Wafer test parameter optimization method and system

Abstract

The invention provides a wafer test parameter optimization method and system, and the method comprises the steps: dividing all test parameters into a plurality of to-be-optimized parameters and neighborhood parameters of each to-be-optimized parameter through a prediction value of a next round of wafer test indexes, and determining a difference gradient between the parameter values of each to-be-optimized parameter and the neighborhood parameters of the to-be-optimized parameter; further mapping to obtain a neighborhood space, determining neighborhood aggregation features of all to-be-optimized parameters through the neighborhood space, constructing a parameter optimization model based on a gradient boosting tree algorithm according to the obtained adaptive parameters and the predicted value, and determining optimization confidence of the parameter optimization model according to the neighborhood aggregation features. And the parameter optimization model carries out adaptive optimization on each to-be-optimized parameter based on the optimization confidence. By adopting the scheme of the invention, the test parameters of the crystal grains in the wafer are optimized by establishing the computer-aided optimization model, so that the escape rate of the wafer test is reduced.

Description

Technical Field

[0001] The present application relates to the field of wafer computer-aided testing technology, and more specifically, to a wafer testing parameter optimization method and system. Background Art

[0002] Wafer testing refers to the process of testing the electrical and physical properties of each grain in a semiconductor wafer. The purpose of the test is to evaluate the quality and reliability of the wafer to ensure that the wafer performs well in manufacturing and actual applications. The parameters in the wafer testing process include: current, voltage, frequency, etc.

[0003] In semiconductor manufacturing, the optimization of wafer test parameters is a key link to ensure that the performance of grains meets the standards. Machine learning can be used to analyze historical test data, identify complex patterns and relationships, and achieve more accurate parameter optimization. Traditional machine learning optimization methods usually use statistical methods to adjust test parameters on historical test data. Although the stability of the test results is guaranteed to a certain extent, the traditional machine learning statistical methods cannot statistically analyze the differences in test parameters corresponding to each grain in the wafer, resulting in serious test escapes, making it impossible to accurately determine whether the wafer is qualified. Therefore, how to optimize the test parameters of the grains in the wafer by establishing a computer-aided optimization model to reduce the escape rate of wafer testing has become a difficult problem faced by the industry. Summary of the invention

[0004] The present application provides a wafer test parameter optimization method and system, which can optimize the test parameters of grains in a wafer by establishing a computer-aided optimization model to reduce the escape rate of wafer testing.

[0005] In a first aspect, the present application provides a wafer test parameter optimization method, comprising the following steps:

[0006] Collect test parameters of all dies during wafer testing;

[0007] Based on the predicted values ​​of the next round of wafer test indicators, all test parameters are divided into a plurality of parameters to be optimized and neighborhood parameters of each parameter to be optimized, and then the difference gradient of parameter values ​​between each parameter to be optimized and its neighborhood parameters is determined;

[0008] A preset spatial mapping network synchronously maps all parameters to be optimized, all neighborhood parameters and all difference gradients to the neighborhood space, and then determines the neighborhood aggregation features of all parameters to be optimized based on the mapping vectors of each difference gradient in the neighborhood space;

[0009] Execute the next round of wafer testing operation, obtain the adaptability parameters of the grains corresponding to the parameters to be optimized, build a parameter optimization model based on the gradient boosting tree algorithm according to all the adaptability parameters and the predicted values, and determine the optimization confidence of the parameter optimization model according to the neighborhood aggregation features;

[0010] The parameter optimization model adaptively optimizes each parameter to be optimized based on the optimization confidence.

[0011] In some embodiments, all test parameters are divided into a plurality of parameters to be optimized and the neighborhood parameters of each parameter to be optimized based on the predicted value of the next round of wafer test indicators, specifically including:

[0012] Determine the predicted values ​​of the next round of wafer test indicators;

[0013] Based on the predicted value, all test parameters to be optimized are screened out from all test parameters, thereby obtaining a plurality of parameters to be optimized;

[0014] The neighborhood parameters of each parameter to be optimized are divided according to the distribution characteristics of the grains in the wafer.

[0015] In some embodiments, dividing the neighborhood parameters of each parameter to be optimized according to the distribution characteristics of the grains in the wafer specifically includes:

[0016] Obtain the distribution characteristics of all grains in the wafer;

[0017] Determine the neighborhood range of each parameter to be optimized according to the distribution characteristics;

[0018] A neighborhood parameter of each parameter to be optimized is determined based on the neighborhood range of each parameter to be optimized.

[0019] In some embodiments, determining the difference gradient of parameter values ​​between each parameter to be optimized and its neighboring parameters specifically includes:

[0020] For each parameter to be optimized, determine the difference between the parameter to be optimized and each test parameter in the corresponding neighborhood parameter;

[0021] The difference gradient of the parameter value between the parameter to be optimized and the corresponding neighborhood parameter is determined through all the difference values, and then the difference gradient of the parameter value between each parameter to be optimized and the corresponding neighborhood parameter is obtained.

[0022] In some embodiments, the preset spatial mapping network synchronously maps all parameters to be optimized, all neighborhood parameters, and all difference gradients to the neighborhood space, specifically including:

[0023] Obtain the distribution characteristics of all grains in the wafer;

[0024] Constructing a spatial mapping network based on Gaussian process regression according to the distribution characteristics;

[0025] Amplifying all parameters to be optimized and all neighborhood parameters through the spatial mapping network to obtain all amplified parameters to be optimized and all amplified neighborhood parameters;

[0026] The spatial mapping network constructs a neighborhood space based on all magnified parameters to be optimized, all magnified neighborhood parameters, and all difference gradients.

[0027] In some embodiments, determining the optimization confidence of the parameter optimization model by the neighborhood aggregation feature specifically includes:

[0028] Obtaining a feature set of the parameter optimization model input;

[0029] In the parameter optimization model, the feature importance of the feature set is evaluated by using the neighborhood aggregation feature to obtain the feature importance of all adaptive parameters in the feature set;

[0030] determining an attribution value and a Shapley value for each fitness parameter in the parameter optimization model based on all feature importances;

[0031] A weighted analysis is performed on all attribution values ​​and all Shapley values ​​to obtain the optimization variables of the parameter optimization model, and then a confidence analysis is performed on the optimization variables to obtain the optimization confidence of the parameter optimization model.

[0032] In some embodiments, the parameter optimization model adaptively optimizes each parameter to be optimized based on the optimization confidence, specifically including:

[0033] Obtaining optimization variables of the parameter optimization model;

[0034] If the optimization confidence is greater than or equal to a preset confidence threshold, then for each parameter to be optimized, the parameter to be optimized is adjusted by the optimization variable to obtain the test parameter after the optimization of the parameter to be optimized, and then the test parameters after all the parameters to be optimized are obtained to complete the optimization of the wafer test parameters.

[0035] In a second aspect, the present application provides a wafer test parameter optimization system, comprising:

[0036] The acquisition module is used to collect the test parameters of all dies during the wafer testing process;

[0037] A processing module, used for dividing all test parameters into a plurality of parameters to be optimized and neighborhood parameters of each parameter to be optimized based on the predicted values ​​of the next round of wafer test indicators, and then determining the difference gradient of parameter values ​​between each parameter to be optimized and its neighborhood parameters;

[0038] The processing module is further used to synchronously map all parameters to be optimized, all neighborhood parameters and all difference gradients to the neighborhood space by a preset spatial mapping network, and then determine the neighborhood aggregation features of all parameters to be optimized based on the mapping vectors of each difference gradient in the neighborhood space;

[0039] The processing module is further used to perform the next round of wafer testing operations, obtain the adaptability parameters of the grains corresponding to the parameters to be optimized, build a parameter optimization model based on the gradient boosting tree algorithm according to all the adaptability parameters and the predicted values, and determine the optimization confidence of the parameter optimization model according to the neighborhood aggregation features;

[0040] An execution module is used for the parameter optimization model to adaptively optimize each parameter to be optimized based on the optimization confidence.

[0041] In a third aspect, the present application provides a computer device, comprising a memory and a processor, wherein the memory stores codes, and the processor is configured to obtain the codes and execute the above-mentioned wafer test parameter optimization method.

[0042] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned wafer test parameter optimization method is implemented.

[0043] The technical solution provided by the embodiments disclosed in this application has the following beneficial effects:

[0044] The wafer test parameter optimization method and system provided by the present application include:

[0045] In an embodiment of the present application, first, the test parameters of all grains in the wafer testing process are collected; based on the predicted values ​​of the next round of wafer testing indicators, all the test parameters are divided into multiple parameters to be optimized and neighborhood parameters of each parameter to be optimized, and then the difference gradient of the parameter value between each parameter to be optimized and its neighborhood parameter is determined; all the parameters to be optimized, all the neighborhood parameters and all the difference gradients are synchronously mapped to the neighborhood space by a preset spatial mapping network, and then the neighborhood aggregation characteristics of all the parameters to be optimized are determined based on the mapping vectors of each difference gradient in the neighborhood space; the next round of wafer testing operations is performed to obtain the adaptability parameters of the grains corresponding to each parameter to be optimized, and a parameter optimization model based on the gradient boosting tree algorithm is constructed according to all the adaptability parameters and the predicted values, and the optimization confidence of the parameter optimization model is determined by the neighborhood aggregation characteristics; the parameter optimization model adaptively optimizes each parameter to be optimized based on the optimization confidence.

[0046] It can be seen that, firstly, the present application divides the test parameters into parameters to be optimized and neighborhood parameters through the predicted values ​​of the test indicators, and then obtains the neighborhood aggregation features, which is conducive to locating the faulty grains to reduce the test escapes. When performing the next round of testing on the faulty grains, the neighborhood aggregation features can also be used as a reference for parameter optimization to optimize the adaptability of the test parameter optimization; secondly, the next round of testing is carried out and the adaptive parameters are obtained, which can provide real-time feedback on the changes of each parameter to be optimized during the next round of testing, so that the parameter optimization model can adjust the optimization strategy as the test indicators change, thereby effectively reducing the escape rate; then, a parameter optimization model is constructed based on the adaptive parameters and the predicted values, which is conducive to improving the accuracy and adaptability of the parameter optimization model, thereby further reducing the test escape rate; finally, the optimization confidence of the parameter optimization model is determined through the neighborhood aggregation features, and then the optimization confidence is adaptively optimized for each parameter to be optimized, which not only improves the accuracy of the test, but also facilitates the identification and correction of the deviation of the test parameters and reduces the escape rate; in summary, the present application scheme can optimize the test parameters of the grains in the wafer, thereby reducing the escape rate of the wafer test. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is an exemplary flow chart of a wafer test parameter optimization method according to some embodiments of the present application;

[0048] Figure 2 is a schematic diagram of a process for determining a difference gradient according to some embodiments of the present application;

[0049] Figure 3 is a schematic diagram of a process for determining neighborhood aggregation features according to some embodiments of the present application;

[0050] Figure 4 is a schematic diagram of the structure of a wafer test parameter optimization system according to some embodiments of the present application;

[0051] Figure 5 It is a structural schematic diagram of a computer device for implementing a wafer test parameter optimization method according to some embodiments of the present application. DETAILED DESCRIPTION

[0052] In order to better understand the technical solution of the present application, the technical solution of the present application will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0053] refer to Figure 1 , which is an exemplary flow chart of a wafer test parameter optimization method according to some embodiments of the present application. The wafer test parameter optimization method 100 mainly includes the following steps:

[0054] In step 101, test parameters of all dies during wafer testing are collected.

[0055] It should be noted that the test parameters in this application refer to the comprehensive values ​​of the historical test parameters of the grains in the wafer from the beginning of wafer testing to the current testing stage, wherein the test parameters may be static leakage current or reverse breakdown voltage or output voltage or other electrical performance parameters.

[0056] In specific implementation, the collection of test parameters of all grains during the wafer testing process can be achieved in the following manner, namely: for each grain in the wafer, monitor the test of the grain, and automatically collect the historical test parameters of all test stages from the start of the wafer test to the current moment, and then use the comprehensive value obtained by analyzing all historical test parameters as the test parameter of the grain, and then obtain the test parameter of each grain in the wafer; as a preferred embodiment, the comprehensive value can be obtained by analyzing all the test parameters that have been tested using the autoregressive analysis model in the prior art. In other embodiments, other existing models can also be used for analysis, which will not be repeated here; it should also be noted that the wafer test in the present application includes multiple test stages.

[0057] In step 102, all test parameters are divided into a plurality of parameters to be optimized and neighborhood parameters of each parameter to be optimized based on the predicted values ​​of the next round of wafer test indicators, and then the difference gradient of the parameter value between each parameter to be optimized and its neighborhood parameters is determined.

[0058] In some embodiments, dividing all test parameters into multiple parameters to be optimized and neighborhood parameters of each parameter to be optimized based on the predicted values ​​of the next round of wafer test indicators can be implemented by the following steps:

[0059] Determine the predicted values ​​of the next round of wafer test indicators;

[0060] Based on the predicted value, all test parameters to be optimized are screened out from all test parameters, thereby obtaining a plurality of parameters to be optimized;

[0061] The neighborhood parameters of each parameter to be optimized are divided according to the distribution characteristics of the grains in the wafer.

[0062] In a specific implementation, the test parameter that needs to be optimized refers to the test parameter whose parameter value has not reached the predicted value. Therefore, based on the predicted value, all test parameters that need to be optimized are screened out from all test parameters, and then a plurality of parameters to be optimized are obtained, that is: the parameter values ​​of all test parameters are compared with the test, and then all test parameters with parameter values ​​less than the predicted value are taken as parameters to be optimized, thereby obtaining a plurality of parameters to be optimized.

[0063] The distribution characteristics can be obtained during wafer fabrication.

[0064] In some embodiments, dividing the neighborhood parameters of each parameter to be optimized according to the distribution characteristics of the grains in the wafer can be implemented in the following manner, namely:

[0065] Obtain the distribution characteristics of all grains in the wafer;

[0066] Determine the neighborhood range of each parameter to be optimized according to the distribution characteristics;

[0067] A neighborhood parameter of each parameter to be optimized is determined based on the neighborhood range of each parameter to be optimized.

[0068] It should be noted that in the present application, one crystal grain corresponds to one test parameter, one parameter to be optimized corresponds to one test parameter, and thus one parameter to be optimized corresponds to one crystal grain.

[0069] In a specific implementation, the distribution characteristics include the spatial coordinates of each grain in the wafer and the regional density of the grains in the wafer, and the distribution characteristics of all the grains in the wafer are obtained, that is, the spatial coordinates of each grain in the wafer and the regional density of the grains in the wafer are obtained by using the grain boundary density measurement method, grain size measurement method and electron backscatter diffraction technology in the prior art, and the obtained spatial coordinates of each grain in the circle and the regional density of the grains in the wafer are used as the distribution characteristics of all the grains in the wafer; the neighborhood range of each parameter to be optimized is determined according to the distribution characteristics, that is, the spatial coordinates of each grain in the wafer are used as the neighborhood range of each parameter to be optimized ... and the neighborhood range of each parameter to be optimized is determined according to the distribution characteristics. The regional density of the grains in the target and wafer is input into the visual segmentation algorithm in the existing machine learning algorithm to obtain the neighborhood range of the grains corresponding to each parameter to be optimized in the wafer, wherein the regional density is linearly related to the neighborhood range; the neighborhood parameters of each parameter to be optimized are determined based on the neighborhood range of each parameter to be optimized, that is: for the grains corresponding to each parameter to be optimized, a plurality of neighborhood grains of the grains are obtained by dividing the neighborhood range of the grains in the visual segmentation algorithm, and then the test parameters of all the neighborhood grains are used as the neighborhood parameters of the parameters to be optimized, thereby obtaining the neighborhood parameters of each parameter to be optimized.

[0070] Preferably, in some embodiments, determining the predicted value of the next round of wafer test indicators can be achieved by using the following steps:

[0071] Construct an indicator prediction model based on BP neural network;

[0072] The trained BP neural network-based indicator prediction model is used to analyze and obtain the predicted values ​​of the next round of wafer test indicators.

[0073] It should be noted that the indicator prediction model based on BP neural network constructed in the present application specifically includes: data preparation, BP neural network architecture design based on back propagation algorithm, model training, model evaluation and model deployment, wherein the BP neural network architecture includes: input layer, hidden layer, output layer and activation functions of each network layer. The use of the indicator prediction model based on BP neural network can improve the confidence of the predicted value.

[0074] In specific implementation, the following steps can be used to analyze and obtain the predicted values ​​of the next round of wafer test indicators using the trained BP neural network-based indicator prediction model: first, collect the test indicator set of the wafer test history, and divide the test indicator set into a training data set and a verification data set; second, use the training data set and the verification data set to train and verify the BP neural network-based indicator prediction model respectively, and the confidence and prediction deviation value of the indicator prediction model can be obtained through model verification; then, use the trained BP neural network-based indicator prediction model to analyze and obtain the first prediction value of the next round of wafer test indicators; finally, obtain the prediction deviation value of the indicator prediction model, and determine the prediction value of the next round of wafer test indicators through the prediction deviation value and the first prediction value.

[0075] It should be noted that the first prediction value is a preliminary prediction value of the next round of wafer test indicators. The first prediction value combined with the deviation of the indicator prediction model can obtain the predicted value of the next round of wafer test indicators, that is: as a preferred embodiment, the predicted value of the next round of wafer test indicators is determined by the predicted deviation value and the first prediction value. The following formula can be used, namely: predicted value = N*first predicted value-(1-N)*prediction deviation value, where N is the confidence of the trained BP neural network-based indicator prediction model, and the confidence value is between 0.6-0.9.

[0076] In some embodiments, reference Figure 2 As shown, this figure is a schematic diagram of the process of determining the difference gradient shown in some embodiments of the present application. In this embodiment, the difference gradient of the parameter value between each parameter to be optimized and its neighboring parameters can be determined by the following steps:

[0077] In step 1021, for each parameter to be optimized, the difference value between the parameter to be optimized and each test parameter in the corresponding neighborhood parameter is determined;

[0078] In step 1022, the difference gradient of the parameter value between the parameter to be optimized and the corresponding neighborhood parameter is determined through all the difference values, and then the difference gradient of the parameter value between each parameter to be optimized and the corresponding neighborhood parameter is obtained.

[0079] In addition, in the specific implementation, the difference value of each test parameter in the parameter to be optimized and the corresponding neighborhood parameter can be determined in the following way, that is: for each test parameter in the corresponding field parameter, the difference between the parameter value of the parameter to be optimized and the test parameter is used as the difference value between the parameter to be optimized and the test parameter, and then the difference value between the parameter to be optimized and each test parameter in the corresponding neighborhood parameter is obtained.

[0080] It should be noted that the difference gradient in the present application is a measure of the degree of difference in parameter values ​​between the parameter to be optimized and the corresponding neighborhood parameter. The larger the difference gradient, the greater the difference in parameter values ​​between the parameter to be optimized and the corresponding neighborhood parameter. The smaller the difference gradient, the smaller the difference in parameter values ​​between the parameter to be optimized and the corresponding neighborhood parameter. The difference gradient can be used to adjust the optimization strategy of the test parameters in real time to reduce the escape rate of wafer testing. As a preferred embodiment, the difference gradient of the parameter values ​​between the parameter to be optimized and the corresponding neighborhood parameter can be determined by all the difference values. The following method is used, namely: first, all the difference values ​​are sorted in order from small to large; then, the difference between every two adjacent difference values ​​is calculated; finally, the average value of all calculated difference values ​​is used as the difference gradient of the parameter values ​​between the parameter to be optimized and the corresponding neighborhood parameter. In other embodiments, other methods can also be used to determine the difference gradient, which is not limited here.

[0081] In step 103, a preset spatial mapping network synchronously maps all parameters to be optimized, all neighborhood parameters and all difference gradients to the neighborhood space, and then determines the neighborhood aggregation features of all parameters to be optimized based on the mapping vectors of each difference gradient in the neighborhood space.

[0082] In some embodiments, the following steps may be used to synchronously map all parameters to be optimized, all neighborhood parameters, and all difference gradients to the neighborhood space by a preset spatial mapping network, namely:

[0083] Obtain the distribution characteristics of all grains in the wafer;

[0084] Constructing a spatial mapping network based on Gaussian process regression according to the distribution characteristics;

[0085] Amplifying all parameters to be optimized and all neighborhood parameters through the spatial mapping network to obtain all amplified parameters to be optimized and all amplified neighborhood parameters;

[0086] The spatial mapping network constructs a neighborhood space based on all magnified parameters to be optimized, all magnified neighborhood parameters, and all difference gradients.

[0087] Among them, in the specific implementation, constructing a spatial mapping network based on Gaussian process regression according to the distribution characteristics can be implemented in the following manner, namely: establishing a Gaussian process regression model, conditionally constraining the Gaussian process regression model through the regional density of grains in the distribution characteristics (that is, taking the regional density as the constraint amount of the Gaussian distribution function in the Gaussian process regression model), and obtaining a spatial mapping network, wherein the spatial mapping network includes an input layer, a hidden layer, an output layer and a Gaussian distribution function; amplifying all parameters to be optimized and all neighborhood parameters through the spatial mapping network to obtain all amplified parameters to be optimized and all amplified neighborhood parameters can be implemented in the following manner, namely: in the input layer of the spatial mapping network, amplifying the output weight of the Gaussian distribution function through the regional density of grains in the wafer, and inputting all parameters to be optimized and all neighborhood parameters into the input layer of the spatial mapping network for amplification, forward propagating all parameters to be optimized and all neighborhood parameters through the spatial mapping network, and obtaining all amplified parameters to be optimized and all amplified neighborhood parameters through the output layer of the spatial mapping network. In other embodiments, other methods can also be used to obtain them, which are not limited here.

[0088] In addition, in the specific implementation, the neighborhood space is an information scatter plot of all the parameters to be optimized and all the neighborhood parameters after magnification. The neighborhood space can amplify the difference between the parameters to be optimized and the neighborhood parameters to facilitate subsequent processing. As a preferred embodiment, the spatial mapping network constructs a neighborhood space based on all the magnified parameters to be optimized, all the magnified neighborhood parameters and all the difference gradients. It can be implemented in the following way, namely: the distribution weight of the spatial mapping network can be set according to the proportion of each difference gradient, and the positions of all the parameters to be optimized and all the neighborhood parameters in the spatial mapping network can be adjusted by setting the distribution weight, so as to draw an information scatter plot, and use the drawn information scatter plot as the neighborhood space. In other embodiments, other methods can also be used to determine the domain space, which is not limited here.

[0089] It should be noted that the spatial mapping network (Spatial Transformer Networks, STN) in this application is a deep learning model that aims to enhance the network's adaptability to geometric transformations.

[0090] In some embodiments, reference Figure 3 As shown, this figure is a schematic diagram of the process of determining neighborhood aggregation features shown in some embodiments of the present application. In this embodiment, the neighborhood aggregation features of all parameters to be optimized based on the mapping vectors of each difference gradient in the neighborhood space can be determined by the following steps:

[0091] In step 1031, all difference gradients are plotted as gradient change curves;

[0092] In step 1032, a mapping vector of each difference gradient in the neighborhood space is determined based on the gradient change curve;

[0093] In step 1033, the distances between different parameters to be optimized in the neighborhood space are obtained;

[0094] In step 1034, neighborhood aggregation features of all parameters to be optimized are determined based on all distances and all mapping vectors.

[0095] Among them, in the specific implementation, all the difference gradients are plotted as gradient change curves, which can be achieved in the following manner, namely: the Matplotlib library in the prior art can be used to plot all the difference gradients as the curve obtained by using the gradient change curve, wherein the gradient change curve can intuitively show all the difference gradients, and in addition, in the neighborhood space, the difference gradient is also amplified as the parameters to be optimized and the neighborhood parameters are amplified, which is conducive to real-time monitoring of the difference changes in the test parameters of the grains during the wafer test process, thereby facilitating the screening of faulty grains, and further facilitating the optimization of faulty grains and reducing test escapes; based on the gradient change curve, the mapping vectors of each difference gradient in the neighborhood space can be determined in the following manner, namely: the gradient() function in the Matplotlib library can be used to obtain the tangent and slope of each difference gradient in the gradient change curve, and then determine the direction of the mapping vector of each difference gradient in the neighborhood space according to the obtained tangent and slope, and use the value corresponding to each difference gradient as the value of the mapping vector of the corresponding difference gradient in the neighborhood space, and then obtain the mapping vector of each difference gradient in the neighborhood space.

[0096] In addition, in a specific implementation, the distances between different parameters to be optimized in the neighborhood space are obtained. For example, the Euclidean distance between each parameter to be optimized in the neighborhood space can be used as the distance between the corresponding two parameters to be optimized, thereby obtaining the distances between different parameters to be optimized. In other embodiments, other methods can also be used to obtain the distances, which will not be described here.

[0097] In addition, in the specific implementation, the neighborhood aggregation features of all parameters to be optimized based on all distances and all mapping vectors can be achieved in the following manner, namely: for every two parameters to be optimized in the neighborhood space, first, the distance between the two parameters to be optimized is obtained; secondly, the sum of the adjacent absolute values ​​of the mappings corresponding to the two parameters to be optimized is calculated; then, the product of the distance between the two parameters to be optimized and the calculated sum is used as the aggregation feature between the two parameters to be optimized, and then the aggregation feature between every two parameters to be optimized in the neighborhood space is obtained; finally, the variance of all aggregation features is used as the neighborhood aggregation feature of all parameters to be optimized. In other embodiments, other methods can also be used to determine the neighborhood aggregation features of all parameters to be optimized, which are not limited here.

[0098] It should be noted that the neighborhood aggregation feature in the present application is an indicator value for measuring the degree of aggregation between different parameters to be optimized. In addition, the neighborhood aggregation feature can also reversely represent the degree of difference between different parameters to be optimized, that is, the larger the neighborhood aggregation feature, the smaller the difference between the parameters to be optimized, and the smaller the neighborhood aggregation feature, the larger the difference between the parameters to be optimized. By identifying the changes in the differences between test parameters through the neighborhood aggregation feature, it is helpful to locate the faulty grains. When performing the next round of testing on the faulty grains, the neighborhood aggregation feature can be used as a reference for parameter optimization to optimize the accuracy of the test parameters in wafer testing.

[0099] In step 104, the next round of wafer testing operation is performed to obtain the adaptability parameters of the grains corresponding to the parameters to be optimized, and a parameter optimization model based on the gradient boosting tree algorithm is constructed according to all the adaptability parameters and the predicted values, and the optimization confidence of the parameter optimization model is determined by the neighborhood aggregation features.

[0100] It should be noted that executing the next round of wafer testing operations refers to performing adaptability testing on the wafer, and obtaining the adaptability parameters of the grains corresponding to each parameter to be optimized means using the collected test parameters as the adaptability parameters of the grains in the wafer when performing adaptability testing on the wafer. Among them, the adaptability test is a test process that optimizes subsequent test conditions by analyzing historical test data, thereby improving the accuracy and efficiency of the test. In this application, when performing the second round of testing on the wafer, an adaptability test is used to ensure the test quality and reduce the test cost.

[0101] In some embodiments, constructing a parameter optimization model based on a gradient boosting tree algorithm according to all the adaptability parameters and the predicted values ​​can be implemented by the following steps:

[0102] Normalize all the adaptability parameters, and use all the normalized adaptability test parameters as feature sets;

[0103] The predicted value is used as an evaluation quantity, and a parameter optimization model based on a gradient boosting tree algorithm is initialized by the evaluation quantity and the feature set.

[0104] It should be noted that, in this embodiment, all the adaptability parameters are normalized to ensure that the different features of the input model are on the same scale. Standardization (Z-score) can be used to normalize the parameter values ​​of all the adaptability parameters, and all the normalized adaptability test parameters are used as the feature set, which can reduce the scale differences between different features of the input model to improve the stability and efficiency of model training.

[0105] In specific implementation, the predicted value is used as an evaluation quantity, and a parameter optimization model based on the gradient boosting tree algorithm is initialized by the evaluation quantity and the feature set. The following method is used, namely: first, a basic decision tree model is created, and the loss function of the decision tree model is established using the mean square error; second, the evaluation quantity and the feature set are divided into a training set, and the decision tree model is trained using the training set; then, iterative training is performed, and new trees are added in each iteration based on the gradient boosting tree algorithm; finally, the loss function value of the decision tree model is continuously verified using the divided verification set during the iteration process, and when the loss function value is the smallest, the trained model is used as the parameter optimization model based on the gradient boosting tree algorithm.

[0106] It should be noted that the parameter optimization model in the present application is a machine learning model used to optimize test parameters during wafer testing. The test parameters of the grains in the wafer can be adjusted through the parameter optimization model. The parameter optimization model can analyze the relationship between the historical test parameters and the current test parameters of the wafer test, identify and correct the deviations between the test parameters, so as to improve the accuracy of the test and reduce test escapes.

[0107] In some embodiments, determining the optimization confidence of the parameter optimization model by the neighborhood aggregation feature can be implemented by the following steps:

[0108] Obtaining a feature set of the parameter optimization model input;

[0109] In the parameter optimization model, the feature importance of the feature set is evaluated by using the neighborhood aggregation feature to obtain the feature importance of all adaptive parameters in the feature set;

[0110] determining an attribution value and a Shapley value for each fitness parameter in the parameter optimization model based on all feature importances;

[0111] A weighted analysis is performed on all attribution values ​​and all Shapley values ​​to obtain the optimization variables of the parameter optimization model, and then a confidence analysis is performed on the optimization variables to obtain the optimization confidence of the parameter optimization model.

[0112] In specific implementation, the feature importance of the feature set is evaluated through the neighborhood aggregation feature, and the feature importance of all adaptive parameters in the feature set can be obtained in the following manner, namely: in the parameter optimization model, for each branch node of the gradient boosting tree (one branch node corresponds to an adaptive parameter in the feature set), the difference between the Gini index of the branch node before iteration and the Gini index of the branch node after iteration is used as the weight for evaluating the feature importance of the feature set, and then the neighborhood aggregation feature is introduced. The neighborhood aggregation feature is an indicator for measuring the degree of aggregation of different parameters to be optimized in the neighborhood set. Therefore, the neighborhood The product of the aggregated feature and the weight is used as the feature importance of the feature value corresponding to the branch node, and then the feature importance of all adaptive parameters in the feature set is obtained; wherein, the Gini index is an indicator for measuring the importance of branch features in the corresponding model in the parameter optimization model. In the parameter optimization model, the smaller the Gini index, the more concentrated the feature values ​​of the branch nodes are and the smaller the differences are. In addition, the feature importance is an indicator for measuring the predictive role of the feature values ​​of the branch nodes in the parameter optimization model. According to the ranking of the feature importance, redundant node features can be eliminated to reduce the overfitting of the parameter optimization model and improve the optimization ability of the parameter optimization model.

[0113] In addition, in the specific implementation, the attribution value and Shapley value of each adaptability parameter in the parameter optimization model based on the importance of all features can be determined in the following manner, namely: in the parameter optimization model, for each adaptability parameter, the feature importance of the adaptability parameter is quantified as an attribution value through the feature importance evaluation mechanism in the gradient boosting tree algorithm, and then the attribution value of each adaptability parameter in the parameter optimization model is obtained, wherein the attribution value refers to the specific contribution of the features of the branch node to the output of the tree model, which can be quantified by the feature importance; in addition, the Shapley value refers to an indicator for the fair distribution of feature importance. The determination of the Shapley value is based on cooperative game theory, which can be used to determine the contribution of each adaptability parameter on the branch in the parameter optimization model. The SHAP library in Python can be used to solve the Shapley value of each adaptability parameter to obtain the Shapley value of each adaptability parameter in the parameter optimization model.

[0114] In specific implementation, the optimization variable refers to the dynamic adjustment amount of the parameter optimization model to different test parameters. As a preferred embodiment, all attribution values ​​and all Shapley values ​​are weighted analyzed to obtain the optimization variables of the parameter optimization model. The optimization variables can be implemented in the following manner, namely: weights can be assigned to the attribution values ​​and Shapley values ​​of each adaptability parameter according to the feature importance evaluation mechanism in the gradient boosting tree algorithm, and then all attribution values ​​and all Shapley values ​​are weightedly summed according to the assigned weights, and the value obtained by the weighted sum is used as the optimization variable of the parameter optimization model. In other embodiments, other methods can also be used to determine the optimization variables, which will not be repeated here.

[0115] It should be noted that the parameter optimization model in the present application optimizes different test parameters by adjusting the attribution value and the Shapley value allocation weights, that is, different test parameters correspond to different adjustment amounts, which is conducive to the parameter optimization model to adaptively optimize different parameters to be optimized.

[0116] In addition, in the specific implementation, the optimization confidence reflects the degree of trust of the parameter optimization model in the dynamic adjustment of different test parameters. As a preferred embodiment, the confidence analysis of the optimization variables is performed to obtain the optimization confidence of the parameter optimization model. It can be achieved in the following ways. For example, the cross-validation method in the prior art can be used to perform confidence analysis on the optimization variables. For example, the grid search or random search in the parameter optimization technology can be used to determine the performance of the parameter optimization model in dynamically adjusting different test parameters according to the optimization variables, thereby evaluating the confidence of the optimization variables. In other embodiments, other methods can also be used to perform confidence analysis on the optimization variables to obtain the optimization confidence of the parameter optimization model, which is not limited here.

[0117] In step 105, the parameter optimization model adaptively optimizes each parameter to be optimized based on the optimization confidence.

[0118] In some embodiments, the parameter optimization model can adaptively optimize each parameter to be optimized based on the optimization confidence by using the following steps:

[0119] Obtaining optimization variables of the parameter optimization model;

[0120] If the optimization confidence is greater than or equal to a preset confidence threshold, then for each parameter to be optimized, the parameter to be optimized is adjusted by the optimization variable to obtain the test parameter after the optimization of the parameter to be optimized, and then the test parameters after all the parameters to be optimized are obtained to complete the optimization of the wafer test parameters.

[0121] In specific implementation, the parameter to be optimized is adjusted by the optimization variable to obtain the test parameter after the parameter to be optimized is that the parameter optimization model adds the optimization variable to the parameter value of the parameter to be optimized to obtain the test parameter after the parameter to be optimized is optimized; the confidence threshold is the judgment quantity of the optimization confidence. When the optimization confidence is greater than or equal to the preset confidence threshold, the optimization variable corresponding to the optimization confidence can be used to adjust the parameter value of the parameter to be optimized.

[0122] In addition, in some embodiments, if the optimization confidence is less than a preset confidence threshold, the corresponding parameters in the parameter optimization model are updated, and the optimization variables of the parameter optimization model are re-determined until the optimization confidence is greater than or equal to the preset confidence threshold.

[0123] It should be noted that in the present application, the partial derivative (gradient) of the loss function under the corresponding parameters can be calculated by the gradient descent method, and then the parameters are updated in the opposite direction of the gradient. In other embodiments, the stochastic gradient descent method, AdaGrad (Adaptive Gradient) and other existing technologies can be used to update the corresponding parameters in the parameter optimization model, which will not be repeated here.

[0124] On the other hand, in some embodiments, the present application provides a wafer test parameter optimization system, referring to Figure 4 , which is a schematic diagram of the structure of a wafer test parameter optimization system according to some embodiments of the present application, the wafer test parameter optimization system 400 includes: a collection module 401, a processing module 402 and an execution module 403, which are respectively described as follows:

[0125] The acquisition module 401 in this application is mainly used to collect the test parameters of all the dies during the wafer testing process;

[0126] Processing module 402, in the present application, the processing module 402 is mainly used to divide all test parameters into multiple parameters to be optimized and neighborhood parameters of each parameter to be optimized based on the predicted value of the next round of wafer test indicators, and then determine the difference gradient of the parameter value between each parameter to be optimized and its neighborhood parameter;

[0127] The processing module 402 in the present application is also used to synchronously map all parameters to be optimized, all neighborhood parameters and all difference gradients to the neighborhood space by a preset spatial mapping network, and then determine the neighborhood aggregation features of all parameters to be optimized based on the mapping vectors of each difference gradient in the neighborhood space;

[0128] The processing module 402 in the present application is also used to perform the next round of wafer testing operations, obtain the adaptability parameters of the grains corresponding to the parameters to be optimized, build a parameter optimization model based on the gradient boosting tree algorithm according to all the adaptability parameters and the predicted values, and determine the optimization confidence of the parameter optimization model by the neighborhood aggregation features;

[0129] Execution module 403, in the present application, execution module 403 is mainly used for the parameter optimization model to adaptively optimize each parameter to be optimized based on the optimization confidence.

[0130] In addition, the present application also provides a computer device, which includes a memory and a processor, the memory stores codes, and the processor is configured to obtain the codes and execute the above-mentioned wafer test parameter optimization method.

[0131] In some embodiments, reference Figure 5 , which is a schematic diagram of the structure of a computer device for implementing a wafer test parameter optimization method according to some embodiments of the present application. The wafer test parameter optimization method in the above embodiment can be Figure 5 The computer device 500 shown in the figure is implemented, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503 and at least one communication interface 504.

[0132] The processor 501 may be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0133] The communication bus 502 may be used to transmit information between the above-mentioned components.

[0134] The memory 503 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a disk or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 503 may exist independently and be connected to the processor 501 via the communication bus 502. The memory 503 may also be integrated with the processor 501.

[0135] Among them, the memory 503 is used to store the program code for executing the solution of the present application, and the execution is controlled by the processor 501. The processor 501 is used to execute the program code stored in the memory 503. The program code may include one or more software modules. The wafer test parameter optimization method in the above embodiment can be implemented by one or more software modules in the program code in the processor 501 and the memory 503.

[0136] The communication interface 504 uses any transceiver or other device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0137] In a specific implementation, as an embodiment, a computer device may include multiple processors, each of which may be a single-CPU processor or a multi-CPU processor. The processor here may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0138] The above-mentioned computer device can be a general-purpose computer device or a special-purpose computer device. In a specific implementation, the computer device can be a desktop computer, a portable computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device or an embedded device. The embodiment of the present application does not limit the type of computer device.

[0139] In addition, the present application also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned wafer test parameter optimization method is implemented.

[0140] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0141] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A wafer test parameter optimization method, characterized in that: The steps include: Collect test parameters of all dies during wafer testing; Based on the predicted values ​​of the next round of wafer test indicators, all test parameters are divided into a plurality of parameters to be optimized and neighborhood parameters of each parameter to be optimized, and then the difference gradient of parameter values ​​between each parameter to be optimized and its neighborhood parameters is determined; A preset spatial mapping network synchronously maps all parameters to be optimized, all neighborhood parameters and all difference gradients to the neighborhood space, and then determines the neighborhood aggregation features of all parameters to be optimized based on the mapping vectors of each difference gradient in the neighborhood space; Execute the next round of wafer testing operation, obtain the adaptability parameters of the grains corresponding to the parameters to be optimized, build a parameter optimization model based on the gradient boosting tree algorithm according to all the adaptability parameters and the predicted values, and determine the optimization confidence of the parameter optimization model according to the neighborhood aggregation features; The parameter optimization model adaptively optimizes each parameter to be optimized based on the optimization confidence.

2. The method according to claim 1, characterized in that Based on the predicted values ​​of the next round of wafer test indicators, all test parameters are divided into multiple parameters to be optimized and the neighborhood parameters of each parameter to be optimized include: Determine the predicted values ​​of the next round of wafer test indicators; Based on the predicted value, all test parameters to be optimized are screened out from all test parameters, thereby obtaining a plurality of parameters to be optimized; The neighborhood parameters of each parameter to be optimized are divided according to the distribution characteristics of the grains in the wafer.

3. The method according to claim 2, characterized in that The neighborhood parameters of each parameter to be optimized are divided according to the distribution characteristics of the grains in the wafer, including: Obtain the distribution characteristics of all grains in the wafer; Determine the neighborhood range of each parameter to be optimized according to the distribution characteristics; A neighborhood parameter of each parameter to be optimized is determined based on the neighborhood range of each parameter to be optimized.

4. The method according to claim 1, characterized in that Determining the difference gradient of parameter values ​​between each parameter to be optimized and its neighboring parameters specifically includes: For each parameter to be optimized, determine the difference between the parameter to be optimized and each test parameter in the corresponding neighborhood parameter; The difference gradient of the parameter value between the parameter to be optimized and the corresponding neighborhood parameter is determined through all the difference values, and then the difference gradient of the parameter value between each parameter to be optimized and the corresponding neighborhood parameter is obtained.

5. The method according to claim 1, characterized in that The preset spatial mapping network maps all parameters to be optimized, all neighborhood parameters and all difference gradients to the neighborhood space synchronously, including: Obtain the distribution characteristics of all grains in the wafer; Constructing a spatial mapping network based on Gaussian process regression according to the distribution characteristics; Amplifying all parameters to be optimized and all neighborhood parameters through the spatial mapping network to obtain all amplified parameters to be optimized and all amplified neighborhood parameters; The spatial mapping network constructs a neighborhood space based on all magnified parameters to be optimized, all magnified neighborhood parameters, and all difference gradients.

6. The method according to claim 1, characterized in that Determining the optimization confidence of the parameter optimization model by the neighborhood aggregation feature specifically includes: Obtaining a feature set of the parameter optimization model input; In the parameter optimization model, the feature importance of the feature set is evaluated by using the neighborhood aggregation feature to obtain the feature importance of all adaptive parameters in the feature set; determining an attribution value and a Shapley value for each fitness parameter in the parameter optimization model based on all feature importances; A weighted analysis is performed on all attribution values ​​and all Shapley values ​​to obtain the optimization variables of the parameter optimization model, and then a confidence analysis is performed on the optimization variables to obtain the optimization confidence of the parameter optimization model.

7. The method according to claim 1, characterized in that The parameter optimization model adaptively optimizes each parameter to be optimized based on the optimization confidence, specifically including: Obtaining optimization variables of the parameter optimization model; If the optimization confidence is greater than or equal to a preset confidence threshold, then for each parameter to be optimized, the parameter to be optimized is adjusted by the optimization variable to obtain the test parameter after the optimization of the parameter to be optimized, and then the test parameters after all the parameters to be optimized are obtained to complete the optimization of the wafer test parameters.

8. A wafer test parameter optimization system, characterized in that: include: The acquisition module is used to collect the test parameters of all dies during the wafer testing process; A processing module, used for dividing all test parameters into a plurality of parameters to be optimized and neighborhood parameters of each parameter to be optimized based on the predicted values ​​of the next round of wafer test indicators, and then determining the difference gradient of parameter values ​​between each parameter to be optimized and its neighborhood parameters; The processing module is further used to synchronously map all parameters to be optimized, all neighborhood parameters and all difference gradients to the neighborhood space by a preset spatial mapping network, and then determine the neighborhood aggregation features of all parameters to be optimized based on the mapping vectors of each difference gradient in the neighborhood space; The processing module is further used to perform the next round of wafer testing operations, obtain the adaptability parameters of the grains corresponding to the parameters to be optimized, build a parameter optimization model based on the gradient boosting tree algorithm according to all the adaptability parameters and the predicted values, and determine the optimization confidence of the parameter optimization model according to the neighborhood aggregation features; An execution module is used for the parameter optimization model to adaptively optimize each parameter to be optimized based on the optimization confidence.

9. A computer device, comprising a memory and a processor, wherein the memory stores a code, characterized in that: The processor is configured to obtain the code and execute the wafer test parameter optimization method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the wafer test parameter optimization method according to any one of claims 1 to 7 is implemented.

Images