A three-dimensional feature recognition method, device, equipment and storage medium

Abstract

The application discloses a three-dimensional feature recognition method, device, equipment and storage medium. The method comprises the following steps: acquiring a first electron microscopic image; generating first frequency domain information corresponding to the first electron microscopic image; analyzing the first frequency domain information based on a three-dimensional feature recognition model to obtain feature data of a three-dimensional structure corresponding to the first electron microscopic image; and outputting the feature data of the three-dimensional structure corresponding to the first electron microscopic image. It can be seen that, by means of the three-dimensional feature recognition model, the first frequency domain information of the first electron microscopic image generated by the CD-SEM can be directly analyzed and processed, without additional hardware modification of the existing detection equipment, so that the detection efficiency of the feature data of the three-dimensional structure can be improved, and the detection cost can be reduced. Moreover, the feature data of the three-dimensional structure is captured in the frequency domain, so that the key features of the data can be more directly extracted, efficient and accurate feature recognition is realized, and the accuracy of micro-nano structure analysis is effectively improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, device and storage medium for recognizing three-dimensional features. Background Technology

[0002] In integrated circuit manufacturing processes, inspection technologies, such as critical dimension scanning electron microscopy (CD-SEM), are of great importance. They can perform high-precision inspection of the feature data of three-dimensional structures (i.e., micro-nano structures) designed and manufactured at the micrometer and nanometer scales. Therefore, they play an important role in the measurement of critical dimensions of patterns, process control and defect control, and design rule verification.

[0003] However, in practical applications, the detection technology for the three-dimensional features of micro and nano structures has certain limitations, resulting in high implementation costs and low efficiency. Summary of the Invention

[0004] This application provides a method, apparatus, device, and storage medium for recognizing three-dimensional features, so as to improve the detection efficiency of feature data of three-dimensional structures and reduce detection costs.

[0005] In a first aspect, embodiments of this application provide a method for recognizing three-dimensional features, including:

[0006] Acquire the first electron micrograph;

[0007] Generate the first frequency domain information corresponding to the first electron micrograph;

[0008] The first frequency domain information is analyzed based on the three-dimensional feature recognition model to obtain the feature data of the three-dimensional structure corresponding to the first electron micrograph.

[0009] Output the feature data of the three-dimensional structure corresponding to the first electron micrograph.

[0010] Optionally, before acquiring the first electron micrograph, the method further includes:

[0011] Constructing three-dimensional models of micro and nano structures;

[0012] Generate a second electron micrograph corresponding to the three-dimensional model;

[0013] Generate the second frequency domain information corresponding to the second electron micrograph;

[0014] The model is trained based on the second frequency domain information, and the three-dimensional feature recognition model is obtained after the training is completed.

[0015] Optionally, generating the second frequency domain information corresponding to the second electron micrograph includes:

[0016] Extract the grayscale information from the second electron micrograph;

[0017] The grayscale information of the second microscopic image is decomposed and transformed using a spectrum analyzer to obtain the second frequency domain information.

[0018] Optionally, the step of training the model based on the second frequency domain information to obtain the three-dimensional feature recognition model after training includes:

[0019] Using the second frequency domain information as input data for multiple decision tree models, and the feature data in the second frequency domain information as output data for the multiple decision tree models, the multiple decision tree models are trained to obtain multiple trained decision tree models.

[0020] A random forest model is generated based on the multiple trained decision tree models, and the random forest model is used as the three-dimensional feature recognition model.

[0021] Optionally, the step of analyzing the first frequency domain information based on a three-dimensional feature recognition model to obtain feature data of the three-dimensional structure corresponding to the first electron micrograph includes:

[0022] The first frequency domain information is analyzed based on the multiple trained decision tree models in the random forest model to obtain the feature data output by the multiple trained decision tree models respectively.

[0023] By integrating the feature data output by the multiple trained decision tree models, the feature data of the three-dimensional structure corresponding to the first electron micrograph is obtained.

[0024] Optionally, after training the model based on the second frequency domain information and obtaining the three-dimensional feature recognition model, the method includes:

[0025] Obtain a validation set, which includes electron micrographs with three-dimensional feature annotations and electron micrographs without three-dimensional feature annotations;

[0026] The three-dimensional recognition structure model is verified based on the electron micrograph with three-dimensional feature annotations to obtain a first verification result;

[0027] Based on the first verification result, the three-dimensional recognition structure model is adjusted to obtain the adjusted three-dimensional feature recognition model;

[0028] The adjusted three-dimensional feature recognition model is verified based on the electron micrograph without three-dimensional feature annotations to obtain a second verification result.

[0029] Optionally, generating the second electron micrograph corresponding to the three-dimensional model includes:

[0030] Using a Monte Carlo simulator, simulation calculations are performed based on pre-set simulation conditions to generate the initial electron micrograph corresponding to the three-dimensional model.

[0031] A third electron microscopy image is acquired, wherein the feature data of the three-dimensional structure corresponding to the third electron microscopy image is the same as the feature data of the micro / nano structure.

[0032] The initial electron micrograph is corrected based on the third electron micrograph to obtain the second electron micrograph.

[0033] Secondly, embodiments of this application provide a three-dimensional feature recognition device, comprising:

[0034] The image acquisition module is used to acquire the first electron micrograph;

[0035] The first frequency domain information generation module is used to generate the first frequency domain information corresponding to the first electron micrograph;

[0036] The feature data acquisition module is used to analyze the first frequency domain information based on the three-dimensional feature recognition model to obtain the feature data of the three-dimensional structure corresponding to the first electron micrograph.

[0037] The feature data output module is used to output the feature data of the three-dimensional structure corresponding to the first electron micrograph.

[0038] Thirdly, embodiments of this application provide an electronic device, the device including: a processor, a memory, and a system bus;

[0039] The processor and the memory are connected via the system bus;

[0040] The memory is used to store a program, the program including instructions that, when executed by the processor, cause the processor to perform any of the implementation steps of the three-dimensional feature recognition method described above.

[0041] Fourthly, embodiments of this application provide a computer-readable storage medium storing instructions that, when executed on an electronic device, cause the electronic device to perform any of the implementation steps of the above-described three-dimensional feature recognition method.

[0042] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0043] In this embodiment, after acquiring the first electron microscope image, first frequency domain information corresponding to the first electron microscope image can be generated. Then, based on a three-dimensional feature recognition model, the first frequency domain information is analyzed to obtain the feature data of the three-dimensional structure corresponding to the first electron microscope image, and the feature data of the three-dimensional structure corresponding to the first electron microscope image is output. It is evident that, with the help of the three-dimensional feature recognition model, the first frequency domain information of the first electron microscope image generated by CD-SEM can be directly analyzed and processed to obtain the feature data of the three-dimensional structure without requiring additional hardware modifications to existing detection equipment. Therefore, the detection efficiency of the feature data of the three-dimensional structure can be improved, and the detection cost can be reduced. Furthermore, capturing the feature data of the three-dimensional structure in the frequency domain allows for more intuitive extraction of key features from the data, achieving efficient and accurate feature recognition, and effectively improving the accuracy of micro-nano structure analysis. Attached Figure Description

[0044] Figure 1 A flowchart illustrating a three-dimensional feature recognition method provided in an embodiment of this application;

[0045] Figure 2 A schematic diagram of a three-dimensional model of a micro / nano structure provided in an embodiment of this application;

[0046] Figure 3 A schematic diagram of an initial electron micrograph generated by simulation, provided for an embodiment of this application;

[0047] Figure 4 A schematic diagram illustrating the grayscale information of a second electron microscope image provided in an embodiment of this application;

[0048] Figure 5 This is a schematic diagram of a second frequency domain information provided in an embodiment of this application;

[0049] Figure 6 This is a schematic diagram of the structure of a three-dimensional feature recognition model provided in an embodiment of this application;

[0050] Figure 7 A schematic diagram illustrating a first verification result provided in an embodiment of this application;

[0051] Figure 8 A schematic diagram illustrating a second verification result provided in an embodiment of this application;

[0052] Figure 9 This is a schematic diagram of the structure of a three-dimensional feature recognition device provided in an embodiment of this application. Detailed Implementation

[0053] As mentioned earlier, in practical applications, the detection technology for the three-dimensional features of micro and nano structures has certain limitations, resulting in high implementation costs and low efficiency.

[0054] For example, multi-angle scanning electron microscopy (SEM) imaging and reconstruction techniques can currently be used to detect the three-dimensional features of micro / nano structures. This requires adding multiple detectors at different angles to the SEM or rotating the sample during the detection process. However, adding multiple detectors at different angles is costly, requires additional communication and customization with the instrument manufacturer, and results in low detection efficiency. Rotating the sample is also slow and unsuitable for real-time detection in production processes. Alternatively, the micro / nano structure sample can be sliced ​​to observe its three-dimensional features, but this method is slow and causes physical damage to the sample. Atomic force microscopy (AFM) can also be used to detect the three-dimensional features of the sample, but this method is slow, has limited applicability, and is sensitive to the detection environment. Another option is to use optical metrology to analyze the aggregate features of the sample using light interference and scattering, which can provide some analysis of the three-dimensional structure. However, its data processing and spectral analysis are relatively complex, and generating and matching complex models requires considerable computational resources and time, making it unsuitable for rapid product detection on production lines.

[0055] Based on this, in order to solve the above problems, this application provides a method for recognizing three-dimensional features, including: after acquiring a first electron microscope image, generating first frequency domain information corresponding to the first electron microscope image, and then analyzing the first frequency domain information based on a three-dimensional feature recognition model to obtain feature data of the three-dimensional structure corresponding to the first electron microscope image, and outputting the feature data of the three-dimensional structure corresponding to the first electron microscope image.

[0056] As can be seen, by using a 3D feature recognition model, the first electron micrograph generated by CD-SEM can be directly analyzed and processed to obtain the feature data of the 3D structure without requiring additional hardware modifications to existing detection equipment. Therefore, this improves the detection efficiency of 3D structure feature data and reduces detection costs. Furthermore, capturing the feature data of the 3D structure in the frequency domain allows for more intuitive extraction of key features, achieving efficient and accurate feature recognition and effectively improving the precision of micro / nano structure analysis. In addition, the above scheme does not require processing of the sample to be tested, ensuring non-destructive testing and significantly improving detection efficiency and economic benefits.

[0057] It should be noted that the embodiments of this application do not limit the executing entity of the three-dimensional feature recognition method. For example, the three-dimensional feature recognition method of this application embodiment can be applied to data processing devices such as terminal devices or servers. The terminal device can be an electronic device such as a smartphone, computer, personal digital assistant (PDA), or tablet computer. The server can be a standalone server, a cluster server, or a cloud server.

[0058] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0059] Figure 1 This is a flowchart illustrating a three-dimensional feature recognition method provided in an embodiment of this application. (In conjunction with...) Figure 1 As shown, the three-dimensional feature recognition method provided in this application embodiment uses a corresponding three-dimensional feature recognition device as the execution subject to describe the specific implementation of the scheme. The three-dimensional feature recognition method may include the following steps S101-S104.

[0060] S101: Acquire the first electron micrograph.

[0061] Here, the first electron micrograph refers to the scanning electron micrograph generated by the CD-SEM to be examined. In practical applications, the CD-SEM can send this first electron micrograph to a 3D feature recognition device in real time, so that the 3D feature recognition device can perform feature analysis of the 3D structure. Alternatively, the first electron micrograph can be stored in a data storage device, and the 3D feature recognition device can read the first electron micrograph from the data storage device.

[0062] In this embodiment, to detect the feature data of the three-dimensional structure corresponding to the first electron micrograph, a three-dimensional feature recognition model can be constructed before acquiring the first electron micrograph, and feature data can be automatically extracted through the three-dimensional feature recognition model. For ease of understanding, the following description is provided in conjunction with the accompanying drawings and a possible implementation method.

[0063] As one possible implementation, the process of acquiring a three-dimensional feature recognition model may specifically include the following steps 11-14.

[0064] Step 11: Construct a three-dimensional model of the micro / nano structure.

[0065] The micro / nano structures can include at least one of the following: one-dimensional arrays of photoresist lines, two-dimensional arrays of geometric patterns, and other three-dimensional structures with micron- or nanometer-scale features formed on a wafer substrate by photolithography. These structures encompass, but are not limited to, complex morphological features such as high aspect ratio trenches, stepped shapes, columnar structures, conical structures, nanowires, and quantum dots. They also include interface features in multilayer thin film stacked structures, variations in film thickness, and microscopic features such as defects, edge roughness, surface protrusions, and etching residues that may occur during manufacturing.

[0066] The three-dimensional model of a micro / nano structure may include at least one of the following: a digital representation of geometric features defined in the form of a polygonal mesh, a digital representation of geometric features defined in a voxel data format, a digital representation of geometric features defined by a parametric method, and other digital representations generated by computer modeling techniques, stored in digital format, and capable of realistically reflecting the geometric features of the target object within the required accuracy range.

[0067] In practical implementation, creating a 3D model of a micro / nano structure refers to the process of using computer software to create and write a digital representation that accurately reflects the geometric characteristics of the micro / nano structure. For example, combining... Figure 2 As shown, a three-dimensional model of a micro / nano structure is, for example, a digital representation in the form of a polygonal grid of a two-dimensional geometric array of trapezoidal photoresist on a wafer substrate generated by computer modeling technology, wherein the base length and height of the trapezoidal photoresist are both 15 nanometers.

[0068] Step 12: Generate the second electron micrograph corresponding to the three-dimensional model.

[0069] In practical implementation, firstly, a Monte Carlo simulator can be used to perform simulation calculations based on pre-set simulation conditions to generate an initial electron micrograph corresponding to the three-dimensional model.

[0070] Here, the Monte Carlo simulator is a high-precision simulator that uses the Monte Carlo algorithm to simulate the interaction between an electron beam and materials. This simulator can generate simulated images corresponding to actual CD-SEM imaging conditions based on a given 3D model and simulation conditions. It should be noted that the electron beam images calculated by the Monte Carlo simulator have been experimentally verified to facilitate the subsequent prediction and extraction of 3D structural parameters from actual electron beam scanning images.

[0071] The simulation condition information refers to the relevant parameters used in the simulation, such as material information and set electron beam parameters (e.g., accelerating voltage, incident angle, and detector configuration). The Monte Carlo algorithm described above is a numerical calculation method based on random sampling and statistical analysis, which can predict system behavior by simulating the motion trajectories and interaction processes of a large number of particles.

[0072] Combination Figure 3 As shown, using the Monte Carlo simulator, it is possible to base on Figure 2 The simulation of the three-dimensional model of the micro / nano structure shown generates the corresponding initial electron micrograph. In this embodiment of the application, Figure 2 In section d), the white solid line box represents the mirror boundary condition in the simulation condition information. Therefore, the Monte Carlo simulator operates within the white solid line box. Thus, although a three-dimensional model of a two-dimensional geometric array of trapezoidal photoresist is established, the three-dimensional model used in the actual simulation can be equivalent to a one-dimensional line array three-dimensional model of trapezoidal photoresist.

[0073] Furthermore, a third electron micrograph can be obtained, and the feature data of the three-dimensional structure corresponding to this third electron micrograph is the same as the feature data of the micro / nano structure. In other words, the three-dimensional structure corresponding to the third electron micrograph has the same three-dimensional features as the micro / nano structure.

[0074] In this way, the initial electron micrograph can be corrected based on the third electron micrograph to obtain the second electron micrograph. That is, by comparing and correcting the initial electron micrograph obtained from the simulation with the third electron micrograph corresponding to the same three-dimensional features, a second electron micrograph with better simulation results can be obtained, which helps to use the second electron micrograph for subsequent model training to obtain a model with better performance.

[0075] Step 13: Generate the second frequency domain information corresponding to the second electron micrograph.

[0076] In practical implementation, the grayscale information of the second electron micrograph can be extracted first. Here, the grayscale information can be represented in at least one of the following data presentation formats: grayscale distribution curve, grayscale distribution surface, and other data presentation formats extracted from the second electron micrograph that can reflect the brightness intensity distribution of different pixels in the image.

[0077] For example, combining Figure 4 As shown, if the second electron microscope image exhibits periodic characteristics along its width (i.e., horizontal direction), then choosing the grayscale distribution curve along the width direction as the data representation form of grayscale information is more appropriate. Based on this, the grayscale distribution curve of the second electron microscope image along its height (i.e., vertical direction) can be generated by averaging the second electron microscope image along its width.

[0078] Next, a spectrum analyzer can be applied to decompose and transform the grayscale information of the second electron micrograph to obtain the second frequency domain information. The spectrum analyzer can be implemented using at least one of the following algorithms: Fourier transform, discrete Fourier transform, fast Fourier transform, and other algorithms that can convert signals from the time or spatial domain to the frequency domain. Correspondingly, the second frequency domain information refers to the data describing the frequency components of the signal and their corresponding amplitude and phase. It reveals the energy distribution and periodicity characteristics of the signal at different frequencies. The second frequency domain information can intuitively display the frequency patterns in the signal; high-frequency components are usually related to details and rapid changes, while low-frequency components correspond to smoothness and overall trends.

[0079] For example, combining Figure 5 As shown, by applying the Fast Fourier Transform method provided by the spectrum analyzer, it is possible to... Figure 4 The grayscale information of the second electron micrograph is decomposed and transformed to generate second frequency domain information for describing the frequency components of the signal and their corresponding amplitude and phase.

[0080] It should be noted that, in the embodiments of this application, by repeating the above steps 11-13 multiple times, the second frequency domain information of the second electron micrographs corresponding to a large number of different three-dimensional models can be obtained. By combining this information with the feature data of the three-dimensional structure and the corresponding simulation condition information and storing it in a data file, a frequency domain information library can be formed.

[0081] In this way, by using the second frequency domain information to train the model, the geometric feature information of the sample can be captured in the frequency domain. Subsequently, the model can be used to extract the key features of the data more intuitively, achieve efficient and accurate feature recognition, and effectively improve the accuracy of micro and nano structure analysis.

[0082] Step 14: Train the model based on the second frequency domain information. After training, a three-dimensional feature recognition model is obtained.

[0083] The three-dimensional feature recognition model can be implemented using at least one of the following algorithms: distance-based matching algorithm, feature similarity analysis-based matching algorithm, machine learning algorithm, deep learning algorithm, and other algorithms that analyze and predict the three-dimensional features of a three-dimensional structure through a pre-built frequency domain feature library.

[0084] In practical implementation, the second frequency domain information can be used as input data for multiple decision tree models, and the feature data in the second frequency domain information can be used as output data for multiple decision tree models to train multiple decision tree models, resulting in multiple trained decision tree models. Then, a random forest model can be generated based on the multiple trained decision tree models, and the random forest model can be used as a 3D feature recognition model.

[0085] For example, combining Figure 6 As shown, 100 decision tree models can be trained independently using the above method. Each decision tree model is trained based on independent samples, learning the relationship between input features and target variables, and the learned relationship is saved and fixed in the form of a parameter file. Here, independent samples can refer to each set of information in the aforementioned frequency domain information database, namely, a three-dimensional model, the frequency domain information of the corresponding second electron micrograph of the three-dimensional model, and the feature data of the three-dimensional structure corresponding to the three-dimensional model.

[0086] Furthermore, after obtaining the 3D feature recognition model, it can be further verified and fine-tuned to improve model performance. For ease of understanding, the following description, in conjunction with the accompanying drawings and one possible implementation method, will illustrate this.

[0087] As one possible implementation, the verification process for the three-dimensional feature recognition model may specifically include the following steps 15-18.

[0088] Step 15: Obtain a validation set, which may include electron micrographs with 3D feature annotations and electron micrographs without 3D feature annotations.

[0089] In this embodiment, two types of electron micrographs, one with three-dimensional feature annotations and the other without, i.e., two types of electron micrographs with known three-dimensional features and unknown three-dimensional features, can be used to verify the three-dimensional feature recognition model, thereby evaluating the model's accuracy and generalization ability.

[0090] Electron micrographs with three-dimensional feature annotations can include at least one of the following: a second electron micrograph generated by the above simulation, an electron micrograph generated using a standard three-dimensional structure under CE-SEM, and other electron micrographs with clearly defined three-dimensional features. Electron micrographs without three-dimensional feature annotations are those whose three-dimensional features are not clearly defined. The aforementioned three-dimensional feature annotations can include at least one of the following: height, width, angle, and other geometric parameters used to describe the shape and dimensional characteristics of the three-dimensional structure.

[0091] Step 16: Verify the three-dimensional recognition structure model based on the electron micrograph with three-dimensional feature annotations to obtain the first verification result.

[0092] In this embodiment, firstly, frequency domain information corresponding to an electron micrograph with three-dimensional feature annotations can be generated. In this embodiment, the process of generating frequency domain information corresponding to an electron micrograph with three-dimensional feature annotations can be referred to the generation method of the second frequency domain information in step 13 above, and will not be repeated here.

[0093] Next, model verification can be performed based on the frequency domain information corresponding to the electron microscope image with 3D feature annotations. That is, the frequency domain information corresponding to the electron microscope image with 3D feature annotations is input into the 3D feature recognition model, and the 3D feature recognition model outputs feature data of the 3D structure corresponding to the electron microscope image with 3D feature annotations. Then, the difference between this feature data and the 3D feature annotations can be compared to obtain the first verification result. For example, taking the sidewall tilt angle of photoresist as an example, the difference between the result output by the 3D feature recognition model and the 3D feature annotations for the electron microscope image with 3D feature annotations, i.e., the first comparison result, can be obtained as follows: Figure 7 As shown.

[0094] Step 17: Adjust the three-dimensional recognition structure model based on the first verification result to obtain the adjusted three-dimensional feature recognition model.

[0095] In the embodiments of this application, the adjustment process of the three-dimensional recognition structure model can be achieved by any existing or future method for fine-tuning the model, which will not be elaborated here.

[0096] Step 18: Verify the adjusted three-dimensional feature recognition model based on electron micrographs without three-dimensional feature annotations to obtain the second verification result.

[0097] In this embodiment, frequency domain information corresponding to an electron microscope image without three-dimensional feature annotation can be generated first. In this embodiment, the process of generating frequency domain information corresponding to an electron microscope image without three-dimensional feature annotation can be referred to the generation method of the second frequency domain information in step 13 above, and will not be repeated here.

[0098] Next, model verification can be performed based on the frequency domain information corresponding to the electron microscope image without 3D feature annotations. That is, the frequency domain information corresponding to the electron microscope image without 3D feature annotations is input into the adjusted 3D feature recognition model, and the adjusted 3D feature recognition model outputs feature data of the 3D structure corresponding to the electron microscope image without 3D feature annotations. Then, the difference between this feature data and the actual 3D feature result can be compared to obtain the second verification result. For example, still taking the 3D feature annotation as the sidewall tilt angle of photoresist, the difference between the result output by the 3D feature recognition model and the actual 3D feature result for the electron microscope image without 3D feature annotations, i.e., the second verification result, can be obtained as follows: Figure 8 As shown.

[0099] S102: Generate the first frequency domain information corresponding to the first electron micrograph.

[0100] In this embodiment of the application, the process of generating the first frequency domain information can be referred to the method of generating the second frequency domain information in step 13 above, and will not be repeated here.

[0101] S103: Analyze the first frequency domain information based on the three-dimensional feature recognition model to obtain the feature data of the three-dimensional structure corresponding to the first electron micrograph.

[0102] As one possible implementation, the aforementioned three-dimensional feature recognition model is the random forest model mentioned in step 14. Accordingly, during the analysis of the first frequency domain information based on this random forest, the first frequency domain information can be analyzed based on multiple trained decision tree models in the random forest model to obtain the feature data output by each of the multiple trained decision tree models. Then, the feature data output by each of the multiple trained decision tree models can be integrated to obtain the feature data of the three-dimensional structure corresponding to the first electron micrograph.

[0103] As can be seen, in the model prediction phase, it is no longer necessary to traverse the entire training database. Instead, the model's input data can be passed to multiple decision trees in the random forest model. Each tree can independently make predictions based on the rules learned during training. Then, the model as a whole averages the predictions from these trees to obtain the final prediction value. This integration of the matching predictions from multiple trees smooths out the errors of individual trees, thus contributing to the stability and accuracy of 3D feature recognition. Furthermore, compared to traditional 3D measurement techniques, the algorithm using the model for detection is more efficient. The random forest model avoids comparing each sample individually, significantly reducing computational resource consumption and greatly improving processing speed, thereby meeting the high real-time requirements of measurement in real-world production environments.

[0104] Furthermore, as mentioned earlier, the 3D feature recognition model can be fine-tuned to improve its performance. Therefore, in this embodiment, the analysis of the first frequency domain information can be specifically implemented using the adjusted 3D feature recognition model.

[0105] As one possible implementation, during the analysis of the first frequency domain information, other models can be used to extract features from the learned frequency domain information, thereby obtaining the feature data of the three-dimensional structure corresponding to the first electron micrograph. In specific implementation, using other models to extract features from the learned frequency domain information can be manifested as feature analysis, pattern recognition, and similarity evaluation of the frequency domain information. Feature analysis of the frequency domain information refers to analyzing different frequency components and their corresponding amplitude and phase information; pattern recognition refers to identifying the feature patterns of the data; and similarity evaluation refers to comparing the features and feature patterns obtained from the above analysis with the pre-constructed frequency domain information database to find matching information from the database to determine the feature data of the three-dimensional structure corresponding to the first electron micrograph.

[0106] S104: Output the feature data of the three-dimensional structure corresponding to the first electron micrograph.

[0107] Based on the aforementioned steps S101-S104, it is clear that in this embodiment, after acquiring the first electron microscope image, first frequency domain information corresponding to the first electron microscope image can be generated. Then, based on a three-dimensional feature recognition model, this first frequency domain information is analyzed to obtain the feature data of the three-dimensional structure corresponding to the first electron microscope image, and the feature data of the three-dimensional structure corresponding to the first electron microscope image is output. It is evident that, with the aid of a three-dimensional feature recognition model, the first frequency domain information of the first electron microscope image generated by CD-SEM can be directly analyzed and processed to obtain the feature data of the three-dimensional structure without requiring additional hardware modifications to existing detection equipment. Therefore, the detection efficiency of the feature data of the three-dimensional structure can be improved, and the detection cost reduced. Furthermore, capturing the feature data of the three-dimensional structure in the frequency domain allows for more intuitive extraction of key features from the data, achieving efficient and accurate feature recognition, and effectively improving the accuracy of micro-nano structure analysis.

[0108] Furthermore, based on the three-dimensional feature recognition method provided in the above embodiments, this application embodiment can also provide a three-dimensional feature recognition device. The three-dimensional feature recognition device will now be described in conjunction with the embodiments and accompanying drawings.

[0109] Figure 9 This is a schematic diagram of the structure of a three-dimensional feature recognition device provided in an embodiment of this application. (Combined with...) Figure 9 As shown, the three-dimensional feature recognition device 900 provided in this application embodiment may include:

[0110] Image acquisition module 901 is used to acquire a first electron micrograph;

[0111] The first frequency domain information generation module 902 is used to generate the first frequency domain information corresponding to the first electron micrograph;

[0112] The feature data acquisition module 903 is used to analyze the first frequency domain information based on the three-dimensional feature recognition model to obtain the feature data of the three-dimensional structure corresponding to the first electron micrograph.

[0113] The feature data output module 904 is used to output the feature data of the three-dimensional structure corresponding to the first electron micrograph.

[0114] Optionally, the three-dimensional feature recognition device 900 further includes:

[0115] The 3D model building module is used to build 3D models of micro and nano structures.

[0116] The image generation module is used to generate a second electron micrograph corresponding to the three-dimensional model;

[0117] The second frequency domain information generation module is used to generate the second frequency domain information corresponding to the second electron micrograph;

[0118] The model training module is used to train the model based on the second frequency domain information, and the three-dimensional feature recognition model is obtained after the training is completed.

[0119] Optionally, the second frequency domain information generation module is specifically used for:

[0120] Extract the grayscale information from the second electron micrograph;

[0121] The grayscale information of the second microscopic image is decomposed and transformed using a spectrum analyzer to obtain the second frequency domain information.

[0122] Optionally, the model training module is specifically used for:

[0123] Using the second frequency domain information as input data for multiple decision tree models, and the feature data in the second frequency domain information as output data for the multiple decision tree models, the multiple decision tree models are trained to obtain multiple trained decision tree models.

[0124] A random forest model is generated based on the multiple trained decision tree models, and the random forest model is used as the three-dimensional feature recognition model.

[0125] Optionally, the feature data acquisition module is specifically used for:

[0126] The first frequency domain information is analyzed based on the multiple trained decision tree models in the random forest model to obtain the feature data output by the multiple trained decision tree models respectively.

[0127] By integrating the feature data output by the multiple trained decision tree models, the feature data of the three-dimensional structure corresponding to the first electron micrograph is obtained.

[0128] Optionally, the three-dimensional feature recognition device 900 further includes:

[0129] A validation set acquisition module is used to acquire a validation set, which includes electron micrographs with three-dimensional feature annotations and electron micrographs without three-dimensional feature annotations.

[0130] The first verification module is used to verify the three-dimensional recognition structure model based on the electron micrograph with three-dimensional feature annotations, and obtain a first verification result;

[0131] The model adjustment module is used to adjust the three-dimensional recognition structure model based on the first verification result to obtain the adjusted three-dimensional feature recognition model.

[0132] The second verification module is used to verify the adjusted three-dimensional feature recognition model based on the electron micrograph without three-dimensional feature annotation, and obtain a second verification result.

[0133] Optionally, the image generation module is specifically used for:

[0134] Using a Monte Carlo simulator, simulation calculations are performed based on pre-set simulation conditions to generate the initial electron micrograph corresponding to the three-dimensional model.

[0135] A third electron microscopy image is acquired, wherein the feature data of the three-dimensional structure corresponding to the third electron microscopy image is the same as the feature data of the micro / nano structure.

[0136] The initial electron micrograph is corrected based on the third electron micrograph to obtain the second electron micrograph.

[0137] Furthermore, embodiments of this application also provide an electronic device, including: a processor, a memory, and a system bus;

[0138] The processor and the memory are connected via the system bus;

[0139] The memory is used to store one or more programs, the one or more programs including instructions that, when executed by the processor, cause the processor to perform any of the implementation steps of the three-dimensional feature recognition method described above.

[0140] Furthermore, embodiments of this application also provide a computer-readable storage medium storing instructions that, when executed on an electronic device, enable any of the above-described steps of the three-dimensional feature recognition method to be implemented.

[0141] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application. It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on describing the differences from other embodiments. The same or similar parts between the various embodiments can be referred to mutually.

[0142] The system disclosed in the embodiments is described in a relatively simple manner because it corresponds to the method disclosed in the embodiments. For relevant details, please refer to the method section.

[0143] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0144] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for recognizing three-dimensional features, characterized in that, include: Acquire the first electron micrograph; Generate the first frequency domain information corresponding to the first electron micrograph; The first frequency domain information is analyzed based on the three-dimensional feature recognition model to obtain the feature data of the three-dimensional structure corresponding to the first electron micrograph. Output the feature data of the three-dimensional structure corresponding to the first electron micrograph.

2. The method for recognizing three-dimensional features according to claim 1, characterized in that, Before acquiring the first electron micrograph, the method further includes: Constructing three-dimensional models of micro and nanostructures; Generate a second electron micrograph corresponding to the three-dimensional model; Generate the second frequency domain information corresponding to the second electron micrograph; The model is trained based on the second frequency domain information, and the three-dimensional feature recognition model is obtained after the training is completed.

3. The method for recognizing three-dimensional features according to claim 2, characterized in that, The generation of the second frequency domain information corresponding to the second electron micrograph includes: Extract the grayscale information from the second electron micrograph; The grayscale information of the second microscopic image is decomposed and transformed using a spectrum analyzer to obtain the second frequency domain information.

4. The method for recognizing three-dimensional features according to claim 2, characterized in that, The step of training the model based on the second frequency domain information, and obtaining the three-dimensional feature recognition model after training, includes: Using the second frequency domain information as input data for multiple decision tree models, and the feature data in the second frequency domain information as output data for the multiple decision tree models, the multiple decision tree models are trained to obtain multiple trained decision tree models. A random forest model is generated based on the multiple trained decision tree models, and the random forest model is used as the three-dimensional feature recognition model.

5. The method for recognizing three-dimensional features according to claim 4, characterized in that, The analysis of the first frequency domain information based on the three-dimensional feature recognition model to obtain the feature data of the three-dimensional structure corresponding to the first electron micrograph includes: The first frequency domain information is analyzed based on the multiple trained decision tree models in the random forest model to obtain the feature data output by the multiple trained decision tree models respectively. By integrating the feature data output by the multiple trained decision tree models, the feature data of the three-dimensional structure corresponding to the first electron micrograph is obtained.

6. The method for recognizing three-dimensional features according to claim 2, characterized in that, After training the model based on the second frequency domain information and obtaining the three-dimensional feature recognition model, the method includes: Obtain a validation set, which includes electron micrographs with three-dimensional feature annotations and electron micrographs without three-dimensional feature annotations; The three-dimensional recognition structure model is verified based on the electron micrograph with three-dimensional feature annotations to obtain a first verification result; Based on the first verification result, the three-dimensional recognition structure model is adjusted to obtain the adjusted three-dimensional feature recognition model; The adjusted three-dimensional feature recognition model is verified based on the electron micrograph without three-dimensional feature annotations to obtain a second verification result.

7. The method for recognizing three-dimensional features according to claim 2, characterized in that, Generating the second electron micrograph corresponding to the three-dimensional model includes: Using a Monte Carlo simulator, simulation calculations are performed based on pre-set simulation conditions to generate the initial electron micrograph corresponding to the three-dimensional model. A third electron microscopy image is acquired, wherein the feature data of the three-dimensional structure corresponding to the third electron microscopy image is the same as the feature data of the micro / nano structure. The initial electron micrograph is corrected based on the third electron micrograph to obtain the second electron micrograph.

8. A three-dimensional feature recognition device, characterized in that, include: The image acquisition module is used to acquire the first electron micrograph; The first frequency domain information generation module is used to generate the first frequency domain information corresponding to the first electron micrograph; The feature data acquisition module is used to analyze the first frequency domain information based on the three-dimensional feature recognition model to obtain the feature data of the three-dimensional structure corresponding to the first electron micrograph. The feature data output module is used to output the feature data of the three-dimensional structure corresponding to the first electron micrograph.

9. An electronic device, characterized in that, The device includes: a processor, a memory, and a system bus; The processor and the memory are connected via the system bus; The memory is used to store a program, the program including instructions that, when executed by the processor, cause the processor to perform the steps of the three-dimensional feature recognition method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program, which, when executed by a terminal device, implements the steps of the three-dimensional feature recognition method according to any one of claims 1 to 7.

Images