Semiconductor wafer 3D volumetric inspection with increased throughput and accuracy

By using milling and imaging of multiple cross-sectional surfaces in the inspection volume of semiconductor wafers to form an average image slice stack, combining moving average values ​​and sparse imaging, the problem of low signal-to-noise ratio and difficult to determine complex structural parameters in the prior art is solved. Achieve 3D volume inspection with high accuracy and high speed.

CN119013757APending Publication Date: 2024-11-22CARL ZEISS SMT GMBH
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202380032436.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2022-04-07
Filing Date
2023-03-28
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The prior art is difficult to achieve high accuracy and high speed in 3D volume inspection of semiconductor wafers, and conventional slicing and image methods require the destruction of the wafer, and the signal-to-noise ratio is low, making it difficult to effectively determine the parameters of complex semiconductor structures.

Method used

By using milling and imaging of multiple cross-sectional surfaces in the inspection volume of semiconductor wafers, an average image slice stack is formed, the moving average value is used to reduce noise, improve image acquisition speed, and optimize the image formation process through sparse imaging and multi-modal image acquisition .

Benefits of technology

The high accuracy and high processing volume of 3D object parameter determination is achieved, which reduces image acquisition time, improves signal-to-noise ratio, avoids chip damage, and enhances the detection ability of complex structures.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119013757A_ABST
    Figure CN119013757A_ABST
Patent Text Reader

Abstract

The present invention relates to a three-dimensional circuit pattern inspection method for an inspection volume at an inspection site of a semiconductor wafer, and more particularly, to a method for determining parameters of a 3D object, such as a HAR structure, in an inspection volume of a semiconductor wafer with increased precision and accuracy, a computer program product and a corresponding semiconductor inspection system. The method employs milling and imaging a plurality of cross-sectional surfaces in an examination volume and forming an average image slice stack therefrom. Thereby, inspection parameters of a 3D object from an average image slice stack are determined with high accuracy and high robustness. The method, computer program product and apparatus can be used for quantitative metering, defect detection, process monitoring, defect inspection and inspection of integrated circuits within semiconductor wafers.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a method for three-dimensional circuit pattern inspection of an inspection volume at an inspection site of a semiconductor wafer, and more particularly to a method, a computer program product and a corresponding semiconductor inspection system for determining parameters of a 3D object (such as a HAR structure) in an inspection volume of a semiconductor wafer with increased precision and accuracy. The method employs milling and imaging of multiple cross-sectional surfaces in the inspection volume and forms an average image slice stack therefrom. Thereby, inspection parameters of the 3D object from the average image slice stack are determined with high accuracy and high robustness. The method, computer program product and apparatus can be used for quantitative metrology, defect detection, process monitoring, defect inspection and inspection of integrated circuits within semiconductor wafers. Background Art

[0002] Semiconductor structures are among the finest man-made structures and have very few defects. These very rare defects are the features that defect detection or defect inspection or quantitative metrology equipment is looking for. The manufactured semiconductor structures are based on prior knowledge, such as from design data, and are manufactured from a limited number of materials and processes. Furthermore, semiconductor structures are manufactured in a series of layers parallel to the surface of the silicon wafer substrate. For example, in logic type samples, metal lines run parallel in the metal layers, while HAR (high aspect ratio) structures and vias run perpendicular to the metal or alternating conductive / non-conductive layers. The angles between metal lines in different layers are 0° or 90°. On the other hand, in the case of VNAND type structures, their cross-sections are known to be circular on average and arranged in a regular raster perpendicular to the silicon wafer surface. During manufacturing, a large number of three-dimensional semiconductor structures are generated in the wafer, where the manufacturing process is affected by several factors. In general, the edge shape, area or overlay position of the semiconductor structure is affected by the nature of the materials involved, the photolithographic exposure or any other manufacturing step involved, such as etching, polishing, deposition or implantation.

[0003] In integrated circuit manufacturing, feature sizes are becoming smaller. Currently, the minimum feature size or critical dimension is less than 10nm, such as 7nm or 5nm, and will approach below 3nm in the near future. Recently, even a minimum feature size of 1nm has been achieved. Therefore, it becomes challenging to measure the edge shape of the pattern with high precision, and to determine the size of the structure or the roughness of the line edge. The measurement resolution of the charged particle system is generally limited by the sampling grating of the individual image points on the sample or the residence time of each pixel, and the diameter of the charged particle beam. The sampling grating resolution can be set in the imaging system and can be adapted to the charged particle beam diameter on the sample. The general grating resolution is 2nm or lower, but the grating resolution limit can be reduced without physical limitations. The charged particle beam diameter has a limited size, which depends on the selected charged particle type, the charged particle beam operating conditions and the charged particle lens system used. The beam resolution is limited to about half of the beam diameter. The resolution can be lower than 3nm, for example, lower than 2nm or even lower than 1nm.

[0004] As the feature size of integrated semiconductor circuits becomes smaller and as the resolution requirements of charged particle imaging systems increase, the inspection and 3D analysis of three-dimensional integrated semiconductor structures in wafers become increasingly challenging. Semiconductor wafers have a diameter of 300 mm and consist of several parts, so-called dies, each of which contains at least one integrated circuit pattern (such as for a memory chip or a processor chip). Semiconductor wafers undergo about 1000 process steps, and more than 100 parallel layers are formed within the semiconductor wafer, including transistor layers, wiring intermediate layers and interconnect layers, and in memory devices there are 3D arrays of multiple memory cells.

[0005] A common method for generating 3D tomographic data at the nanometer (nm) level from a semiconductor sample is the so-called slice and image method, which is performed, for example, by a dual beam device. The slice and image method is described in patent WO2020 / 244795A1. According to the method of WO2020 / 244795A1, a 3D volume inspection is obtained at an inspection sample extracted from a semiconductor wafer. The disadvantage of this method is that the wafer must be destroyed to obtain the inspection sample. As described in patent WO2021 / 180600A1, this disadvantage has been solved by using a slice and image method at an inclined angle into the surface of the semiconductor wafer. According to this method, at least a first inspection site is determined, and a 3D volume image of the inspection volume is obtained by slicing and imaging multiple cross-sectional surfaces of the inspection volume. In the first example of precise measurement, a large number of N cross-sectional surfaces of the inspection volume are generated, where N exceeds 100, or even more average image slices. For example, in a volume with a lateral size of 5 µm and a slice distance of 5 nm, 1000 slices are milled and imaged. For the alignment and registration of cross-sectional averaged image slices, a number of different methods have been proposed. For example, reference markers or so-called fiducials can be used, or feature-based alignment can be used. However, according to recent requirements and in many application examples, it has been shown that these methods need to be further improved for the alignment and registration of multiple cross-sectional image slices.

[0006] According to several inspection tasks, full 3D volumetric images with high accuracy and high throughput are required. In one example, the inspection task is to determine a set of specific parameters of semiconductor objects (e.g. high aspect ratio (HAR) structures) inside the inspection volume with high precision. For fast image acquisition with fast scan time, the signal-to-noise ratio (SNR) is very low. Previous techniques use, for example, long dwell times, i.e. increasing the dwell time per pixel. In other examples, single image acquisition is replaced by fast scanning and adding several images to obtain similar SNR of a single scan with long dwell time. These methods have several disadvantages. First, high SNR image acquisition is very time-consuming, slowing down volumetric image acquisition to only a few inspection sites per hour. Second, the cumulative charging of primary charged particles and the resulting high current of primary charged particles lead to high charging effects in the wafer sample, which deteriorates the image generation. Third, during long image acquisitions, the inspection system will experience significant drift. As the resolution requirements increase (currently below 5 nm), drift becomes more important. In the near future, the demand for resolution will be below 3 nm, below 2 nm or even lower. Therefore, drifts or vibrations already present in the order of a few nm during the image acquisition of the 2D averaged image slice stack can lead to unacceptable shearing.

[0007] Furthermore, in some applications, the method described in WO 2021 / 180600 A1 does not provide sufficient information for determining a set of parameters of a complex semiconductor structure.

[0008] Therefore, the object of the present invention is to provide a further improvement of the slicing and imaging method described in WO 2021 / 180600 A1. In general, the object of the present invention is to provide a wafer inspection method for inspecting three-dimensional semiconductor structures in an inspection volume with higher accuracy, more information and higher speed. Summary of the invention

[0009] These objects are solved by the invention as described in the claims and clauses and by the examples given in the embodiments of the invention.

[0010] This patent application claims priority to U.S. Provisional Patent Application No. 63 / 328,418 filed on April 7, 2022, the disclosure of which is incorporated herein by reference in its entirety.

[0011] The present invention is an improvement in a slice and image method for generating a 3D volume image of an inspection volume of a wafer sample. The study of semiconductor objects includes reconstructing the 3D shape of the semiconductor object of interest from a 3D volume image formed by a stack of 2D average image slices. The stack of 2D average image slices is obtained by imaging a series of cross-sectional surfaces with a charged particle beam imaging device, such as a scanning electron microscope (SEM) or a helium ion microscope (HIM), and by repeatedly milling the cross-sectional surfaces in the inspection volume of the wafer sample with a FIB. A typical stack of 2D average image slices of a 3D volume image may contain between a few hundred average image slices up to more than 1000 average image slices.

[0012] In one example, the charged particle beam imaging device is arranged axially perpendicular to the wafer sample surface, and the 2D averaged image slice contains a cross section of a semiconductor object of interest (e.g., a memory (VNAND) structure such as a high aspect ratio (HAR) channel). In one example, the wafer sample is provided by a whole wafer, and the cross-sectional surface is milled by a FIB configured at an oblique angle to the wafer surface.

[0013] According to a first embodiment of the present invention, a method for forming an image of a stack of M average image slices with increased throughput and accuracy is provided. The method for forming an image of a stack of M average image slices according to the first embodiment utilizes acquisition of a large number of N cross-sectional images through multiple cross-sectional surfaces of an inspection volume by rapid milling and rapid image scanning. According to the first embodiment, a lengthy process of acquiring high-quality images for multiple cross-sectional surfaces is no longer required. Instead, a stack of M high-quality image slices is determined based on a moving average of the first plurality of N cross-sectional images. According to the first embodiment, less image scanning time is used to acquire each cross-sectional image, and a large number of cross-sectional images are generated, each of which has a higher noise level.

[0014] In one example, the number M is less than the number of N cross-sectional images, where M <= N multiplied by a coefficient A, where the coefficient A>=3, such as A=5, A=7, or A=20 or greater. The number of cross-sectional images N is generally between the number of multiple cross-sectional images Q and twice the number of cross-sectional surfaces, where 2xQ=>N=>Q. The image acquisition method of a stack of M average image slices according to the first embodiment relies on general characteristics of the semiconductor object of interest, such as the HAR channel, which is expected to change only slightly in a predetermined direction. By calculating a moving average or average in a predetermined direction for multiple cross-sectional images, noise can be effectively reduced and image acquisition time can be reduced. In addition, by generating more and denser cross-sections with smaller distances between cross-sections, the milling time required to generate cross-sectional surfaces can be more effectively utilized.

[0015] The variation of the semiconductor object of interest is usually limited between each subset of cross-sectional images of the cross-sectional surfaces at small milling distances. Therefore, the alignment and compensation of lateral drift are computationally simple operations, whether by optimally determining the best matching condition or by determining the contrast or edge slope in each average image slice. By acquiring a large number of N cross-sectional images of different cross-sectional surfaces formed in the inspection volume, the drift of the carrier or charged particle beam column can be determined from the cross-sectional images and can be effectively removed to determine a moving average. In one example, multiple cross-sectional surfaces are FIB milled through an inspection volume with a small z-distance of, for example, 2 nm, 3 nm or 5 nm. Each cross-sectional surface is imaged at least once by a fast scanning operation with a charged particle beam imaging device, and for each cross-sectional surface, at least one cross-sectional image is obtained. After or during image acquisition, a moving average of a subset of cross-sectional images is determined. During the determination of the average value, the image contrast is evaluated, for example, by calculating the gradient in the moving average image; if an overall low gradient or low contrast is detected in a particular direction, the determination of the average value is optimized by including the lateral drift in the cross-sectional image. Thus, the effect of the drift in the particular direction is reduced. During optimization of drift compensation, the maximum expected drift can be limited to reduce computation time and increase robustness of the method. Drift can be determined as an absolute drift of each cross-sectional image relative to a reference, or as a relative drift relative to, for example, a previous cross-sectional image. Different registration or alignment methods can be used to increase the speed and performance of drift compensation. In some cases, manual verification of drift compensation can be performed, and erroneous image shifts can be discarded.

[0016] In one example, the texture filling of the cross-sectional surface may vary. For example, the semiconductor structure may include several different stacks or layers of semiconductor features, which are stacked on top of each other. The cross-sectional image may therefore include image fragments of different stacks. For example, by milling at an inclined angle, the cross-sectional surface passes through a first VNAND stack and at least a second VNAND stack. The position of the transition zone between the two stacks gradually changes on the cross-sectional image. Subsequent cross-sectional images share image fragments or areas from the same stack with the same texture (e.g., an area showing a channel cross section of the upper VNAND stack), but also have areas with different textures (e.g., stack transitions). In these cases, the alignment used to determine the lateral offset should be limited to areas showing the same texture so that the moving average is not affected.

[0017] In one example, the formation of M average image slices is achieved by performing a one-dimensional (1D) numerical convolution from a large number of N cross-sectional images with a convolution kernel. In one example, a large number of N cross-sectional images form a 3D data set with abscissas x and y for each image, and different z coordinates correspond to predetermined directions. The moving average or average is described by convolution with a 1D convolution kernel on the z coordinate of the 3D data set. The convolution kernel can be a function of a predetermined number of cross-sectional images, such as a moving rectangular function (rect-function) or a moving Gaussian kernel in the z direction. Other convolution kernels for calculating different weighted moving average point-wise median filters are also feasible. Other examples of convolution kernels correspond to filters for finding the minimum of a norm or matching function according to a given noise model. The width of the convolution kernel can be predetermined or can be adapted according to the local noise level or the required SNR of the generated M average image slices. In other examples, machine learning-based algorithms can be applied to optimize the calculation of the moving average.

[0018] According to a first embodiment, less image scanning time is used to acquire each cross-sectional image, and many cross-sectional images are generated. For each new image scan, any milling artefacts of the previous cross-sectional image are removed, and defects or artifacts of the milling operation vary with different cross-sectional surfaces. According to a first embodiment of the present invention, at least some of these milling artifacts are averaged out as a moving average is formed. The depth of field of a charged particle beam imaging system is quite low, and it is difficult to achieve perfect focus. By averaging multiple cross-sectional images, the focus defects of individual cross-sectional images are averaged, and an average image slice with higher clarity is generated.

[0019] According to a first embodiment of the invention, many cross-sectional images are acquired with reduced scanning time, thereby increasing the noise level. A stack of M average image slices is formed by averaging. Thus, the increased noise level of the rapidly acquired cross-sectional images is reduced, and an average image slice with increased SNR is formed.

[0020] During the determination of the moving average, global drifts can be detected and compensated, while individual deviations of, for example, the trajectory of a HAR channel can still be detected. Further alignment and registration of the cross-sectional images at fiducials or image features present in the semiconductor is also possible. Image acquisition can be performed by exchanging image scanning and milling operations, or by interleaving imaging and milling. The calculation of the average image slice can be performed during image acquisition according to the slice and image method, thereby reducing the memory requirements for storing a plurality of N cross-sectional images.

[0021] The method according to the first embodiment can further be applied to generate training data for inspection methods using machine learning.Annotations can be transferred from an average image slice with a low noise level to a set of cross-sectional images for averaging, each having a large noise level and obtained by fast image scanning.

[0022] According to a second embodiment of the present invention, an image forming method of a stack of M average image slices with increased throughput and accuracy is provided, and the accuracy of image formation is even further improved. According to the second embodiment, image acquisition and milling operations are interleaved and performed simultaneously, and image acquisition of N cross-sectional images and milling of N cross-sectional surfaces are performed simultaneously. In one example, milling with FIB is performed by arranging the FIB beam in the direction of the YZ plane, which has a milling angle GF relative to the y-axis. The FIB scans along a first direction (x direction) perpendicular to the FIB beam. At the same time, image acquisition using an imaging charged particle beam is performed by rapidly scanning the cross-sectional surface in the x direction and stepping in the y direction by an imaging charged particle beam. Thus, a large number of N cross-sectional images are obtained, wherein each cross-sectional image of each interleaved milling and imaging operation contains a first area according to the cross-sectional surface before the actual milling operation, and a second area according to the cross-sectional surface after the actual milling operation, and each cross-sectional image thus contains image areas at different z positions in the cross-sectional image stack. The large number of N cross-sectional images again forms a 3D data set with x and y coordinates for each image and different z coordinates corresponding to the two regions within each cross-sectional image. Thereby, the influence of drift between different cross-sectional images can be reduced even further. The method according to the second embodiment can be combined with the method according to the first embodiment and a stack of M average image slices can be formed by determining a moving average or an average as described in the first embodiment.

[0023] According to an example, other cross-sectional images can be selectively added by, for example, repeating the image scan. For example, a first set of cross-sectional images can be obtained in parallel with the milling operation described in the second example, and a second set of cross-sectional images can be obtained between two consecutive milling operations.

[0024] By limiting the performance of time-consuming image scanning operations to selected regions of interest, the throughput of inspection methods for 3D semiconductor objects of interest in an inspection volume of a wafer sample may be further improved.

[0025] According to a third embodiment of the present invention, the image formation method of the stack of M averaged image slices is further improved and the image formation throughput is even further increased. According to the third embodiment, the time to obtain each cross-sectional image is reduced by limiting the image scan to at least one region of interest (ROI).

[0026] Full-size 3D volume data generation using a slicing and imaging approach provides the most complete information about the inspection volume in the wafer sample. However, the slicing and imaging approach generates an inconveniently large amount of data. On the other hand, a fully reconstructed inspection volume typically contains a large amount of redundant information that is not useful for the specific semiconductor object of interest and is not required for the measurement or set of measurements. At the same time, image quality, resolution and pixel size near the features of interest are typically affected, reducing acquisition / processing time and data volume. It is therefore desirable to be able to adjust imaging and milling parameters to limit image scanning to cross-sections that intersect the area of ​​the volume of interest inside the inspection volume. Using the method according to the third embodiment, the total acquisition time of relevant 3D volume data for the semiconductor object of interest can be even further reduced.

[0027] A method for inspecting a 3D semiconductor object of interest in an inspection volume of a wafer sample according to a third embodiment comprises the following steps: defining at least one first volume of interest in an inspection volume of a semiconductor wafer, the first volume of interest being within the inspection volume. In one example, a region of interest (ROI) is formed by at least one volume of interest inside the inspection volume, and the ROI of subsequent cross-sectional images is determined based on an overlap with the at least one volume of interest. In one example, the volume of interest is buried deep in the inspection volume below the wafer surface.

[0028] The method further comprises the steps of: milling a plurality of cross-sectional surfaces through the inspection volume, and acquiring a plurality of cross-sectional image segments by performing an image scanning operation of the cross-sectional surface segments within the at least one first volume of interest. In one example, the milling is performed at an oblique angle to the wafer surface, and at least the first volume of interest is oriented perpendicular to the wafer surface.

[0029] In one example, the at least first volume of interest is defined according to CAD data. The method may further comprise the step of aligning the at least first volume of interest within the inspection volume according to at least a first image scan of a first cross-sectional surface within the inspection volume. From the first image scan, the exact position of the inspection volume within the wafer can be obtained and registered, and the position of the volume of interest can be accurately determined. In one example, the aligning step comprises further realigning the at least first volume of interest at a second or further image scan of at least a second or further cross-sectional surface within the inspection volume. Thereby, the position of the volume of interest can be accurately maintained even in the presence of, for example, milling deviations.

[0030] For example, in a regularly arranged HAR channel, the region of interest may include only one row or one column of HAR channels. First, by performing a milling operation using a FIB beam, the region of interest becomes accessible for imaging. Thus, a cross-sectional surface covering at least one ROI is formed. After detecting the region of interest, the image acquisition performed by imaging the charged particle beam can be reduced to the at least one ROI. Thus, the time for image acquisition is reduced. The at least one ROI may have a complex shape, such as a cross or T-shape formed by a row and a column of HAR channels. The at least one ROI may also be formed by several separate ROIs, which are distributed on each cross-sectional surface. The ROI may also be a function of the z coordinate or depth of the cross-sectional surface formed in the inspection volume of the wafer, and may follow the 3D shape of the semiconductor object of interest. The z-correlation of the size, shape, and position of the ROI may be determined based on prior information of the semiconductor object of interest. The tracking of unknown semiconductor objects of interest may be guided by model-based assumptions or by using machine learning-based techniques to track the trajectory of structures in the inspection volume.

[0031] In one example, changes in the semiconductor of interest are expected to be confined to certain areas within an inspection volume. This expectation can be triggered by a priori information, which includes CAD information or information obtained from other inspection volumes. According to an example of the method of the third embodiment, the throughput can be increased even further by adaptively varying the distance between two adjacent cross-sectional surfaces. The method thus comprises the following steps: Based on the prior information of the semiconductor object of interest within the at least first volume of interest, adjusting the distance between two adjacent cross-sectional surfaces. In one example, two adjacent volumes of interest are defined and the alignment of different layer stacks of the semiconductor sample can be determined.

[0032] In one example, the at least first volume of interest is determined during milling of a plurality of cross-sectional surfaces into the examination volume. According to the method, the step of defining the at least first volume of interest comprises the following steps:

[0033] a) acquiring a first image scan of a first cross-sectional surface;

[0034] b) selecting a first cross-sectional image segment in the first image scan, the first cross-sectional image segment containing a cross section of the semiconductor object of interest;

[0035] c) milling a second cross-sectional surface into the inspection volume of the wafer sample;

[0036] d) selecting a second cross-sectional image segment by projecting the first cross-sectional image segment onto the second cross-sectional surface;

[0037] e) acquiring a second cross-sectional image segment of the second cross-sectional surface segment;

[0038] f) Repeating steps c) to e) by milling other cross-sectional surfaces and selecting and acquiring other cross-sectional image segments until a predefined interruption criterion is met.

[0039] The first cross-sectional image segment corresponds to a cross section through a volume of interest within the first image scan. In the first image scan, the cross section through the volume of interest is determined by a cross section of the semiconductor object of interest, with a boundary region within the cross section of the semiconductor object of interest. The cross section of the semiconductor object of interest may be determined by image processing, pattern recognition, template matching, or machine learning operations, or a combination of these methods, and by utilizing CAD information.

[0040] In one example, the centroid of the cross section of the semiconductor object of interest corresponds to the centroid of the first cross-sectional image segment. After milling the second or another cross-sectional surface, the first cross-sectional image segment is projected onto the second or another cross-sectional surface, and an image of the second or another cross-sectional surface having the first cross-sectional image segment is obtained. The cross section of the semiconductor object of interest is again determined by image processing, pattern recognition, template matching or machine learning operations, or a combination of these methods, and the centroid of the semiconductor object of interest is determined. If the centroid of the semiconductor object of interest deviates from the center of the first projected cross-sectional image segment, the first cross-sectional image segment is adjusted to form a second or adjusted cross-sectional image segment, wherein the centroid of the semiconductor object of interest coincides with the center of the second, adjusted, projected cross-sectional image segment. This adjustment method is applied from one cross-sectional surface to another cross-sectional surface, and the volume of interest is continuously adjusted through the milling operation and image acquisition of the entire inspection volume.

[0041] In one example, the method further comprises the step of adjusting the position or size of the first cross-sectional image segment according to the second cross-sectional image segment, wherein the cross-sectional position and size of the semiconductor object of interest in the second cross-sectional image segment are analyzed by image processing.

[0042] In one example of step e), projecting the first cross-sectional image segment onto the second cross-sectional surface is performed in a predetermined direction, for example parallel to the direction of the HAR channel. In this example, the predetermined direction is oriented perpendicular to the wafer surface.

[0043] In an example of a method according to the third embodiment, a rapid cross-sectional image scan of a cross-sectional surface is obtained by rapid image scanning, and a cross-sectional image with a reduced SNR is generated. In the cross-sectional image with a low SNR, a region of interest of the volume of interest is determined, and only the region of interest generates an image acquisition with an increased dwell time and an increased SNR. The steps of rapid image acquisition and determining the region of interest in the rapidly acquired cross-sectional image with a low SNR can be repeated.

[0044] In one example, the method of the third embodiment includes the step of generating a 3D volume image of the 3D semiconductor object of interest. The 3D volume image must not be limited to the volume of interest (which includes the 3D semiconductor object of interest), but can be formed for the entire inspection volume. In one example, the 3D volume image of the inspection volume includes a 3D volume image of at least a first volume of interest, and enhanced 3D image data from a priori information outside the at least first volume of interest. The enhanced 3D image data can be obtained from CAD data or from a previously acquired 3D volume image of a reference volume.

[0045] The method according to the third embodiment can be combined with the method according to the first or second embodiment described above. Thereby, forming M average image slices from N cross-sectional images is simplified to determining a moving average within the at least one volume of interest. Furthermore, in an interleaved operation, the first volume of interest can be located in the region of a first z position before the actual milling operation, and the second volume of interest can be located in the region of a second z position after the actual milling operation.

[0046] According to a fourth embodiment of the present invention, the image formation method of the stack of M averaged image slices is further improved, and the processing volume of image acquisition is even further improved. According to the fourth embodiment, a sparse imaging method is applied. The effect of sparse imaging is similar to that of the method according to the third embodiment, but with multiple small ROIs. For example, in multiple HAR channels, image acquisition can usually be reduced to a few HAR channels or even details of the HAR channels. The method according to the fourth embodiment relies on prior knowledge about the semiconductor object of interest in the inspection volume. During the sparse image acquisition process, the image acquisition is reduced to sparse image sampling positions, and each average image slice is filled with enhancement information based on prior knowledge. For example, the enhancement information can be derived based on the CAD model of the semiconductor object of interest, which is modified to best match the image data of the sparse image sampling position. The sparse image sampling position can be predetermined based on prior information, for example, based on the CAD data of the semiconductor object of interest. The method according to the fourth embodiment can be combined with the method according to any one of the first to third embodiments described above, and the processing volume of the wafer inspection task can be improved.

[0047] Using the rapid image acquisition method of the first embodiment of the present invention, the imaging parameters of the charged particle imaging system can be optimized to detect specific semiconductor objects of interest. The optimization may, for example, include selecting a detection method for detecting the interaction products produced at each cross-sectional surface. In some instances, the following disadvantage may occur, namely, the enhancement of the detection of specific semiconductor features is achieved at the expense of other semiconductor features. For example, when imaging a semiconductor memory wafer, the detection of HAR channel boundaries may conflict with the parameters required for accurate detection of word line layers. Alternatively, the detection of accurate HAR channel boundaries may conflict with the ability to detect groove etching boundaries. This problem becomes more serious when destructive milling is performed according to the slice and image method, where it is impossible to reacquire images for cross-sectional surface layers that have been removed.

[0048] According to a fifth embodiment of the present invention, the rapid imaging method of the stack of M average image slices is further improved, and the accuracy of image acquisition is even further improved. The method of the fifth embodiment adopts multimodal image acquisition, including at least the first and second imaging operation modes of the charged particle beam imaging system, to improve the image formation accuracy of the stack of M average image slices. In an example, the change of imaging setting or imaging mode can be used to distinguish the imaging structure by element composition, while being less sensitive to morphological contrast or surface effects. The imaging mode can be modified and combined, and the information in the image scan acquired rapidly is enhanced with information from different imaging modes.

[0049] According to a first example, the detector characteristics are adjusted at least during the imaging mode operation. For example, the detector gain is adjusted to a low signal level, and during the second imaging mode operation, the detector gain is adjusted to a high signal level. Thereby, the signal obtained by the detector is adjusted to, for example, different ROIs in the inspection task. According to another example, the cutoff energy of secondary or backscattered electrons is adjusted. Thereby, for example, imaging can be simplified to certain depths, or certain materials, within the sample below the cross-sectional surface.

[0050] According to a second example, different imaging modes are achieved by using separate or different detectors. For example, the first detector and the second detector can be configured in different angular segments and configured to determine topographic contrast. For example, the first detector can be configured to collect large angle scattered electrons, while the second, in-lens detector can be configured to collect small angle backscattered electrons. For example, the first detector can be an electron detector, while the second detector can be an X-ray detector. Using multimodal image acquisition, the image formation of the stack of M flat image slices can be further improved by obtaining additional information.

[0051] In one example, the method includes the step of changing the detection mode between subsequent image scanning operations, wherein the change in the detection mode includes at least one of a change in dynamic range, a change in interaction product energy range, or a change in interaction product type.

[0052] The method according to the fifth embodiment may be combined with the method for each imaging mode of multimodal imaging according to any one of the first to fourth embodiments. For example, different convolution kernels may be applied to the z stack of the cross-sectional image of each imaging mode. Thus, a high SNR may be achieved in parallel in all imaging modes. Different multimodal imaging modes may also be applied to selected ROIs or to sparse imaging locations. According to the first to fourth embodiments, the additional information of multimodal imaging may be utilized to further enhance the required information related to the semiconductor object of interest and the accuracy of image formation without reducing the throughput increase.

[0053] According to a sixth embodiment, an inspection system for 3D inspection of a semiconductor object of interest in an inspection volume of a wafer is provided. The inspection system is configured to perform a method according to any one of the first to fifth embodiments, including a wafer sample holder and a carrier for holding and positioning a wafer sample. The inspection system includes a dual-beam device, which includes a FIB column and a charged particle beam imaging system, such as a scanning electron microscope (SEM) or a helium ion microscope (HIM). In another example, a correction electron microscope with multiple detectors for angular resolved imaging of interaction products is applied. This correction electron microscope is described in German patent application No. 102021212203.5 filed on October 28, 2021, which is incorporated herein by reference. The dual-column configuration of the dual-beam device is at an angle and forms an intersection of the optical axes of the columns. The carrier is configured to position and hold a sample, which is included in an inspection volume near the intersection. The dual-beam device further includes at least one detector for secondary or backscattered charged particles. In one example, the dual beam device further comprises at least one second detector for acquiring a second signal according to the multimodal imaging operation. The inspection system further comprises a control unit for controlling the operation of the dual beam device. The inspection system further comprises a memory for storing software program code, operating instruction program code, and a memory for storing acquired digital image data. The inspection system further comprises a processor configured to execute processing instructions for determining a moving average. The inspection system is further configured to receive prior information, such as CAD information.

[0054] According to an example, a charged particle imaging system of a dual beam device is configured to perform different imaging modes. Different imaging modes may include different detectors, or detectors configured for different detector settings. The detector may include a grating with a variable gate voltage. Thus, the detector setting may be changed by selecting a repulsive field to cut off low energy charged interaction products. The detector may include a variable analog gain factor, which is configured to adjust the analog signal before conversion to a digital signal. Thus, weak signals with higher accuracy and detail may be obtained.

[0055] An example of a charged particle imaging system includes a scanning deflector and a scanning controller configured to perform different image scanning operations, such as linear averaging, variation of dwell time, and variation of the scanning path, such as a zigzag scanning strategy, or a zigzag scanning strategy.

[0056] An example of a charged particle imaging system configured to perform different imaging modes according to the fifth embodiment includes a charged particle source and multiple particle optical elements configured to vary the acceleration voltage, beam current, charged particle beam angle, numerical aperture of the charged particle beam, or cross-sectional shape of the charged particle beam.

[0057] One example of a charged particle imaging system includes correction components for chromatic and spherical aberrations that allow inspection with higher resolution at lower electron voltages below 1 keV, for example below 500 eV.

[0058] According to an embodiment of the present invention, the acquisition time of a 3D volume image of a semiconductor object of interest by a slicing and imaging method is reduced, and the image accuracy is increased. As a result, the throughput of the inspection task is increased, and an inspection system for 3D inspection of a semiconductor object of interest with increased throughput and increased accuracy is provided. An inspection system for volume inspection of a semiconductor wafer with high accuracy and throughput is provided. The inspection system is configured for milling and imaging of multiple cross-sectional surfaces in an inspection volume, and for determining inspection parameters of a 3D object from multiple cross-sectional surface images. The present invention provides an apparatus and method for 3D inspection of an inspection volume in a wafer with high accuracy and high throughput, and for determining a set of parameters for semiconductor features inside the inspection volume. The method and a system configured to perform the method can be used for quantitative metrology, defect detection, process monitoring, defect inspection, and integrated circuit inspection within a semiconductor wafer. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The present invention described in the examples and embodiments is not limited to the embodiments and embodiments, but can be implemented by those skilled in the art through various combinations or modifications. The present invention will be more fully understood with reference to the following drawings:

[0060] Figure 1A diagram showing an inspection system for 3D volume inspection having a dual beam arrangement is shown.

[0061] Figure 2 A method for volume inspection in a wafer using tilted cross-section milling and imaging by a dual beam setup is described.

[0062] Figure 3 An example of averaging image slices is described.

[0063] Figure 4 A diagram showing a method according to a first embodiment is shown.

[0064] Figure 5 Graph showing the number of cross-sectional surfaces and a data stack of cross-sectional images.

[0065] Figure 6 A first example of calculation of an average image slice is shown.

[0066] Figure 7 A second example of calculation of the average image slice is shown.

[0067] Figure 8 A third example showing the calculation of the average image slice.

[0068] Fig. 9 An example of scanning acquisition of a cross-sectional image in parallel with a milling operation according to the second embodiment is shown.

[0069] Fig.10 Another example of the calculation of the average image slice is shown.

[0070] Fig.11 A method with a limited volume of interest according to a third embodiment is shown.

[0071] Fig.12 A first example of the third embodiment is shown.

[0072] Fig.13 A second example of the third embodiment is shown.

[0073] Fig.14 Another example of the third embodiment is shown, with dynamic adjustment of the region of interest.

[0074] Fig.15 Another example of the third embodiment is shown, in which a determination of overlay or alignment error is performed.

[0075] Fig.16 A method according to a fourth embodiment is shown.

[0076] Fig.17 An example of the fourth embodiment of sparse image acquisition is shown.

[0077] Fig.18 A diagram shows an inspection system for 3D volume inspection with a corrected electron microscope having a dual-beam arrangement. DETAILED DESCRIPTION

[0078] Throughout the drawings and description, the same reference numerals are used to describe the same or similar features or components.The coordinate system is selected so that the wafer surface 55 coincides with the XY plane.

[0079] Recently, the study of 3D inspection volumes in semiconductor wafers has proposed a slicing and imaging method, which is applicable to the inspection volume inside the wafer. Thereby, a 3D volume image is generated at the inspection volume inside the wafer in a so-called "wedge cutting" method or a wedge cutting geometry without removing the sample from the wafer. The slicing and imaging method is applied to inspection volumes with a size of a few micrometers, for example, 5 micrometers to 10 micrometers, or up to 50 micrometers in a wafer with a diameter of 200 mm or 300 mm. V-grooves or wedge-shaped grooves are milled in the top surface of the integrated semiconductor wafer to form a cross section at an inclined angle to the top surface. The 3D volume image of the inspection volume is obtained, for example, at a limited number of measurement positions, such as at representative positions of the bare die, such as at a process control monitor (PCM), or at positions identified by other inspection tools. The slicing and imaging method only locally destroys the wafer, and other bare dies can still be used, or the wafer can still be used for further processing. Methods and inspection systems based on 3D volumetric image generation have been described in patent case WO 2021 / 180600A1, which is incorporated herein by reference in its entirety. The present invention is an improvement and extension of methods and inspection systems based on 3D volumetric image generation, where higher throughput is required. One of the achievements of the invention of the slicing and imaging method for inspecting semiconductor devices is that the imaging time required to obtain multiple average image slices is reduced.

[0080] The present invention is applicable to semiconductor devices having high aspect ratios, and / or composed of semiconductor elements in multiple layers located within the device. The manufacture of such devices strongly relies on the ability to characterize the semiconductor elements in 3D. Full-scale 3D tomography according to the improved method and apparatus of the present invention uses improved slicing and imaging techniques with increased throughput.

[0081] Figure 11 shows an inspection system of the present invention. According to a first embodiment, an improved wafer inspection system 1000 for 3D volume inspection is disclosed. The improved wafer inspection system 1000 for high throughput 3D volume inspection is configured for a slicing and imaging method using a dual beam device 1 under a wedge cut geometry. For a wafer 8, a number of measurement positions are defined in a position map or inspection list generated from an inspection tool or from design information, including measurement positions 6.1 and 6.2. The wafer 8 is placed on a wafer support platform 15. The wafer support platform 15 is mounted on a stage 155 using actuators and position control 21. Actuators for the wafer stage 155 and means for precision control 21, such as laser interferometers, are known in the art. A control unit 16 receives information about the actual position of the wafer stage 155 and is configured to control the wafer stage 155 and adjust the measurement position 6.1 of the wafer 8 at the intersection 43 of the dual beam device 1. The dual beam device 1 comprises a FIB column 50 having a FIB optical axis 48 and a charged particle beam (CPB) imaging system 40 having an optical axis 42. At the intersection 43 of the two optical axes of the FIB and the CPB imaging system, the wafer surface 55 is arranged at an inclination angle GF to the FIB axis 48. The FIB axis 48 and the CPB imaging system axis 42 include an angle GFE. Figure 1 The normal to the wafer surface 55 is represented by the z-axis in the coordinate system of . A focused ion beam (FIB) 51 is generated by a FIB column 50 and impinges on the surface 55 of the wafer 8 at an angle GF. At the inspection position 6.1, an inclined cross-sectional surface is milled into the wafer by ion beam milling at a predetermined y position at about an inclined angle GF, which is controlled by the stage 155 and the position control 21. Figure 1 In the example of , the tilt angle GF is about 30°. Due to beam divergence of a focused ion beam (e.g., a gallium ion beam), or due to variable material properties relative to milling along the cross-sectional surface, the actual tilt angle of the inclined cross-sectional surface may deviate from the tilt angle GF by up to 1° to 4°. Using the charged particle beam imaging system 40, an image of the milled surface can be acquired. Figure 1 In the example of , the charged particle beam imaging system 40 is arranged with its charged particle beam 44 perpendicular to the wafer surface 55 and parallel to the z-axis. In other configurations, the optical axis 42 of the charged particle beam imaging system 40 is arranged at an angle relative to the z-axis.

[0082] During imaging, the scanning unit of the charged particle beam imaging system 40 scans a charged particle beam 44 along a scanning path at the measurement position 6.1 on the wafer cross-sectional surface, and generates secondary particles and backscattered particles. The particle detector 17.1 and the optional internal particle detector 17.2 collect at least some of the secondary particles and / or backscattered particles, and transmit the particle count to the control unit 19. There may also be other detectors for other types of interaction products (such as X-rays or photons). The control unit 19 controls the charged particle beam imaging column 40 and the FIB column 50, and is connected to the control unit 16 to control the position of the wafer mounted on the wafer support platform 15 via the wafer carrier 155. The operation control unit 2 communicates with the control unit 19, which triggers the placement and alignment of the measurement position 6.1 of the wafer 8 at the intersection 43 via the wafer carrier movement, and triggers the repeated operation of FIB milling, image acquisition and carrier movement. The control unit 19 and the operation control unit 2 include a memory for storing instructions in the form of software code, and at least one processor to execute the instructions during operation, for example, to perform the methods described in the first to fourth embodiments. The memory is further configured to store digital image data. The operation control unit 2 may further include a user interface or an interface to other communication interfaces to receive instructions, prior information and transmit inspection results.

[0083] Each new cross-sectional surface is milled by the FIB beam 51 and imaged by a charged particle imaging beam 44 (which may be, for example, a scanning electron beam for a SEM or a helium ion beam for a helium ion microscope (HIM)).

[0084] The operation control unit 2 is configured to perform a 3D inspection within the inspection volume 160 in the wafer 8. The operation control unit 2 is further configured to reconstruct the characteristics of the semiconductor structure of interest from the 3D volume image. In one example, the features and 3D positions of the semiconductor of interest (e.g., the positions of the HAR structures) are detected by an image processing method from, for example, the HAR centroid. 3D volume image generation including image processing methods and feature-based alignment is further described in patent case WO2020 / 244795A1, which is incorporated herein by reference.

[0085] Figure 2 The details of the slicing and imaging method in the wedge-cut geometry are further explained. By repeating the slicing and imaging method in the wedge-cut geometry, a plurality of J cross-sectional average image slices including average image slices of the cross-sectional surfaces 52, 53.i, ..., 53.J are generated, and a 3D volume image of the inspection volume 160 is generated at the inspection position 6.1 of the wafer 8. Figure 2Illustrated is a wedge cut geometry in the example of a 3D memory stack. The cross-sectional surfaces 53.1, ..., 53.J are milled with the FIB beam 51 at an angle GF of about 30° to the wafer surface 55, but other angles GF, for example between GF=20° and GF=60° are possible. Figure 2 The situation is illustrated when the surface 52 is a new cross-sectional surface that was last milled by the FIB 51. The cross-sectional surface 52 is scanned by, for example, the SEM beam 44, and a high-resolution cross-sectional average image slice is generated. The cross-sectional average image slice includes a first cross-sectional image feature formed by the intersection with a high aspect ratio (HAR) structure or a via (for example, the first cross-sectional image feature of the HAR structures 4.1, 4.2 and 4.3), and a second cross-sectional image feature formed by the intersection with the layer L.1, ..., LM, which contains, for example, SiO2, SiN- or tungsten lines. Parts of the lines are also called "word lines". The maximum number of layers M is typically greater than 50, for example greater than 100, or even greater than 200. The HAR structures and layers extend throughout most of the inspection volume in the wafer, but may include gaps. The HAR structures typically have a diameter of about or below 100 nm, for example about 80 nm, or for example 40 nm. The HAR structures are arranged in a regular, for example hexagonal grating, with a pitch of about less than 300 nm, for example even less than 200 nm. Thus, the cross-sectional average image slices contain first cross-sectional image features as intersections or cross-sections of the HAR structure at different depths (Z) at respective XY locations. In the case of a cylindrical vertical memory HAR structure, the first cross-sectional image features obtained are circular or elliptical structures located at different depths determined by the location of the structure on the inclined cross-sectional surface 52. The memory stack extends in the Z direction perpendicular to the wafer surface 55. The thickness d or minimum distance d between two adjacent cross-sectional average image slices is adjusted to a value typically on the order of a few nanometers, such as 30 nm, 20 nm, 10 nm, 5 nm, 4 nm or even less. Once the material layer of a predetermined thickness d is removed by FIB, the next cross-sectional surface 53.i, ..., 53.J is exposed and can be used for imaging using the charged particle imaging beam 44.

[0086] The plurality of J cross-sectional average image slices obtained in this way covers the inspection volume of the wafer 8 at the measuring position 6.1 and is used to form a 3D volume image with a high 3D resolution of, for example, less than 10 nm, preferably less than 5 nm. Figure 2 ) typically has a lateral extension LX=LY=5 μm to 15 μm in the xy plane and a depth LZ of 2 μm to 15 μm below the wafer surface 55. However, the extension can also be greater and amount to, for example, 50 μm.

[0087] Figure 3An average image slice 331.i generated by the imaging charged particle beam 44 and corresponding to the i-th cross-sectional surface 53.i is shown. The average image slice 311.i includes an edge line 315 between the oblique cross section at the edge coordinate yi and the wafer surface 55. To the right of the edge, the average image slice 331.i shows several cross sections 307.1, ..., 307.S through the HAR structure, which intersect the cross-sectional surface 301.i. In addition, the average image slice 331.i includes cross sections 313.1 to 313.3 of several word lines at different depths or z positions.

[0088] According to a first embodiment, an image forming method of a stack of M average image slices with increased throughput and accuracy is provided. A data stack of M high-quality average image slices is determined from the average of a first plurality of N cross-sectional images. According to the first embodiment, less image scanning time is used to acquire each cross-sectional image and generate many cross-sectional images, each of which has a higher noise level. High-quality average image slices are generated by averaging a subset of cross-sectional images using a convolution kernel. The image acquisition method of the stack of M average image slices according to the first embodiment relies on typical characteristics of the semiconductor object of interest, such as the HAR channel, which is expected to change only slightly in a predetermined direction. By calculating a moving average or average in the predetermined direction for many cross-sectional images, noise can be effectively reduced and image acquisition time can be reduced. In addition, the milling time used to generate cross sections can be more efficiently used to generate more and denser cross-sectional images, where there is a smaller distance between cross sections.

[0089] The method according to the first embodiment is not sensitive to errors in the milling and imaging processes. Less image scanning time is used to acquire each cross-sectional image and generate many cross-sectional images, from which average image slices are calculated. For each new image scan, any milling products of the previous cross-sectional image are removed, and defects or products of the milling operation can change with different cross-sectional surfaces. According to the first embodiment of the present invention, at least some of these milling products are averaged out with the formation of a moving average, and polishing of the cross-sectional surface is no longer required. The depth of field of the charged particle beam imaging system is very low, and it is difficult to achieve perfect focus. By averaging multiple cross-sectional images, the focusing defects of each cross-sectional image can be averaged and a higher definition average image slice can be generated, and time-consuming precise focusing can be omitted.

[0090] The method according to the first embodiment is shown in Figure 4 middle.

[0091] In step A1, an examination position 6.i is selected and positioned and is maintained at the intersection 43 of the dual beam device 1. An examination volume 160 is defined and processing parameters for acquiring a 3D volume image are defined. These parameters include, for example, the number Q of cross-sectional surfaces to be milled, the number N of cross-sectional images to be acquired, and the size or extension B of a one-dimensional convolution kernel, and a predetermined direction, in which the one-dimensional convolution kernel is to be applied. Other parameters are the dwell time and other parameters of the image acquisition for acquiring the cross-sectional images. The parameters may also further include a critical value for interrupting or terminating the method.

[0092] In step A2, the slice and image method is performed until a critical value is reached. The critical value is reached if a predetermined control condition is met, such as the appearance of a specific depth within the semiconductor wafer sample or a specific layer within a volume of interest within the semiconductor wafer sample.

[0093] In step A2.1, multiple cross-sectional surfaces are milled by FIB, and N cross-sectional images are acquired by a charged particle beam imaging system. The cross-sectional images are stored in a memory P.

[0094] In optional step A2.3, it is repeatedly assessed whether the examination volume boundary has been reached or the required information has been collected. If a predefined threshold is reached, the slice and image method is aborted.

[0095] The results are as follows Figure 5 As shown, a HAR channel in a 3D memory stack is taken as an example. Figure 5 a shows a plurality of cross-sectional surfaces 53.1 to 53.N passing through an inspection volume 160 having a plurality of HAR structures 309 at a tilt angle GF. Each cross-sectional surface intersects a plurality of HAR structures having cross sections 307. Figure 5 b schematically illustrates an image data stack 163 with a plurality of cross-sectional images 311.1 to 311.N in a memory P, each cross-sectional image 311.i corresponding to an image acquired from one of the N cross-sectional surfaces 53.1 to 53.N. Each cross-sectional image 311.i is aligned, for example after a first alignment, at a reference point or alignment mark (not shown) or by a precision stage.

[0096] In step A2.2, a sequence of subsets of cross-sectional images 311.1 ... N is processed and an average image slice is calculated from each subset of cross-sectional images. During the processing, a moving average is calculated for each subset of cross-sectional images.

[0097] The formation of M average image slices can be described as a mathematical convolution of a 3D data image stack of cross-sectional images with a 1D convolution kernel. In one example, a large number of N cross-sectional images form a 3D data set with abscissas x and y for each image, and different z coordinates corresponding to the predetermined direction. The moving average or average is described by convolving with a 1D convolution kernel on the z coordinate of the 3D data stack. The convolution kernel can be a function of a predetermined number of cross-sectional images, such as a moving rectangular function or a moving Gaussian kernel in the z direction. Other convolution kernels for calculating different weighted moving average point-wise median filters are also feasible. Other examples of convolution kernels correspond to filters for finding the minimum of a norm or matching function according to a given noise model. The width of the convolution kernel can be determined during step A1, or can be adaptive according to the local noise level or the SNR required for the average image slice. For example, if the noise level is too high, the extension B of the convolution kernel can be increased. However, the method is not limited to the classical calculation of convolution using a convolution kernel, but a machine learning-based algorithm can be applied to optimally perform the calculation of the average image slice.

[0098] Figure 6 illustrate Figure 5 A first example of method step A2.2 in an example of . For a first subset 165.1 of four cross-sectional images from i=1 to I=4, an average value is calculated and a first average image slice 331.1 with j=1 is written to the memory S. Then, an average is continued for a second subset 165.2 of four cross-sectional images from i=2 to i=5, and a second average image slice 331.2 with j=2 is written to the memory S. By iterating the method steps until the last subset 165.M, a data stack 167 of average image slices containing j=1 to M average image slices 331.j is finally generated in the memory S.

[0099] In step A3, a semiconductor object of interest in the inspection volume is explored and parameters of the semiconductor object of interest are determined. For example, the semiconductor object of interest is formed by a plurality of repeated semiconductor structures, such as Figure 2 The HAR structure shown in . The parameters may be, for example, volume, lateral position, dimension such as distance or diameter, coverage area, mean value of these features, and maximum deviation from the mean value of the individual structures. The parameters or features of interest are ultimately attributed to the inspection location and stored in the memory S of the control unit 2 or written to the inspection file.

[0100] The data stack 167 of average image slices 331.1 ...M generated according to step A2.2 can be obtained according to different examples. Figure 6In a first example, a moving average is determined and a large number M of average image slices are generated. This operation is equivalent to a mathematical convolution in the direction of index i of the data stack 163 of cross-sectional images 311.1 to 311.N. The convolution kernel may be a weighted sum of the image intensities of each group of cross-sectional images 311 at a given coordinate (x, y). The weights may be the same or different. An instance of a convolution kernel is given by a rectangular function (with equal weights) or a Gaussian function. The width of the convolution kernel (i.e. the number B of each subset of cross-sectional images over which the average is calculated) may be B=3 or greater, preferably greater than 9 or 11. Larger values ​​are also possible, for example B of about 50 or more. By increasing the extension of the average convolution kernel, the time for rapid image acquisition may be even further reduced and the throughput may even be even further increased without increasing the image noise in the average image slice.

[0101] Figure 7 Here, the number M of averaged image slices 311 is further reduced by performing the averaging operation only on every second, every third, or usually every nth subset of cross-sectional images, for example: Figure 7 The number n in is 4. Herein, the extension of the average convolution kernel is denoted by B=5, and each average image slice 331 is calculated from a subset 165.j of B=5 cross-sectional images of the data stack 163. In one example, the value B can be higher. The average convolution kernel of this example still has overlap on the cross-sectional image slices from the data stack 163, so that data from some cross-sectional average image slices are used to calculate two different average image slices 331.j. In this example, the convolution kernel can be a weighted sum with unequal weights, such as a symmetric trigonometric function or a Gaussian function.

[0102] Figure 8 Another example is described. Herein, the number M of average image slices is further reduced, and the subsets of cross-sectional images used for calculating the average image slices no longer show any overlap.

[0103] In the entire example, it is not necessary to obtain the entire data stack 163 of the cross-sectional images 311.1 ... N at once; in one example, the calculation of the average image slices 331.1 ... M is performed in parallel with the image acquisition, and not all cross-sectional images 311.1 ... N must be stored in the memory P. Once the moving average image slice 331.j is calculated, the cross-sectional images 311.i that are no longer needed can be overwritten by the new cross-sectional images. As a result, the memory size of the memory P can be limited.

[0104] In one example, the number M of average image slices is approximately equal to the number N of cross-sectional images, and the number M of average image slices is derived from the average convolution kernel expanded to B. In another example, the number M is significantly smaller than the number N of cross-sectional images, where M <= A x N; the coefficient A >= 3, such as A=5, A=7 or A=20 or more.

[0105] In the above example, cross-sectional average image slices 311.i are obtained from each cross-sectional surface 53.i by fast scanning image acquisition. However, it is also possible that the number N of cross-sectional average image slices 311.1 ... N deviates from the number Q of cross-sectional surfaces 53.1 ... Q. For example, the number N of cross-sectional images is generally between the number Q of a plurality of cross-sectional surfaces 53 divided by 2 and twice the number of cross-sectional surfaces 53, where 2 x Q => N => Q / 2. For example, fast milling is applied, and fast image acquisition is applied only to each second cross-sectional surface, where N < Q. For example, each cross-sectional surface 53.1 ... Q is imaged at least once by a fast scanning operation with a charged particle beam imaging device 40, and for each cross-sectional surface 53.1 ... Q, at least one cross-sectional image 311.1 ... N is obtained, where N>Q.

[0106] Between each subset of cross-sectional images of cross-sectional surfaces of small milling distances, the variation of the semiconductor object of interest is usually very limited. In one example, a plurality of cross-sectional surfaces 53.1 ... Q are milled by the FIB 50 through the inspection volume 160, which have, for example, a small z distance of less than 5 nm, 3 nm, or even less than 2 nm. By acquiring a large number of N cross-sectional images 311.1 ... N of different cross-sectional surfaces 53.1 ... Q formed in the inspection volume 160, the drift of the stage 155 or the charged particle beam imaging column 40 can be determined from the cross-sectional images 311.1 ... N, and effectively removed in the calculation process of the moving mean or average value by determining the best matching condition of the cross-sectional image subset. In one example, the image contrast of the average image slice 331.j is evaluated, for example, by calculating the gradient. If an overall low gradient or low contrast is detected in a particular direction, the determination of the average value is optimized by including a lateral offset to the cross-sectional images of the cross-sectional image subset. Thus, the influence of drift in this particular direction can be reduced.

[0107] During optimization of drift compensation, the maximum expected drift can be limited to reduce computation time and increase robustness of the method. Drift can be determined as an absolute drift for each cross-sectional image relative to a reference, or as a relative drift relative to, for example, a previous cross-sectional image. Different registration or alignment methods can be employed to improve the speed and performance of drift compensation. In some cases, manual verification of drift compensation can be performed and erroneous image shifts can be discarded.

[0108] According to a first embodiment of the present invention, a large number of N cross-sectional images are obtained with a reduced scan time, thereby increasing the noise level. A stack of M average image slices is formed by averaging. Thus, the increased noise level of the rapidly acquired cross-sectional images can be reduced, and an average image slice with an increased SNR is formed.

[0109] During the determination of the moving average, global drifts can be detected and compensated, while individual deviations of the trajectory of, for example, a HAR channel can still be detected. Further alignment and registration of the cross-sectional images at image feature fiducials is also possible. According to the slice and image method, the calculation of the average image slice can be performed during image acquisition, thereby reducing the memory requirements for storing a plurality of N cross-sectional images.

[0110] The method according to the first embodiment can also be applied to generate training data for an inspection method using machine learning. Annotations can be transferred from an average image slice with a low noise level to a set of cross-sectional images for averaging, each image having a large noise level and acquired by rapid image scanning. Therefore, high-precision annotations can be performed on high-quality average image slices, and the annotation information can be transferred to the set of cross-sectional images to train a machine learning algorithm to extract information from cross-sectional images with a large noise level acquired by rapid image scanning.

[0111] According to a second embodiment, the effects of drift are even further reduced. According to a second embodiment, image acquisition of N cross-sectional images and milling of N cross-sectional surfaces are performed simultaneously in parallel. In one example, milling of the FIB is performed by scanning the FIB beam 51 along a first direction (e.g., the x-direction). At the same time, image acquisition is performed by scanning the imaging charged particle beam 44 across the cross-sectional surfaces in the same x-direction and by stepping from line scan to line scan in the y-direction. Fig. 9Interleaved milling and image acquisition at three different times during a milling operation are shown. At a first time t1, only a small portion of the cross-sectional surface 53 is milled by the FIB 51, and at the back end of the milling operation, an image scan 57 is started by scanning in the x direction and stepping in the y direction. Therefore, the image scan at time t1 is performed mostly on the previous cross-sectional surface 61 before milling. At a second time t2, about half of the next cross-sectional surface 63 is milled, and the image scan 57 is performed on the previous cross-sectional surface 61 and the newly milled cross-sectional surface area 63 after milling. At time t3, the milling of the FIB 51 is almost completed. Therefore, each cross-sectional image 311 of each interleaved milling and imaging operation includes a first area 61 according to the cross-sectional surface before the actual milling operation, and a second area 63 according to the cross-sectional surface after the actual milling operation. Areas at different z positions are separated by lines 67. Each cross-sectional image 311 therefore includes image areas at different z positions. The large number of N cross-sectional images again forms a 3D data stack with x and y coordinates for each image and different z coordinates corresponding to the two regions within each cross-sectional image. In this way, the influence of drift between different cross-sectional images can be reduced even further. The method according to the second embodiment can be combined with the method according to the first embodiment and a stack of M average image slices can be formed by determining a moving average or average, as described in the first embodiment and as described in the example of FIG. Fig.10 In the figure, the same reference numerals are used. Fig.10 An example is described. The moving average or mean of the average image slices 331.m of the data stack 167 of average image slices is formed by a subset 165.m of the cross-sectional images of the data stack 163 of cross-sectional images 311.1 ... N. According to an example, the subset 165.m of the cross-sectional images also only contains the parts of the cross-sectional images 311.1 ... N that are located within a specific z range of the convolution kernel used to calculate a specific average image slice 331.m.

[0112] According to an example, more cross-sectional images can be selectively added by, for example, repeated image scanning. For example, a first set of cross-sectional images can be obtained in parallel with the milling operation described in the second example, and a second set of cross-sectional images can be obtained between two subsequent milling operations.

[0113] According to a third embodiment of the present invention, the throughput of a method for inspecting a 3D semiconductor object of interest in an inspection volume of a wafer sample may be further improved by limiting the performance of time-consuming image scanning operations to selected regions of interest. Fig.11 A method according to the third embodiment is described.

[0114] Step A11 is after step A1 according to the first example. In addition to step A1 of the first example, a volume of interest within the inspection volume of the wafer sample is selected. This selection may be performed by external input, such as from CAD data, or by user input from, for example, a graphical user interface. For example, a user selects a volume of interest containing at least one semiconductor object of interest. Step A11 may further include adjusting the distance of the cross-sectional surface according to the selected volume of interest.

[0115] In step A12, the slice and image method is performed until a critical value is reached.

[0116] In optional step A12.0, a first cross-sectional image is acquired and an intersection point between the volume of interest and the first cross-sectional image is determined, thereby recording the position of the volume of interest.

[0117] To determine and select the volume of interest, some "reference" structures of the semiconductor object within the first cross-sectional image may be used, such as the number of word lines or other layers (see Figure 2 ), or other notable features. Determination of the volume of interest may also include using machine learning detection to identify specific features (e.g., transistors, contact points, intersections of two or more components). This determination may be further improved by using fiducials or alignment marks.

[0118] In step A12.1, the next cross-sectional surface is FIB milled, and the region of interest is determined according to the intersection of the cross-sectional surface and the volume of interest.

[0119] In step A12.2, a cross-sectional image is obtained by scanning the region of interest using a charged particle beam imaging system. The cross-sectional image of the region of interest within the volume of interest is analyzed, and if selected, the volume of interest is suitable for the next image acquisition step. If selected, the image scanning of the corrected cross-sectional image of the region of interest within the actual cross-sectional surface is repeated. The final cross-sectional image of the region of interest of the actual cross-sectional surface is stored in the memory S.

[0120] In optional step A12.3, it is evaluated whether the boundaries of the volume of interest have been reached or the required information has been collected. If a predefined threshold is reached, the slice and image method is aborted.

[0121] In step A13, step A3 according to the first example is followed, except that the investigation and determination of the parameters of the semiconductor object of interest is limited to the selected volume of interest.

[0122] Fig.12Schematic illustration of an example of restricted image acquisition. A plurality of cross-sectional surfaces 53.i within an inspection volume 160 are shown, which are subsequently milled to a wafer surface 55 by FIB at an angle GF. The extension of the inspection volume is LX x LY x LZ. Within the inspection volume 160, a volume of interest 171 is defined. In this simplified example, the volume of interest has an extension of LX x LYI x LZ. In this example, the charged particle imaging beam 44 of the imaging system is arranged perpendicular to the cross-sectional surfaces 53 (or perpendicular to the FIB axis 48, see Figure 1 ). Thus, a higher resolution can be achieved during imaging. The acquisition of the cross-sectional image by the charged particle imaging beam 44 is limited to the region of interest 173.i of each cross-sectional surface 53.i (only the first region of interest 173.1 is indicated). The extension of the region of interest in the y direction is thus limited to LYI which is significantly smaller than L. The image scan of the charged particle imaging beam 44 is adjusted for each cross-sectional surface 53.i, and the time interval for image acquisition can be reduced.

[0123] Fig.13 a Illustration of an example of a buried volume of interest used to study HAR channels. In this paper, the volume of interest is bounded by two z-planes, z1 and z2, and is limited to contain only three HAR channels. Fig.13 b shows a top view of five consecutive cross-sectional surfaces 53 and the corresponding regions of interest 173.1 to 173.5. The time interval of the image acquisition is reduced. In an example, lateral drift may occur between the milling of, for example, cross-sectional surfaces 53.3 and 53.4, but it is not critical because the region of interest 173.4 can therefore be adjusted to compensate for any lateral drift.

[0124] Fig.14An example of frequent or dynamic adjustment of a region of interest 173 to cover a volume of interest 171 within an inspection volume 160 is illustrated. The volume of interest of this example is dynamically adjusted in step A12.2 to cover a semiconductor object of interest, such as a HAR channel 309. In a first cross-sectional surface 53.1, a first region of interest 173.1 is determined to cover a semiconductor object of interest (here, a cross section through the HAR channel 309). The region of interest 173.1 is selected to include a cross section through the HAR channel 309 with a specific boundary area to cover the expected lateral position change of the HAR channel 309. A second region of interest 173.2 is then selected based on the first region of interest 173.1, and image slices of the second cross-sectional surface 53.2 are analyzed, such as by an image processing method or a pattern recognition method. In this example, the cross section through the HAR channel 309 is no longer located at the center of the second region of interest 173.2, and the subsequent region of interest 173.3 is adaptively changed to a larger extension and a new center position at the center of the cross section through the HAR channel 309 in the region of interest 173.2. After analyzing the cross-sections through the HAR channel 309 inside the image acquired from the region of interest 173.3, the size of the subsequent region of interest 173.3 may be reduced again, for example, when the cross-sections through the HAR channel 309 have not changed in the cross-sectional surface 53.3 relative to the cross-sectional surface 53.2 through the HAR channel 309. By iteratively adjusting the size or position of the region of interest 173 from one slice to another, the time-consuming operation of scanning image acquisition may be reduced.

[0125] Fig.15 Another example of the third embodiment with two volumes of interest 171.1 and 171.2 is shown. In some applications, the relevant information to be extracted by the inspection task is the spatial relationship between two semiconductor objects of interest 309.1 and 309.2 within the inspection volume 160. In this example, a first average image slice 175.1 of a first region of interest of the first volume of interest 171.1 and a second average image slice 175.2 of a second region of interest of the second volume of interest 171.2 are generated, and the spatial relationship between the semiconductor objects of interest 309.1 and 309.2 can be determined. In this example, the spatial relationship is the alignment or overlay accuracy dy of the two stacks of HAR channels 309.1 and 309.2.

[0126] According to an example, the method according to the third embodiment comprises a further step A14 (see Fig.11 ), according to which the measurement results are displayed to the user via a user display. The measured image data stack is displayed within the examination volume, and the missing image information from outside the volume of interest is enhanced by, for example, CAD information.

[0127] With the third embodiment, an inspection method with increased throughput is provided, which includes automatic adjustment of milling and imaging parameters during the slicing and imaging method, which allows to obtain a data stack of optimized image slices for a specific inspection task (e.g., looking for a specific defect, focusing on a specific IC component or its components, or focusing on a specific location within a semiconductor wafer). The adjustment of the imaging parameters includes limiting the scanning image acquisition to a cross-sectional surface segment within at least one volume of interest, while faster milling will form a larger cross-sectional surface. In one example, image acquisition is performed with high resolution and long dwell time within the volume of interest, and image acquisition is continued with low resolution and short dwell time outside the volume of interest. Thereby, a low-resolution overview image of the inspection volume is generated, which includes a high-resolution volumetric image of the volume of interest. The low-resolution overview image can be enhanced by enhancement information according to step A14.

[0128] The third embodiment may be combined with the method according to the first or second embodiment, or both. Thereby, an inspection method with further increased throughput is provided. The reduction of the time interval for image acquisition according to the third embodiment benefits from limiting the image acquisition to a small volume of interest within the inspection volume. However, in some instances, suppression of the volume of interest is not possible. The method according to the fourth embodiment of the present invention overcomes the problem of volumetric image acquisition of a large volume of interest. According to the fourth embodiment, a sparse image acquisition method is applied. Fig.16 A method according to a fourth embodiment is described in detail.

[0129] Step A21 is after step A11 of the third example. In addition to step A11, step A21 further comprises selecting the density of the sparse image to be acquired. For example, the sparse image acquisition may be performed only on less than 10% of the cross-sectional surface in the inspection volume. For example, the sparse image acquisition may be performed on about 30% or 50% of the cross-sectional surface in the inspection volume.

[0130] In step A22, the slice and image method is performed until a critical value is reached.

[0131] In step A22.1, a first or next cross-sectional surface is milled by FIB.

[0132] In step A22.2, a first or next plurality of sparse image regions for sparse image acquisition are acquired from the cross-sectional surface milled in step A22.1. The sparse images are stored in a memory S.

[0133] In step A22.3, the image regions are registered and a subsequent plurality of sparse image regions is determined. A determination can be obtained from the registered plurality of image regions such that, for example, an even distribution of the sparse image regions over the examination volume is maintained. Thus, coverage of the examination volume with the image regions is maintained. The sparse image regions can also be determined based on a priori information (e.g., CAD information) and can be determined to cover special features of interest within the examination volume.

[0134] In step A24, a 3D volume image of the inspection volume is calculated based on the sparse image data obtained in step A22. The calculation can be performed by a mathematical method or a machine learning method, and the machine learning method is trained to fill in the missing data to supplement the sparse image data. The mathematical method may include interpolation, extrapolation, or averaging. In another example, the missing data used to supplement the sparse image data is obtained from CAD information. For example, a 3D volume image can be derived by fitting a parametric model description of the semiconductor object of interest to the sparse image data.

[0135] In step A23, step A3 according to the first example is followed.

[0136] The fourth embodiment can be combined with the methods according to the first to third embodiments. Fig.17 a illustrates an example of a method according to a fourth embodiment. Inside the examination volume 160, a large volume of interest 171 is defined. A plurality of cross-sectional surfaces 53 are milled (only one is indicated). Fig.17 b represents the five cross-sectional surfaces 53.1 to 53.5 in transverse xy coordinates. A different set of sparse imaging regions 181 is defined for each cross-sectional surface, and the sparse image segments are acquired by a charged particle imaging beam. Based on the sparse image segments with different distributions, a 3D volume image of the volume of interest can be extrapolated, as described in step A24 above.

[0137] One disadvantage of the slice and image approach is the destruction of the wafer sample within the inspection volume 160. After the cross-section surface is milled by removing material above the cross-section surface, additional images of the removed area in the inspection volume cannot be acquired. This can be challenging in some instances of embodiments of the present invention. The fifth embodiment of the present invention overcomes some of these problems with rapid image acquisition by utilizing a multimodal imaging approach.

[0138] The method of the fifth embodiment uses multimodal image acquisition, including the operation of at least the first and second imaging modes of the charged particle beam imaging system to perform image formation of a data stack of cross-sectional images or an average image slice. The images obtained by the different imaging modes can be normalized, and the average image slice is calculated as described in the first embodiment of the present invention. In another example, different sparse image regions according to the fourth embodiment can be obtained using different detector characteristics, and the signals obtained by the detector are adjusted to different semiconductor features. The images obtained by the different imaging modes can complement each other.

[0139] According to a first example, detector characteristics are adjusted for operation in different imaging modes. For example, during operation in a first imaging mode, the detector gain is adjusted to a low signal level, and during operation in a second imaging mode, the detector gain is adjusted to a high signal level. According to another example, the cutoff energy for secondary or backscattered electrons is adjusted. Thus, for example, imaging can be simplified to certain depths within the sample below a cross-sectional surface or certain materials.

[0140] In one example, the method includes the step of changing the detection mode between subsequent image scanning operations, wherein the change in the detection mode includes at least one of a change in dynamic range, a change in interaction product energy range, or a change in interaction product type.

[0141] In one example, the accuracy of a metrology task can be improved by acquiring a series of images in the same field of view to extend the dynamic range of an image sample. Each image in the series is acquired using a different detector setting for a single detector. For example, with a given detector brightness / contrast (or gain / offset) setting, the detector can measure a specific range of charged particle counts. The charged particle counts are converted to, for example, an 8-bit format by an analog-to-digital converter. Therefore, differences in weak signals are lost due to conversion losses. On the other hand, if the detector gain is too high, strong signals will produce detector overflow and differences in strong signals will also be lost. In order to extend the minimum and maximum range of detector electron counts, the offset and gain can be adjusted so that different ranges of signals can be detected with high resolution. Thus, an extension of the SEM signal resolution is achieved.

[0142] According to the second example, different imaging modes are implemented in parallel by utilizing spatially separated or different detectors. For example, the first detector and the second detector may be configured in different angular segments and configured to determine the morphological contrast. For example, the first detector may be configured to collect large-angle scattered electrons, and the second in-lens detector may be configured to collect small-angle backscattered electrons. For example, the first detector may be an electron detector and the second detector may be an X-ray detector. For example, the first detector may be configured to detect low-energy interaction products, and the second detector may be configured to detect high-energy interaction products. The interaction products may be secondary electrons or backscattered charged particles of a charged particle beam imaging system. For the second example with the first and second detectors, the accuracy of the desired information and image formation about the semiconductor object of interest is further improved without reducing the throughput improvement according to the first to fourth embodiments.

[0143] With multimodal image acquisition, the image formation of the stack of M averaged image slices is further improved by acquiring additional information. The method according to the fifth embodiment may be combined with the method according to any one of the first to fourth embodiments.

[0144] In a sixth embodiment of the present invention, an inspection system for 3D inspection of a semiconductor object of interest is described. Figure 1 A first example of an inspection system 1000 for 3D inspection in an inspection volume of a wafer sample 8 is shown. The inspection system 1000 is configured to perform a method according to any one of the first to fifth embodiments. The inspection system 1000 includes a wafer sample support platform 15 and a stage 155 to hold and position the wafer sample 8. The inspection system includes a dual beam device 1, which includes a FIB column 50 and a charged particle beam imaging system 40, such as a scanning electron microscope (SEM) or a helium ion microscope (HIM). The dual beam device 1 further includes at least one detector 17.1 for secondary or backscattered charged particles. In one example, the dual beam device 1 further includes at least one second in-lens detector 17.2 for acquiring a second signal in a multimodal imaging operation according to the fifth embodiment. The inspection system further includes a control unit 19 to control the operation of the dual beam device 1. The control unit 19 is configured to provide instructions to a scanning unit of the charged particle beam imaging system 40, so that the image scanning operation can be limited to a region of interest or a sparse imaging region according to the third or fourth embodiment of the present invention. The inspection system 1000 further comprises an operation control unit 2 having a memory for storing software program codes, operation instruction program codes, and a memory for storing acquired digital image data received from the detectors 17.1 and 17.2. The operation control unit 2 further comprises a processor configured to execute processing instructions for calculating a moving average. The operation control unit 2 of the inspection system 1000 is also configured to receive a priori information, such as CAD information, via an interface. Figure 1 Further details of the inspection system 1000 are provided.

[0145] Fig.18 A second example of an inspection system 1000 for 3D inspection in an inspection volume of a wafer sample 8 is shown, wherein the inspection system 1000 is used in conjunction with Figure 1 The same reference numerals and references are used in the description of Figure 1 Description. In a second example, a charged particle beam imaging system 40 is formed by a corrected electron microscope having a plurality of detectors 17.1 and 17.3 for angularly resolved imaging of applied interaction products. Such a corrected electron microscope is described in international patent application PCT / EP2022 / 079753 filed on October 25, 2022, which is incorporated herein by reference. The corrected electron microscope comprises an electron source 1301, a collimation system 1405, a beam splitter 1500, and a reflector element 1415. Chromatic aberration, spherical aberration, and field curvature are corrected using the reflector element 1415. The corrected electron microscope 40 further comprises an in-lens detector 17.2 and an objective lens 1102 having a deflection scanner 35. Backscattered electrons and secondary electrons 9 are collected by the objective lens 1102 and at least partially guided via a beam splitter to a projection system 1605, to electron detectors 17.1 and 17.3.

[0146] The FIB column 50 and charged particle beam imaging system 40 of the dual beam apparatus 1 are arranged at an angle and form an intersection 43 of the optical axis 48 and the column. The stage 155 is configured to position and hold a wafer sample 8 including an inspection volume proximate to the intersection 43.

[0147] According to an example, the charged particle imaging system of the dual beam device is configured to perform different imaging modes according to the fifth embodiment. Different imaging modes may include different detectors, or detectors configured for different detector settings. The detector may include a grid 1607 with a variable grid voltage. Therefore, the detector setting can be changed by selecting a repulsive field to cut off low-energy charged interaction products. The detector may include a variable analog gain factor configured to adjust the analog signal before conversion into a digital signal. Therefore, weak signals with higher accuracy and detail can be obtained. The detector may include a plurality of detector elements 17.1, 17.3 for angularly resolved detection of interaction products. The charged particle imaging system 40 formed by a corrected electron microscope allows inspection at a lower electron voltage below 1keV, for example below 500eV or even below 300eV, with higher resolution.

[0148] The charged particle imaging system 40 according to the sixth embodiment includes a scanning deflector 35 and a scanning controller 23, which is configured to perform different image scanning operations, such as the execution of line averaging, changes in dwell time, changes in scanning paths, such as a zigzag scanning strategy or a zigzag scanning strategy. The scanning controller 23 is also configured to perform different image scanning operations according to the first to fourth embodiments of the present invention. The different scanning configurations include fast image scanning according to the first embodiment, image scanning operations perpendicular to the FIB beam 51 according to the second embodiment, limited scanning operations limited to at least one region of interest according to the third embodiment, or limited scanning operations configured for image scanning different sets of sparse imaging regions 181 according to the fourth embodiment.

[0149] An example of a charged particle imaging system 40 configured to perform different imaging modes according to the fifth embodiment includes a charged particle source 1301 and particle optical elements 1405 and 1102, which are configured to change the acceleration voltage, beam current, charged particle beam angle, numerical aperture of the charged particle beam, or cross-sectional shape of the charged particle beam.

[0150] Throughout the specification, the invention is described for inspection within an inspection volume of a wafer sample.The wafer sample 8 may be formed by a wafer, for example a 300 mm wafer, or by small wafer sample pieces extracted from a wafer.

[0151] A typical acquisition of an examination volume with a lateral extension of about LX=10 µm x LY=10 µm and a depth of LZ=10 µm requires slicing and imaging of, for example, 2000 cross-sectional image slices with a dwell time of about 2 µs per pixel. Thus, a typical time to acquire a 3D volume image with a high lateral resolution of about 2 nm may exceed 24 hours. Typically, the image acquisition time is between 12 hours and 30 hours. On the other hand, the milling time required for this examination volume is about one hour or even less, almost regardless of the number of cross-sectional surfaces to be milled. Using the method according to an embodiment of the invention, the image acquisition time is significantly reduced, for example at least by a factor of two or more. For example, by averaging a cross-sectional image set comprising at least four rapidly scanned cross-sectional images, the dwell time is reduced by a quarter and the SNR of the cross-sectional image slices is increased. Even if the number of cross-sectional images is increased, for example, by a factor of two, the total image acquisition time is reduced by a factor of two. The dwell time per pixel can be reduced even further, for example below 0.1 µs, and the averaging kernel can be increased to cover, for example, 25 or 30 cross-sectional images. Assuming that the semiconductor object of interest varies slowly in a predetermined direction (e.g., the z-direction perpendicular to the wafer surface), a slight resolution drop in the predetermined direction by averaging is generally acceptable, and an average image slice with a large SNR is obtained. As a result, the reduction in image acquisition time may be a larger factor, such as a factor of 5, 8, or even larger. The noise level is indirectly proportional to the dwell time (i.e., the square root of the dwell time), and by averaging a set of cross-sectional images, at least a similar SNR of an image slice with a larger dwell time is achieved. In addition, fast image scanning with a reduced per-pixel dwell time is less sensitive to charging effects, so the SNR of the average image slice is improved even more.

[0152] According to one example, a volume of interest may be determined and image acquisition may be limited to the region of interest within each cross-sectional surface. According to another example, image acquisition is performed only at sparse regions or segments of the cross-sectional surface. According to both examples, image acquisition time for high quality cross-sectional image slices is reduced by a factor of two or more. In one example, fast image acquisition and averaging is combined with limited regions of interest to even further reduce image acquisition time.

[0153] As described above, image registration or alignment of multiple images for calculating an average image is performed, for example, at a fiducial or alignment feature. Such an alignment feature can be a two-dimensional structure prefabricated by charged particle beam induced sputtering of a metal layer deposition and alignment features in the metal layer. Such an alignment feature typically has a greater extension, such as a line width of 20 nm or greater, such as 50 nm, 100 nm or even greater. Other examples of alignment features include existing features present on a semiconductor wafer, such as alignment marks. Other examples of alignment features include three-dimensional alignment features, such as line structures, crosses or holes milled into the wafer.

[0154] Compared to semiconductor features with dimensions below 10 nm or even smaller (e.g., 5 nm), which typically require an image resolution of 2 nm or less or a sampling distance of, e.g., 1 nm, such larger extensions do not require high-resolution image acquisition. Between the acquisition of each cross-sectional image, drift may produce lateral displacements between consecutive images. Even with the coarse alignment features described above, image-to-image lateral displacements can be determined with high accuracy of less than 1 nm without sacrificing image acquisition speed.

[0155] In a first example according to the first and second embodiments, in each quickly obtained cross-sectional image with a large noise level, an image of the alignment reference point can be obtained at the same high resolution as the semiconductor feature. In order to accurately calculate the center position of the reference point, the pixels covering the coarse reference point can be averaged by a lateral convolution kernel. Therefore, each image is locally averaged at the image area containing the image data of the alignment reference point, and the reference position of the image is obtained with high precision from the locally averaged image data of the alignment reference point. The image noise is reduced by lateral averaging, so that, for example, the center position of the alignment reference point can be obtained with high precision. Then, an average image of the semiconductor feature is obtained by averaging multiple images with high lateral overlap accuracy.

[0156] In a second example according to the third embodiment, fast imaging is obtained by restricting image acquisition to a region of interest. In a third example, multiple sparse image segments of the semiconductor features of interest are obtained in each image scan. In addition to image acquisition of the region of interest or sparse image regions, an image region containing alignment fiducials can be included within each image acquisition. Thereby, images of the region of interest or sparse image regions can be laterally aligned with high precision. In one example, the speed of image acquisition is further increased by reducing the image acquisition time of the image segments of the alignment fiducials. As described above, alignment fiducials are typically larger than the semiconductor features of interest and do not require high resolution. Due to the statistical nature of noise, the time average and the overall average are the same. A low-resolution image of the alignment fiducials can thus be obtained by fast scanning and lateral pixel merging or averaging as described above, or by changing the scan image acquisition of the alignment fiducials to a lower resolution at a lower scan frequency, or averaging several low-resolution images of the alignment fiducials.

[0157] The invention described in the embodiments can be described by the following examples:

[0158] Example 1: A method for volumetric inspection of a 3D semiconductor object of interest in an inspection volume of a wafer sample, comprising:

[0159] Acquiring a first number N of cross-sectional images by fast milling and fast image scanning through a plurality of cross-sectional surfaces of the inspection volume;

[0160] Forming a second number M of average image slices from the first number N of cross-sectional images by calculating a moving average in a predetermined direction of the semiconductor object of interest, wherein the moving average of each average image slice is calculated from a subset of the first number N of cross-sectional images.

[0161] Example 2: The method according to Example 1, wherein the second number M is less than the first number N, where M < N, preferably M <= A x N, where A >= 3, for example A is between 3 and 11, A = 20 or even A is about 50. Thus, for example, any of the following relationships can hold: 3 ≤ A ≤ 11, or 3 ≤ A ≤ 20, or 45 ≤ A ≤ 55.

[0162] Example 3: The method according to Example 1, wherein the second number M is almost equal to the first number N. Preferably, M < N and / or M / N ≤ 0.95, or M / N ≤ 0.98, or M / N ≤ 0.99.

[0163] Example 4: The method according to any one of Examples 1 to 3, comprising sequentially performing fast milling and fast image scanning of a new cross-sectional surface at different times to acquire a cross-sectional image.

[0164] Example 5: A method as described in any one of Examples 1 to 3, comprising performing rapid milling and rapid image scanning of a new cross-sectional surface in parallel at the same time to acquire a cross-sectional image, wherein the scanning direction of the charged particle imaging beam is perpendicular to the direction of the focused ion beam (FIB), and wherein the cross-sectional image includes a first area corresponding to a first cross-sectional surface before the actual rapid milling operation, and a second area corresponding to a second deeper cross-sectional surface according to the actual rapid milling.

[0165] Example 6: The method of any one of Examples 1 to 5, wherein the semiconductor object of interest is at least one HAR channel having a predetermined direction oriented perpendicular to the top surface of the wafer sample.

[0166] Example 7: The method as described in any one of Examples 1 to 6, comprising performing the fly milling at an inclined angle to the top surface of the sample, wherein the inclined angle GF is equal to or less than 45°, for example, equal to or less than 30°.

[0167] Example 8: The method as described in any one of Examples 1 to 7 comprises calculating a moving average of the second number of cross-sectional images by convolving the cross-sectional image with a one-dimensional convolution kernel parallel to the predetermined direction.

[0168] Example 9: The method of Example 8, wherein the one-dimensional convolution kernel extends over at least B=3 consecutive cross-sectional images, preferably between B=3 and B=20 consecutive cross-sectional images, and has the form of a weighted sum, a rectangular function, or a Gaussian distribution function. Preferably, the following relationship holds: 3 ≤ B ≤ 20.

[0169] Example 10: The method as described in Example 8 or 9 includes selecting the extension B of the one-dimensional convolution kernel according to the noise level in the cross-sectional image to obtain a predetermined signal-to-noise ratio (SNR) in the average image slice, wherein a larger extension B for higher noise levels is, for example, B=11, B=20 or higher.

[0170] Example 11: The method as described in any one of Examples 1 to 10 includes compensating for lateral drift of the cross-sectional images of the plurality of cross-sectional images during calculation of the moving average, wherein the lateral drift direction is perpendicular to the predetermined direction.

[0171] Example 12: A method as described in Example 11, comprising determining the lateral drift in the cross-sectional image in the average image slice based on the low gradient or low contrast in the drift direction, and optimizing the calculation of the average image slice with maximum isotropic contrast or gradient by averaging at least one cross-sectional image and including the lateral drift.

[0172] Example 13: The method of Example 11 further comprises determining a lateral drift relative to a reference position obtained from the image segments of the alignment fiducials.

[0173] Example 14: The method of Example 13 further comprises horizontal averaging by pixel binning or horizontal convolution of the image segments of the aligned reference points with an averaging convolution kernel.

[0174] Example 15: The method as described in any one of Examples 1 to 14, wherein the method further comprises the following step: adjusting the distance between two adjacent cross-sectional surfaces according to previous information of the semiconductor object of interest within the at least first volume of interest.

[0175] Example 16: The method as described in any one of Examples 1 to 15 further comprises the following step: adjusting a milling angle of a subsequently milled cross-sectional surface according to previous information of the semiconductor object of interest within the at least first volume of interest.

[0176] Example 17: The method as described in any one of Examples 1 to 16 further includes the following steps: changing the detection mode between subsequent rapid image scans, wherein the change in the detection mode includes at least one of a change in dynamic range, a change in energy range of interaction products, or a change in type of interaction products.

[0177] Example 18: A method for inspecting a 3D semiconductor object of interest in an inspection volume of a wafer sample, comprising:

[0178] defining at least a first volume of interest in an inspection volume of a semiconductor wafer, the first volume of interest being within the inspection volume;

[0179] milling a plurality of cross-sectional surfaces through the inspection volume;

[0180] A plurality of cross-sectional image segments are acquired by performing an image scanning operation of a cross-sectional surface segment within the at least first volume of interest, wherein each cross-sectional image segment is determined according to a region of interest formed by an intersection of a cross-sectional surface and the at least first volume of interest.

[0181] Example 19: The method of Example 18, comprising performing milling at an oblique angle to the wafer surface, and wherein the at least first volume of interest is oriented perpendicular to the wafer surface.

[0182] Example 20: The method of Example 18 or 19, comprising defining the at least first volume of interest based on CAD data.

[0183] Example 21: The method as described in any one of Examples 18 to 20 further comprises the following step: aligning the at least first volume of interest within the inspection volume based on at least a first image scan of a first cross-sectional surface within the inspection volume.

[0184] Example 22: The method of Example 21, wherein the aligning step comprises further realigning the at least first volume of interest at at least second or further image scans of second or further cross-sectional surfaces within the inspection volume.

[0185] Example 23: The method of any one of Examples 18 to 19, wherein the step of defining the at least first volume of interest comprises the following steps:

[0186] acquiring a first image scan of a first cross-sectional surface;

[0187] selecting a first cross-sectional image segment in the first image scan, the first cross-sectional image segment including a cross-section of a semiconductor object of interest;

[0188] milling a second cross-sectional surface into the inspection volume of the wafer sample;

[0189] selecting a second cross-sectional image segment by projecting the first cross-sectional image segment onto the second cross-sectional surface;

[0190] Acquire a second cross-sectional image segment of the second cross-sectional surface segment;

[0191] Steps c) to e) are repeated by milling further cross-sectional surfaces and selecting and acquiring further cross-sectional image segments until a predefined interruption criterion is met.

[0192] Example 24: The method as described in Example 23, wherein, in step e), projecting the first cross-sectional image segment onto the second cross-sectional surface is performed along the predetermined direction.

[0193] Example 25: The method as described in Example 23 or 24 further includes the following steps: adjusting the position and size of the first cross-sectional image segment according to the second cross-sectional image segment, wherein the cross-sectional position and size of the semiconductor object of interest within the second cross-sectional image segment are analyzed through image processing.

[0194] Example 26: The method of any one of Examples 18 to 25, wherein the semiconductor object of interest is a HAR channel, and the predetermined direction is oriented perpendicular to the wafer surface.

[0195] Example 27: The method as described in any one of Examples 18 to 26 further includes the following step: generating a 3D volume image of the 3D semiconductor object of interest.

[0196] Example 28: The method of Example 27, wherein the step of generating a 3D volume image of the 3D semiconductor object of interest comprises the step of generating a 3D volume image of the at least first volume of interest, and the step of generating enhanced 3D image data outside the at least first volume of interest.

[0197] Example 29: A method as described in Example 28, wherein the enhanced 3D image data is obtained from CAD data.

[0198] Example 30: The method as described in any one of Examples 18 to 29 further comprises the following step: adjusting the distance between two adjacent cross-sectional surfaces according to previous information of the semiconductor object of interest within the at least first volume of interest.

[0199] Example 31: The method as described in any one of Examples 18 to 30 further comprises the following step: forming a plurality of image slice segments from the plurality of cross-sectional image segments by determining a moving average along a predetermined direction.

[0200] Example 32: The method as described in any one of Examples 18 to 31 further includes the following steps: changing the detection mode between two subsequent image scanning operations, wherein the change in the detection mode includes at least one of a change in dynamic range, a change in energy range of interaction products, or a change in type of interaction products.

[0201] Example 33: A method for inspecting a 3D semiconductor object of interest in an inspection volume of a wafer sample, comprising:

[0202] milling a plurality of cross-sectional surfaces through the inspection volume;

[0203] Selecting multiple sparse regions for each cross-sectional surface;

[0204] acquiring a plurality of sparse cross-sectional image segments by performing an image scanning operation of the plurality of sparse regions of each cross-sectional surface;

[0205] A 3D volumetric image of the examination volume is generated from the plurality of sparse cross-sectional image segments.

[0206] Example 34: The method of Example 33, wherein the step of generating the 3D volume image comprises image interpolation between the plurality of sparse cross-sectional image segments.

[0207] Example 35: The method of Example 34, wherein the image interpolation between the plurality of sparse cross-sectional image segments is based on CAD data of the 3D semiconductor object of interest.

[0208] Example 36: The method of Example 33, wherein the selection of the plurality of sparse regions for each cross-sectional surface varies from cross-sectional surface to cross-sectional surface.

[0209] Example 37: The method of any one of Examples 33 to 36, wherein the selection of the plurality of sparse regions of each cross-sectional surface is performed based on image analysis of a plurality of previously acquired sparse cross-sectional image segments.

[0210] Example 38: The method of any one of Examples 33 to 37, wherein the selection of the plurality of sparse regions is performed based on CAD data.

[0211] Example 39: The method of any one of Examples 33 to 38, wherein the milling is performed at an oblique angle to the wafer surface.

[0212] Example 40: A method as described in any one of Examples 33 to 39, wherein the method further comprises the following step: adjusting the distance between two adjacent cross-sectional surfaces based on previous information of the semiconductor object of interest.

[0213] Example 41: A method as described in any one of Examples 33 to 40, further comprising the following steps: changing the detection mode between two image scanning operations, wherein the change in the detection mode includes at least one of a change in dynamic range, a change in energy range of interaction products, or a change in the type of interaction products.

[0214] Example 42: The method of any one of Examples 33 to 41 further comprises compensating for lateral drift of each of the plurality of sparse regions of each cross-sectional surface.

[0215] Example 43: The method of Example 42, comprising determining a lateral drift relative to a reference position obtained from an image segment aligned with a fiducial.

[0216] Example 44: The method of Example 43 further comprises obtaining an image segment of the alignment reference point having a lower resolution.

[0217] Example 45: A method for inspecting a 3D semiconductor object of interest in an inspection volume of a wafer sample, comprising:

[0218] milling a plurality of cross-sectional surfaces through the inspection volume;

[0219] scanning each cross-sectional surface using an imaging charged particle beam, wherein interaction products are generated during the scan and include secondary electrons, backscattered charged particles, photons, or X-rays in different energy ranges;

[0220] A plurality of cross-sectional images are formed from the detection signals of at least the first and a second interaction product.

[0221] Example 46: The method of Example 45, wherein the first interaction product and the second interaction product are selected according to any of the following:

[0222] a) selected first and second energy ranges of interaction products; or

[0223] b) a first and a second dynamic range of a selected interaction product; or

[0224] c) First and second types of interaction products.

[0225] Example 47: The method of Example 45 or 46, wherein during the step of forming the plurality of cross-sectional images, the cross-sectional images obtained from the first detector are enhanced by information received from the second detector.

[0226] Example 48: The method of Example 45 or 46, wherein during the step of forming the plurality of cross-sectional images, the cross-sectional images obtained from the first image scan are enhanced by information received from the second image scan.

[0227] Example 49: A semiconductor inspection apparatus for inspection of a 3D semiconductor object of interest in an inspection volume of a wafer sample, comprising:

[0228] A wafer sample stage, used for holding and positioning the wafer sample;

[0229] a dual-column microscope comprising a focused ion beam (FIB) column and a charged particle beam imaging column, forming a common intersection of the optical axis of the FIB column and the optical axis of the charged particle beam imaging column;

[0230] a control unit configured to control the dual-column microscope to perform a sectioning and imaging method in the examination volume;

[0231] A processor and a memory with software code installed, the processor being configured to perform any of the above method steps.

[0232] Example 50: The semiconductor inspection apparatus of Example 49 further comprises at least first and second detectors configured to perform multimodal image acquisition.

[0233] Example 51: The semiconductor inspection apparatus as described in Example 49 or 50, wherein the charged particle beam imaging column is formed by a correction electron microscope having a correction member for correcting chromatic aberration and spherical aberration.

[0234] However, the present invention described in the examples and embodiments is not limited to the above examples but can be implemented by those skilled in the art through various combinations or modifications thereof.

[0235] A list of reference numerals is provided below:

[0236] 1Dual beam device

[0237] 2Operation control unit

[0238] 4 Cross-sectional image features

[0239] 6 Measurement location

[0240] 8 wafer samples

[0241] 15 Wafer support platform

[0242] 16-stage control unit

[0243] 17 Secondary or backscattered electron detector

[0244] 19Control Unit

[0245] 21 Position Sensor

[0246] 23 Scan Controller

[0247] 40 Charged Particle Beam (CPB) Imaging System

[0248] 42 Optical axis of the imaging system

[0249] 43 intersection

[0250] 44 Imaging Charged Particle Beams

[0251] 48Fib optical axis

[0252] 50FIB column

[0253] 51 Focused Ion Beam

[0254] 52 cross-sectional surface

[0255] 53 cross-sectional surface

[0256] 55 Wafer top surface

[0257] 57 Scanning path of imaging beam

[0258] 61Surface image area before milling

[0259] 63 Surface image area after milling

[0260] 67 Lines or steps between different areas

[0261] 155 wafer stage

[0262] 160 Check volume

[0263] 163 Image Data Stack

[0264] 165 cross-sectional image subsets

[0265] 167 image slice stack

[0266] 171Follow volume

[0267] 173 Areas of Interest

[0268] 175 Average images of the region of interest

[0269] 181 Sparse area of ​​image acquisition

[0270] Measured cross-sectional image of the 307HAR structure

[0271] 309HAR Structure

[0272] 311 cross-sectional image slices

[0273] 313 Cross-section through word line

[0274] 315 with surface edge

[0275] 331 Average image slices

Claims

1. A method for volumetric inspection of a 3D semiconductor object of interest in an inspection volume of a wafer sample, comprising: - acquiring a first number N of cross-sectional images by fast milling and fast image scanning through a plurality of cross-sectional surfaces of the examination volume; - forming a second number M of average image slices from the first number N of cross-sectional images by calculating a moving average in a predetermined direction of the semiconductor object of interest, wherein the moving average of each average image slice is calculated from a subset of the first number N of cross-sectional images. 2 . The method as claimed in claim 1 , wherein the second number M is smaller than the first number N, wherein M<N, in particular M≤A×N, wherein the factor A≥3.

3. The method of claim 1, wherein the second number M is almost equal to the first number N, in particular M < N and M / N ≤ 0.

95. 4 . The method of claim 1 , further comprising sequentially performing rapid milling of a new cross-sectional surface and rapid image scanning at different times to acquire a cross-sectional image.

5. The method of any one of claims 1 to 3, further comprising performing rapid milling and rapid image scanning of a new cross-sectional surface in parallel at the same time to acquire a cross-sectional image, wherein a scanning direction of the charged particle imaging beam is perpendicular to a direction of a focused ion beam (FIB), and wherein the cross-sectional image comprises a first region corresponding to a first cross-sectional surface before an actual rapid milling operation; and a second region corresponding to a second deeper cross-sectional surface according to the actual rapid milling operation.

6. The method of any one of claims 1 to 5, wherein the semiconductor object of interest is at least one HAR channel having a predetermined direction oriented perpendicular to a top surface of the wafer sample. 7 . The method according to claim 1 , further comprising performing the fly milling at an inclined angle to the top surface of the sample, wherein an inclined angle GF ≤ 45°, in particular GF ≤ 30°. 8 . The method according to claim 1 , further comprising calculating a moving average of the second number of cross-sectional images by convolving the cross-sectional images with a one-dimensional convolution kernel parallel to the predetermined direction.

9. The method of claim 8, wherein the one-dimensional convolution kernel extends over at least B=3 consecutive cross-sectional images and has a form of a weighted sum, a rectangular function or a Gaussian distribution function.

10. The method of claim 8 or 9, further comprising selecting an extension B of the one-dimensional convolution kernel according to a noise level in the cross-sectional image, so as to obtain a predetermined signal-to-noise ratio (SNR) level in an average image slice. 11 . The method according to claim 1 , further comprising compensating for a lateral drift of the cross-sectional images of the plurality of cross-sectional images during calculation of the moving average, wherein the lateral drift direction is perpendicular to the predetermined direction.

12. The method of claim 11, further comprising determining the lateral drift in the cross-sectional image in the average image slice based on the low gradient or low contrast in the drift direction, and optimizing the calculation of the average image slice with maximum isotropic contrast or gradient by averaging at least one cross-sectional image and including the lateral drift.

13. The method of claim 11, further comprising determining the lateral drift relative to a reference position obtained from an image segment aligned with a fiducial.

14. The method of claim 13, further comprising horizontal averaging by pixel binning or horizontal convolution of the image segments of the aligned reference points with an averaging convolution kernel.

15. The method of any one of claims 1 to 14, wherein the method further comprises the step of adjusting a distance between two adjacent cross-sectional surfaces according to previous information of the semiconductor object of interest within the at least first volume of interest.

16. The method of any one of claims 1 to 15, further comprising the step of adjusting a milling angle of a subsequently milled cross-sectional surface based on previous information of the semiconductor object of interest within the at least first volume of interest.

17. The method as described in any one of claims 1 to 16 further comprises the following steps: changing the detection mode between subsequent rapid image scans, wherein the change in the detection mode comprises at least one of a change in dynamic range, a change in energy range of interaction products, or a change in type of interaction products.

18. A method for inspecting a 3D semiconductor object of interest in an inspection volume of a wafer sample, comprising: - defining at least a first volume of interest in an inspection volume of the semiconductor wafer, the first volume of interest being within the inspection volume; - milling a plurality of cross-sectional surfaces through the inspection volume; - acquiring a plurality of cross-sectional image segments by performing an image scanning operation of a cross-sectional surface segment within the at least first volume of interest, wherein each cross-sectional image segment is determined according to a region of interest formed by the intersection of the cross-sectional surface and the at least first volume of interest.

19. The method of claim 18, comprising performing milling at an oblique angle to a wafer surface, and wherein the at least first volume of interest is oriented perpendicular to the wafer surface.

20. The method of claim 18 or 19, comprising defining the at least first volume of interest based on CAD data.

21. The method of any one of claims 18 to 20, further comprising the step of aligning the at least first volume of interest within the inspection volume based on at least a first image scan of a first cross-sectional surface within the inspection volume.

22. The method of claim 21, wherein the aligning step comprises further realigning the at least first volume of interest at at least second or further image scans of second or further cross-sectional surfaces within the examination volume.

23. The method of any one of claims 18 to 19, wherein the step of defining the at least first volume of interest comprises the steps of: a) acquiring a first image scan of a first cross-sectional surface; b) selecting a first cross-sectional image segment in the first image scan, the first cross-sectional image segment comprising a cross section of the semiconductor object of interest; c) milling a second cross-sectional surface into the inspection volume of the wafer sample; d) selecting a second cross-sectional image segment by projecting the first cross-sectional image segment onto the second cross-sectional surface; e) acquiring a second cross-sectional image segment of the second cross-sectional surface segment; f) Repeating steps c) to e) by milling other cross-sectional surfaces, selecting and acquiring other cross-sectional image segments until a predefined interruption criterion is met. 24 . The method of claim 23 , wherein in step e), projecting the first cross-sectional image segment onto the second cross-sectional surface is performed along the predetermined direction.

25. The method of claim 23 or claim 24, further comprising the following steps: adjusting the position and size of the first cross-sectional image segment according to the second cross-sectional image segment, wherein the cross-sectional position and size of the semiconductor object of interest within the second cross-sectional image segment are analyzed by image processing.

26. The method of any one of claims 18 to 25, wherein the semiconductor object of interest is a HAR channel, and the predetermined direction is oriented perpendicular to a wafer surface.

27. The method of any one of claims 18 to 26, further comprising the step of generating a 3D volume image of the 3D semiconductor object of interest.

28. The method of claim 27, wherein the step of generating a 3D volume image of the 3D semiconductor object of interest comprises the step of generating a 3D volume image of the at least first volume of interest, and the step of generating enhanced 3D image data outside the at least first volume of interest.

29. The method of claim 28, wherein the enhanced 3D image data is obtained from CAD data.

30. The method of any one of claims 18 to 29, further comprising the step of adjusting a distance between two adjacent cross-sectional surfaces according to previous information of the semiconductor object of interest within the at least first volume of interest.

31. The method of any one of claims 18 to 30, further comprising the step of forming a plurality of image slice segments from the plurality of cross-sectional image segments by determining a moving average along a predetermined direction.

32. The method as described in any one of claims 18 to 31 further comprises the following steps: changing the detection mode between two subsequent image scanning operations, wherein the change in the detection mode comprises at least one of a change in dynamic range, a change in energy range of interaction products, or a change in type of interaction products.

33. A method for inspecting a 3D semiconductor object of interest in an inspection volume of a wafer sample, comprising: - milling a plurality of cross-sectional surfaces through the examination volume; -selecting a plurality of sparse regions for each cross-sectional surface; - acquiring a plurality of sparse cross-sectional image segments by performing an image scanning operation of the plurality of sparse regions of each cross-sectional surface; - generating a 3D volume image of the examination volume from the plurality of sparse cross-sectional image segments.

34. The method of claim 33, wherein the step of generating the 3D volumetric image comprises image interpolation between the plurality of sparse cross-sectional image segments.

35. The method of claim 34, wherein the image interpolation between the plurality of sparse cross-sectional image segments is based on CAD data of the 3D semiconductor object of interest.

36. The method of claim 33, wherein the selection of the plurality of sparse regions for each cross-sectional surface varies with different cross-sectional surfaces.

37. The method of any one of claims 33 to 36, wherein the selection of the plurality of sparse regions of each cross-sectional surface is performed based on image analysis of a plurality of previously acquired sparse cross-sectional image segments.

38. The method of any one of claims 33 to 37, wherein the selection of the plurality of sparse regions is performed based on CAD data.

39. A method as claimed in any one of claims 33 to 38, wherein the milling is performed at an oblique angle to the wafer surface.

40. The method of any one of claims 33 to 39, wherein the method further comprises the step of adjusting the distance between two adjacent cross-sectional surfaces according to previous information of the semiconductor object of interest.

41. The method as described in any one of claims 33 to 40 further comprises the following steps: changing the detection mode between two image scanning operations, wherein the change in the detection mode comprises at least one of a change in dynamic range, a change in energy range of interaction products, or a change in type of interaction products.

42. The method of any one of claims 33 to 41, further comprising compensating for lateral drift of each of the plurality of sparse regions of each cross-sectional surface.

43. A method as claimed in claim 42, comprising determining a lateral drift relative to a reference position obtained from image segments aligned with the fiducials.

44. The method of claim 43, further comprising obtaining an image segment of the alignment fiducial having a lower resolution.

45. A method for inspecting a 3D semiconductor object of interest in an inspection volume of a wafer sample, comprising: - milling a plurality of cross-sectional surfaces through the examination volume; - scanning each cross-sectional surface using an imaging charged particle beam, wherein during the scan interaction products are generated which include secondary electrons, backscattered charged particles, photons or X-rays in different energy ranges; - forming a plurality of cross-sectional images from the detection signals of at least the first and second interaction products.

46. ​​The method of claim 45, wherein the first interaction product and the second interaction product are selected based on any of the following: -a) a first and a second energy range of selected interaction products; or -b) a first and a second dynamic range of a selected interaction product; or -c) First and second types of interaction products.

47. A method as claimed in claim 45 or 46, wherein during the step of forming the plurality of cross-sectional images, the cross-sectional images obtained from the first detector are enhanced by information received from the second detector.

48. The method of claim 45 or 46, wherein during the step of forming the plurality of cross-sectional images, the cross-sectional images obtained from the first image scan are enhanced by information received from the second image scan.

49. A semiconductor inspection apparatus for inspecting a 3D semiconductor object of interest in an inspection volume of a wafer sample, comprising: - Wafer sample stage, used to hold and position the wafer sample; - a dual-column microscope comprising a focused ion beam (FIB) column and a charged particle beam imaging column, forming a common intersection of the optical axis of the FIB column and the optical axis of the charged particle beam imaging column; - a control unit configured to control the dual-column microscope to perform a sectioning and imaging method in the examination volume; - A processor and a memory with installed software code, the processor being configured to perform any of the above method steps.

50. The semiconductor inspection apparatus of claim 49, further comprising at least first and second detectors configured to perform multi-mode image acquisition.

51. The semiconductor inspection apparatus according to claim 49 or 50, wherein the charged particle beam imaging column is formed of a correction electron microscope having a correction member for correcting chromatic aberration and spherical aberration.

Citation Information

Cited By

  • Double-path interferometer switching method and device and computer storage medium

    CN120970478A