Method and system for obtaining measurements of on-wafer semiconductor structures

Abstract

The invention relates to a method (10) for obtaining a measurement of a semiconductor structure (30) on a wafer (72), comprising: obtaining a volumetric imaging data set (22) of the wafer (72), comprising a plurality of 2D profile images (24, 26); obtaining a contour (32) of the semiconductor structure (30) in the 2D profile image (24, 26); one or more measurement specifications (28) indicating features in the first 2D cross-sectional image (24) with respect to a contour (32) of the semiconductor structure (30); propagating the one or more measurement specifications (28) indicated in the first 2D profile image (24) to a further 2D profile image (26); and obtaining a measurement of the semiconductor structure (30) by evaluating one or more measurement specifications (28) in the first 2D profile image (24) and the further 2D profile image (26).

Description

[0001] Related applications

[0002] This application claims priority to German Patent Application No. 10 2023 109947.7, filed on April 19, 2023, which is incorporated herein by reference in its entirety. Technical Field

[0003] This invention relates to a method and system for obtaining measurements of semiconductor structures on a wafer. More particularly, this invention relates to a method for obtaining a volumetric imaging dataset of a wafer containing semiconductor structures, and to a corresponding computer-readable medium, computer program product, and system. This method, computer-readable medium, computer program product, and system can be used for quantitative metrology, defect detection, defect inspection, critical dimension inspection, and process window verification. Background Technology

[0004] The fabrication of wafers containing semiconductor structures is a complex sequence of deposition and removal of physical materials at nanometer-scale resolution. Therefore, extracting measurements such as the inter- and intra-structural distances of the fabricated 3D structures is crucial for monitoring the process.

[0005] Wafers made of thin silicon wafers serve as substrates for microelectronic devices containing semiconductor structures embedded within and on the wafer. Semiconductor structures are constructed layer by layer using repetitive processing steps involving repeated chemical, mechanical, thermal, and optical processes. The size, shape, and layout of semiconductor structures and patterns are influenced by a variety of factors. One of the most critical steps is the photolithography process.

[0006] Photolithography is a process used to create patterns on a substrate. The pattern to be printed on the substrate surface is generated using computer-aided design (CAD). Based on the design, a photomask is generated for each layer, containing a magnified image of the computer-generated pattern to be etched into the substrate. The photomask can be further adjusted, for example, using optical proximity correction techniques. During the printing process, the illumination image projected from the photomask is focused onto a thin film of photoresist formed on the substrate. Semiconductor chips that power mobile phones or tablet computers contain, for example, about 80 to 120 patterned layers.

[0007] As the semiconductor industry's integration density continues to increase, photolithography masks must image increasingly smaller structures onto wafers. The aspect ratios and number of layers in integrated circuits are constantly increasing, and structures are increasingly evolving towards a third (vertical) dimension. Currently, the height of memory stacks exceeds tens of micrometers. In contrast, feature sizes are becoming smaller. The smallest feature size, or critical size, is below 10 nm, such as 7 nm or 5 nm, and will approach below 3 nm in the near future. As the complexity and size of semiconductor structures grow to the 3D dimension, the lateral dimensions of integrated semiconductor structures become smaller. Generating small structure sizes imaged onto wafers requires photolithography masks or templates for nanoimprint lithography with smaller structural or patterned elements.

[0008] Due to the minute size of the patterned elements in photolithographic masks or templates and the complexity of semiconductor structure fabrication processes, it is impossible to eliminate errors during wafer fabrication. Therefore, wafer inspection, examination, and metrology play a crucial role in monitoring defects during semiconductor process control. Traditionally, measurements of 2D semiconductor structures were performed manually by experts. However, because the semiconductor structure on a wafer is 3D, 3D measurements are necessary for more precise monitoring of the production process.

[0009] To obtain 3D imaging datasets, such as imaging datasets containing multiple 2D cross-sections of a wafer, destructive imaging techniques (using focused ion beam scanning electron microscopy (FIB-SEM)) can be used. The resulting imaging datasets are very dense (e.g., containing thousands of images), thus posing challenges in terms of scalability, robustness, and reproducibility of measurements.

[0010] For example, WO 2021 / 083581 A1 discloses a method for measuring shape deviations of 3D HAR structures in FIB-SEM tomography. This method involves generating a template representing cross-sectional image features of the HAR structure of interest and detecting instances of that template in 2D cross-sectional images of an imaging dataset. For example, the detected instances are assigned to different 3D HAR structures based on the distance between the center coordinates of instances in adjacent 2D cross-sectional images. Based on the detected instances assigned to the same 3D HAR structure, the surface of the 3D HAR structure is reconstructed, and parameters characterizing the geometry of the entire semiconductor structure are obtained.

[0011] However, this method requires significant effort from the user to generate templates of the structures of interest, such as through manual annotation. Furthermore, it relies on finding the corresponding structure in different 2D cross-sectional images, which is prone to errors because the structure may move or even disappear in different 2D cross-sectional images. Finally, the method requires generating 3D surfaces from multiple boundary coordinates, which is computationally very expensive.

[0012] US 2020 / 0173772 A1 discloses a method for 3D reconstruction of HAR features generated from wedge cuts using harmonics and fast Fourier transform. The computational cost of this method is high due to the generation of 3D surfaces.

[0013] WO 2022 / 223229 A1 discloses a method for measuring semiconductor structures by detecting contours in cross-sectional images and analyzing parameters of these contours, namely displacement from an ideal location or deviations in radius, diameter, area, or shape. However, these measurements cannot be flexibly specified or adapted to the needs of the user or application.

[0014] Therefore, the object of this invention is to provide a rapid calculation method for measuring 3D semiconductor structures. Another object of this invention is to reduce the workload for users of this method. A further object is to allow users to easily, quickly, and flexibly define the required measurements and to accurately calculate the defined measurements. Another object of this invention is to improve the accuracy of the method used for measuring 3D semiconductor structures. Another object of this invention is to provide a high-accuracy wafer defect detection method. Another object of this invention is to provide a method for inspecting critical dimensions of semiconductor structures on a wafer. Another object of this invention is to provide a method for process window verification. Another object of this invention is to increase yield during wafer quality control or quality assurance processes. Another object of this invention is to minimize the operation time of quality control.

[0015] The stated object is achieved by means of the invention specified in the independent claims. Advantageous embodiments and further developments of the invention are described in detail in the dependent claims. Summary of the Invention

[0016] This invention relates to methods, computer-readable media, computer program products, and systems for obtaining measurements of semiconductor structures on a wafer.

[0017] The first embodiment of the present invention relates to a method for obtaining measurements of a semiconductor structure on a wafer, particularly a computer-implemented method. This method includes: a) obtaining a volumetric imaging dataset of the wafer, comprising a plurality of 2D cross-sectional images; b) obtaining the outline of a semiconductor structure in the 2D cross-sectional images of the imaging dataset; c) identifying one or more measurement specifications of features relating to the outline of the semiconductor structure in a first 2D cross-sectional image of the imaging dataset; d) propagating the one or more measurement specifications identified in the first 2D cross-sectional image to another 2D cross-sectional image of the imaging dataset; and e) obtaining measurements of the semiconductor structure by evaluating the first 2D cross-sectional image and the other 2D cross-sectional images of the imaging dataset.

[0018] 3D measurements are obtained from multiple 2D measurements in different 2D cross-sectional images of the imaging dataset, rather than from 3D reconstruction of the semiconductor structure. This significantly reduces computation time because computationally intensive 3D reconstruction of the semiconductor structure is not required. Furthermore, the user workload is reduced because the measurement specifications are specified only in a single 2D cross-sectional image and automatically propagate to other 2D cross-sectional images. Therefore, there is no need for time-consuming template generation of the semiconductor structure from different 2D cross-sectional images. Since the measurement specifications automatically propagate to other 2D cross-sectional images, there is no need to use error-prone trial-and-error methods to find instances of the same semiconductor structure in different 2D cross-sectional images.

[0019] Measurements of the acquired semiconductor structures can be used in a variety of different ways. For example, the tilt or twist of a three-dimensional feature, such as a HAR structure like a memory hole or pillar, can be measured by interpolating the centroid of the HAR structure in different 2D cross-sections with a straight line and measuring the angle between the interpolation line and the vertical line. In another instance, the ellipticity, mean radius, tilt, angle, or curvature of the axis of the HAR structure (e.g., the axis of a pillar) can be measured by comparing corresponding parameters in 2D cross-sectional images. In yet another instance, the minimum distance between two 3DHAR structures can be calculated by measuring the 2D distance between two features in different 2D cross-sectional images of an imaging dataset and selecting the minimum distance. With a specified baseline value (e.g., a critical size value), the distance between features or the length of a feature can be measured and compared to the baseline value. Defects can be detected based on the resulting measurements. For process window verification, wafer slices can be produced using lithography processes with different manufacturing parameters. The wafer cross-section can then be examined by comparing measurements or detecting defects, and the optimal manufacturing parameters can be selected.

[0020] The term "defect" refers to a local deviation from the prior definition of a semiconductor structure. For example, a defect in a semiconductor structure may cause related semiconductor devices to malfunction. For instance, based on detected defects, the photolithography process may be improved, or the wafer may be repaired. For example, a detected bridging defect indicates insufficient etching, so the etching amount should be increased; a detected broken line indicates over-etching, so the etching amount should be reduced; a persistent defect indicates a defective photomask, so the photomask must be inspected; a detected missing structure suggests suboptimal material deposition, so the material deposition should be modified.

[0021] According to an example of a first embodiment of the present invention, an imaging dataset is obtained by focused ion beam scanning electron microscopy.

[0022] According to an example of a first embodiment of the present invention, the chip is a memory chip, such as comprising RAM, DRAM, or NAND structures. Due to the simple structure of such memory chips, measurement specifications can be effectively indicated in a first 2D cross-sectional image and propagated to other 2D cross-sectional images. Therefore, computation time and user workload are reduced.

[0023] According to one embodiment of the present invention, the measurement specifications are derived from a group including feature locations, feature distances, and feature dimensions. Features of the semiconductor structure's contour may include points, lines, or curves defined relative to the contours or contour segments of one or more semiconductor structures in the first 2D cross-sectional image. For example, features of the semiconductor structure's contour may be derived from a group including points on the contour or contour segment, regions defined by the contour or contour segment, segments of the contour, and centroids of the contour or contour segment. Therefore, various dimensions of the semiconductor structure can be measured by specifying measurement specifications relating to features defined relative to the contour or contour segment in the first 2D cross-sectional image. This increases the flexibility of the method and reduces the user's workload. In one example, the measurement specifications are used to indicate critical dimensions of the semiconductor structure to ensure printability on the wafer.

[0024] In a preferred embodiment, the measurement specifications in the first 2D profile image are constituted by feature distances, and the features of the contour are constituted by points relative to the contour, specifically contour points. This means that the measurement specifications in the first 2D profile image only indicate the distances between points relative to the contour, specifically the distances between contour points. Thus, the measurement specifications can be specified in a particularly simple, accurate, and flexible manner, simplifying the propagation of the measurement specifications and requiring less computation time.

[0025] In a preferred embodiment, in step d), one or more measurement specifications indicated in the first 2D profile image are propagated to another 2D profile image of the imaging dataset by associating features of the semiconductor structure contour in the first 2D profile image with corresponding features in the semiconductor structure contour in another 2D profile image, wherein features of the semiconductor structure contour in the first 2D profile image define one or more measurement specifications.

[0026] Instead of recalculating the measurement specifications in each additional 2D profile image, the contour features of the semiconductor structure in the first 2D profile image (e.g., contour points that define one or more measurement specifications) are directly propagated to the additional 2D profile images. This simplifies and significantly accelerates the calculation of measurement specifications.

[0027] According to an example of the first embodiment of the present invention, in step d), one or more measurement specifications indicated in the first 2D profile image are propagated to another 2D profile image of the imaging dataset by associating the contour of the semiconductor structure in the first 2D profile image with the corresponding contour of the same semiconductor structure in another 2D profile image, and by associating the features of the contour of the semiconductor structure in the first 2D profile image with the corresponding features in the associated contour of the semiconductor structure in another 2D profile image.

[0028] In this way, measurement specifications can be quickly and automatically propagated to other 2D profile images, thereby reducing computation time and user workload.

[0029] Furthermore, the accuracy of measurement propagation is improved because measurement specifications or features are not directly propagated to another 2D cross-sectional image, which can be error-prone. Instead, the contour of the semiconductor structure is propagated to another 2D cross-sectional image. Feature points defined relative to the contour can then be calculated with sub-pixel accuracy based on the propagated contour in the other 2D cross-sectional image. For example, the measurement specification includes measuring the distance between two contour points of the semiconductor structure in the 2D cross-sectional image. Instead of directly propagating each contour point indicated in the first 2D cross-sectional image to the other 2D cross-sectional image, e.g., by pattern matching, the position of each contour point relative to the contour (e.g., the direction from the centroid of the contour) is calculated. The relevant contour is then obtained in the other 2D cross-sectional image. Based on these relevant contours, contour points are calculated relative to the relevant contours (e.g., contour points located in the same direction from the centroid of the relevant contour). Thus, two contour points can be calculated with sub-pixel accuracy, and the distance between them can be measured with greater precision.

[0030] Furthermore, it enables contour association on 2D profile images with greater accuracy (e.g., by using a tracking algorithm that considers all 2D profile images simultaneously), rather than relying on trial and error to associate contours between every two individual images.

[0031] According to one aspect of this example of the first embodiment of the present invention, by applying a contour extraction method to a first 2D cross-sectional image and another 2D cross-sectional image, the contour of the semiconductor structure in the first 2D cross-sectional image and the other 2D cross-sectional image is obtained. Thus, the contour of the semiconductor structure in the first 2D cross-sectional image and the other 2D cross-sectional image can be obtained automatically and with high accuracy, for example, by using machine learning methods.

[0032] The outline of the semiconductor structure in the first 2D cross-sectional image and other 2D cross-sectional images can be obtained in a variety of ways.

[0033] According to one aspect of this embodiment of the invention, the contours of semiconductor structures in the first 2D cross-sectional image and the other 2D cross-sectional image are obtained by applying an object detection or image segmentation algorithm to the first 2D cross-sectional image and the other 2D cross-sectional image. In the case of the segmentation algorithm, the contours of the semiconductor structures in the first 2D cross-sectional image and the other 2D cross-sectional image are obtained by calculating the segment boundaries obtained using the segmentation algorithm. In the case of the object detection algorithm, the contours of the semiconductor structures in the first 2D cross-sectional image and the other 2D cross-sectional image can be represented by bounding boxes obtained by the object detection algorithm, or they can be obtained from the object detection results by a contour extraction method, for example, applied to the obtained bounding boxes. Thus, the contours of semiconductor structures in the first 2D cross-sectional image and the other 2D cross-sectional image can be obtained automatically and with high accuracy, for example, using machine learning methods. For example, the machine learning method can be trained on annotated 2D cross-sectional images or marked design files or some standard object detection or segmentation database available online.

[0034] According to one aspect of this example of the first embodiment of the present invention, the outline of the semiconductor structure in the first 2D cross-sectional image (24) and the other 2D cross-sectional image is obtained by applying an instance segmentation algorithm to the first 2D cross-sectional image (24) and the other 2D cross-sectional image. The instance segmentation algorithm provides the area of ​​the semiconductor structure and also assigns instance numbers to the detected image segments (e.g., each column cross-section is assigned a different instance number in the first 2D cross-sectional image and the other 2D cross-sectional image). In this way, pixels belonging to the same type of semiconductor structure but belonging to different instances can be distinguished. This simplifies the propagation of segmented instances to other 2D cross-sectional images because the segmentation already corresponds to different semiconductor structure instances. Therefore, the accuracy of this method is improved.

[0035] The outline of the semiconductor structure in the first 2D cross-sectional image and other 2D cross-sectional images can be represented in different ways, such as curves of area or boundaries.

[0036] In one aspect of this example of the first embodiment of the present invention, the contour of the semiconductor structure in the first 2D cross-sectional image and another 2D cross-sectional image is represented by contour points, for example, by edge pixels of the semiconductor structure in the 2D cross-sectional image or by subsampled contours. This simplifies the representation of the contour. This allows for rapid calculation of relevant contours and features in another 2D cross-sectional image.

[0037] In one aspect of this example of the first embodiment of the present invention, the outline of a semiconductor structure in a first 2D cross-sectional image and a further 2D cross-sectional image is represented by a bounding box. A bounding box is a particularly simple representation of a contour and is commonly used to represent the results of algorithms, such as object detection algorithms. Bounding boxes allow for the rapid and simple computation of relevant contours and features in the further 2D cross-sectional images, for example, through tracking algorithms.

[0038] The outlines of semiconductor structures in the first 2D cross-sectional image and the subsequent 2D cross-sectional images can be correlated in different ways.

[0039] According to one aspect of this embodiment of the invention, associating the contour of a semiconductor structure in a first 2D cross-sectional image with the corresponding contour of the same semiconductor structure in another 2D cross-sectional image includes calculating the matching of contour points of the semiconductor structure in the first 2D cross-sectional image and contour points of the semiconductor structure in the other 2D cross-sectional image, for example, using the Hungarian algorithm. This allows for the calculation of contour associations in the first 2D cross-sectional image and the other 2D cross-sectional image with high accuracy. Furthermore, since only single points are matched between different 2D cross-sectional images instead of entire contours or image segments, computation time can be reduced.

[0040] According to one aspect of this example of the first embodiment of the present invention, the contour of a semiconductor structure in a first 2D cross-sectional image is associated with the corresponding contour of the same semiconductor structure in another 2D cross-sectional image. A tracking algorithm or optical flow algorithm is applied to another 2D cross-sectional image of the imaging dataset to track the contour of the semiconductor structure in the first 2D cross-sectional image. By using tracking or optical flow algorithms to track the contours, the contour association between the first 2D cross-sectional image and the other 2D cross-sectional image is calculated with high accuracy.

[0041] In one example, associating the contour of a semiconductor structure in a first 2D cross-sectional image with the corresponding contour of the same semiconductor structure in another 2D cross-sectional image involves applying a tracking algorithm configured to minimize the number of interruptions in the obtained trajectory, where the trajectory is defined as the path of the semiconductor structure's contour through consecutive 2D cross-sectional images in the imaging dataset. By minimizing trajectory interruptions, particularly obtaining long trajectories, it also includes associated contours with lower confidence. These lower-confidence contours are often due to defects and should not be excluded from the trajectory. Therefore, by evaluating one or more measurement specifications, defects can be detected with higher accuracy.

[0042] In one example, an instance segmentation algorithm is used to obtain the outline of the semiconductor structure in a first 2D cross-sectional image and in another 2D cross-sectional image. Then, a tracking algorithm is used to trace each segmented instance from the first 2D cross-sectional image on the other 2D cross-sectional image. This yields particularly accurate results because the tracking has been simplified, as the instance segmentation algorithm already provides information about individual semiconductor structure instances.

[0043] According to one aspect of this example of the first embodiment of the present invention, associating the contour of a semiconductor structure in a first 2D cross-sectional image with the corresponding contour of the same semiconductor structure in another 2D cross-sectional image involves registering the image dataset to a reference imaging dataset with labeled contours. Thus, contour associations in the first 2D cross-sectional image and the other 2D cross-sectional image are indirectly calculated with high precision via the reference imaging dataset, because registration of the entire imaging dataset can be performed with higher precision than association of a single contour due to the additional information contained in the imaging dataset.

[0044] According to an example of the first embodiment of the present invention, the contours of the semiconductor structures in the first 2D cross-sectional image and the other 2D cross-sectional images are obtained by calculating the 3D segmentation of the semiconductor structure in the imaging dataset and calculating the contours of the segmented semiconductor structure in the first 2D cross-sectional image and the other 2D cross-sectional image from the 3D segmentation. The accuracy of the method is improved by using a 3D segmentation algorithm to obtain the contours for two reasons: first, the 3D segmentation algorithm relies on information from the entire imaging dataset to calculate the 3D segmentation; and second, essentially, the 3D segmentation provides the contour association between different 2D cross-sectional images, and it does not need to be calculated separately. Therefore, in step d), the contour of the semiconductor structure in the first 2D cross-sectional image is associated with the corresponding contour of the same 3D segmentation of the same semiconductor structure in another 2D cross-sectional image, and the features of the contour of the semiconductor structure in the first 2D cross-sectional image are associated with the corresponding features in the relevant contour of the semiconductor structure in the other 2D cross-sectional image, thereby propagating one or more measurement specifications indicated in the first 2D cross-sectional image to another 2D cross-sectional image of the imaging dataset, wherein the features of the contour of the semiconductor structure in the first 2D cross-sectional image define one or more measurement specifications.

[0045] According to an example of a first embodiment of the present invention, at least one measurement specification includes: a contour point of a semiconductor structure profile, wherein the contour point is defined by a specific point relative to the profile, particularly the centroid; and a direction vector indicating the direction of the contour point with respect to the specific point in a first 2D cross-sectional image, wherein associating the contour point in the first 2D cross-sectional image with a corresponding contour point of a related contour in another 2D cross-sectional image includes calculating the intersection of the related contour and the direction vector, which begins at the specific point of the related contour in the other 2D cross-sectional image. Presenting the contour point by means of a specific point defined relative to the profile and a direction vector in the first 2D cross-sectional image allows for more accurate propagation of the contour point to another 2D cross-sectional image because it is independent of the shape of the related contour, the related contour point lies in the same direction as the corresponding specific point in all 2D cross-sectional images, and can be calculated with sub-pixel accuracy.

[0046] According to one embodiment of the invention, at least one measurement specification includes calculating the centroid of the contour of a semiconductor structure in one or more 2D cross-sectional images. This centroid is obtained by analyzing the intensity profile of a 1D cross-section along a region (e.g., a bounding box) surrounding the contour in the one or more 2D cross-sectional images. Thus, the centroid can be calculated with higher accuracy compared to deriving it from the contour of the semiconductor structure in the 2D cross-sectional images. For example, the centroid can be obtained from a point of symmetry of the intensity profile, where the symmetry point can be calculated as the intersection of the intensity profile and its axis of symmetry. For example, the calculated centroid can be used as a measurement specification or as a specific point defining the location of contour points on the contour, as described above.

[0047] According to an example of a first embodiment of the present invention, in step d), propagating one or more measurement specifications indicated in the first 2D profile image to another 2D profile image of the imaging dataset includes generating a confidence score indicating the reliability of the relevant measurement specifications in the first 2D profile image and the other 2D profile image. The confidence score can, for example, indicate a low probability of contour detection or contour association, such as due to the invisibility of the semiconductor structure. The confidence score can be used, for example, by a contour association algorithm, during defect detection, or by a subsequent algorithm that relies on measurements of the obtained semiconductor structure. Therefore, the accuracy of the method or subsequent algorithm is improved.

[0048] According to one embodiment of the present invention, defects are detected by detecting outliers in measurements obtained from measurement specifications in a first 2D profile image and another 2D profile image of the evaluation image dataset. Alternatively, variations in the obtained measurements can be analyzed to detect defects.

[0049] According to one embodiment of the present invention, the inspection target, particularly the target processing volume (e.g., by querying the user or loading from memory), is obtained, and the number of additional 2D profile images and / or measurement specifications is automatically adapted to meet the inspection target. Thus, the algorithm can be adapted to a specific inspection target, i.e., the requirements of the application. The inspection target may include runtime, processing volume, resource constraints, etc.

[0050] Preferably, one or more measurement specifications in the first 2D profile image are indicated via a user interface. This provides the user with a particularly simple and flexible way to indicate measurement specifications. For example, this method allows for the indication of measurements that would be difficult to describe without a user interface.

[0051] According to one aspect, the user interface is configured to allow the user to specify one or more measurement specifications by selecting one or more features of the outline of a semiconductor structure on a first 2D profile image of the imaging dataset.

[0052] In one instance, the user interface is configured to assist the user by automatically calculating modifications to selected features of the semiconductor structure profile used to define the measurement specifications during the specification specification process.

[0053] In one example, the user interface is configured to assist the user during the selection of one or more features by calculating modifications to the profile of the selected features relative to the semiconductor structure. These features may be modified automatically or after user approval. By calculating the modifications to the selected features, measurement specifications can be specified with greater accuracy (e.g., using subpixel accuracy), thereby improving the accuracy of the method.

[0054] According to one embodiment of the present invention, this method further includes using a user interface configured to allow a user to load measurement specifications from memory or a database and / or store measurement specifications in memory or a database. This reduces the user's workload, as the stored measurement specifications can be applied to further imaging datasets or similar use cases.

[0055] According to an example of a first embodiment of the present invention, the method further includes using a user interface configured to propose measurement specifications to a user, which the user can accept, modify, or reject, wherein recommendations for measurement specifications are generated based on measurement specifications previously specified by the user. Measurement specifications may be generated from previously specified measurement specifications for the same imaging dataset, and / or from previously specified measurement specifications for different imaging datasets or use cases. For example, machine learning methods may be used to obtain recommendations for measurement specifications.

[0056] According to an example of a first embodiment of the present invention, the method further includes using a user interface configured to allow a user to specify measurement specifications via natural language processing. Specifically, the method can be used in conjunction with an instance segmentation algorithm for contour detection, which assigns instance numbers to each detected semiconductor structure in a 2D profile image. The user can then easily specify the measurement by referring to a specific instance of the semiconductor structure in the first 2D profile image. This reduces the workload for the user in specifying the measurement specifications.

[0057] According to an example of a first embodiment of the present invention, the method further includes using a visualization device for visualizing features and / or profiles and / or measurement specifications, wherein the visualization indicates the correlation between the features and / or profiles and / or measurement specifications and the semiconductor structure. Thus, users can easily inspect and verify measurements obtained by this method, identify sources of error, and visually detect defects. Therefore, the workload for users is reduced.

[0058] According to an example of a first embodiment of the invention, the method further includes using a visualization device for visualizing features and / or contours and / or measurement specifications, wherein the features and / or contours and / or measurement specifications in the first 2D profile image are distinguishable from features and / or contours and / or measurement specifications propagating in another 2D profile image. Thus, a user can easily distinguish between the first 2D profile image used to indicate measurement specifications and the other 2D profile images.

[0059] According to one embodiment of the invention, the method further includes using a visualization device for visualizing a 3D representation of the propagated measurement specifications. This simplifies the inspection and verification of the obtained measurements and enables visual detection of defects.

[0060] According to an example of a first embodiment of the present invention, the method further includes using a user interface and a visualization device configured to allow a user to browse a 2D profile image containing visual features and / or contours and / or measurement specifications. This simplifies the inspection and verification of the obtained measurements and enables visual detection of defects.

[0061] In one example, the method used to obtain measurements of the semiconductor structure on a wafer is a computer-implemented method.

[0062] A computer-readable medium according to a second embodiment of the present invention has a computer program stored thereon and executable by a computing device, the computer program comprising program code for performing the method as described above according to any one of the first embodiments of the present invention.

[0063] A computer program product according to a third embodiment of the present invention includes instructions, wherein when a computer executes the program, the instructions cause the computer to perform the method described above according to any one of the first embodiments of the present invention.

[0064] A system for obtaining measurements of semiconductor structures on a wafer according to a fourth embodiment of the present invention comprises: an imaging device configured to provide a volumetric imaging dataset including a plurality of 2D cross-sectional images of the wafer; one or more processing devices; and one or more machine-readable hardware storage devices containing instructions executable by the one or more processing devices to perform operations comprising any one of the methods as claimed in the preceding claims. Optionally, the system may include a database and / or a user interface and / or a visualization device.

[0065] While the examples and embodiments of the present invention are described with respect to semiconductor wafers, it should be understood that the present invention is not limited to semiconductor wafers, but can also be applied, for example, to master masks or masks used in semiconductor manufacturing or other manufactured objects. Furthermore, the techniques described herein can be used in conjunction with various 3D imaging techniques, such as X-rays, CT scans, etc.

[0066] The invention described by the examples and embodiments is not limited to these examples and embodiments, and can be implemented by those skilled in the art in different combinations or modifications. Attached Figure Description

[0067] Figure 1 A flowchart illustrating a method for obtaining measurements of semiconductor structures on a wafer according to a first embodiment of the present invention is shown.

[0068] Figure 2 An imaging dataset containing a first 2D profile image and another 2D profile image is shown;

[0069] Figure 3 The image shows a first 2D cross-sectional image containing the outline of the semiconductor structure and three indicated measurement specifications;

[0070] Figure 4 This shows the contour points that define the circular outline using specific points and direction vectors;

[0071] Figure 5 A 2D cross-sectional image showing a semiconductor structure containing the detected imprecise centroid;

[0072] Figure 6 A method is shown for obtaining the centroid of a semiconductor structure in a 2D cross-sectional image by analyzing the intensity distribution;

[0073] Figure 7a and 7b Defect detection in semiconductor structures is visualized by analyzing measurements obtained from measurement specifications in the first 2D profile image and the second 2D profile image of the evaluation imaging dataset.

[0074] Figure 8A method illustrating an example of a first embodiment of the present invention; and

[0075] Figure 9 The system according to the fourth embodiment of the present invention is illustrated schematically. Detailed Implementation

[0076] The following describes advantageous exemplary embodiments of the invention, which are schematically illustrated in the figures. Throughout the figures and description, the same reference numerals are used to describe the same features or components. Dashed lines indicate optional features.

[0077] Figure 1 A flowchart is shown of a method 10 (particularly a computer-implemented method) for obtaining measurements of a semiconductor structure on a wafer according to a first embodiment of the present invention. The method includes: an imaging step 12 to obtain a volumetric imaging dataset of a wafer containing multiple 2D cross-sectional images; a contour generation step 14 to obtain a contour of the semiconductor structure in the 2D cross-sectional images of the imaging dataset; a measurement specification step 16 to indicate one or more measurement specifications relating to the contour of the semiconductor structure in a first 2D cross-sectional image of the imaging dataset; a propagation step 18 to propagate the one or more measurement specifications indicated in the first 2D cross-sectional image to another 2D cross-sectional image of the imaging dataset; and a measurement step 20 to obtain measurements of the semiconductor structure by evaluating one or more measurement specifications in the first 2D cross-sectional image and the other 2D cross-sectional image of the imaging dataset.

[0078] This method provides measurements of 3D semiconductor structures on a wafer at a low computational cost. The measurement specifications are specified only in a single 2D cross-sectional image and automatically propagate to other 2D cross-sectional images, rather than reconstructing the 3D shape of the semiconductor structure. Thus, measurements of the 3D semiconductor structure can be calculated without requiring 3D reconstruction of the semiconductor structure. This method requires minimal user effort, as the user only needs to specify the measurement specifications for the outline of the semiconductor structure in the first 2D cross-sectional image. No template generation is required. For example, the obtained measurements can be used for defect detection in semiconductor structures, verification of critical dimensions, or process window verification.

[0079] Figure 2 An imaging dataset 22 obtained by focused ion beam scanning electron microscopy (FIB-SEM) is shown. Imaging dataset 22 contains the semiconductor structure of a memory wafer. Imaging dataset 22 includes a first 2D cross-sectional image 24 and a further 2D cross-sectional image 26. In measurement specification step 16, one or more measurement specifications 28 are indicated in the first 2D cross-sectional image 24. In propagation step 18, the indicated measurement specifications 28 are propagated to the further 2D cross-sectional image 26 of imaging dataset 22.

[0080] Figure 3The first 2D cross-sectional image 24 shows the outline 32 of the semiconductor structure 30 and three indicated measurement parameters 28. The measurement parameters 28 are defined relative to the features of the outline 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24.

[0081] According to an example of a first embodiment of the invention, the features include specific points, lines, or curves defined relative to the contours 32 or contour segments of one or more semiconductor structures 30 in the first 2D cross-sectional image 24, particularly points (contour points 34) or contour segments on the contour, regions defined by the contours or contour segments, the contours or contour segments, the centroid 36 of the contours or contour segments, etc. Specific points, lines, or curves can be calculated based on the coordinates of the contours 32 or contour segments of one or more semiconductor structures 30 in the first 2D cross-sectional image 24. A contour segment refers to a portion of a contour line 32, for example, a portion defined relative to the intersecting lines, curves, axes, or geometry of one or more contour lines 32, such as a portion of contour 32 relative to an axis of symmetry; or a portion of contour 32 above or below the intersecting lines of contour 32; or a portion of contour 32 contained in a circle of a specific diameter surrounding the centroid 36 of one or more contours 32; or a portion of contour 32 contained within a bounding box of a specific size centered at the centroid 36 or some other point relative to one or more contours 32, etc.

[0082] Features may include, for example, points, lines, or curves defined relative to a single contour 32, such as... Figure 3 The points shown are: contour point 34 or centroid 36, points on the circumcircle or bounding box of contour 32 or contour segment, diameter or chord of contour 32 or contour segment, circumcircle or bounding box of contour 32 or contour segment, etc.

[0083] Features may include, for example, points, lines, or curves defined relative to two or more contours 32 or contour segments, such as points on the connecting line between two centroids 36 of contour 32 or contour segment, points at the center of two or more contours 32 or contour segments, contour points on different contours having the minimum distance between them, connecting lines between contours 32 or contour segments, intersecting lines, or axes of symmetry.

[0084] According to an example of a first embodiment of the present invention, measurement specification 28 is derived from a group containing feature locations, feature distances, and feature dimensions. Measurement specification 28 is used to obtain 2D measurements of cross-sections of the semiconductor structure 30 within a first 2D cross-sectional image 24 and a further 2D cross-sectional image 26. Measurement specification 28 may include feature locations, such as feature coordinates in a coordinate system, or angles of features relative to certain lines, or positions of features relative to specific markers in the first and / or further 2D cross-sectional images 24, 26, such as relative to a specified point, line, or plane. Measurement specification 28 may include feature distances. Distances may be measured using different norms, for example, - Norm, or distance that can be measured on the profile (e.g., the distance between two points on the profile can be measured by the length of the profile segment connecting the two points). Feature distance may include the shortest distance between features on one or more profiles 32 or the shortest distance between a feature and a specified point (e.g., a landmark), a straight line, a curve, or a plane. Measurement specification 28 may include feature dimensions, such as the area of ​​the profile or the area of ​​a profile segment, the length of the profile or the length of the profile segment, or the length of the radius or diameter, or the length of any other straight line or curve defined by one or more profiles (e.g., a straight line or curve connecting two points of a profile, or a straight line or curve connecting points of two or more profiles), or the area or length of the circumcircle or bounding box of one or more profiles. For example, Figure 3 The measurement specification 28 shown defines the distance between contour points 34 or between contour points 34 and centroid 36.

[0085] In one example, the measurement specification 28 in the first 2D profile image 24 only indicates the distances between points defined relative to the profile 32, specifically the distances between profile points 34. The distance between two points can be represented, for example, by a line connecting the two points, such as... Figure 3 As shown. This provides a particularly simple, efficient, and flexible way to indicate measurement specifications 28. For example, it enables the indication of distances that are difficult to describe in any other way.

[0086] According to an example of a first embodiment of the present invention, in step d), by associating the contour 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 with the corresponding contour 32 of the same semiconductor structure 30 in another 2D cross-sectional image 26, and by associating the features of the contour 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 with the corresponding features in the relevant contour 32 of the semiconductor structure 30 in the other 2D cross-sectional image 26, one or more measurement specifications 28 indicated in the first 2D cross-sectional image 24 are propagated to another 2D cross-sectional image 26 of the imaging dataset 22. Therefore, by associating the features of the contour 32 with the corresponding features of the corresponding contour 32 in the other 2D cross-sectional image 26, the measurement specifications 28 defined relative to the contour 32 of the semiconductor structure in the first 2D cross-sectional image 24 can be automatically transferred to the other 2D cross-sectional image 26. Thus, the measurement is propagated completely automatically to... Figure 2 Another 2D cross-sectional image 26 is shown. Measurement specification 28 also propagates with high accuracy because the features of the contour are calculated based on the relevant contour, which can be accomplished with subpixel accuracy.

[0087] The outline 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26 can be obtained in various ways.

[0088] According to one aspect of an embodiment of the invention, contours of semiconductor structures in a first 2D profile image and another 2D profile image are obtained by applying a contour extraction method to the first 2D profile image and the other 2D profile image. Many contour extraction methods are known to those skilled in the art. Contour extraction methods may, for example, include edge detection methods based on image gradients or filters, such as the Canny edge detector, Sobel edge detector, Robert's edge detector, or Gabor filter. Contour extraction methods may also include active contours or snake models, which can be used to obtain closed contours by minimizing the length and / or curvature of the contour while simultaneously fitting the contour to image edges. Connected contours can also be obtained from edge detection using random walk techniques. Contour extraction methods may also include segmentation methods, such as superpixel segmentation or watershed methods. Distance-transformed images measure the distance between each pixel in the image and an edge. Therefore, zero contours in a distance-transformed image represent contours. If prior knowledge is available, contour extraction methods may also include model-based contour extraction methods for detecting contours belonging to a specific object (e.g., a specific semiconductor structure). Generally, machine learning models can be trained to detect edges. They can also be trained to detect object-specific edges, for example, only edges in an image that belong to a specific type of object (e.g., a specific type of semiconductor structure).

[0089] According to one aspect of an embodiment of the present invention, the contour 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26 is obtained by applying an object detection or image segmentation algorithm to the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26. The boundaries of the detected objects or the boundaries of the image segments represent the contours in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26, respectively. Object detection and image segmentation methods are known to those skilled in the art.

[0090] Object detection methods involve detecting instances of a semantically specific object within an image. Classical object detection methods primarily rely on a specific set of features representing the object of interest, such as Scale Invariant Feature Transform (SIFT) features or Histogram of Oriented Gradients (HOG) features. Based on these feature vectors, classification algorithms (such as Support Vector Machines or neural networks) can be trained to distinguish feature vectors belonging to the object of interest from other objects or the background. Pattern recognition approaches can also be used for object detection; for example, Hough transform, particularly for detecting simple geometric shapes such as circles or rectangles, which is typical for cross-sections of semiconductor structures. Modern object detection methods are mostly based on machine learning models, such as deep learning models like convolutional neural networks. During training, deep learning models automatically learn filters to generate features from images best suited for solving the object detection task.

[0091] Image segmentation is commonly used to locate objects and boundaries in an image by dividing it into multiple image regions. More precisely, image segmentation is the process of assigning labels to each pixel in an image so that pixels with the same label share certain features. Classical image segmentation methods group pixels into regions based on some feature, such as intensity, color, or texture. For example, thresholding or clustering methods can be used, which assign pixels to clusters that are closest relative to the features. Region growing methods, such as watershed or superpixel segmentation, can also be used to automatically assign pixels to image regions. Histogram-based image segmentation methods are very efficient and can handle intensity variations. Image segmentation methods can consider image edges by supporting region boundaries that are consistent with image edges. Energy minimization methods, such as those based on solving partial differential equations, like variational methods, or graph-based methods, such as Markov random fields, can be used to minimize an objective function that includes, for example, a feature similarity term that favors regions sharing the same features and an edge term that favors the overlap between image edges and region edges, while ensuring the minimum length and / or curvature of the boundaries. If prior knowledge is available for the object of interest, such as pattern recognition methods, model-based image segmentation methods can be used. Such pattern recognition methods are particularly useful for segmenting simple geometries, such as if a semiconductor structure is circular and / or rectangular. Modern image segmentation methods are primarily based on machine learning models, such as deep learning models, which derive their own set of features best suited for solving image segmentation tasks during the training phase. For example, deep learning models may be based on the U-net architecture.

[0092] According to an example of the first embodiment of the present invention, the outline 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26 is obtained by applying an instance segmentation algorithm to the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26.

[0093] Instance segmentation is the task of detecting and delineating each distinct object of interest appearing in an image. Therefore, instance segmentation algorithms can compute semantic segments of an image (e.g., using image labels of different categories, such as "semiconductor structure" or "background"), and additionally extract distinct instances of the same object category, for example, by assigning specific indices to each object of the same category. Thus, instance segmentation algorithms compute 1) the object category and 2) the number of object instances to which a pixel belongs for each pixel in an image. To simplify the work, prior knowledge can be used for the object of interest, such as the type and appearance of the semiconductor structure of interest. For example, machine learning models can be used to compute instance segmentation. For instance, a machine learning model can be trained based on user annotations or a design file containing a model labeled with semiconductor structures. It can also be trained on segmentation results, such as on superpixel labels, contour extraction results, or watersheds. To improve the performance of the machine learning model, hyperparameter optimization and / or neural network architecture search methods can be employed.

[0094] Various machine learning models can be used for instance segmentation, such as deep learning methods, reinforcement learning methods, and transformers. A transformer machine learning model suitable for instance segmentation is called a "Masked-attention Mask Transformer" and relies on the classification of image segments represented by a C-dimensional feature vector called an "object query," which can be processed by a transformer decoder trained with a set prediction target. A simple meta-architecture consists of three parts: a backbone network that extracts low-resolution features from the image; a pixel decoder that progressively upsamples the low-resolution features in the backbone network output to produce high-resolution per-pixel embeddings; and a transformer decoder that operates on the image features to process the object query. The final binary mask prediction is decoded from the per-pixel embeddings via the object query. The transformer decoder may include a mask attention operator that extracts local features by limiting cross-attention to the foreground region of the predicted mask for each query, rather than focusing on the full feature map. To handle small objects, efficient multi-scale strategies can be employed to utilize high-resolution features. It feeds continuous feature maps from the feature pyramid of the pixel decoder into continuous transformer decoder layers in a cyclic polling manner. For example, alternative machine learning models for instance segmentation are deep learning models such as Masked R-CNN, YOLACT, or TensorMask.

[0095] The outline 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26 can be represented in various ways.

[0096] In any of the examples described herein, the contour point 34 may represent the contour 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the further 2D cross-sectional image 26. Therefore, a contour point is a point located on the contour of the semiconductor structure 30. For example, the contour point 34 is obtained by subsampling the contour 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the further 2D cross-sectional image 26 at a specific density, such as... Figure 3 As shown. Alternatively, contour points 34 can be obtained by detecting edge pixels of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26.

[0097] In any of the examples described herein, the bounding box may represent the outline 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the further 2D cross-sectional image 26. The bounding box may have various shapes surrounding the semiconductor structure 30 in the first 2D cross-sectional image 24 and the further 2D cross-sectional image 26, such as rectangular, circular, elliptical, or any other type of shape. Outline points 34 can be obtained from the bounding box by subsampling multiple lines of the bounding box. In the case of a rectangular bounding box, the edges of the bounding box can be used as outline points 34.

[0098] The outline 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 can be associated with the corresponding outline of the same semiconductor structure in another 2D cross-sectional image 26 in various ways.

[0099] According to one aspect of an example of a first embodiment of the present invention, associating the contour 32 of a semiconductor structure 30 in a first 2D cross-sectional image 24 with the corresponding contour of the same semiconductor structure in another 2D cross-sectional image 26 includes calculating the matching of contour points 34 of the semiconductor structure 30 in the first 2D cross-sectional image 24 with the contour points 34 of the semiconductor structure 30 in the other 2D cross-sectional image 26. This is particularly useful if the contour 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26 is represented by contour points 34 as described above. Such contour points 34 of the contour 32 in the 2D cross-sectional images 24, 26 can be obtained in different ways, for example, by subsampling the contour 32 of the semiconductor structure, by an edge detection algorithm, or by a machine learning algorithm. If the contour 32 is represented by contour points 34, associating the contour 32 in different 2D cross-sectional images 24, 26 requires, for example, finding the matching of contour points 34 in different 2D cross-sectional images 24, 26 using a Hungarian algorithm. The Hungarian Algorithm is a combinatorial optimization algorithm known to those skilled in the art, which solves the assignment problem in polynomial time. The assignment problem involves finding a matching of a specific size in a weighted bipartite graph, where the weighted sum of the edges is minimized. This is applied to the association of contour points 34 of contour 32 in 2D profile images 24 and 26, where the nodes of the graph are contour points 34 and the edges of the graph represent possible associations between contour points 34 in different 2D profile images 24 and 26. The weights of the graph edges can be defined relative to the distances of contour points 34 in the first 2D profile image and the other 2D profile image 26.

[0100] According to one aspect of an example of a first embodiment of the present invention, associating the outline 32 of a semiconductor structure 30 in a first 2D cross-sectional image 24 with the corresponding outline 32 of the same semiconductor structure 30 in another 2D cross-sectional image 26 includes applying a tracking algorithm or an optical flow algorithm to track the outline 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 on another 2D cross-sectional image 26 of the imaging dataset 22.

[0101] Tracking algorithms are known to those skilled in the art. They predict the future positions of one or more moving objects based on the history of their individual positions. The contour 32 of the semiconductor structure 30 in the first 2D profile image 24 can be used to initialize the tracking algorithm, which then tracks the contour 32 on another 2D profile image 26. The tracking algorithm may rely on the results of an object detection algorithm, applying those results to the first 2D profile image 24 and the other 2D profile image 26 to obtain object pointers, such as, for example, bounding boxes, contours, or centroids of segments obtained by an object of interest or instance segmentation algorithm, as described above. In this case, the tracking algorithm finds the trajectory of the object pointers through the first 2D profile image 24 and the other 2D profile image 26 according to some optimization criteria, such as the spatial distance between object pointers in different 2D profile images or the similarity of object pointers in the images according to some similarity metric, such as color, intensity, texture, shape similarity, size, etc. Using the trajectory, the contour 32 can be associated by following the trajectory through different 2D profile images. Other tracking algorithms do not require the results of an object detection algorithm, such as a Kalman filter.

[0102] In particular, to trace the contours 32 of the semiconductor structure 30 using different 2D cross-sectional images 24, 26 (which may contain defects), it is important to prevent trajectory interruptions caused by irregularities in the 2D cross-sectional images 24, 26 due to defects. In fact, it is these irregularities that cause measurement irregularities in the measurement specifications, thus necessitating the detection of defects in the semiconductor structure 30. For this purpose, it is important to associate (almost) every contour 32 (e.g., bounding box) in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26 with a trajectory. In contrast to common object detection methods that discard contours 32 with low similarity metrics, it is important to also consider contours 32 with low similarity metrics, which may be due to defects in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26. In the first step, contours 32 with high similarity metrics are associated with trajectories of different semiconductor structures 30. In the second step, contours 32 with low similarity metrics are assigned to trajectories. In the second step, instead of using an appearance-based similarity metric, the overlap (cross-union ratio) of regions of contour 32 in different 2D profile images 24 and 26 is used to assign contours with low similarity metrics to trajectories. The more a region of contour 32 with a low similarity metric overlaps with a region of contour 32 already assigned to a trajectory (especially for adjacent 2D profile images 24 and 26), the more likely the contour 32 with a low similarity metric belongs to the same trajectory. Additionally, the path of the trajectory can be analyzed to find contours 32 with low similarity metrics that may belong to the same trajectory.

[0103] These interruptions in the trajectory can be detected to detect defects in the semiconductor structure 30 in the 2D profile images 24, 26, rather than to prevent trajectory interruptions caused by irregularities in the 2D profile images 24, 26. For example, such interruptions can be detected by analyzing the path of the trajectory of the contour 32 (e.g., the bounding box).

[0104] Optical flow algorithms are known to those skilled in the art. Optical flow refers to a displacement field that indicates the motion between two or more 2D profile images 24, 26. Based on the optical flow field indicating the displacement between consecutive 2D profile images, the contour 32 or features of contour 32 in the first 2D profile image 24 can be associated with the contour 32 or features of contour 32 in another 2D profile image 26.

[0105] According to one aspect of an embodiment of the present invention, associating the contour 32 of a semiconductor structure 30 in a first 2D cross-sectional image 24 with a corresponding contour 32 of the same semiconductor structure 30 in another 2D cross-sectional image 26 includes registering an imaging dataset 22 to a reference imaging dataset having labeled contours 32. Since the association of the contours 32 of the semiconductor structures in the 2D cross-sectional images of the reference imaging dataset is known, association information can be transferred from the contours 32 in the reference dataset to the registered contours in the imaging dataset 22.

[0106] According to an example of the first embodiment of the present invention, the outlines 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26 are obtained by calculating the 3D segmentation of the semiconductor structure 30 in the imaging dataset 22 and calculating the outlines 32 of the semiconductor structure segmented in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26 based on the 3D segmentation.

[0107] 3D segmentation methods can be obtained by generalizing 2D segmentation methods to volumetric imaging datasets. For example, volumetric voxel representations can be obtained from the first 2D profile image 24 and another 2D profile image 26 of imaging dataset 22, and the segmentation method can be applied to volumetric pixels instead of individual pixels. For example, energy minimization methods or machine learning models, as described above, can also be used for 3D segmentation. The surface mesh of the segmented semiconductor structure containing volumetric pixels is then obtained using the Marching Cubes algorithm. The contour 32 of the semiconductor structure 30 can be obtained as the intersection of the volumetric pixel-based 3D segmentation or the surface mesh with the plane of the 2D profile image.

[0108] The 3D segments (volume pixels or surface mesh segments) obtained within the imaging dataset 22 inherently contain information that the contours in the first 2D profile image 24 and the other 2D profile image 26 belong to the same 3D segment. Therefore, 3D segmentation can be used to associate contours 32 belonging to the same semiconductor structure 30 in the first 2D profile image 24 and the other 2D profile image 26. Thus, in step d), by associating the contours 32 of the semiconductor structure 30 in the first 2D profile image 24 with the corresponding contours 32 of the same 3D segment of the same semiconductor structure 30 in the other 2D profile image 26, and by associating the features of the contours 32 of the semiconductor structure 30 in the first 2D profile image 24 with the corresponding features in the associated contours of the semiconductor structure 30 in the other 2D profile image 26, one or more measurement specifications 28 indicated in the first 2D profile image 24 can be propagated to the other 2D profile image 26 of the imaging dataset 22.

[0109] Associating contour features in different 2D profile images 24 and 26 can be difficult, especially for contour point 34. Associating two contour points 34 on contour 32 in the two 2D profile images 24 and 26 requires accurate detection and association of contour 32 and the positions of contour points 34 on contour 32. To associate contour points 34 with high accuracy, in Figure 4 In one example of a first embodiment of the invention, at least one measurement specification 28 includes: a contour point 34 of a contour 32 of a semiconductor structure 30, wherein the contour point 34 is defined by a specific point 38 relative to the contour 32, particularly the centroid 36; and a direction vector 40 indicating the direction of the contour point 34 relative to the specific point 38 in a first 2D cross-sectional image 24, and wherein associating the contour point 34 in the first 2D cross-sectional image 24 with the corresponding contour point 34 of the associated contour 32 in another 2D cross-sectional image 26 includes calculating the intersection of the associated contour 32 with the direction vector 40, which begins at the specific point 38 of the associated contour 32 in the other 2D cross-sectional image 26. Therefore, instead of directly associating the contour points 34 in the different 2D cross-sectional images 24 and 26, the contour points 34 in the different 2D cross-sectional images 24 and 26 are associated in the following way: 1) Define the contour point 34 relative to a specific point 38 (e.g., centroid 36) of contour 32, and the direction vector 40 relative to the specific point 38 in the first 2D cross-sectional image; 2) Obtain the associated contour for each of the other 2D cross-sectional images 26 as described above; 3) For each associated contour in the other 2D cross-sectional image 26, calculate the specific point 38 from the associated contour 32; 4) For each associated contour in the other 2D cross-sectional image 26, calculate the associated contour point 34 as the intersection of the associated contour 32 and the direction vector 40 starting from the calculated specific point 38 of the associated contour. In this way, the measurement specification 28 can be transmitted between the 2D cross-sectional images 24 and 26. Figure 4 In this measurement specification 28, for example, the distance between the contour point 34 of the first contour 32 and the centroid 36 of a segment of the second contour 32' is included. The segment of the second contour 32' is defined as a portion of the contour located within a circle 42 centered on the contour point 34. The measurement specification 28 can then be transferred to another 2D profile image 26 by: 1) calculating the associated contour points in the other 2D profile image 26 as described above; and 2) calculating the centroid of the associated segment of the second contour 32' located within a circle 42 centered on the associated contour point 34. By calculating associated features of each contour in the other 2D profile image 26 (e.g., the contour point 34 of the first contour 32 and the centroid 36 of the segment of the second contour 32') based on the associated contours instead of matching a single feature, the measurement specification 28 can be transferred to another 2D profile image 26 with high accuracy.

[0110] Accurate calculation of the centroid 36 of the outline 32 of the semiconductor structure 30 in the 2D cross-sectional images 24 and 26 can also be difficult. For example, if the outline 32 is represented by outline points 34 or a bounding box of any shape, it is difficult to calculate the centroid 36 from the outline 32 itself. In addition, the outline 32 in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26 may be inaccurate for various reasons, which makes it difficult to accurately calculate the centroid 36 of the outline 32. Figure 5 2D cross-sectional images 24 and 26 are shown of a semiconductor structure 30 containing a detected imprecise centroid 36.

[0111] To accurately detect the centroid 36 of the semiconductor structure 30 in a 2D cross-sectional image, in Figure 6 In one example of the first embodiment of the present invention, at least one measurement specification 28 includes calculating the centroid 36 of the contour 32 of the semiconductor structure 30 in one or more 2D cross-sectional images 24, 26, wherein the centroid 36 is obtained by analyzing the intensity distribution 46, 48 of 1D cross-sections 54, 56 along the region 44 (e.g., bounding box) surrounding the contour 32 in one or more 2D cross-sectional images 24, 26. Figure 6The diagram above shows the intensity distribution 46 along a horizontal 1D profile 54 within a region 44 surrounding contour 32, where the coordinates along the horizontal 1D profile 54 are indicated on the horizontal axis and the corresponding image intensity on the vertical axis. The diagram below shows the intensity distribution 48 along a vertical 1D profile 56 within a region 44 surrounding contour 32, where the coordinates along the vertical 1D profile 56 are indicated on the horizontal axis and the corresponding image intensity on the vertical axis. The centroid 36 can then be detected, for example, by finding the symmetrical points 50 and 52 of intensity distributions 46 and 48. The symmetrical points 50 and 52 of intensity distributions 46 and 48 are the intersections of intensity distributions 46 and 48 with their axes of symmetry 58 and 60. The axes of symmetry 58 and 60 of intensity distributions 46 and 48 can be found, for example, by calculating the cross-correlation between intensity distributions 46 and 48 and their flipped variations. The maximum cross-correlation value indicates the coordinates of the axes of symmetry 58 and 60, and thus the coordinates of the centroid 36. When contour 32 is represented by a bounding box of any shape, the bounding box can be used as a region 44 surrounding contour 32. When contour 32 is represented by contour points, contour curves, or contour segments of semiconductor structure 30, the region 44 surrounding the contour can be obtained, for example, by calculating a bounding box of any shape from the coordinates of the contour points, contour curves, or contour segments (e.g., by calculating the minimum and maximum horizontal and vertical coordinates of the contour). To improve the accuracy of centroid 36, the bounding box can be centered on the calculated centroid, and this process can be repeated once or multiple times or until convergence.

[0112] The centroid 36 of the contour 32 found according to the described method can be used in different ways to obtain measurements of the semiconductor structure 30, for example, as a feature of the contour 32 used to define the measurement specification 28, or as a specific point 38 used to define the position of the contour point 34 on the contour 32 by means of the specific point 38 and the direction vector 40 as described above, or whenever the centroid 36 of the contour 32 of the semiconductor structure 30 is found in the first 2D cross-sectional image 24 or another 2D cross-sectional image 26.

[0113] According to an example of a first embodiment of the invention, in step d), propagating one or more measurement specifications 28 indicated in the first 2D profile image 24 to another 2D profile image 26 of the imaging dataset 22 includes generating confidence scores indicating the reliability of the associated measurement specifications 28 in the first 2D profile image 24 and the other 2D profile image 26. Thus, the uncertainty of the detection or association of contours 32 and / or feature points within the different 2D profile images 24, 26 can be measured and used in further processing steps or subsequent algorithms. For example, during contour propagation to another 2D profile image, contours detected with high uncertainty can be ignored or weighted. For example, in the case of low confidence scores, propagated measurement specifications 28 can be ignored in the evaluation of measurement specifications 28 (e.g., due to the high uncertainty of the associated contours or features), or the corresponding measurements can be weighted according to the confidence scores. For example, it may be important if the semiconductor structure 30 is not visible in one or more 2D cross-sectional images, such as in the case of partial hollowing, thus indicating that the semiconductor structure 30 is partially invisible due to an incorrect deposition process or due to some defect causing a break in the semiconductor structure 30. If the confidence score is additionally specified, subsequent algorithms can process the obtained measurements with higher accuracy.

[0114] According to one embodiment of the invention, defects are detected by detecting outliers in measurements obtained from measurement specification 28 in the first 2D profile image 24 and another 2D profile image 26 of the evaluation imaging dataset 22. Outliers can be detected statistically, for example, by calculating confidence intervals or p-values ​​based on the obtained measurements. Outliers can also be detected by defining threshold values ​​for the obtained measurements, for example, user-specified measurement values ​​or threshold values, or the minimum or maximum standard deviation relative to the obtained measurements. Defects can be detected by applying a machine learning model to the obtained measurement list. This list may, for example, contain measurements obtained from the evaluation measurement specification, or it may contain 3D locations defining features of the measurement specification. The machine learning model can be trained to distinguish between a measurement list of semiconductor structures without defects and a measurement list of semiconductor structures containing defects. Defects can also be detected visually. In another embodiment, defects are detected by analyzing changes in measurement values ​​obtained from measurement specification 28 in the first 2D profile image 24 and another 2D profile image 26 of the evaluation imaging dataset 22. The greater the variation in the obtained measurement value, the more likely an outlier or defect is to exist in the corresponding semiconductor structure 30 on the wafer. The variation in the obtained measurement value can be estimated, for example, by the variance or standard deviation of the obtained measurement value.

[0115] Figure 7a and 7bDefect detection of semiconductor structure 30 is visualized by analyzing measurements obtained from measurement specifications 28 in the first 2D profile image 24 and the additional 2D profile image 26 of the evaluation imaging dataset 22. Figure 7a A tilted semiconductor structure 30 is shown. A first 2D cross-sectional image 24 and a further 2D cross-sectional image 26 of the semiconductor structure 30 are obtained. A measurement specification 28 is provided in the first 2D cross-sectional image 24 to indicate the location of the centroid 36 of the contour 32 containing the semiconductor structure 30. The contour 32 in the first 2D cross-sectional image 24 and the further 2D cross-sectional image 26 are correlated, and the centroid 36 of the correlated contour 32 is calculated in the first 2D cross-sectional image 24 and the further 2D cross-sectional image 26 (e.g., by analyzing the intensity distribution along the 1D cross-section, as described above). The indicated measurement specification 28 can be evaluated by measuring the location of the centroid 36 in the first 2D cross-sectional image 24 and the further 2D cross-sectional image 26. For example, defects can be detected by specifying one or more critical values ​​for the distance between the location of the centroid 36 in the further 2D cross-sectional image 26 and the location of the centroid 36 in the first 2D cross-sectional image 24. Alternatively, the variation in the position of the centroid 36 in the first 2D profile image 24 and the additional 2D profile image 26 can be measured and used as a defect indicator. Alternatively, the angle between the line connecting the centroid 36 and the perpendicular reference line intersecting the centroid 36 of the contour 32 in the first 2D profile image 24 can be calculated, and an angle threshold can be specified to detect defects.

[0116] Figure 7b The diagram shows a first model semiconductor structure 30 and a second model semiconductor structure 30' based on a design file. The first contour 32 indicated in 2D cross-sectional images 24 and 26 corresponds to the first model semiconductor structure 30 after being printed onto a wafer. The second contour 32' indicated in 2D cross-sectional images 24 and 26 corresponds to the second model semiconductor structure 30' after being printed onto a wafer. Measurement specifications 28 are defined relative to the closest contour points of the first contour 32 and the second contour 32'. Defects 28 can then be detected by analyzing and evaluating the measurement specifications, which are the distances between the closest contour points in each 2D cross-sectional image 24 and 26, for example, by detecting outliers. Outliers can be detected, for example, by calculating confidence intervals or by comparing each distance to a predefined critical value or a critical value defined relative to the average value μ and standard deviation σ of the distances, such as [μ - 3σ, μ + 3σ].

[0117] Ultimately, the user can see the detected defects, for example, by highlighting one or more measurement specifications 28 corresponding to the defects in the first 2D profile image 24 and / or the additional 2D profile image 26. Additionally, the user can be shown measurements obtained by evaluating one or more measurement specifications 28 in the first 2D profile image 24 and / or the additional 2D profile image 26.

[0118] According to one embodiment of the invention, an inspection target, particularly a target processing volume, is obtained, and the number of additional 2D profile images 26 and / or measurement specifications 28 is automatically adjusted to meet the inspection target. For example, a specific time span can be provided for the analysis of the imaging dataset 22. To meet this requirement, the number of additional 2D profile images can be automatically subsampled, or the number of measurement specifications 28 can be automatically reduced to save computation time.

[0119] According to an example of a first embodiment of the present invention, the measurement specifications in the first 2D profile are indicated through a user interface. Specifically, the measurement specifications are marked graphically in the first 2D profile image through the user interface.

[0120] According to one example, the user interface is configured to allow the user to indicate a measurement specification 28 by selecting one or more features of the contour 32 of the semiconductor structure 30 on the first 2D cross-sectional image 24 of the imaging dataset 22. For example, features can be indicated by a simple click on the first 2D cross-sectional image 24. Thus, points can be marked, contours or contour portions can be selected, etc. For example, measurement specifications can be indicated by clicking two or more points, by clicking specific symbols near the contour, etc. For example, distances can be indicated by marking two points or by connecting points with lines, and contour regions can be selected by clicking specific symbols displayed near the contour, etc.

[0121] The user interface can be configured to assist the user during the selection of one or more features by calculating modifications relative to the contour 32 of the semiconductor structure 30. For example, a user-indicated contour point can be automatically corrected by moving the indicated contour point to the nearest contour point 34 on the contour 32 of the semiconductor structure 30, or a user-specified centroid 36 can be automatically corrected by moving the indicated centroid to the nearest centroid 36 of the contour 32 of the semiconductor structure 30 (e.g., calculated by analyzing the intensity distribution along a 1D profile, as described above). User confirmation may be required before modifying one or more selected features. Alternatively, one or more features may be modified automatically.

[0122] According to one embodiment of the invention, the method includes using a user interface configured to allow a user to load measurement specifications 28 from memory or a database and / or store measurement specifications 28 in memory or a database. For example, a design file with pre-specified measurement specifications 28 may be loaded, and the measurement specifications 28 may be automatically transferred to a first 2D profile 24 of an imaging dataset, for example, via a registration method. In one example, after measurements are performed on the semiconductor structure of a wafer, the specified measurement specifications 28 may be stored in memory or a database. The user interface may be configured to directly evaluate measurement specifications 28 specified by the user in the background, while allowing the user to specify additional measurement specifications 28 (streaming mode). Alternatively, the user interface may be configured to allow the user to annotate multiple measurement specifications 28 of interest before starting evaluation (batch mode).

[0123] According to one embodiment of the invention, the method includes using a user interface configured to allow a user to analyze the propagated measurement specification 28 and / or measurements obtained by evaluating the propagated measurement specification 28. For example, the user interface may be configured to allow a user to click on the measurement specification 28 and automatically display all propagated measurement specifications and / or measurements obtained by evaluating the propagated measurement specification 28.

[0124] According to an example of a first embodiment of the invention, method 10 includes using a user interface configured to propose a measurement specification 28 to a user, which can be accepted, modified, or rejected by the user, wherein a suggestion of measurement specification 28 is generated based on a previously specified measurement specification 28 by the user. For example, if a user specifies a measurement specification for measuring the distance between two specified contour points 34 of a pair of contours 32, the user interface can transfer the specified contour points 34 to adjacent pairs of contours 32 of the same type and suggest that the user add the corresponding measurement specification 28. This process is particularly useful in the case of repetitive semiconductor structures, such as those used for memory wafers. Alternatively, machine learning methods can be used to predict additional measurement specifications 28 based on a known history of user-specified measurement specifications 28, such as Markov chains or knowledge graphs. The history of specified measurement specifications 28 can be used to train machine learning models, which may be for the same user to obtain personalized predictions or for different users to obtain general predictions.

[0125] According to one embodiment of the invention, the method includes using a user interface configured to allow a user to specify measurement specification 28 via natural language processing. For example, a description of the measurement specification could be used, such as “find the shortest distance between trench 2 and the nearest channel.” The image segments and the outline 32 in the form of instance indices are particularly useful here because the user can easily refer to instance numbers to define measurement specification 28.

[0126] According to one embodiment of the invention, the method includes using a visualization device for visualizing features and / or contours and / or measurement specifications 28, wherein the visualization indicates the association between features and / or contours 32 and / or measurement specifications 28 and semiconductor structure 30. For example, all features and / or contours 32 and / or measurement specifications 28 associated with the same semiconductor structure 30 are visualized in the same manner, such as using the same colors; however, features and / or contours 32 and / or measurement specifications 28 associated with different semiconductor structures 30 are visualized in different manner, such as using different colors.

[0127] According to an example of a first embodiment of the invention, the method includes using visualization means for visualizing features and / or contours 32 and / or measurement specifications 28, wherein the features and / or contours 32 and / or measurement specifications 28 in the first 2D profile image 24 differ from those propagating in another 2D profile image 26. For example, the first 2D profile image 24 may be visualized using opaque colors, while the other 2D profile image 26 may be visualized using translucent colors, or vice versa.

[0128] According to an example of a first embodiment of the present invention, the method 10 includes using a visualization device for visualizing a 3D representation of the propagated measurement specification 28. Features of the contour 32 in the first 2D cross-sectional image 24 and further 2D cross-sectional images 26 can be visualized in a 3D view, rather than visualizing the measurement specification 28 in multiple first 2D cross-sectional images 24 and further 2D cross-sectional images 26. For example, the features of the contour 32 can be interpolated to achieve the 3D representation. For example, in Figure 7a In this process, the centroid 36 can be interpolated to obtain 3D lines.

[0129] According to an example of a first embodiment of the present invention, the method 10 includes using a user interface and a visualization device configured to allow a user to browse a 2D profile image containing visual features and / or contours 32 and / or measurement specifications 28.

[0130] Figure 8A method 10' according to an example of a first embodiment of the present invention is shown. In step S1, a volumetric imaging dataset 22 containing 2D profile images is obtained. In step S2, the contour of a semiconductor structure in the first 2D profile image is obtained, for example, by parsing primitives (e.g., simple geometric shapes such as circles and rectangles). For example, the contour can be represented by contour points. Alternatively, an instance segmentation algorithm can be applied to the first 2D profile image. In step S3, one or more 2D profile images are displayed to a user to allow the user to visually evaluate the data and identify the measurement specifications of interest. In step S4, the user interface is configured to allow the user to query a database entry containing the identified measurement specifications of interest. If this database entry exists (Yes 62), the measurement specifications of the database entry are retrieved from the database in step S5. If this database entry does not exist (No 64), in step S6, the user interface is configured to allow the user to indicate the measurement specifications regarding the contour features obtained in step S2 on the first 2D profile image of the imaging dataset. In step S7, the user interface is configured to calculate modifications to the selected features of the contour, for example, by moving the indicated contour point to the nearest point on the contour, or by moving the indicated centroid to the calculated centroid of the contour. For example, this can be achieved through... Figure 6The described method calculates the centroid. In step S8, the indicated measurement specification is stored in the current set of measurement specifications. In step S9, the user is asked if additional measurement specifications are needed. If yes, the method continues to step S6. If no, in step S10, the measurement plan for the current set is stored in the database. In step S11, the retrieved set of measurement values ​​(after step S5) or the stored set of current measurement values ​​(after step S10) is loaded from the database. In step S12, the measurement specification is automatically propagated to the first 2D profile image. As previously mentioned, in step S13, the measurement specification is automatically propagated to other 2D profile images. The measurement specification can be propagated to all other 2D profile images or to selected other 2D profile images. The user can be prompted to indicate the selection, for example by indicating a specific 2D profile image or by indicating a subsampling rate (e.g., every 3rd 2D profile image). Alternatively, the user interface can automatically select 2D profile images relative to a given processing volume requirement. For example, measurement specifications can be propagated to another 2D profile image by applying an instance segmentation algorithm to that image. A tracking algorithm can then be used to trace segmented instances in the first 2D profile image across the second 2D profile image to obtain associated contours in each image. Measurement specifications can be propagated from the first 2D profile image to the second 2D profile image by identifying corresponding features of associated contours in the first and second 2D profile images. For example, contour points can be represented by specific points and orientation vectors in the first 2D profile image, as described above. Figure 4 As described above, the aforementioned contour points can be associated with corresponding contour points in another 2D cross-sectional image by finding the intersection of the direction vectors starting from a corresponding specific point of the associated contour in another 2D cross-sectional image. In step S14, measurements are obtained by evaluating the measurement specifications propagating in the first cross-sectional image and the other 2D cross-sectional image. For example, measurements are displayed to the user by listing the measured values ​​or by graphically displaying the measurements in the image dataset in a 2D or 3D view, or by displaying a statistical evaluation. The associated contours and / or associated features and / or associated measurement specifications can be indicated using the same representation, such as the same color or line type or texture. Optionally, the user interface can be configured to allow the user to specify rules for detecting defects, such as threshold values ​​or confidence intervals. Alternatively, defects can be detected automatically, for example, based on statistical methods for detecting outliers or using machine learning methods as described above. Detected defects can be highlighted. In step S15, method 10' ends.

[0131] Figure 9A system 70 according to a fourth embodiment of the present invention is schematically illustrated, which can be used to obtain measurements of semiconductor structures on a wafer 72. System 70 includes an imaging device 74 and a processing device 76. The imaging device 74 is coupled to the processing device 76, for example, via cable or wirelessly. The imaging device 74 is configured to acquire a volumetric imaging dataset 22 containing 2D cross-sectional images of the wafer 72. An exemplary implementation of the imaging device 74 is a focused ion beam scanning electron microscope (FIB-SEM).

[0132] Imaging device 74 can provide imaging dataset 22 to processing device 76. Processing device 76 includes processor 78, which is implemented, for example, as a CPU or GPU. Processor 78 can receive imaging dataset 22 via interface 80. Processor 78 can load program code from memory 82. Processor 78 can execute program code. When executing program code, processor 78 performs techniques such as those described herein, for example, obtaining measurements of semiconductor structures in a wafer; detecting defects in the imaging dataset of the wafer; propagating measurement specifications indicated in a first 2D cross-sectional image to another 2D cross-sectional image; detecting or associating the contour and / or features of semiconductor structures in 2D cross-sectional images; training and / or applying machine learning models for segmentation, object detection, tracking, or defect detection; calculating the centroid of the semiconductor structure contour by analyzing intensity distribution, etc. For example, once the program code is loaded from memory 82, processor 78 can execute... Figure 1 or Figure 8 The method is illustrated. System 70 may optionally include a user interface 84, for example, for indicating measurement specifications, defect detection rules, natural language processing, viewing associated measurement specifications, etc. System 70 may optionally include a database 86. Database 86 may be used, for example, to load a collection of measurement specifications, training data, or pre-trained machine learning models. System 70 may optionally include a visualization device 88 for visualizing measurements and / or measurement specifications and / or defects and / or profile associations and / or feature associations, etc., to the user.

[0133] The methods disclosed herein can be used, for example, during research and development, during high-volume wafer manufacturing, or for process window verification or enhancement. Additionally, the methods disclosed herein can also be used, for example, for defect detection of X-ray imaging datasets containing semiconductor structures after packaging semiconductor devices for delivery.

[0134] In this specification, references to "an embodiment," "an example," or "an aspect" mean that a particular feature, structure, or characteristic described in the associated embodiment, example, or aspect is included in at least one embodiment, example, or aspect. Therefore, the terms "according to an embodiment," "according to an example," or "according to an aspect" appearing in this specification do not necessarily refer exclusively to the same embodiment, example, or aspect, but may refer to the same embodiment, example, or aspect. Furthermore, in one or more embodiments, one skilled in the art will understand that particular features or characteristics can be combined in any suitable manner.

[0135] Furthermore, while some embodiments, instances, or aspects described herein include some but not all features in other embodiments, instances, or aspects, combinations of features in different embodiments, instances, or aspects are intended to be within the scope of the claims and form different embodiments, as those skilled in the art will understand.

[0136] This invention may be described by the following terms:

[0137] 1. A computer-implemented method 10 for obtaining measurements of a semiconductor structure 30 on a wafer 72, the method comprising:

[0138] a. Obtain a volumetric imaging dataset 22 of the wafer 72, which contains multiple 2D cross-sectional images 24 and 26;

[0139] b. Obtain the outline 32 of the semiconductor structure 30 in the 2D cross-sectional images 24 and 26 of the imaging dataset 22;

[0140] c. Specify one or more measurement specifications 28 for the features of the outline 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 of the imaging dataset 22;

[0141] d. Propagate one or more measurement specifications 28 indicated in the first 2D profile image 24 to another 2D profile image 26 of the imaging dataset 22; and

[0142] e. Measurements of the semiconductor structure 30 are obtained by evaluating one or more measurement specifications 28 in the first 2D profile image 24 and another 2D profile image 26 of the imaging dataset 22.

[0143] 2. The method as described in Clause 1, wherein the measurement specification 28 is derived from a group that includes feature location, feature distance, and feature size.

[0144] 3. The method as described in Clause 1 or 2, wherein the features of the outline 32 of the semiconductor structure 30 include points, lines or curves defined relative to the outline 32 or outline segments of one or more semiconductor structures 30 in the first 2D cross-sectional image 24.

[0145] 4. The method as described in Clause 3, wherein the features of the contour 32 of the semiconductor structure 30 are derived from a group of points comprising the contour 32 or contour segments, a region defined by the contour 32 or contour segments, a contour or a segment of the contour 32, and a centroid 36 of the contour 32 or contour segments.

[0146] 5. The method as described in any of the preceding clauses, wherein in step d, one or more measurement specifications 28 indicated in the first 2D cross-sectional image 24 are propagated to the other 2D cross-sectional image 26 by associating the contour 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 with a corresponding contour 32 of the same semiconductor structure 30 in another 2D cross-sectional image 26, and by associating features of the contour 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 with corresponding features in the associated contour of the semiconductor structure 30 in the other 2D cross-sectional image 26.

[0147] 6. The method as described in Clause 5, wherein the contour 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26 is obtained by applying the contour extraction method to the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26.

[0148] 7. The method as described in Clause 5, wherein the outline 32 of the semiconductor structure 30 in the first 2D profile image 24 and the other 2D profile image 26 is obtained by applying an object detection or image segmentation algorithm to the first 2D profile image 24 and the other 2D profile image 26.

[0149] 8. The method as described in Clause 5, wherein the outline 32 of the semiconductor structure 30 in the first 2D profile image 24 and the other 2D profile image 26 is obtained by applying an instance segmentation algorithm to the first 2D profile image 24 and the other 2D profile image 26.

[0150] 9. The method as described in any one of Clauses 5 to 8, wherein the outline of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26 is represented by the outline point 34.

[0151] 10. The method as described in any one of clauses 5 to 8, wherein the outline of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the further 2D cross-sectional image 26 is represented by a bounding box.

[0152] 11. The method as described in any one of Clauses 5 to 10, wherein associating the contour 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 with the corresponding contour 32 of the same semiconductor structure 30 in the other 2D cross-sectional image 26 includes calculating the mappings of the contour points 34 of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the contour points 34 of the semiconductor structure 30 in the other 2D cross-sectional image 26.

[0153] 12. The method as described in any one of clauses 5 to 10, wherein the outline 32 of the semiconductor structure 30 in the first 2D profile image 24 is associated with a corresponding outline 32 of the same semiconductor structure 30 in the other 2D profile image 26, and a tracking algorithm or optical flow algorithm is applied to the other 2D profile image 26 of the imaging dataset 22 to track the outline 32 of the semiconductor structure 30 in the first 2D profile image 24.

[0154] 13. The method as described in any one of clauses 5 to 10, wherein associating the outline 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 with the corresponding outline 32 of the same semiconductor structure 30 in the other 2D cross-sectional image 26 includes registering the image dataset 22 to a reference imaging dataset with marked outlines.

[0155] 14. The method of any one of clauses 1 to 4, wherein the outline 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26 is obtained by calculating the 3D segmentation of the semiconductor structure 30 in the imaging dataset 24 and calculating the outline 32 of the segmented semiconductor structure in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26 from the 3D segmentation.

[0156] 15. The method as described in Clause 14, wherein in step d, one or more measurement specifications 28 indicated in the first 2D profile image 24 are propagated to the other 2D profile image 26 by associating the contour 32 of the semiconductor structure 30 in the first 2D profile image 24 with the corresponding contour 32 of the same 3D segment of the same semiconductor structure 30 in the other 2D profile image 26, and by associating the features of the contour 32 of the semiconductor structure 30 in the first 2D profile image 24 with the corresponding features in the relevant contour 32 of the semiconductor structure 30 in the other 2D profile image 26.

[0157] 16. The method as described in any one of clauses 5 to 15, wherein at least one measurement specification 28 comprises: a contour point 34 of a contour 32 of a semiconductor structure 30, wherein the contour point 34 is defined by a specific point 38 relative to the contour 32, particularly the centroid 36; and a direction vector 40 indicating the direction of the contour point 34 with respect to the specific point 38 in the first 2D cross-sectional image 24, and wherein associating the contour point 34 in the first 2D cross-sectional image 24 with a corresponding contour point 34 of a related contour 32 in another 2D cross-sectional image 26 comprises calculating the intersection of the related contour 32 and the direction vector 40, the direction vector starting at the specific point 38 of the related contour 32 in the other 2D cross-sectional image 26.

[0158] 17. The method as described in any of the preceding clauses, wherein at least one measurement specification 28 includes calculating the centroid 36 of the profile 32 of the semiconductor structure 30 in one or more 2D cross-sectional images 24, 26, and wherein the centroid 36 is obtained by analyzing the intensity distribution 46, 48 of 1D cross-sections 54, 56 along a region 44 surrounding the profile 32 in the one or more 2D cross-sectional images 24, 26.

[0159] 18. The method as described in any of the preceding clauses, wherein in step d, propagating one or more measurement specifications 28 indicated in the first 2D profile image 24 to another 2D profile image 26 of the imaging dataset 22 includes generating confidence scores indicating the reliability of the relevant measurement specifications 28 in the first 2D profile image 24 and the other 2D profile image 26.

[0160] 19. The method as described in any of the preceding clauses, wherein the defect is detected by detecting outliers in measurements obtained from measurement specifications 28 in the first 2D profile image 24 and the additional 2D profile image 26 of the image dataset 22.

[0161] 20. The method as described in any of the preceding clauses, wherein a target processing volume is obtained, and wherein the number of additional 2D profile images 26 and / or the number of measurement specifications 28 are automatically adapted to meet the target processing volume.

[0162] 21. The method as described in any of the preceding clauses further includes using a user interface 84 configured to allow a user to indicate a measurement specification 28 by selecting one or more features of the contour 32 of the semiconductor structure 30 on the first 2D profile image 24 in the imaging dataset 22, and wherein the user interface 84 is configured to assist the user by calculating modifications to the selected one or more features of the contour 32 of the semiconductor structure 30 during the selection of the one or more features.

[0163] 22. The method described in any of the preceding clauses further includes using a user interface 84 configured to allow a user to load the measurement specification 28 from memory or a database 86 and / or store the measurement specification 28 in the memory or the database 86.

[0164] 23. The method described in any of the preceding clauses further includes using a user interface 84 configured to present a measurement specification 28 to a user, which can be accepted, modified, or rejected by the user, wherein a recommendation for the measurement specification 28 is generated based on the measurement specification 28 previously specified by the user.

[0165] 24. The method described in any of the preceding clauses further includes using a user interface 84 configured to allow a user to specify the measurement specification 28 via natural language processing.

[0166] 25. The method as described in any of the preceding clauses, wherein the imaging dataset 22 is obtained by focused ion beam scanning electron microscopy.

[0167] 26. The method as described in any of the preceding clauses, wherein the chip 72 is a memory chip.

[0168] 27. A computer-readable medium having a computer program stored thereon and executable by a computing device, the computer program comprising code for performing the method 10 as described in any of the preceding clauses.

[0169] 28. A computer program product comprising instructions, wherein when a computer executes the program, the instructions cause the computer to perform the method 10 as described in any of the preceding clauses.

[0170] 29. A system 70 for obtaining measurements of a semiconductor structure 30 on a wafer 72, comprising:

[0171] - Imaging device 74, configured to provide a volumetric imaging dataset 22 containing multiple 2D cross-sectional images 24, 26 of the wafer 72;

[0172] - or multiple processing devices 76;

[0173] - One or more machine-readable hardware storage devices containing instructions executable by the one or more processing devices 76 to perform operations including any of the methods 10 described in the preceding method clauses.

[0174] In summary, the present invention relates to a method 10 for obtaining measurements of a semiconductor structure 30 on a wafer 72, comprising: obtaining a volumetric imaging dataset 22 of the wafer 72, which includes a plurality of 2D cross-sectional images 24, 26; obtaining a contour 32 of the semiconductor structure 30 in the 2D cross-sectional images 24, 26; indicating one or more measurement specifications 28 of features relating to the contour 32 of the semiconductor structure 30 in the first 2D cross-sectional image 24; propagating the one or more measurement specifications 28 indicated in the first 2D cross-sectional image 24 to another 2D cross-sectional image 26; and obtaining measurements of the semiconductor structure 30 by evaluating the one or more measurement specifications 28 in the first 2D cross-sectional image 24 and the other 2D cross-sectional image 26.

[0175] List of reference numerals

[0176] 10 methods

[0177] 12 Imaging Steps

[0178] 14. Contour Generation Steps

[0179] 16 Measurement Standard Procedures

[0180] 18. Dissemination Steps

[0181] 20 Measurement Steps

[0182] 22 Imaging Datasets

[0183] 24 First 2D cross-sectional image

[0184] 26. Additional 2D cross-sectional images

[0185] 28 Measurement Specifications

[0186] 30', 30' semiconductor structure

[0187] 32' and 32' outlines

[0188] 34 contour points

[0189] 36. Center of mass

[0190] 38 specific points

[0191] 40 Direction Vector

[0192] 42 circles

[0193] 44 areas

[0194] Intensity distribution of 46 and 48

[0195] Symmetrical points of 50 and 52

[0196] Sections 54 and 56, 1D

[0197] 58, 60 symmetry axes

[0198] 62 is

[0199] 64 No

[0200] 66 is

[0201] 68 No

[0202] 70 System

[0203] 72 chips

[0204] 74 Imaging Device

[0205] 76. Processing equipment

[0206] 78 processor

[0207] 80 interface

[0208] 82MB memory

[0209] 84 User Interface

[0210] 86 Database

[0211] 88 Visualization Devices

Claims

1. A method (10) for obtaining measurements of a semiconductor structure (30) on a wafer (72), comprising: a. Obtain a volumetric imaging dataset (22) of the wafer (72), which contains multiple 2D profile images (24, 26); b. Obtain the outline (32) of the semiconductor structure (30) in the 2D profile images (24, 26) of the imaging dataset (22); c. Specify one or more measurement specifications (28) for the features of the outline (32) of the semiconductor structure (30) in the first 2D profile image (24) of the imaging dataset (22); d. Propagate the one or more measurement specifications (28) indicated in the first 2D profile image (24) to another 2D profile image (26) of the imaging dataset (22); and e. Measurements of the semiconductor structure (30) are obtained by evaluating one or more measurement specifications (28) in the first 2D profile image (24) and the additional 2D profile image (26) of the imaging dataset (22).

2. The method of claim 1, wherein the measurement specification (28) is derived from a group that includes feature location, feature distance, and feature size.

3. The method of claim 1 or 2, wherein the features of the outline (32) of the semiconductor structure (30) comprise points, lines or curves defined by the outline (32) or outline segment of one or more semiconductor structures (30) in the first 2D cross-sectional image (24).

4. The method of any of the preceding claims, wherein the features of the outline of the semiconductor structure (30) are derived from a group of points comprising the outline (32) or outline segments, regions defined by the outline (32) or outline segments, segments of the outline or outline (32), and centroids (36) of the outline (32) or outline segments.

5. The method as claimed in any of the preceding claims, wherein the measurement specification (28) is constituted by a feature distance, and wherein the features of the contour are constituted by points defined relative to the contour, in particular contour points.

6. The method as claimed in any of the preceding claims, wherein the contour (32) of the semiconductor structure (30) in the first 2D cross-sectional image (24) and the other 2D cross-sectional image (26) is obtained by applying a contour extraction method to the first 2D cross-sectional image (24) and the other 2D cross-sectional image (26).

7. The method as claimed in any of the preceding claims, wherein the outline (32) of the semiconductor structure (30) in the first 2D profile image (24) and the other 2D profile image (26) is obtained by applying an object detection or image segmentation algorithm to the first 2D profile image (24) and the other 2D profile image (26).

8. The method as claimed in any of the preceding claims, wherein the outline (32) of the semiconductor structure (30) in the first 2D profile image (24) and the other 2D profile image (26) is obtained by applying an instance segmentation algorithm to the first 2D profile image (24) and the other 2D profile image (26).

9. The method as claimed in any of the preceding claims, wherein the contour (32) of the semiconductor structure (30) in the first 2D cross-sectional image (24) and the further 2D cross-sectional image (26) is represented by contour points (34).

10. The method as claimed in any of the preceding claims, wherein the outline (32) of the semiconductor structure (30) in the first 2D cross-sectional image (24) and the further 2D cross-sectional image (26) is represented by a bounding box.

11. The method of any of the preceding claims, wherein in step d), the one or more measurement specifications (28) indicated in the first 2D profile image (24) are propagated to the other 2D profile image (26) of the imaging dataset (22) by associating the features of the contour (32) of the semiconductor structure (30) in the first 2D profile image (24) with corresponding features in the contour of the semiconductor structure (30) in the other 2D profile image (26), wherein the one or more measurement specifications are defined with respect to the features of the contour (32) of the semiconductor structure (30) in the first 2D profile image (24).

12. The method as claimed in any of the preceding claims, wherein in step d), the one or more measurement specifications (28) indicated in the first 2D profile image (24) are propagated to the other 2D profile image (26) of the imaging dataset (22) by associating the contour (32) of the semiconductor structure (30) in the first 2D profile image (24) with the corresponding contour (32) of the same semiconductor structure (30) in another 2D profile image (26) by associating the features of the contour (32) of the semiconductor structure (30) in the first 2D profile image (24) with the corresponding features in the associated contour of the semiconductor structure (30) in another 2D profile image (26).

13. The method of claim 12, wherein associating the contour (32) of the semiconductor structure (30) in the first 2D cross-sectional image (24) with the corresponding contour (32) of the same semiconductor structure (30) in the other 2D cross-sectional image (26) comprises calculating a mapping of the contour points (34) of the semiconductor structure (30) in the first 2D cross-sectional image (24) and the contour points (34) of the semiconductor structure (30) in the other 2D cross-sectional image (26).

14. The method of claim 12, wherein the contour (32) of the semiconductor structure (30) in the first 2D profile image (24) is associated with the corresponding contour (32) of the same semiconductor structure (30) in the other 2D profile image (26) and a tracking algorithm or optical flow algorithm is applied to the other 2D profile image (26) of the imaging dataset (22) to track the contour (32) of the semiconductor structure (30) in the first 2D profile image (24).

15. The method of claim 12, wherein associating the contour (32) of the semiconductor structure (30) in the first 2D cross-sectional image (24) with the corresponding contour (32) of the same semiconductor structure (30) in the other 2D cross-sectional image (26) comprises registering the image dataset (22) to a reference imaging dataset having marked contours.

16. The method of any one of claims 1 to 5, wherein the outline (32) of the semiconductor structure (30) in the first 2D cross-sectional image (24) and the other 2D cross-sectional image (26) is obtained by calculating the 3D segmentation of the semiconductor structure (30) in the imaging dataset (24) and calculating the outline (32) of the segmented semiconductor structure in the first 2D cross-sectional image (24) and the other 2D cross-sectional image (26) from the 3D segmentation.

17. The method of claim 16, wherein in step d), one or more measurement specifications (28) indicated in the first 2D profile image (24) are propagated to the other 2D profile image (26) by associating the contour (32) of the semiconductor structure (30) in the first 2D profile image (24) with the corresponding contour (32) of the same 3D segment of the same semiconductor structure (30) in the other 2D profile image (26) and by associating the feature of the contour (32) of the semiconductor structure (30) in the first 2D profile image (24) with the corresponding feature in the relevant contour (32) of the semiconductor structure (30) in the other 2D profile image (26).

18. The method of any one of claims 11 to 17, wherein at least one measurement specification (28) comprises: a contour point (34) of a contour (32) of a semiconductor structure (30), wherein the contour point (34) is defined by a specific point (38) relative to the contour (32), in particular the centroid (36); and a direction vector (40) indicating the direction of the contour point (34) with respect to the specific point (38) in the first 2D cross-sectional image (24), and wherein associating the contour point (34) in the first 2D cross-sectional image (24) with a corresponding contour point (34) of a related contour (32) in another 2D cross-sectional image (26) comprises calculating the intersection of the related contour (32) and the direction vector (40), the direction vector starting at the specific point (38) of the related contour (32) in the other 2D cross-sectional image (26).

19. The method of any of the preceding claims, wherein at least one measurement specification (28) comprises calculating the centroid (36) of the profile (32) of a semiconductor structure (30) in one or more 2D profile images (24, 26), and wherein the centroid (36) is obtained by analyzing the intensity distribution (46, 48) of a 1D profile (54, 56) along a region (44) surrounding the profile (32) in the one or more 2D profile images (24, 26).

20. The method of any of the preceding claims, wherein in step d), propagating one or more measurement specifications (28) indicated in the first 2D profile image (24) to another 2D profile image (26) of the imaging dataset (22) includes generating confidence scores indicating the reliability of the relevant measurement specifications (28) in the first 2D profile image (24) and the other 2D profile image (26).

21. The method as claimed in any of the preceding claims, wherein defects are detected by detecting outliers in measurements obtained from the measurement specifications (28) in the first 2D profile image (24) and the additional 2D profile image (26) of the image dataset (22).

22. The method as claimed in any of the preceding claims, wherein an inspection target is obtained, in particular a target processing volume, and wherein the number of additional 2D profile images (26) and / or the number of measurement specifications (28) are automatically adapted to satisfy the inspection target.

23. The method as claimed in any of the preceding claims, wherein the one or more measurement specifications in the first 2D profile image are indicated by a user interface (84).

24. The method of claim 23, wherein the user interface (84) is configured to allow a user to indicate one or more measurement specifications (28) by selecting one or more features of the outline (32) of the semiconductor structure (30) on the first 2D profile image (24) of the imaging dataset (22).

25. The method of claim 23 or 24, wherein the user interface (84) is configured to assist the user by automatically calculating modifications to selected features of the profile (32) of the semiconductor structure (30) for defining the one or more measurement specifications (28) during the specification of the measurement specifications (28).

26. The method of any of the preceding claims further comprises using a user interface (84) configured to allow a user to indicate a measurement specification (28) by selecting one or more features of the contour (32) of the semiconductor structure (30) on the first 2D profile image (24) in the imaging dataset (22), and wherein the user interface (84) is configured to assist the user by calculating modifications to the selected one or more features of the contour (32) of the semiconductor structure (30) during the selection of the one or more features.

27. The method of any of the preceding claims further includes using a user interface (84) configured to allow a user to load the measurement specification (28) from memory or a database (86) and / or store the measurement specification (28) in memory or a database (86).

28. The method of any of the preceding claims further includes using a user interface (84) configured to present a measurement specification (28) to a user, the measurement specification being acceptable, modifiable or rejected by the user, wherein a suggestion for the measurement specification (28) is generated based on a measurement specification (28) previously indicated by the user.

29. The method of any of the preceding claims further includes using a user interface (84) configured to allow a user to specify the measurement specification (28) via natural language processing.

30. The method of any of the preceding claims, wherein the imaging dataset (22) is obtained by focused ion beam scanning electron microscopy.

31. The method as claimed in any of the preceding claims, wherein the wafer (72) is a memory wafer.

32. The method as claimed in any of the preceding claims, wherein the method (10) for obtaining measurements of the semiconductor structure (30) on the wafer (72) is a computer-implemented method.

33. A computer-readable medium having a computer program stored thereon and executable by a computing device, the computer program comprising code for performing the method (10) as claimed in claim 32.

34. A computer program product comprising instructions that, when executed by a computer, cause the computer to perform the method (10) as claimed in claim 32.

35. A system (70) for obtaining measurements of a semiconductor structure (30) on a wafer (72), comprising: -Imaging device (74), which is configured to provide a volumetric imaging dataset (22) containing 2D cross-sectional images (24, 26) of the wafer (72); - One or more processing devices (76); - One or more machine-readable hardware storage devices containing instructions executable by the one or more processing devices (76) to perform operations comprising any one of the methods (10) as described in the preceding method claims.

Images