Automated measurement method for electron micrographs obtained in semiconductor processes

Abstract

The present disclosure relates to an automatic measurement method for electron micrograph images obtained in a semiconductor process. The method comprises: obtaining an electron micrograph image to be analyzed as a first image; obtaining a region selected by a user in the first image as a first region, and obtaining information of a manual measurement operation performed by the user in the first region; obtaining a value of a specified image feature of the first region, and selecting one from a plurality of preset region matching algorithms as a first algorithm based on the specified image feature value; obtaining a value of a specified edge feature related to the manual measurement operation information, and selecting one from a plurality of preset edge recognition algorithms as a second algorithm based on the specified edge feature value; and recognizing at least one second region from the first image by using the first algorithm, recognizing an edge to be measured in the second region by using the second algorithm, and performing automatic measurement by using the edge.

Description

Technical Field

[0001] This disclosure relates to the field of semiconductor process inspection technology, and in particular to an automated measurement method for electron micrographs obtained in semiconductor processes. Background Technology

[0002] In the development and mass production of semiconductor processes, the analysis of the fine structure of devices is a crucial step in supporting process development, ensuring process stability, and improving product yield. Since the structural dimensions involved in process development are typically smaller than micrometers, and advanced processes involve structural dimensions reaching several nanometers or even smaller, this process is highly dependent on electron microscopy images. Currently, commonly used electron microscopy imaging methods in the industry include scanning electron microscopy (SEM), transmission electron microscopy (TEM), and scanning transmission electron microscopy (STEM). Among them, SEM, with its high resolution and large depth of field, is often used to observe the surface and near-surface structures of devices; TEM can penetrate samples to obtain internal microstructural information and is a core means of analyzing the multilayer structure and interface features of devices; STEM combines the characteristics of SEM and TEM, enabling elemental analysis while acquiring high-resolution images. The electron microscopic images obtained by these imaging methods are the core basis for the analysis of key structures such as the gate structure of logic devices, the memory cell array of memory devices, the photosensitive elements of imaging devices, and the epitaxial layers of power devices.

[0003] Current methods for analyzing these electron micrographs still primarily rely on engineers manually measuring their geometric features. However, this approach suffers from several insurmountable drawbacks: 1. Inefficiency: During the semiconductor process development and mass production stages, massive electron micrographs need to be processed. Manually measuring feature dimensions and statistically analyzing structural parameters image by image is extremely time-consuming and cannot meet the needs of rapid analysis of large-scale data. 2. Limited data volume: Due to the inefficiency of manual operation, engineers can only select a small number of samples from a massive amount of images for analysis, resulting in the analysis results failing to fully reflect the overall state of the process and easily overlooking potential process anomalies; 3. High dependence on subjectivity: The results of manual measurement are highly dependent on the personal judgment of engineers. For example, there are subjective differences in the identification of feature edges and the selection of measurement start and end points. Measurement results from different engineers or the same engineer at different times are prone to deviation, resulting in poor data consistency. 4. Experience limitations: Novice engineers lack a deep understanding of semiconductor structure and image features, which can easily lead to misjudgments during manual measurement, resulting in low accuracy and poor reliability of the extracted data. Meanwhile, the experience of senior engineers is difficult to quickly replicate and promote.

[0004] Therefore, we hope that there will be corresponding improvement plans. Summary of the Invention

[0005] In view of this, one or more embodiments of the present disclosure provide an automated measurement method and computing device for electron micrographs obtained in semiconductor processes.

[0006] According to a first aspect of this disclosure, an automated measurement method for electron micrographs obtained in semiconductor processes is provided, comprising: The electron micrograph to be analyzed is obtained as the first image; Obtain the area selected by the user in the first image as the first region, and obtain information about the manual measurement operations performed by the user in the first region; Obtain the value of a specified image feature of the first region, and based on the specified image feature value, select one from a plurality of preset region matching algorithms as the first algorithm; Obtain the value of a specified edge feature related to the manual measurement operation information, and based on the specified edge feature value, select one from a set of preset edge recognition algorithms as a second algorithm; and At least one second region is identified from a first image using a first algorithm, an edge to be measured is identified in the second region using a second algorithm, and the edge is used for automatic measurement.

[0007] In one example, the method further includes, before identifying at least one second region from the first image using the first algorithm: Based on the specified image feature values, determine the values ​​of specified parameters of the first algorithm; and / or, Based on the specified edge feature value, the value of the specified parameter of the second algorithm is determined.

[0008] In one example, the method further includes obtaining the geometric parameters to be measured selected by the user before obtaining the values ​​of specified image features of the first region; The specified image features and / or the specified edge features are determined at least based on the geometric parameters; and / or the region matching algorithm and / or the edge recognition algorithm are determined at least based on the geometric parameters.

[0009] In one example, the method further includes, after performing automatic measurement using the edge: Obtain information about the user's editing operations on the results of the automatic measurement; Obtain the values ​​of the specified image features of the second region involved in the editing operation and / or the values ​​of the specified edge features related to the editing operation; and Based on the value of the specified image feature of the second region, the first algorithm is reselected or the value of the specified parameter of the first algorithm is adjusted, and / or based on the value of the specified edge feature related to the editing operation, the second algorithm is reselected or the value of the specified parameter of the second algorithm is adjusted.

[0010] In one example, based on the specified image feature values, one of a set of preset region matching algorithms is selected as the first algorithm, including: Based on the specified image feature values ​​and a preset first selection condition, one of a plurality of preset region matching algorithms is selected as the first algorithm.

[0011] In one example, obtaining the value of a specified edge feature associated with the manual measurement operation information includes: When the manual measurement operation information includes information about the measurement point input by the user, the value of the specified edge feature within a predetermined range around the measurement point is obtained.

[0012] In one example, based on the specified edge feature value, one of a set of preset edge recognition algorithms is selected as the second algorithm, including: Based on the specified edge feature value, and based on the preset second selection condition, one of the preset edge recognition algorithms is selected as the second algorithm.

[0013] In one example, the specified image feature includes at least one of the following: Noise level, contrast distribution, structure density, edge sharpness, edge strength, information entropy changes between different sub-regions, interlayer contrast characteristics, interlayer interface blurring degree, morphological features, and image feature encoding obtained by a specified deep learning model.

[0014] In one example, the region matching algorithm includes at least one of the following: Adaptive threshold segmentation algorithm, region growing algorithm, texture matching algorithm, multi-threshold layering algorithm, morphological feature matching algorithm, and encoding vector spacing algorithm.

[0015] In one example, the specified edge feature includes at least one of the following: The grayscale gradient of the edge, the edge width, the blur degree of the edge, the grayscale difference of the edge, the edge shape, and the area of ​​the closed region of the edge.

[0016] In one example, the edge recognition algorithm includes at least one of the following: Improved Canny operator algorithm, adaptive Sobel operator algorithm, gray-level gradient peak localization algorithm, gray-level transition point localization algorithm, contour extraction algorithm.

[0017] According to a second aspect of this disclosure, a computing device is provided, comprising: a processor; and a memory having executable code stored thereon, which, when executed by the processor, causes the processor to perform the method described in the first aspect above.

[0018] According to a third aspect of this disclosure, a computer program product is provided, including executable code that, when executed by a processor of an electronic device, causes the processor to perform the method described in the first aspect above.

[0019] According to a fourth aspect of this disclosure, a non-transitory machine-readable storage medium is provided, on which executable code is stored, which, when executed by a processor of an electronic device, causes the processor to perform the method described in the first aspect above.

[0020] Therefore, the method according to this disclosure, compared with existing electron microscopy image measurement methods, can balance efficiency, accuracy, versatility and ease of use, and meet the needs of semiconductor process development and mass production stages for rapid and accurate measurement of fine structures. Attached Figure Description

[0021] The above and other objects, features and advantages of this disclosure will become more apparent from the more detailed description of exemplary embodiments thereof taken in conjunction with the accompanying drawings, wherein like reference numerals generally denote like parts.

[0022] Figure 1 A schematic flowchart of an automated measurement method for electron micrographs obtained in a semiconductor process, according to at least one embodiment of the present disclosure, is shown.

[0023] Figures 2 to 8 Several exemplary schematic diagrams are shown, respectively illustrating user-performed region selection and manual measurement operations according to various embodiments of the present disclosure.

[0024] Figure 9 An exemplary schematic diagram showing the results of automated measurements according to at least one embodiment of the present disclosure is provided.

[0025] Figure 10 An exemplary schematic diagram of a batch measurement statistical report according to at least one embodiment of the present disclosure is shown.

[0026] Figure 11A schematic flowchart of some steps in an automated measurement method for electron micrographs obtained in a semiconductor process, according to at least one embodiment of the present disclosure, is shown.

[0027] Figure 12 A schematic diagram of the structure of a computing device according to at least one embodiment of the present disclosure is shown. Detailed Implementation

[0028] Preferred embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While preferred embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0029] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0030] It should be understood that terms such as "first," "second," etc., used herein are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first," "second," or similar terms may explicitly or implicitly include one or more of the stated features.

[0031] As mentioned earlier, electron micrographs in semiconductor processes are still mainly measured manually by engineers. To address the shortcomings of manual measurement, the industry has tried some alternative solutions, including traditional image recognition methods (such as edge detection) and deep learning-based image analysis solutions.

[0032] For example, traditional edge detection-based image recognition methods can extract edge information from images using traditional algorithms such as the Sobel operator, Canny operator, and Laplacian operator, thereby enabling the measurement of feature dimensions, etc.

[0033] For example, deep learning-based image analysis methods can automatically identify and measure feature structures in electron micrographs by constructing neural network models and training them with a large amount of labeled data.

[0034] However, this disclosure finds that these methods still have the following obvious problems: First, for measurement schemes based on traditional image recognition methods: 1. Poor versatility: Semiconductor electron microscopy images are complex and diverse. For example, the noise level of SEM images, the contrast difference of TEM images, and the image features of different device structures are all different. A single edge detection algorithm is difficult to adapt to all scenarios. An algorithm that performs well in one type of image may miss or falsely detect edges in another type of image. 2. Low recognition accuracy: Traditional algorithms are not robust to problems such as image noise and uneven contrast. Although they can provide high-precision recognition results in specific optimized scenarios, interference from multiple stages in the semiconductor process, including manufacturing, prototyping, and imaging, can occur. For example, when there is noise interference in the image or the structural edges are blurred, edge positioning deviations are prone to occur, leading to insufficient measurement accuracy. 3. High barrier to entry: The efficient and accurate application of this type of method requires users to have a solid theoretical foundation in image analysis. They need to manually adjust the algorithm parameters (such as threshold, operator size and value) according to different image scenarios. However, the core professional field of semiconductor process engineers is process development and device design. They lack knowledge of image analysis and find it difficult to master and apply this type of method. Second, regarding measurement schemes for deep learning-based image analysis methods: 1. Heavy reliance on labeled data: The performance of deep learning models is highly dependent on the quantity and quality of labeled data. In terms of quantity, thousands or even tens of thousands of electron micrographs under different scenes and process conditions need to be collected and labeled, resulting in extremely high data collection costs. In terms of quality, the labeling work needs to be completed by senior engineers to ensure the accuracy of labeled information such as feature edges and measurement areas, and the time and labor costs of high-quality labeling far exceed those of ordinary manual measurement. 2. High initial investment: In addition to labeled data, the development and training of deep learning models require a professional team of algorithm engineers, as well as high-performance computing hardware (such as GPU servers), resulting in extremely high time and financial costs in the early stages. 3. Time lag: When semiconductor processes are updated (such as device structure adjustment and process upgrade) or measurement requirements change (such as the addition of new feature parameter measurements), it is necessary to re-collect labeled data, adjust the model structure and train, and the whole process takes several weeks or even months, which cannot quickly respond to new process or measurement requirements. 4. Limited generalization ability: Even if the model performs well in existing labeled data scenarios, its recognition accuracy will drop significantly when faced with new scenarios not included in the training (such as images of new device structures or abnormal processes), making it difficult to adapt to the rapid iteration of semiconductor processes.

[0035] In view of this, this disclosure proposes a novel automated measurement method for electron micrographs obtained in semiconductor processes. This method automatically analyzes the features related to the user's manual measurement operations and selects an appropriate algorithm to perform automated measurement of the electron micrographs. Compared with existing electron micrograph measurement methods, this method can balance efficiency, accuracy, versatility and ease of use, and meet the needs of rapid and accurate measurement of fine structures in the semiconductor process development and mass production stages.

[0036] The disclosed solution does not limit the type of electron microscopy image and is applicable to various electron microscopy images (e.g., SEM images, TEM images, or STEM images). The disclosed solution also does not limit the semiconductor process and the type of semiconductor device fabricated, nor the type of measurement. It can be widely applied in the process development and mass production stages of various semiconductor devices, such as logic devices, memory devices, image sensors, and power devices, to achieve efficient and accurate measurement of various geometric features (such as critical dimensions (CD) and other geometric parameters) of various structures within the device.

[0037] At least some embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings.

[0038] Figure 1 A schematic flowchart of an automated measurement method for electron micrographs obtained in a semiconductor process, according to at least one embodiment of the present disclosure, is shown. Additionally, unless otherwise indicated herein, "a plurality of" refers to two or more.

[0039] like Figure 1 As shown, in step S101, the electron micrograph to be analyzed is obtained as the first image.

[0040] For example, in the graphical user interface (GUI) of the measurement software provided by the method according to the embodiments of the present disclosure, the engineer, i.e. the user, opens the electron micrograph to be analyzed (such as a SEM image, TEM image, or STEM image, etc.), and the method of the embodiments of the present disclosure loads and obtains the electron micrograph as the first image.

[0041] In some examples, the method according to embodiments of this disclosure can also initialize measuring tools (such as ruler calibration, image zoom in / out functions, etc.) to determine the correspondence between the pixel size of the first image and its actual physical size. For example, ruler calibration can be achieved by reading the ruler built into the first image or based on a pixel-to-physical size conversion relationship input by the user.

[0042] Then, in step S102, the region selected by the user in the first image is obtained as the first region, and information on the manual measurement operation performed by the user in the first region is obtained.

[0043] For example, in the GUI provided by the method according to the embodiments of the present disclosure, the user can select (e.g., box selection) the area to be analyzed (i.e., "area of ​​interest") in the first image by dragging the mouse or other means according to measurement needs.

[0044] For example, when measuring the gate CD of a logic device, the user can select a region containing a single gate structure to guide subsequent measurement actions. This avoids the measurement method performing invalid analysis on irrelevant structures (such as the substrate region) in the first image, thus improving measurement efficiency. This operation is consistent with the user's daily manual measurement process, requiring no learning of new operating logic.

[0045] Following a user action, the method of this embodiment obtains the region selected by the user in the first image as the first region. For example, the image coordinate range of the region selected by the user can be recorded.

[0046] In some examples, the user can select multiple regions of interest based on measurement needs. Accordingly, the first region obtained in step S102 is a plurality of first regions, each corresponding to a region selected by the user.

[0047] For example, these regions of interest can have multiple levels. A user can first select an upper-level region of interest, and then select one or more lower-level regions of interest within that upper-level region. For instance, a user can first select a region of repeating cells in a logic circuit as the upper-level region of interest, and then select one of the transistor structures within that circuit as its lower-level region of interest.

[0048] After selecting a region of interest, the user can perform manual measurements within that region, for example, by clicking with a mouse. For instance, any point entered by the user (e.g., a mouse click) for measurement can be referred to as a measurement point. Accordingly, the method of this embodiment obtains information about the manual measurement operations performed by the user in the first region in step S102, such as information about each measurement point, like the coordinates of the start and end points of the measurement clicked by the user. For example, the manual measurement operation information may include information about all input actions performed by the user during manual measurement (such as mouse actions or other peripheral device (e.g., keyboard) input actions).

[0049] For example, a user can use the mouse to click on both sides of the structure to be measured to determine the start and end positions of the measurement, completing a manual CD measurement. This operation is consistent with the user's daily manual measurement process, requiring no learning of new operational logic. Accordingly, the manual measurement operation information obtained in step S102 includes the coordinate information of the measurement start and end points clicked by the user.

[0050] This disclosure does not limit the content of the user's manual measurement operations or the geometric features and parameters of the measurements. The geometric features to be measured may include, for example: the distance between two points or two lines (the corresponding geometric parameters may be dimensions in a single dimension, such as width, thickness, etc.), the angle formed by two lines, the contour of a specific structure (such as the contour of a void defect, the corresponding geometric parameters may be such as defect density), the height of a one-dimensional line in a two-dimensional image, and so on.

[0051] Additionally, users can perform one or more manual measurement operations in the area of ​​interest as needed; for example, they can measure both the width of one structure and the length of another. Accordingly, the method of this embodiment can obtain information on all manual measurement operations performed by the user in step S102.

[0052] Furthermore, in some examples, users can select the geometric parameters to be measured before performing manual measurements. Therefore, the method of this disclosure embodiment can also obtain the geometric parameters selected by the user. The method of this disclosure embodiment can also set different measurement scenarios for different geometric parameters, and, as needed, can provide automated assistance for manual measurements in some measurement scenarios, as described below. Figure 3 Examples of measurements are described.

[0053] In some examples, as needed, the specified image features in step S103 and / or the specified edge features in step S104 can be determined at least based on the geometric parameters. Additionally, as needed, multiple preset region matching algorithms in step S103 and / or multiple preset edge recognition algorithms can be determined at least based on the geometric parameters. This allows for more accurate selection of appropriate region matching algorithms and / or edge recognition algorithms.

[0054] Figures 2 to 8 Several examples of user-performed region selection and manual measurement operations are provided. For ease of understanding, Figures 2 to 8 All images are presented using simplified diagrams instead of actual electron microscope images. The points clicked by the user for measurement, as described below, can all be referred to as measurement points.

[0055] Figure 2 This is an example of performing a single CD measurement in, for example, an SEM image. The yellow rectangle represents the "region of interest" selected by the user (which contains, for example, one gate structure), the blue dots represent the start and end points of the measurement clicked by the user (the left edge point and the right edge point), and the green line segments represent the manually measured CD marking lines.

[0056] Figure 3This is an example of performing multiple CD measurements. The yellow rectangle represents the "region of interest" selected by the user (which contains, for example, one gate structure), the blue dots represent the measurement range clicked by the user, and the green lines represent a set of CD result markers obtained by the method of this embodiment of the disclosure for automatic measurement.

[0057] Figure 4 This is an example of measuring extreme values ​​of the spacing. The yellow rectangle represents the "region of interest" selected by the user (which contains, for example, one gate structure), the blue dots represent the measurement range clicked by the user, and the green line segments are the annotation lines for the CD extreme value results obtained automatically by the method of this embodiment. Although Figure 4 The value shown is the minimum spacing in the horizontal direction, but the orientation of the measurement can be horizontal, vertical, or any specific orientation as needed, and the extreme value type can be the maximum or minimum value.

[0058] Figure 5 This is an example of angle measurement. The yellow rectangle represents the "area of ​​interest" selected by the user, the blue dots represent the start and end edges of the measurement clicked by the user, and the green line segment and angle symbol represent the result label line of an angle measured from the input point, which can be measured by the method of this disclosure embodiment.

[0059] Figure 6 This is an example of boundary measurement. The yellow rectangle represents the "area of ​​interest" selected by the user, the blue dots represent multiple reference points clicked by the user, and the green line segment represents the boundary automatically detected by the method of this embodiment based on the user-clicked reference points.

[0060] Figure 7 This is an example of measuring the spacing between feature points or feature lines. The yellow rectangle represents the user-selected "region of interest," the red horizontal line is the horizontal reference line set by the user, and the blue dots are the target feature points set by the user. The green line segment represents the distance from the blue dots to the red horizontal line. Users can use horizontal or vertical reference lines, feature points, and feature regions as needed, and can set the measurement for horizontal spacing / vertical spacing / in-plane spacing, etc. Furthermore, users can mix and match features at different levels as needed, for example... Figure 7 The red horizontal lines represent global features, while the blue feature points represent features within a region of interest (such as a gate structure).

[0061] Figure 8 This is an example of region segmentation measurement. The yellow rectangle represents the "region of interest" selected by the user, the blue dots represent one or more "foreground points" or "reserved points" clicked by the user, the red dots represent one or more "background points" or "exclusion points" clicked by the user, and the green area represents the segmented region automatically obtained from the input points, which can be automatically performed by the method of this disclosure embodiment.

[0062] In addition, with Figure 8 Similarly, defect measurements can also be performed. For example, in cases where it is necessary to measure the density of void defects in a structure, the user can select a region of interest containing typical defects, including the structure of interest, such as an epitaxial layer. Then, within this region of interest, the user can manually click on one or more points within the defects to mark the void defects, and manually click on one or more points outside the defects to mark the non-defective parts of the structure. The void defect region can be automatically analyzed using these two sets of input points, which can be performed automatically by the method of this disclosure embodiment.

[0063] Back Figure 1 After step S102, step S103 is performed, in which the value of a specified image feature of the first region is obtained, and based on the specified image feature value, one of a plurality of preset region matching algorithms is selected as the first algorithm.

[0064] In some examples, the specified image features can be one or more pre-defined image features as needed. For example, the number and type of the specified image features can be pre-defined based on the geometric parameters to be measured and / or a pre-defined region matching algorithm.

[0065] For example, the specified image features may include at least one of the following: noise level, contrast distribution, structure density, edge sharpness, edge strength, information entropy change between different sub-regions, interlayer contrast features, interlayer interface blurring degree, morphological features, and image feature encoding obtained by a specified deep learning model.

[0066] For example, at least one of the following methods can be used as needed: pixel brightness statistics, Sobel edge detection, image information entropy calculation, encoding by a pre-trained specified deep learning model, etc., to automatically analyze and obtain the values ​​of specified image features in the first region.

[0067] In some examples, multiple region matching algorithms can be preset as needed to cover various measurement scenarios and ensure the accuracy of region matching under different conditions. For example, region matching algorithms with optimal region matching accuracy can be preset for different imaging methods (such as SEM / TEM / STEM, etc.), different device types (such as logic / memory / imaging / power devices, etc.), and / or different image qualities (noise levels, contrast differences, etc.).

[0068] For example, the region matching algorithm may include at least one of the following: adaptive threshold segmentation algorithm, region growing algorithm, texture matching algorithm, multi-threshold hierarchical algorithm, morphological feature matching algorithm, and encoding vector spacing algorithm.

[0069] In some examples, a first algorithm can be selected from a set of preset region matching algorithms based on the specified image feature value and a preset first selection condition. For example, the first selection condition can be used to filter out the algorithm that best matches the specified image feature value from the preset multiple region matching algorithms, that is, the algorithm with the highest accuracy in region matching given the specified image feature value.

[0070] For example, the first selection condition may include at least one conditional expression. The conditional expression may include, for example, a logical judgment constructed based on the specified image feature values.

[0071] For example, the conditional expression may include at least one of the following: a threshold comparison expression that compares the corresponding specified image feature value with its threshold; a logical combination expression that combines multiple threshold comparison expressions; a judgment expression that calculates the total score by weighting or summing the corresponding multiple specified image feature values ​​or their corresponding scores equally or in a weighted manner, such as selecting different algorithms based on the total score falling into different intervals.

[0072] Furthermore, in some examples, the method of this disclosure embodiment can also determine the values ​​of specified parameters of the first algorithm based on the specified image feature values, as needed. For example, the specified parameters may include key parameters of the first algorithm, such as the number of histogram partitions, the threshold parameter for edge detection, the optimal window size for two-dimensional region matching, and the value of the adaptive threshold. Specified parameters can be pre-set for multiple candidate region matching algorithms.

[0073] The determined first algorithm and its optional parameter values ​​can form a "region identification configuration" for the user's region of interest (i.e., the first region). This configuration ensures that the method of this disclosure embodiment only performs subsequent analysis on regions that are consistent with the region characteristics selected by the user, avoiding interference from irrelevant regions.

[0074] Then, in step S104, the value of the specified edge feature related to the manual measurement operation information is obtained, and based on the specified edge feature value, one of the preset edge recognition algorithms is selected as the second algorithm.

[0075] Understandable, although Figure 1 The example given shows that step S104 is executed after step S103, but this disclosure is not limited to this. Steps S103 and S104 can be executed in any order. For example, step S103 can be executed after step S104, or steps S103 and S104 can be executed in parallel, or so on.

[0076] In some examples, the specified edge features can be one or more edge features preset as needed. For example, the number and type of the specified edge features can be preset based on the geometric parameters to be measured and / or a preset edge recognition algorithm.

[0077] For example, the specified edge features may include at least one of the following: grayscale gradient of the edge, edge width, edge blurring degree, grayscale difference of the edge, edge shape, and area of ​​the edge closed region.

[0078] For example, at least one of the following methods can be used as needed: thresholding, brightness template matching, or encoding distance using a pre-trained deep learning model, to automatically analyze and obtain the values ​​of specified edge features related to the manual measurement operation information. For example, if the manual measurement operation information includes information about measurement points input by the user (such as mouse clicks), the values ​​of the specified edge features within a predetermined range around the measurement points can be obtained.

[0079] In some examples, multiple edge recognition algorithms can be preset as needed to cover various measurement scenarios and ensure the accuracy of edge recognition under different conditions. For example, edge recognition algorithms with optimal edge recognition accuracy can be preset for different imaging methods (such as SEM / TEM / STEM, etc.), different device types (such as logic / memory / imaging / power devices, etc.), and / or different image qualities (noise levels, contrast differences, etc.).

[0080] For example, the edge recognition algorithm may include at least one of the following: improved Canny operator algorithm, adaptive Sobel operator algorithm, gray-level gradient peak localization algorithm, gray-level transition point localization algorithm, and contour extraction algorithm.

[0081] In some examples, a second edge recognition algorithm can be selected from a set of preset edge recognition algorithms based on the specified edge feature value and a preset second selection condition. For example, the second selection condition can be used to filter out the algorithm that best matches the specified edge feature value from the preset edge recognition algorithms, that is, the algorithm with the highest accuracy in edge recognition given the specified edge feature value.

[0082] For example, the second selection condition may include at least one conditional expression. The conditional expression may include, for example, a logical judgment constructed based on the specified edge characteristic value.

[0083] For example, the conditional expression may include at least one of the following: a threshold comparison expression that compares the corresponding specified edge feature value with its threshold; a logical combination expression that combines multiple threshold comparison expressions; a judgment expression that calculates the total score by equally weighting or weighted summing the corresponding multiple specified edge feature values ​​or their corresponding scores, such as selecting different algorithms based on the total score falling into different intervals.

[0084] Furthermore, in some examples, the method of this disclosure embodiment can also determine the values ​​of specified parameters of the second algorithm based on specified edge feature values ​​as needed. For example, the specified parameters may include key parameters of the second algorithm, such as gradient thresholds, edge smoothing coefficients, etc. Specified parameters can be pre-set for multiple candidate edge recognition algorithms.

[0085] The determined second algorithm and its optional parameter values ​​can form an "edge localization configuration" for edges related to user manual measurements. This configuration ensures that the method of this disclosure embodiment can accurately locate edges that correspond to the edge features measured by the user manually.

[0086] Then, in step S105, at least one second region is identified from the first image using a first algorithm, the edge to be measured is identified in the second region using a second algorithm, and the edge is automatically measured.

[0087] Understandable, although Figure 1 The example given shows that steps S103, S104, and S105 are executed sequentially. However, this disclosure is not limited to this, and steps S103, S104, and S105 can be executed in any feasible order as needed. For example, regardless of the execution order of steps S104, the operation of identifying at least one second region from the first image using the first algorithm in step S105 can be performed after step S103; or, regardless of the execution order of steps S103, the operation of identifying the edge to be measured in the second region using the second algorithm in step S105 and automatically measuring the edge can be performed after step S104.

[0088] For example, the "region identification configuration" obtained in step 103 can be combined with the "edge positioning configuration" obtained in step 104 to form an automated measurement scheme, which can be applied to the current image to be measured, i.e., the first image, so as to automatically identify all target edges in the first image that meet the features, thereby calculating the measurement parameters and outputting the results accordingly.

[0089] For example, a first algorithm can be used to identify regions in the first image that match specified image features of the first region as the second region. In some cases, it is possible to select whether to detect a single matching region or all matching regions as the second region, depending on the needs, to adapt to different scenarios.

[0090] When detecting a single matching region, only the best matching region among all candidate regions (i.e., the region with the highest matching degree, such as the highest similarity, the shortest cluster distance, or the highest total score) is output as the second region. When detecting all matching regions, all regions that meet the conditions set in the region recognition configuration (i.e., the first algorithm) (such as similarity threshold, cluster distance threshold, etc.) are output as the second region.

[0091] Then, within each identified second region, the second algorithm can be used to identify edges that match the specified edge features related to the user's manual measurement as the edges to be measured. For example, the edge closest to the aforementioned specified edge feature value (e.g., the sum of the values ​​or the total score) can be detected as the edge to be measured. Then, automatic measurement is performed using the detected edges, such as calculating CD. In some examples, the automatic measurement results can be displayed on the GUI provided by the method of this disclosure embodiment for user verification, such as... Figure 9 shown.

[0092] Figure 9 An exemplary schematic diagram showing the results of automated measurements according to at least one embodiment of the present disclosure is provided. Figure 9 Corresponding to the above Figure 2 The scenario of manually measuring a single point CD.

[0093] like Figure 9 As shown, automated measurement was performed on the first image, resulting in five yellow rectangles, which represent five second regions automatically identified using region recognition configuration. These regions are then compared with... Figure 2 The user-selected "area of ​​interest" shown is consistent with each of them, each containing, for example, one gate structure; additionally, each yellow rectangle contains a green line segment, which represents the CD measurement marking line between the edges automatically identified using the second algorithm within each area, and is consistent with... Figure 2 The CD markings shown are consistent with those obtained by user manual measurement.

[0094] In addition, in some examples, all measurement results in the first image can be statistically analyzed (e.g., individual CD values, average, standard deviation, maximum, minimum, etc.) and output in tabular or chart form. Measurement markers (such as edge positions, CD values, etc.) can also be overlaid on the first image for easy user verification.

[0095] In addition, in some examples, if the user needs to measure a batch of electron micrographs under the same process and imaging conditions, the "region identification configuration" and "edge positioning configuration" generated in steps S103 and S104 can be saved as a "measurement recipe". In subsequent use, only the batch of images needs to be imported and the recipe loaded; the method of this embodiment can automatically apply the same measurement scheme to complete the automated measurement of all images without requiring repeated manual operation by the user.

[0096] Figure 10 An exemplary schematic diagram of a batch measurement statistical report according to at least one embodiment of the present disclosure is shown.

[0097] Figure 10 This is an example of a statistical process control (SPC) analysis curve for gate CD obtained by automatic measurement. In this curve, the X-axis represents the number of each image imported in batches, the Y-axis represents the corresponding CD mean, the yellow control line represents a set of upper and lower limits of process control as the qualification limit, and the red control line represents another set of upper and lower limits of process control as the control limit. Figure 10 This demonstrates intuitively the data output capability of the method for batch measurement in the embodiments of this disclosure.

[0098] The method of this disclosure allows for the automated measurement of data from a first image or other images or batches of images. This data can be output in various applicable analytical charts to visually demonstrate the statistical characteristics and trends of the measured parameters, supporting semiconductor process analysis. These analytical charts include, but are not limited to: scatter plots, box plots, indicator completion tables, profile curves, and profile power spectral density plots.

[0099] Furthermore, in some examples, to further improve measurement accuracy and cope with complex image scenes (such as images with a few abnormal structures or blurred edges), the method of this disclosure embodiment can also be used in... Figure 1 After the automatic measurement in step S105, a process for user editing and iterative optimization of the results is added, such as... Figure 11 exemplified.

[0100] Figure 11 A schematic flowchart of some steps in an automated measurement method for electron micrographs obtained in a semiconductor process, according to at least one embodiment of the present disclosure, is shown.

[0101] like Figure 11 As shown above, in the aforementioned Figure 1 The process may further include step S1106 after step S105, wherein information on the editing operations performed by the user on the results of the automatic measurement is obtained.

[0102] For example, the editing operation may include at least one of deleting erroneous measurement marks, supplementing missed structural measurement points, and fine-tuning edge positioning.

[0103] In some examples, the automatic measurement results can be displayed in the GUI provided by the method of this disclosure embodiment, and an interactive editing interface (including function buttons such as "delete mark", "manual adjustment", "recalculate" etc.) can be provided. In addition, one or more blurry areas that need to be corrected can be marked.

[0104] Users can view the automatic measurement results and measurement markers on the images output in the GUI. If they find individual erroneous results (such as misidentifying noise as structural edges, missing structural parts, etc.), they can edit them using the mouse. For example, users can delete erroneous measurement markers, manually add the start and end points of measurements for missed structures, and fine-tune the edge positioning. In this process, users only need to operate on the specific measurement results of interest and do not need to modify other automatically identified results. Accordingly, the method of this embodiment can receive information about the interactive editing operations performed by the user on the automatic measurement results.

[0105] Then, the specified image features and / or specified edge features can be re-analyzed based on the editing operation information.

[0106] like Figure 11 As shown, in cases where the editing operation involves a matching problem in a second region, such as when a user deletes an incorrect measurement marker, step S1107 is performed, wherein the value of the specified image feature of the second region involved in the editing operation is obtained; in cases where the editing operation involves an edge recognition problem, such as when a user fine-tunes the edge positioning, step S1108 is performed, wherein the value of the specified edge feature related to the editing operation is obtained. In cases where the editing operation involves both of the above problems, steps S1107 and S1108 are performed. Figure 11 The dashed arrows in the text indicate that these steps are selected to be executed based on the user's editing actions, and are not always required.

[0107] Then, based on the values ​​of the specified image features and / or specified edge features obtained again, the first algorithm (“region identification configuration”) obtained in step S103 and / or the second algorithm (“edge localization configuration”) obtained in step S104 are iteratively optimized.

[0108] like Figure 11As shown, after step S1107, step S1109 is executed, wherein, based on the value of the specified image feature of the second region, the first algorithm is reselected or the value of the specified parameter of the first algorithm is adjusted. After step S1108, step S1110 is executed, wherein, based on the value of the specified edge feature related to the editing operation, the second algorithm is reselected or the value of the specified parameter of the second algorithm is adjusted.

[0109] For example, if a user deletes the measurement marker of a certain noisy area, the method of this embodiment can automatically adjust the corresponding parameters of the first algorithm to enhance the suppression capability of this type of noise; if a user fine-tunes the positioning position of a certain edge, the method of this embodiment can recalculate the gray-scale gradient features of the edge and automatically adjust the threshold of the second algorithm.

[0110] Then, the optimized first and second algorithms can be reapplied to the current image to obtain new measurement results. For example... Figure 11 As shown, after step S1109 and / or S1110, the process returns to step S105.

[0111] Users can edit the new automatic measurement results again, thus repeating steps S1106 to S1110 accordingly, until the measurement results meet the accuracy requirements. If similar images need to be measured subsequently, the optimized configuration can be updated to the original "measurement recipe" to ensure continuous improvement in batch measurement accuracy.

[0112] Based on the same principles and configuration process, the methods of this disclosure can be applied to more complex measurement combinations and broader feature extraction, and are not limited to the measurement types or levels in the examples above.

[0113] To facilitate understanding of the technical solutions of the embodiments of this disclosure, some specific examples of automated measurements performed by the methods of the embodiments of this disclosure are given below. It should be understood that the various specific values ​​and descriptions given in the following examples are merely exemplary and for reference.

[0114] Example 1: Automated Measurement of SEM Images of Gate CDs of Logic Devices During the development phase of 14nm logic device technology, a semiconductor equipment manufacturer needs to measure the gate width (CD) of the gate structure in SEM images. The images contain multiple gate arrays, have slight noise, and some gate edges are slightly blurred due to process or measurement reasons.

[0115] First, the user opens the SEM image to be analyzed. The method of this embodiment reads the scale information (1 pixel = 0.2 nm) of the image and completes pixel-physical size calibration.

[0116] Then, the user can drag and select a region containing a complete gate structure (this region represents typical characteristics, has no obvious noise interference, and has clear gate features) as the region of interest.

[0117] Then, within the selected area of ​​interest, the user clicks the left edge of one of the gates (the abrupt change point from dark to bright grayscale) with the left mouse button, and then clicks the right edge of the gate (the abrupt change point from bright to dark grayscale). The method of this embodiment records the manually measured CD value as 28.5nm and saves the coordinates of the start and end points and the corresponding grayscale features.

[0118] Then, the method of this embodiment automatically analyzes and obtains the specified image features of the region of interest: the gate region is a bright area, the substrate is a dark area, the grayscale contrast is approximately 80 (0-255 grayscale range), and the noise standard deviation is approximately 5. The method of this embodiment selects an "adaptive threshold segmentation algorithm" (suitable for regions with obvious light and dark contrast) from the algorithm library, automatically calculates the optimal threshold window size as 15×15 pixels (determined based on noise level, balancing noise suppression and edge preservation), and generates a region recognition configuration.

[0119] Then, the method of this embodiment automatically analyzes and obtains the specified edge features related to the start and end points measured manually by the user: the peak grayscale gradient of the left edge is 60 (grayscale change rate), and the edge width is about 3 pixels (slight blur); the peak grayscale gradient of the right edge is 55, and the edge width is about 3 pixels. The method of this embodiment selects the "improved Canny operator" (which has better positioning accuracy for blurred edges than traditional operators) from the algorithm library, automatically sets the lower threshold of gradient to 30 (higher than the noise gradient to avoid false detection), the upper threshold to 60 (matching the peak gradient of the left edge), and the edge smoothing coefficient to 2 (reducing the interference of noise on edge positioning), and generates the edge positioning configuration.

[0120] Then, the method of this embodiment applies the above-described region identification configuration and edge positioning configuration to automatically identify all gate regions (a total of 10 gates) that match the features of the region of interest in the entire SEM image, and accurately locates the left and right edges of each gate to calculate the CD value. The output results show that the average CD value of the 10 gates is 28.4nm, the maximum value is 29.9nm, and the minimum value is 27.0nm. At the same time, the edge position of each gate is marked with a green line on the image, and the CD value is marked with blue numbers.

[0121] Then, when the user viewed the results, they found that one of the gates had a 0.5nm deviation in the right edge of the system positioning due to blurred edges (the CD value showed 29.0nm, but it should actually be 28.5nm). The user deleted the incorrect measurement mark of the gate by right-clicking the mouse, and then manually clicked the correct right edge of the gate to complete the editing of the automatic measurement results.

[0122] Then, the method of this embodiment records the user's editing operations, automatically analyzes the grayscale gradient features of the blurred edge (gradient peak value is 45, edge width is 5 pixels), automatically adjusts the gradient lower limit threshold of the improved Canny operator to 25, adjusts the smoothing coefficient to 3, and optimizes the edge positioning configuration. The optimized region recognition configuration and edge positioning configuration are reapplied to the current image, and the CD values ​​of 10 gates are recalculated. The results show that the CD value of the blurred gate is corrected to 28.5nm, the standard deviation of the CD of all gates is reduced, and the measurement accuracy is significantly improved. After the user confirms that the results meet the requirements, the optimized configuration is saved as "14nm logic device gate CD measurement recipe" for subsequent batch measurement of SEM images under the same process conditions.

[0123] When the semiconductor equipment manufacturer needs to perform CD measurements on 50 SEM images of the gates of 14nm logic devices in the same batch, the user can directly import all images and load the saved "14nm logic device gate CD measurement recipe". The method of this embodiment automatically applies the configuration in the recipe, completing the batch measurement of all 50 images within 1 minute, acquiring CD data for 500 gates, and generating a statistical report (including the mean and standard deviation of gate CD in a single image, as well as the overall CD distribution histogram and process stability analysis curve of the 50 images). Compared with traditional manual measurement (15 minutes per image, 12.5 hours for 50 images), the efficiency is improved by 750 times, and all measurement data are generated based on a unified standard, with a consistency error of less than 0.1nm for repeated experimental data, which is far better than the error of more than 1.0nm in manual measurement.

[0124] Example 2: TEM image thickness measurement of multilayer structures of storage devices (such as NAND Flash) A memory chip company needs to perform TEM image measurements on the thickness of the oxide layer, nitride layer, and polysilicon layer of a NAND Flash device. The images show small contrast differences in the multilayer structure and blurring of the interlayer interfaces.

[0125] First, after opening the TEM image to be analyzed, the user selects the region of interest containing the complete multilayer structure, and manually clicks the upper and lower interfaces of the oxide layer (to complete a thickness measurement).

[0126] Then, the method of this embodiment automatically analyzes the interlayer contrast characteristics (the gray value of the oxide layer is about 120, the gray value of the nitride layer is about 80, and the gray value of the polysilicon layer is about 150) and the interface blurring degree (the interface width is about 2 pixels) of the region of interest, and accordingly selects the "multi-threshold layering algorithm" as the region matching algorithm and the "gray-level jump point positioning algorithm" as the edge recognition algorithm from the algorithm library.

[0127] Then, the method of this embodiment utilizes the selected region matching algorithm and edge recognition algorithm to quickly identify all multi-layer structure regions in the image, accurately measure the thickness of each layer (measurement error <0.5nm), support interactive adjustment of the blurred interface, and finally save the recipe for subsequent batch TEM image measurements.

[0128] Example 3: Measurement of epitaxial layer defect density in STEM images of power devices (such as IGBTs) A power device company needs to perform STEM image measurement on the defect (such as void) density of the epitaxial layer of IGBTs. The defect size in the image is small (50-200nm) and is easily confused with noise.

[0129] First, after opening the STEM image to be analyzed, the user selects the region of interest containing typical void defects, which includes the structure of interest, such as the epitaxial layer. Then, the user manually clicks one or more points within the defect to mark the void defect, and manually clicks one or more points outside the defect to mark the non-defect part.

[0130] Then, the method of this embodiment automatically analyzes the specified image features of the void defect to obtain a gray value that is 20-30 lower than the surrounding epitaxial layer, an irregular circle shape, and an area of ​​about 10,000 pixels². Based on this, the "morphological feature matching algorithm" and the "adaptive noise filtering + contour extraction algorithm" are selected from the algorithm library as the region matching algorithm and the "defect edge recognition algorithm" as the defect edge recognition algorithm.

[0131] Then, the method of this embodiment uses the selected algorithm to traverse the entire STEM image, identify each cavity defect, count the number of defects, calculate the defect density (defects / mm²), and mark the location and size of each defect.

[0132] After users interactively edit and delete misidentified noise points, they can save the recipe for batch statistical analysis. Compared to manual statistical analysis (which takes about 30 minutes per image), efficiency is improved by about 20 times, and defect identification accuracy is increased from about 85% to about 98%.

[0133] Therefore, in at least some embodiments of this disclosure, the familiar manual measurement process of semiconductor process engineers can be used as a foundation. The engineer's normal manual measurement actions serve as anchor points for correct results, converting these actions (such as selecting areas of interest, clicking measurement start and end points, etc.) into reusable automatic recognition configurations. Algorithms automatically match analysis strategies and parameters adapted to user operations. Thus, after an engineer completes a manual measurement, all information in the image matching the user's input features can be automatically analyzed and extracted, achieving accurate automated measurement. The generated strategies and parameters can be reused as automated analysis recipes for subsequent batch automated analysis in the same scenario. Furthermore, based on the same approach, interactive result optimization can be used to adapt to changes in the process scenario based on minimal user feedback.

[0134] Therefore, it enables rapid and accurate measurement of single or batch electron micrographs without the need for complex preliminary data annotation or professional image analysis knowledge, thus lowering the barrier to entry and improving measurement efficiency and data consistency.

[0135] Specifically, compared with the prior art, the method of the present disclosure can bring the following significant technical effects: 1. Low barrier to entry and suited to engineers' habits: This method can be designed based on the manual measurement process that engineers are familiar with. Engineers only need to perform routine measurement operations. They do not need to master professional image analysis algorithms or deep learning knowledge, nor do they need to make complex parameter adjustments. This greatly reduces the barrier to entry and can be quickly promoted and applied in process engineering teams. 2. No prior labeling or training required, fast response speed: This method does not require the collection of massive amounts of labeled data in advance, nor does it require special model training. The measurement configuration can be automatically generated with just one manual measurement operation by the engineer. The time from operation to output result is only a few seconds to tens of seconds, which can quickly respond to new processes or measurement requirements (such as CD measurement of new device structures), and solve the problem of time lag in deep learning methods. 3. High measurement efficiency and comprehensive data coverage: After an engineer completes a manual operation, this method can automatically analyze all structures in the image that meet the characteristics, realizing "one operation, batch measurement", which improves efficiency by tens to hundreds of times compared with manual measurement; at the same time, the efficient measurement capability supports comprehensive analysis of massive images, avoiding the data limitations of manual sampling analysis, and can more comprehensively reflect the process status. 4. High accuracy and strong data consistency: This method automatically analyzes the image features corresponding to user operations to generate suitable algorithms and parameter configurations, avoiding subjective differences in manual measurement. At the same time, the interactive result optimization function can further correct individual abnormal results to ensure measurement accuracy. In addition, reusing the "measurement recipe" can achieve unified standard measurement of batch images, and the data consistency is significantly better than that of manual measurement. 5. High versatility and adaptability to complex scenarios: This method can automatically adapt to different imaging methods (SEM / TEM / STEM), different device types (logic / storage / imaging / power devices), and different image qualities (noise level, contrast difference) through the dynamic adaptation logic of "user operation-feature analysis-algorithm matching", without the need to develop separate algorithms for each scenario; at the same time, the interactive optimization function can deal with complex images under abnormal process conditions, and its generalization ability far exceeds that of traditional edge detection and deep learning methods. 6. Reusable configuration reduces repetitive work: The generated "measurement recipe" can be saved and reused. Subsequent batch image measurements under the same process conditions do not require repeated manual operations, further reducing repetitive work for engineers and improving work efficiency. 7. No specific data required, eliminating intellectual property concerns: This method automates engineers' manual workflows directly, without relying on specific images or detailed structural or process information for configuration. The data involved is controllable, eliminating intellectual property protection risks in the applied semiconductor scenarios.

[0136] Figure 12 A schematic diagram of a computing device according to at least one embodiment of the present disclosure is shown, which can be used to implement the above-described automated measurement method for electron micrographs obtained in semiconductor processes.

[0137] See Figure 12 The computing device 1200 includes a memory 1210 and a processor 1220.

[0138] Processor 1220 may be a multi-core processor or may contain multiple processors. In some embodiments, processor 1220 may include a general-purpose main processor and one or more special coprocessors, such as a graphics processing unit (GPU), a digital signal processor (DSP), etc. In some embodiments, processor 1220 may be implemented using custom circuitry, such as an application-specific integrated circuit (ASIC) or a field-programmable gate array (FPGA).

[0139] Memory 1210 may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. ROM may store static data or instructions required by processor 1220 or other modules of the computer. Permanent storage devices may be read-write storage devices. Permanent storage devices may be non-volatile storage devices that retain stored instructions and data even when the computer is powered off. In some embodiments, permanent storage devices use mass storage devices (e.g., magnetic or optical disks, flash memory) as permanent storage devices. In other embodiments, permanent storage devices may be removable storage devices (e.g., floppy disks, optical drives). System memory may be a read-write storage device or a volatile read-write storage device, such as dynamic random access memory. System memory may store some or all of the instructions and data required by the processor during operation. Furthermore, memory 1210 may include any combination of computer-readable storage media, including various types of semiconductor memory chips (DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and disks and / or optical disks may also be used. In some embodiments, memory 1210 may include a removable storage device that is readable and / or writable, such as a laser disc (CD), a read-only digital multifunction optical disc (e.g., DVD-ROM, dual-layer DVD-ROM), a read-only Blu-ray disc, an ultra-high density optical disc, a flash memory card (e.g., SD card, mini SD card, Micro-SD card, etc.), a magnetic floppy disk, etc. Computer-readable storage media do not contain carrier waves or transient electronic signals transmitted wirelessly or via wired connections.

[0140] The memory 1210 stores executable code, which, when processed by the processor 1220, enables the processor 1220 to execute the automated measurement method for electron micrographs obtained in semiconductor processes, as described above.

[0141] The automated measurement method for electron micrographs obtained in semiconductor processes according to this disclosure has been described in detail above with reference to the accompanying drawings.

[0142] Furthermore, the method according to this disclosure can also be implemented as a computer program or computer program product, which includes computer program code instructions for performing the steps defined in the above-described method of this disclosure.

[0143] Alternatively, this disclosure may be implemented as a non-transitory machine-readable storage medium (or computer-readable storage medium, or machine-readable storage medium) storing executable code (or computer program, or computer instruction code) that, when executed by a processor of an electronic device (or computing device, server, etc.), causes the processor to perform the steps of the method described above according to this disclosure.

[0144] Those skilled in the art will also understand that the various exemplary logic blocks, modules, circuits, and algorithm steps described in connection with the disclosure herein can be implemented as electronic hardware, computer software, or a combination of both.

[0145] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems and methods according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0146] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. An automated measurement method for electron micrographs obtained in semiconductor processes, comprising: The electron micrograph to be analyzed is obtained as the first image; Obtain the area selected by the user in the first image as the first region, and obtain information about the manual measurement operations performed by the user in the first region; Obtain the value of a specified image feature of the first region, and based on the specified image feature value, select one from a plurality of preset region matching algorithms as the first algorithm; Obtain the value of a specified edge feature related to the manual measurement operation information, and based on the specified edge feature value, select one from a set of preset edge recognition algorithms as a second algorithm; as well as At least one second region is identified from a first image using a first algorithm, an edge to be measured is identified in the second region using a second algorithm, and the edge is used for automatic measurement.

2. The method according to claim 1, further comprising, before identifying at least one second region from the first image using the first algorithm: Based on the specified image feature values, determine the values ​​of the specified parameters of the first algorithm; And / or, Based on the specified edge feature value, the value of the specified parameter of the second algorithm is determined.

3. The method according to claim 1, further comprising, before obtaining the value of the specified image feature of the first region: Obtain the geometric parameters to be measured selected by the user; The specified image features and / or the specified edge features are determined at least based on the geometric parameters; and / or the region matching algorithm and / or the edge recognition algorithm are determined at least based on the geometric parameters.

4. The method according to claim 1, further comprising, after performing automatic measurement using the edge: Obtain information about the user's editing operations on the results of the automatic measurement; Obtain the values ​​of the specified image features of the second region involved in the editing operation and / or the values ​​of the specified edge features related to the editing operation; as well as Based on the value of the specified image feature of the second region, the first algorithm is reselected or the value of the specified parameter of the first algorithm is adjusted, and / or based on the value of the specified edge feature related to the editing operation, the second algorithm is reselected or the value of the specified parameter of the second algorithm is adjusted.

5. The method according to claim 1, wherein, Based on the specified image feature values, one algorithm is selected from a set of preset region matching algorithms as the first algorithm, including: Based on the specified image feature values ​​and a preset first selection condition, one of a plurality of preset region matching algorithms is selected as the first algorithm; and / or, Obtaining the value of a specified edge feature related to the manual measurement operation information includes: When the manual measurement operation information includes information about the measurement point input by the user, the value of the specified edge feature within a predetermined range around the measurement point is obtained; and / or, Based on the specified edge feature value, select one from a set of preset edge recognition algorithms as the second algorithm, including: Based on the specified edge feature value, and based on the preset second selection condition, one of the preset edge recognition algorithms is selected as the second algorithm.

6. The method according to claim 1, wherein, The specified image feature includes at least one of the following: Noise level, contrast distribution, structure density, edge sharpness, edge strength, information entropy changes between different sub-regions, interlayer contrast characteristics, interlayer interface blurring degree, morphological features, and image feature encoding obtained from a specified deep learning model; and / or, The region matching algorithm includes at least one of the following: Adaptive threshold segmentation algorithm, region growing algorithm, texture matching algorithm, multi-threshold layering algorithm, morphological feature matching algorithm, and encoding vector spacing algorithm.

7. The method according to claim 1, wherein, The specified edge feature includes at least one of the following: The grayscale gradient of the edge, the edge width, the degree of blurring of the edge, the grayscale difference of the edge, the edge shape, and the area of ​​the closed region of the edge; and / or, The edge recognition algorithm includes at least one of the following: Improved Canny operator algorithm, adaptive Sobel operator algorithm, gray-level gradient peak localization algorithm, gray-level transition point localization algorithm, contour extraction algorithm.

8. A computing device, comprising: processor; as well as A memory having executable code stored thereon, which, when executed by the processor, causes the processor to perform the method as described in any one of claims 1 to 7.

9. A computer program product comprising executable code that, when executed by a processor of an electronic device, causes the processor to perform the method as claimed in any one of claims 1 to 7.

10. A non-transitory machine-readable storage medium having executable code stored thereon, which, when executed by a processor of an electronic device, causes the processor to perform the method as claimed in any one of claims 1 to 7.

Images