, and specifically includes the following steps: Randomly extracting and returning the historical experimental process data according to a preset repetition number to obtain a plurality of Bootstrap samples; Constructing a plurality of second partial least square regression models and a plurality of second regression coefficient matrices according to the least square regression algorithm and the plurality of Bootstrap samples; Solving the plurality of second regression coefficient matrices in reverse according to the preset expected dependent variable and the plurality of second partial least square regression models to obtain a plurality of estimated values; Generating a confidence interval according to the plurality of estimated values; Screening the group of candidate feasible solutions according to the confidence interval to obtain the process parameters; The recipe and report generation module is configured to generate a silicon carbide epitaxy process recipe and an evaluation report according to the process parameters.

8. A silicon carbide epitaxy process recipe generation apparatus, characterized by, The silicon carbide epitaxy process recipe generation device includes a memory and at least one processor, and the memory stores instructions; At least one processor calls the instructions in the memory to enable the silicon carbide epitaxy process recipe generation device to perform each step of the silicon carbide epitaxy process recipe generation method according to any one of claims 1-6.

Images