Parameter prediction model construction method, prediction method and prediction system for CD-SEM measurement

Abstract

The invention discloses a parameter prediction model construction method, a prediction method and a prediction system for CD-SEM measurement, and the model construction method comprises the steps: obtaining a data set which comprises at least one data set with a proximity effect feature, each data set comprises standard data and an SE signal set, the standard data comprises depth-to-width ratio data and / or height data, and the SE signal group in the data group with the proximity effect characteristic is a secondary electronic signal generated by performing electron beam scanning on a geometric structure with an adjacent structure; and training a pre-constructed neural network model by taking the SE signal group in the same data group in the data set as input and taking the standard data in the same data group as output to obtain a parameter prediction model for predicting depth-to-width ratio data and / or height data in CD-SEM measurement. According to the method, a hardware structure does not need to be modified, so that the problem of complexity of the hardware structure and the problem of reliability influence do not exist.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor measurement technology, and in particular to a parameter prediction model construction method, a prediction method and a prediction system for CD-SEM measurement. Background Art

[0002] Critical dimension scanning electron microscopy (CD-SEM) is a key technology in the semiconductor industry for measuring pattern structures on wafers or masks. With the development of the integrated circuit industry, the size of device structures has become smaller and smaller, and device structures have also evolved into three-dimensional. CD-SEM faces greater challenges, especially in three-dimensional (3D) metrology, where the height and aspect ratio of device structures are important parameters for characterizing three-dimensionality. However, CD-SEM cannot directly measure these important parameters, making it very inconvenient to obtain these parameters.

[0003] For the measurement of height and aspect ratio, there are several methods for measurement in the prior art: 1. Use images with two or more different incident electron beam angles, but this method increases the complexity of the measurement equipment and takes more time to collect images; 2. Use an exponential function to link the BSE (backscattered electron) signal intensity with the depth of the HAR (high aspect ratio) hole, but this method usually requires the CD SEM to work in a high acceleration voltage state, which is easy to damage the sample. In addition, compared with the secondary electron (SE) image, the image quality of the detection through BSE is lower. 3. By tilting the sample worktable, the electron beam multi-angle incident image is obtained, but the mechanical movement of the worktable brings instability and strict alignment problems. In response to the measurement of height and aspect ratio, Applied Materials has developed and implemented electron beam tilting on its product V4i+CD-SEM for height measurement. After calibration, two sets of images can be obtained by calibrating the low-angle (5°) beam and the high-angle (14°) beam, which are used to measure the sidewall edge width to calculate the height. However, the approach taken by Applied Materials V4i+ would complicate the CD SEM hardware and introduce reliability issues. Summary of the invention

[0004] The embodiment of the present invention provides a parameter prediction model construction method, prediction method and prediction system for CD-SEM measurement, so as to solve the problem that it is difficult to measure the aspect ratio and height parameters of the device structure in the prior art.

[0005] According to a first aspect of the present invention, there is provided a method for constructing a parameter prediction model for CD-SEM measurement, comprising:

[0006] Acquire a data set, wherein the data set includes at least one data set having a proximity effect feature, each data set includes standard data and an SE signal set, wherein the standard data includes aspect ratio data and / or height data, and the SE signal set in the data set having the proximity effect feature is a secondary electron signal generated by performing electron beam scanning on a geometric structure having adjacent structures;

[0007] The SE signal group in the same data group in the data set is used as input, and the standard data in the same data group is used as output, and the pre-constructed neural network model is trained to obtain a parameter prediction model for predicting aspect ratio data and / or height data in CD-SEM measurement.

[0008] The parameter prediction model construction method of the present invention trains the constructed neural network model by taking the SE signal group as input and the standard data corresponding to the SE signal group, that is, the corresponding aspect ratio data and / or height data as output, so that the obtained parameter prediction model can predict the aspect ratio data and / or height data of the corresponding device structure based on the SE signal. Compared with the method of the prior art, there is no need to modify the hardware structure, so there is no problem of complicating the hardware structure, there is no problem of reliability impact, and there is no need to adjust the working parameters of the CD-SEM, which will not cause damage to the sample and will not affect the quality of the detected image. The data set used in the training of the parameter prediction model of the present invention is a data set with a proximity effect feature, so that parameters related to the proximity effect feature can be predicted. After precision detection test, when the parameter prediction model constructed by the present invention predicts the aspect ratio and height, the average relative error between the aspect ratio prediction value and the standard value is: 2.4%, the percentage of prediction cases within 20% is 100.0%, and the average relative error between the height prediction value and the standard value is: 1.9%, and the percentage of prediction cases within 10% is 99.5%.

[0009] In some embodiments, the method for constructing the neural network model comprises:

[0010] Construct the network architecture of the neural network model based on the Identity module and the Dense module;

[0011] A neural network model is built according to the constructed network architecture and the models library under the Keras library that encapsulates TensorFlow. During construction, the compile function is used to compile the model, the loss function is specified as mse, Adam is specified as the optimizer of the neural network, and the metrics function is specified as mse.

[0012] Therefore, through such a setting, the network architecture of the parameter prediction model of the present invention adopts a fully connected layer to replace the convolution layer and the pooling layer. Compared with the traditional residual network that is more suitable for image processing, the modified model architecture is more suitable for nonlinear regression, so that the prediction results obtained by the neural network model after training are more accurate and the training efficiency is also higher.

[0013] In some embodiments, the network architecture of the constructed neural network model is configured to consist of two basic modules connected sequentially, a batch normalization layer, a fully connected layer, and a linear activation function, wherein each basic module is formed by sequentially connecting two Identity modules after a Dense module;

[0014] The Identity module includes three basic units connected in sequence, wherein each basic unit is composed of a fully connected layer, a batch normalization layer and a ReLU activation function connected in sequence, and a Dropout layer is inserted after the second basic unit;

[0015] The Dense module includes three basic units connected in sequence, wherein a Dropout layer is inserted after the second basic unit, and a residual path unit is respectively connected after the first basic unit and the second basic unit, and the residual path unit is composed of a fully connected layer and a batch normalization layer stacked in sequence, and the output of the residual path unit and the result between the batch normalization layer and the ReLU activation function in the third basic unit in the Dense module are configured to perform a tensor sum operation.

[0016] Therefore, through such settings, the network architecture can be further made more suitable for nonlinear regression, thereby improving the accuracy of prediction results and training efficiency.

[0017] In some embodiments, the data set is configured by performing SEM simulation imaging on a geometric structure having adjacent structures constructed in a simulated manner, obtaining standard data according to parameters of the constructed geometric structure, and obtaining SE signals according to parameters of the simulated imaging.

[0018] Therefore, through such a setting, the simulated structure can be used to replace the actual product for testing, which can greatly reduce the training cost. In addition, the parameters of the simulated structure are highly controllable, and device structures with various parameter combinations can be provided as data sets, providing different types of data samples for model training.

[0019] In some embodiments, the data set includes a training set and a test set, and the training set and the test set are obtained by performing random number division on several data groups in the data set;

[0020] When training the constructed neural network model, the constructed neural network model is trained using the data group in the training set to obtain model parameters, and the obtained model parameters are optimized using the data group in the test set.

[0021] Therefore, through such a setting, the data set can be divided, the model can be trained using the training set, and the model parameters can be optimized using the test set to ensure the accuracy of the parameter prediction model obtained through training. At the same time, the data set can be shuffled so that the data groups in the data set are no longer coherent, thereby reducing the parameter prediction model's consideration of data coherence during training, so that the final parameter prediction model can be more suitable for more different scenarios, thereby improving the prediction accuracy of the parameter prediction model.

[0022] In some embodiments, the method uses the SE signal group in the same data group in the data set as input and the standard data in the same data group as output to train the pre-constructed neural network model to obtain the parameter prediction model, including:

[0023] According to a preset number of training times, a data group in a training set is selected, and a pre-constructed neural network model is trained using a SE signal group in the same data group as input and a standard data in the same data group as output to obtain a first model parameter;

[0024] Update the neural network model according to the first model parameter to obtain a first neural network model;

[0025] Inputting the SE signal group in the data group in the test set into the first neural network model, and obtaining the prediction result of the first neural network model on the input SE signal group;

[0026] The loss function value is used to evaluate the error between the prediction result and the standard data in the corresponding data group, and the first neural network model is optimized according to the neural network optimizer and the error to obtain a parameter prediction model.

[0027] Therefore, through such settings, the settings in the neural network training process can be improved, and the MSE specified by the loss function can be used as an indicator to monitor the training effect in real time during the training process, and the neural network model can be iteratively optimized to improve the accuracy of the final parameter prediction model.

[0028] In some embodiments, before training the constructed neural network model, the data set is also preprocessed, including normalizing the data set.

[0029] Therefore, by setting it up like this, the bias of the simulator or the electron microscope itself can be avoided. In addition, data normalization can solve the problem of inconsistent scales of different features, making the impact of different features on the model more balanced. In addition, normalization can also prevent the problem of gradient disappearance or explosion, thereby accelerating the convergence of the model.

[0030] In some embodiments, the method further includes determining the prediction accuracy of the parameter prediction model based on the data set and the parameter prediction model, including:

[0031] Taking the SE signal group in the selected data group in the data set as input, inputting it into the parameter prediction model, and obtaining the prediction data output by the parameter prediction model;

[0032] The prediction accuracy of the parameter prediction model is determined by the metrics function based on the predicted data and the standard data in the data group where the input SE signal group is located.

[0033] Therefore, through such settings, the prediction accuracy of the trained parameter prediction model can be further determined as an accuracy reference.

[0034] According to a second aspect of the present invention, there is provided a training device for a parameter prediction model for CD-SEM measurement, comprising:

[0035] A data acquisition module, used to acquire a data set, wherein the data set includes at least one data group with a proximity effect feature, each data group includes standard data and an SE signal group, the standard data includes aspect ratio data and / or height data, and the SE signal group in the data group with the proximity effect feature is a secondary electron signal generated by electron beam scanning of a geometric structure with adjacent structures;

[0036] The training module is used to train the pre-built neural network model with the SE signal group in the same data group in the data set as input and the standard data in the same data group as output, so as to obtain a parameter prediction model for predicting aspect ratio data and / or height data in CD-SEM measurement.

[0037] According to a third aspect of the present invention, a parameter prediction method for CD-SEM measurement is provided, wherein the parameter prediction model obtained by the parameter prediction model construction method for CD-SEM measurement described in the first aspect and the SE signal of the device structure are used to predict the aspect ratio and / or height of the device structure.

[0038] According to a fourth aspect of the present invention, there is provided a parameter prediction system for CD-SEM measurement, comprising:

[0039] A prediction module is used to predict the aspect ratio and / or height of the device structure based on the SE signal of the device structure and the parameter prediction model obtained by the parameter prediction model construction method for CD-SEM measurement described in any one of claims 1 to 8. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0041] Figure 1 A flowchart of a method for constructing a parameter prediction model for CD-SEM measurement according to an embodiment of the present invention;

[0042] Figure 2 It is a schematic diagram of the overall structure of a geometric structure simulated and constructed when acquiring a data set in a method for constructing a parameter prediction model for CD-SEM measurement according to an embodiment of the present invention;

[0043] Figure 3 It is a schematic diagram of the cross-sectional structure of a geometric structure simulated and constructed when acquiring a data set in a method for constructing a parameter prediction model for CD-SEM measurement according to an embodiment of the present invention;

[0044] Figure 4 It is a comparison diagram of SE signals of structures with different bottom widths;

[0045] Figure 5 Comparison diagram of SE signals of structures with different side wall angles;

[0046] Figure 6 It is a comparison diagram of SE signals of structures with different heights;

[0047] Figure 7 This is a flow chart of step S11 in a method for constructing a parameter prediction model for CD-SEM measurement according to an embodiment of the present invention;

[0048] Figure 8 A schematic diagram of a network architecture of a neural network model in a method for constructing a parameter prediction model for CD-SEM measurement according to an embodiment of the present invention;

[0049] Fig. 9 This is a flow chart of step S12 in a method for constructing a parameter prediction model for CD-SEM measurement according to an embodiment of the present invention;

[0050] Fig.10A flowchart of a method for constructing a parameter prediction model for CD-SEM measurement according to another embodiment of the present invention;

[0051] Fig.11 for Fig.10 A flowchart of step S13 in the method for constructing a parameter prediction model for CD-SEM measurement according to the embodiment shown;

[0052] Fig.12 is a scatter plot of the aspect ratio in the standard data and the aspect ratio in the predicted data;

[0053] Fig.13 is a relative error distribution diagram between the aspect ratio in the standard data and the aspect ratio in the predicted data;

[0054] Fig.14 is a scatter plot of the height in the standard data and the height in the predicted data;

[0055] Fig.15 It is the relative error distribution diagram between the height in the standard data and the height in the predicted data;

[0056] Fig.16 It is a principle block diagram of a training device for a parameter prediction model for CD-SEM measurement according to one embodiment of the present invention;

[0057] Fig.17 It is a schematic structural diagram of an embodiment of an electronic device of the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application may be combined with each other.

[0060] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include" and "comprise" include not only those elements, but also other elements that are not explicitly listed, or also include elements inherent to such processes, methods, articles or equipment. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the existence of other identical elements in the process, method, article or equipment that includes the elements.

[0061] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0062] Figure 1 The overall steps of the method for constructing a parameter prediction model for CD-SEM measurement according to one embodiment of the present invention are schematically shown. Figure 1 As shown, the parameter prediction model construction method for CD-SEM measurement of the present invention includes the following steps:

[0063] Step S11: Acquire a data set, wherein the data set includes at least one data set having a proximity effect feature, each data set includes standard data and an SE signal set, wherein the standard data includes aspect ratio data and / or height data, and the SE signal set in the data set having the proximity effect feature is a secondary electron signal generated by electron beam scanning a geometric structure having adjacent structures;

[0064] Step S12: Taking the SE signal group in the same data group in the data set as input and the standard data in the same data group as output, the pre-constructed neural network model is trained to obtain a parameter prediction model for predicting aspect ratio data and / or height data in CD-SEM measurement.

[0065] Step S11 is a step of acquiring a data set. Specifically, the data set includes several groups of data groups, each of which represents the structural parameters of a device structure and the SE signal obtained by the corresponding SEM simulation imaging, that is, each of which includes standard data and a corresponding SE signal group, and the standard data includes aspect ratio data and / or height data. Among them, at least one data group is a data group with proximity effect characteristics, and the SE signal group in the data group with proximity effect characteristics is a secondary electron signal generated by electron beam scanning of a geometric structure with adjacent structures. Specifically, a two-dimensional array can be used to represent the data set. For example, each row represents a data group, the first two columns of each row are standard data, and the third column and the following are SE signal groups, thereby realizing a one-to-one correspondence between the SE signal and its corresponding aspect ratio data and height data.

[0066] Each data set in the data set can be obtained by measuring and testing actual samples. In some possible implementations, the data set with proximity effect characteristics in the embodiments of the present invention can be set to be obtained by performing SEM simulation imaging on a geometric structure with adjacent structures constructed by simulation. Figure 2 and Figure 3 As shown, Figure 2 and Figure 3 The overall structural diagram and cross-sectional structural diagram of the simulated geometric structure with adjacent structures are schematically shown. Among them, h represents the structure height, bcd represents the bottom width of the structure, swa represents the side wall angle, r represents the fillet of the top and bottom, d represents the spacing between adjacent structures, and scanpath represents the electron probe scanning path. Generally speaking, it can be considered that the arc of the fillet is tangent to the adjacent edge, so the arc position of the fillet can be obtained based on swa and r. In CD-SEM metrology, the influence of the proximity effect cannot be ignored, so the proximity effect performance of different structure spacing, side wall angles and heights is analyzed. The benchmark structure parameters used are: bcd = 20, swa = 86°, d = 20, h = 100. When changing the values ​​of d, swa or h, the other parameters remain unchanged. All the above scanning signal profiles are derived from SE signals. Exemplarily, the data set exported in the embodiment of the present invention includes 5906 data groups, that is, 5906 rows of data groups, wherein the first column is aspect ratio data, the second column is height data, and each subsequent column is SE signal, with a total number of columns being 245.

[0067] Figures 4 to 6 The figure schematically shows the comparison diagram of SE signals corresponding to the changes of different parameters. Figure 4 : is a comparison diagram of SE signals of structures with different bottom widths. Specifically, the d values ​​in these structures are set to 20, 30, 50, 100 and 200 nm respectively. Figure 4 It can be seen that the SE signal generally tends to weaken as the line structure spacing decreases. This is because when the distance between adjacent lines is close, the SE signal is more easily blocked, that is, the proximity effect is more obvious. Figure 5 The SE signal comparison diagram of structures with different side wall angles. Specifically, the swa values ​​in these structures are 86°, 87°, 88°, 89° and 90° respectively. Figure 5 It can be seen that the SE signal peak decreases with the increase of swa value, which is due to the increased possibility of SE signal being blocked by adjacent structures. The SE signal peak shifts to the right with the increase of swa, which is due to the increase of upper edge width. In addition, when swa is relatively small, a second peak appears in the signal curve. When swa increases to 88°, the second peak is almost invisible. Figure 6Figure 2 is a comparison of SE signals of structures with different heights. The h values ​​in these structures are 70, 80, 90 and 100 nm respectively. Figure 6 It can be seen that the peak value of the SE signal decreases as the height decreases, and the peak position shifts to the right. This is because when the swa is fixed, different heights will lead to different upper groove widths. In addition, it can be found that the path between the upper edge and the lower edge of the SE signal from both sides increases as the height decreases. This is because the decrease in height makes the adjacent structure less likely to block the SE signal. Furthermore, in the present invention, by obtaining a data set with proximity effect characteristics for training, it is possible to predict parameters related to the proximity effect characteristics.

[0068] In some possible implementations, the data set can be split to form a training set and a test set, wherein the training set and the test set are obtained by performing random number division on several groups of data in the data set. Specifically, in the data set, the training set accounts for 90%, and the remaining 10% is for testing. When training the constructed neural network model, the constructed neural network model is trained with the divided training set, and the model is optimized with the divided test set when training the neural network model, and the prediction accuracy of the parameter prediction model is determined after the training is obtained. It can be understood that in the present invention, the data set is obtained by performing SEM simulation imaging on the geometric structure with adjacent structures constructed by simulation, and then there will be a certain coherence between the groups of data in the data set, so that when training, the neural network model may take this coherence into consideration, thereby affecting the prediction accuracy of the parameter prediction model obtained by training. By disrupting the entire data set by using random number division, the training set used for training can be no longer coherent, thereby reducing the consideration of coherence of the neural network model during training, so that the parameter prediction model finally obtained can be more suitable for more different scenarios, and the prediction accuracy of the parameter prediction model is improved. Furthermore, by obtaining the test set in this way, there is no need to further obtain additional data sets for testing, thereby improving the overall training and testing efficiency.

[0069] Step S13 is a step of training a pre-built neural network model using the acquired data set to obtain a parameter prediction model. Specifically, the pre-built neural network is constructed based on a network architecture constructed based on an Identity module and a Dense module. Figure 7 The steps of constructing a pre-built neural network model in a parameter prediction model construction method for CD-SEM measurement according to an embodiment of the present invention are schematically shown. Figure 7 As shown, it can be specifically implemented by including the following steps:

[0070] Step S21: constructing a network architecture of a neural network model based on the Identity module and the Dense module;

[0071] Step S22: construct a neural network model according to the constructed network architecture and the models library under the Keras library that encapsulates TensorFlow. During construction, use the compile function to compile the model, specify the loss function as mse, specify Adam as the optimizer of the neural network, and specify the metrics function as mse.

[0072] In step S21, the network architecture of the neural network model is first constructed. In the present invention, the network architecture of the constructed neural network model is based on the network architecture of the residual network, which is different from the residual network in that the convolutional layer and pooling layer of the residual network are replaced by a fully connected layer. Figure 8 The network architecture of the parameter prediction model for CD-SEM measurement according to one embodiment of the present invention is schematically shown. Figure 8 As shown, it shows the architecture of the Identity module, the architecture of the Dense module and the overall network architecture. Among them, the overall network architecture is set to consist of two basic modules connected in sequence, a batch normalization layer, a fully connected layer and a linear activation function. Among them, each basic module is formed by sequentially connecting two Identity modules after a Dense module. Its Identity module is set to consist of three basic units connected in sequence, each basic unit is composed of a fully connected layer, a batch normalization layer and a ReLU activation function connected in sequence, wherein a Dropout layer is inserted after the second basic unit. The Dense module is set to consist of three basic units connected in sequence, wherein a Dropout layer is inserted after the second basic unit, and a residual path unit is connected after the first basic unit and the second basic unit respectively. The residual path unit consists of a fully connected layer and a batch normalization layer stacked in sequence, and the output of the residual path unit and the result between the batch normalization layer and the ReLU activation function in the third basic unit in the Dense module are configured to perform a tensor sum operation. The network architecture of the present invention is configured in this way. Compared with the network architecture of the traditional neural network model, the traditional residual network is more suitable for image processing, while the network architecture of the present invention is more suitable for nonlinear regression after being improved and optimized.

[0073] After the network architecture is constructed, step S22 is executed. Step S22 is a step of constructing a neural network model based on the constructed network architecture. Specifically, in this embodiment, it can be implemented as constructing a neural network model according to the constructed network architecture and the models library under the Keras library that encapsulates TensorFlow. Keras is an open source neural network library that can run with TensorFlow, CNTK or Theano as the backend. Among them, when building a neural network model, the compile function is used to compile the model, the loss function is specified as mse, Adam is specified as the optimizer of the neural network, the iteration direction of the neural network parameters is controlled, and the metrics function is specified as mse. Loss function, that is, loss function (Loss Function) is an important concept in machine learning and deep learning, which is used to measure the difference or error between the model prediction result and the true result. It is a numerical evaluation indicator that provides a measure of model performance by comparing the model output and the true label. The main function of the optimizer is to make the gradient drop faster and better during the gradient descent process, so as to find the minimum value of the objective function as soon as possible, that is, how to find the optimal "loss" (loss) as quickly as possible. Metrics functions, or evaluation functions, are used to evaluate the performance of the model during and after training. When the model is compiled (compile), the evaluation function is input as a parameter of metrics. The evaluation function is similar to the loss function, but the result of the evaluation function is not used in the training process. The accuracy evaluation function is used to evaluate the accuracy of the model prediction. Among the commonly used regression evaluation indicators, the relative error of the model prediction obtained using MSE as the loss function is small. Specifically, the relative error obtained using MSE is 4.8%, the relative error using MAE is 6.3%, and the relative error using MSLE is 56.6%. For the regression evaluation indicators selected by the metrics function, the same function as the loss function is generally selected to make it easier to observe the training effect.

[0074] When the acquired data set is used to train the pre-built neural network model, in the implementation of dividing the data set into a training set and a test set, the training set can be used to train the neural network model. In some possible implementations, the data set can also be preprocessed before training, and the preprocessing includes normalizing the data set. The data set is normalized to avoid the bias of the simulator or the electron microscope itself. In addition, data normalization can solve the problem of inconsistent scales of different features, so that the influence of different features on the model is more balanced. In addition, normalization can also prevent the problem of gradient disappearance or explosion, thereby accelerating the convergence speed of the model. When training the neural network model, the fit function can be used for training, specifying epochs, batch_size, validation_split, and recording the loss value and verification loss value, i.e., metrics value, during the training process. Among them, epoch refers to the process when a complete data set passes through the neural network once and returns once, that is, an epoch is the process of training all training samples once; batch_size is the number of samples that are input into the model at the same time in each iteration when the model is trained in advance. validation_split is a floating point number between 0 and 1, which is used to specify a certain proportion of data in the data set as the test set. For example, in this embodiment, the training set is specified to account for 90%, and the remaining 10% is the test set, then validation_split is specified to be 0.1. After the parameter prediction model is obtained through training, the parameter prediction model is saved. Fig. 9 The step flow of step S12 of one embodiment of the present invention is schematically shown, referring to Fig. 9 As shown, it can be specifically implemented by including the following steps:

[0075] Step S31: according to a preset number of training times, a data group in a training set is selected, and a pre-constructed neural network model is trained with a SE signal group in the same data group as input and a standard data in the same data group as output to obtain a first model parameter;

[0076] Step S32: updating the neural network model according to the first model parameter to obtain a first neural network model;

[0077] Step S33: inputting the SE signal group in the data group in the test set into the first neural network model, and obtaining the prediction result of the first neural network model on the input SE signal group;

[0078] Step S34: Use the loss function value to evaluate the error between the prediction result and the standard data in the corresponding data group, and optimize the first neural network model according to the neural network optimizer and the error to obtain a parameter prediction model.

[0079] In step S31, the number of samples preset to be simultaneously input into the model can be determined by a specified batch_size value, and the preset number of training times can be determined by a specified epoch value. According to the specified batch_size value and epoch value, the pre-constructed neural network model is trained with the SE signal group in the same data group as input and the standard data in the same data group as output, and the first model parameters of the corresponding trained neural network model can be obtained.

[0080] Step S32 is a step of updating the neural network model based on the first model parameters. After the update, a first neural network model based on the first model parameters is obtained.

[0081] In step S33, by inputting the input data of the data group in the test set, that is, the SE signal group, into the first neural network model, the corresponding prediction result can be obtained.

[0082] Finally, step S34 is executed to optimize the first neural network model based on the prediction results and the standard data. Specifically, the error of the first neural network model can be determined using the loss function specified when the neural network model is constructed. After the error is obtained, the first neural network model is optimized according to the error and the optimizer Adam specified when the neural network model is constructed to obtain a parameter prediction model.

[0083] Fig.10 The overall steps of the method for constructing a parameter prediction model for CD-SEM measurement according to another embodiment of the present invention are schematically shown. Fig.10 As shown, in some possible implementations, the parameter prediction model construction method for CD-SEM measurement of the present invention may also include the following steps:

[0084] Step S13: Determine the prediction accuracy of the parameter prediction model according to the data set and the parameter prediction model.

[0085] Step S13 is a step for detecting the prediction accuracy of the trained parameter prediction model. By determining the prediction accuracy, it can be used as a comparison of the prediction results. Fig.11 The step flow of step S13 of the method for constructing a parameter prediction model for CD-SEM measurement according to one embodiment of the present invention is schematically shown. Fig.11 As shown, step S13 can be specifically implemented as including the following steps:

[0086] Step S41: taking the SE signal group in the selected data group in the data set as input, inputting it into the parameter prediction model, and obtaining the prediction data output by the parameter prediction model;

[0087] Step S42: Determine the prediction accuracy of the parameter prediction model through a metrics function according to the prediction data and the standard data in the data group where the input SE signal group is located.

[0088] Step S41 is a step of using a parameter prediction model to perform simulation prediction. Specifically, in step S41, a test set may be used to perform simulation prediction, and the SE signal group in the test set is used as an input of the parameter prediction model to determine prediction data. The prediction data includes a predicted aspect ratio and / or a predicted height.

[0089] Step S42 is a step of calculating the average error of the predicted data and the corresponding standard data. Specifically, it can be calculated based on the metrics function specified when building the neural network model, so that the error rate of the two can be calculated. Specifically, it is also possible to derive the change of loss with epoch during training, derive the corresponding scatter plot of standard data and predicted data, and derive the distribution map of predicted data error to form visual data for easy comparison and observation.

[0090] For example, the following example shows the entire process of the parameter prediction model construction method for CD-SEM measurement. When building the network architecture and neural network model, in this process, the codes for data processing, deep learning model architecture design, training, verification and testing are all written in Python programming language, using Keras, the high-level API of TensorFlow. MSE is used as the training and verification loss during the training process. The MSE formula is as follows:

[0091]

[0092] The dataset used contains 5906 data sets, of which 90% of the data sets are used as training sets and 10% of the data sets are used as test sets by random number division. In the dataset, the bcd ranges from 20nm to 36nm, swa from 80° to 90°, h from 48nm to 144nm, aspect ratio from 4 to 20, r is fixed at 1nm, and the electron beam scanning tilt angle is fixed at 2.5°.

[0093] After that, the training set can be used to train the neural network model to obtain the parameter prediction model. After obtaining the parameter prediction model, the parameter prediction model can be saved, and the prediction accuracy of the parameter prediction model can be tested using the test set. Specifically, the test results are as follows: Figures 12 to 15 shown. Fig.12 It is a scatter plot of the aspect ratio in the standard data and the aspect ratio in the predicted data, wherein the shaded part indicates that the relative error is within 20%. Fig.13It is the relative error distribution diagram between the aspect ratio in the standard data and the aspect ratio in the predicted data. Fig.14 It is a scatter plot of the height in the standard data and the height in the predicted data, where the shaded part indicates that the relative error is within 10%. Fig.15 It is the relative error distribution diagram between the height in the standard data and the height in the predicted data.

[0094] Prediction result data:

[0095] 1. Depth-to-width ratio: The final training and validation losses were 0.0052 and 0.00035 respectively, and the loss of the test set was 0.00036. The square root of the test loss (RMSE) was 0.0190. By testing the test set (10% of the dataset), the average relative error between the predicted data and the standard data was: 2.4%, and the percentage of predicted cases with an error within 20% was 100.0%.

[0096] Height: The final training and validation losses were 0.0052 and 0.00067 respectively, and the loss of the test set was 0.00068. The square root of the test loss (RMSE) was 0.0262. By testing the test set (10% of the dataset), the average relative error between the predicted data and the standard data was: 1.9%, and the percentage of predicted cases within 10% error was 99.5%.

[0097] Fig.16 The overall module composition of the training device for the parameter prediction model of CD-SEM measurement according to one embodiment of the present invention is schematically shown. Fig.16 As shown, the training device for the parameter prediction model of CD-SEM measurement of the present invention includes the following modules:

[0098] A data acquisition module 1 is used to acquire a data set, wherein the data set includes at least one data group with a proximity effect feature, each data group includes standard data and an SE signal group, the standard data includes aspect ratio data and / or height data, and the SE signal group in the data group with the proximity effect feature is a secondary electron signal generated by electron beam scanning of a geometric structure with adjacent structures;

[0099] The training module 2 is used to train the pre-constructed neural network model with the SE signal group in the same data group in the data set as input and the standard data in the same data group as output, so as to obtain a parameter prediction model for predicting aspect ratio data and / or height data in CD-SEM measurement.

[0100] It should be noted that the implementation process and implementation principle of the training device for the parameter prediction model for CD-SEM measurement in the embodiment of the present invention can be specifically referred to the corresponding description of the above method embodiment, such as the corresponding description of the acquisition of the data set, the training of the model, etc. in the method embodiment, so they are not repeated here. Exemplarily, the training device for the parameter prediction model for CD-SEM measurement in the embodiment of the present invention can be any intelligent device with a processor, including but not limited to computers, smart phones, personal computers, robots, cloud servers, etc.

[0101] The present invention also provides a parameter prediction method for CD-SEM measurement, which uses the parameter prediction model obtained by the parameter prediction model construction method for CD-SEM measurement described in any of the above embodiments and the SE signal of the device structure to predict the depth-to-width ratio and / or height of the device structure. Specifically, the depth-to-width ratio and / or height of the device structure can be predicted by simply inputting the SE signal of the device structure into the parameter prediction model. Compared with the method of the prior art, there is no need to modify the hardware structure, so there is no problem of complicating the hardware structure, there is no problem of affecting reliability, and there is no need to adjust the working parameters of the CD-SEM, so there will be no damage to the sample and no impact on the quality of the detected image.

[0102] The present invention also provides a parameter prediction system for CD-SEM measurement, which includes a prediction module, which is used to predict the aspect ratio and / or height of the device structure based on the SE signal of the device structure and the parameter prediction model obtained by the parameter prediction model construction method for CD-SEM measurement described in any one of the above embodiments.

[0103] In some embodiments, an embodiment of the present invention provides a non-volatile computer-readable storage medium, wherein the storage medium stores one or more programs including execution instructions, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute the parameter prediction model construction method for CD-SEM measurement of any of the above embodiments of the present invention.

[0104] In some embodiments, an embodiment of the present invention further provides a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the parameter prediction model construction method for CD-SEM measurement of any one of the above embodiments.

[0105] In some embodiments, an embodiment of the present invention further provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the parameter prediction model construction method for CD-SEM measurement of any of the above embodiments.

[0106] In some embodiments, the present invention further provides a storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the method for constructing a parameter prediction model for CD-SEM measurement according to any of the above embodiments is implemented.

[0107] Fig.17 FIG. 1 is a schematic diagram of the hardware structure of an electronic device for executing a method for constructing a parameter prediction model for CD-SEM measurement provided by another embodiment of the present application, such as Fig.17 As shown, the device includes:

[0108] One or more processors 310 and memory 320, Fig.17 A processor 310 is taken as an example.

[0109] The apparatus for executing the parameter prediction model construction method for CD-SEM measurement may further include: an input device 330 and an output device 340 .

[0110] The processor 310, the memory 320, the input device 330 and the output device 340 may be connected via a bus or other means. Fig.17 The example of connecting through bus is taken in the following.

[0111] The memory 320 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the parameter prediction model construction method for CD-SEM measurement in the embodiment of the present application. The processor 310 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 320, that is, the parameter prediction model construction method for CD-SEM measurement in the above method embodiment is implemented.

[0112] The memory 320 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created according to the use of the parameter prediction model construction method for CD-SEM measurement, etc. In addition, the memory 320 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 320 may optionally include a memory remotely arranged relative to the processor 310, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0113] The input device 330 may receive input digital or character information and generate signals related to user settings and function control of the image processing device. The output device 340 may include a display device such as a display screen.

[0114] The one or more modules are stored in the memory 320 , and when executed by the one or more processors 310 , the parameter prediction model construction method for CD-SEM measurement in any of the above method embodiments is executed.

[0115] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.

[0116] The electronic devices of the embodiments of the present application exist in various forms, including but not limited to:

[0117] (1) Mobile communication devices: These devices are characterized by their mobile communication functions and their main purpose is to provide voice and data communications. These terminals include: smart phones (such as iPhone), multimedia phones, functional phones, and low-end phones.

[0118] (2) Ultra-mobile personal computer devices: These devices fall into the category of personal computers, have computing and processing capabilities, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, such as iPad.

[0119] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players (such as iPods), handheld game consoles, e-books, as well as smart toys and portable car navigation devices.

[0120] (4) Server: A device that provides computing services. The server consists of a processor, hard disk, memory, system bus, etc. The server is similar to a general computer architecture, but because it needs to provide highly reliable services, it has higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0121] (5) Other electronic devices with data interaction functions.

[0122] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0123] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for constructing a parameter prediction model for CD-SEM measurement, characterized in that: include: Acquire a data set, wherein the data set includes at least one data set having a proximity effect feature, each data set includes standard data and an SE signal set, wherein the standard data includes aspect ratio data and / or height data, and the SE signal set in the data set having the proximity effect feature is a secondary electron signal generated by performing electron beam scanning on a geometric structure having adjacent structures; The SE signal group in the same data group in the data set is used as input, and the standard data in the same data group is used as output, and the pre-constructed neural network model is trained to obtain a parameter prediction model for predicting aspect ratio data and / or height data in CD-SEM measurement.

2. The method according to claim 1, characterized in that The method for constructing the neural network model comprises: Construct the network architecture of the neural network model based on the Identity module and the Dense module; A neural network model is built according to the constructed network architecture and the models library under the Keras library that encapsulates TensorFlow. During construction, the compile function is used to compile the model, the loss function is specified as mse, Adam is specified as the optimizer of the neural network, and the metrics function is specified as mse.

3. The method according to claim 2, characterized in that The network architecture of the constructed neural network model is set to consist of two basic modules connected in sequence, a batch normalization layer, a fully connected layer, and a linear activation function, where each basic module is formed by connecting two Identity modules in sequence after a Dense module; The Identity module includes three basic units connected in sequence, wherein each basic unit is composed of a fully connected layer, a batch normalization layer and a ReLU activation function connected in sequence, and a Dropout layer is inserted after the second basic unit; The Dense module includes three basic units connected in sequence, wherein a Dropout layer is inserted after the second basic unit, and a residual path unit is respectively connected after the first basic unit and the second basic unit, and the residual path unit is composed of a fully connected layer and a batch normalization layer stacked in sequence, and the output of the residual path unit and the result between the batch normalization layer and the ReLU activation function in the third basic unit in the Dense module are configured to perform a tensor sum operation.

4. The method according to claim 1, characterized in that: The data set is set by performing SEM simulation imaging on a geometric structure with adjacent structures constructed in a simulated manner, obtaining standard data according to parameters of the constructed geometric structure, and obtaining SE signals according to parameters of the simulated imaging.

5. The method according to claim 1, characterized in that The data set includes a training set and a test set, and the training set and the test set are obtained by performing random number division on a plurality of data groups in the data set; When training the constructed neural network model, the constructed neural network model is trained using the data group in the training set to obtain model parameters, and the obtained model parameters are optimized using the data group in the test set.

6. The method according to claim 5, characterized in that The method uses the SE signal group in the same data group in the data set as input and the standard data in the same data group as output to train the pre-built neural network model to obtain the parameter prediction model, including: According to a preset number of training times, a data group in a training set is selected, and a pre-constructed neural network model is trained using a SE signal group in the same data group as input and a standard data in the same data group as output to obtain a first model parameter; Update the neural network model according to the first model parameter to obtain a first neural network model; Inputting the SE signal group in the data group in the test set into the first neural network model, and obtaining the prediction result of the first neural network model on the input SE signal group; The loss function value is used to evaluate the error between the prediction result and the standard data in the corresponding data group, and the first neural network model is optimized according to the neural network optimizer and the error to obtain a parameter prediction model.

7. The method according to claim 1, characterized in that It also includes determining the prediction accuracy of the parameter prediction model based on the data set and the parameter prediction model, which includes: Taking the SE signal group in the selected data group in the data set as input, inputting it into the parameter prediction model, and obtaining the prediction data output by the parameter prediction model; The prediction accuracy of the parameter prediction model is determined by the metrics function based on the predicted data and the standard data in the data group where the input SE signal group is located.

8. A training device for a parameter prediction model for CD-SEM measurement, characterized in that: include: A data acquisition module, used to acquire a data set, wherein the data set includes at least one data group with a proximity effect feature, each data group includes standard data and an SE signal group, the standard data includes aspect ratio data and / or height data, and the SE signal group in the data group with the proximity effect feature is a secondary electron signal generated by electron beam scanning of a geometric structure with adjacent structures; The training module is used to train the pre-built neural network model with the SE signal group in the same data group in the data set as input and the standard data in the same data group as output, so as to obtain a parameter prediction model for predicting aspect ratio data and / or height data in CD-SEM measurement.

9. A parameter prediction method for CD-SEM measurement, characterized in that: The parameter prediction model obtained by the parameter prediction model construction method for CD-SEM measurement described in any one of claims 1 to 7 and the SE signal of the device structure are used to predict the aspect ratio and / or height of the device structure.

10. A parameter prediction system for CD-SEM measurement, characterized in that: include: A prediction module, used to predict the aspect ratio and / or height of the device structure according to the SE signal and parameter prediction model of the device structure; Wherein, the parameter prediction model is generated by training the method described in any one of claims 1 to 7 or the device described in claim 8.

Images