Method for determining a distortion-corrected position of a feature in an image imaged with a multi-beam charged particle microscope, corresponding computer program product and multi-beam charged particle microscope

Abstract

This invention discloses a method for determining the distortion-corrected locations of features in an image composed of one or more image patches, each image patch consisting of a plurality of image subfields, each image subfield being imaged by a correlated small beam of a multi-beam charged particle microscope. The method includes the following steps: a) providing a plurality of vector distortion maps for each image subfield, each vector distortion map characterizing the position-related distortion of each pixel in the correlated image subfield; b) identifying features of interest in the image; c) extracting the geometric properties of the feature; d) determining a corresponding image subfield containing the geometric properties of the extracted feature; e) determining one or more locations of the extracted geometric properties of the feature within the determined corresponding image subfield; and f) correcting one or more locations of the extracted geometric features in the image based on the vector distortion maps of the corresponding image subfields, thereby establishing distortion-corrected image data.

Description

Technical Field

[0001] This invention relates to the field of multi-beam charged particle microscopy and related detection tasks. More specifically, this invention relates to a method for determining the distortion-corrected position of a feature in an image composed of one or more image patches, wherein each image patch consists of a plurality of image subfields, and each image subfield is imaged by a corresponding small beam of a multi-beam charged particle microscope. This invention also relates to a computer program product and a multi-beam charged particle microscope. Prior Technology

[0002] As microstructures such as semiconductor devices continue to shrink and become more complex, there is a need to further develop and optimize planar fabrication technologies, as well as inspection systems for the fabrication and inspection of small-sized microstructures. The development and fabrication of semiconductor devices require design verification, such as testing wafers, and planar fabrication technologies are optimized for processing used in reliable, high-throughput manufacturing. Furthermore, there is a growing need to analyze semiconductor wafers for reverse engineering and custom, individual configuration of semiconductor devices. Therefore, high-throughput inspection tools are required for the high-precision testing of microstructures on wafers.

[0003] Typical silicon wafers used to manufacture semiconductor devices have a maximum diameter of 12 inches (300 mm). Each wafer is divided into 30 to 60 repeating regions ("dies") with an area of ​​approximately 800 square millimeters. Semiconductor devices comprise multiple semiconductor structures fabricated in layers on the wafer surface using planar integration techniques. Due to the processes involved, semiconductor wafers typically have flat surfaces. The component dimensions of integrated semiconductor structures extend down to a critical dimension (CD) of 5 nm in the range of several µm, and will gradually decrease in the near future, for example, to component dimensions or critical dimensions (CD) below 3 nm (e.g., 2 nm), or even below 1 nm. With these small structure dimensions, dimensional defects of the critical dimension must be identified over a large area (relative to the structure size) in a short time. For several applications, the requirements for measurement accuracy provided by the inspection device are even higher, for example, two or more orders of magnitude. For example, the width of a semiconductor component must be measured with an accuracy of less than 1 nm, such as 0.3 nm or even finer, and the relative positions of semiconductor structures must be determined with a coverage accuracy of less than 1 nm, such as 0.3 nm or even finer.

[0004] Therefore, one of the objectives of this invention is to provide a charged particle system and a high-throughput charged particle system operation method that allows for high-precision measurement of semiconductor components with an accuracy of less than 1 nm, less than 0.3 nm, or even 0.1 nm.

[0005] Recent developments in charged particle microscopy (CPM) include multi-beam charged particle microscopy (MSEM), such as those disclosed in patents US7244949 and US20190355544. In MSEM, a sample is irradiated by an array of electron microbeams containing, for example, 4 to up to 10,000 electron beams (considered as primary radiation), with each beam separated from its next adjacent beam by a distance of 1–200 micrometers. For example, an MSEM may have approximately 100 separated electron beams or microbeams configured in a hexagonal array, with the microbeams separated by a distance of approximately 10 µm. Multiple primary charged particle microbeams are focused onto the surface of the sample under study, such as a semiconductor wafer fixed to a wafer carrier mounted on a movable platform, via a common objective. During the irradiation of a wafer surface with a primary charged particle beam, interaction products, such as secondary electrons, originate from a plurality of intersections formed by the focal points of the primary charged particle beams. The number and energy of these interaction products depend on the material composition and layout of the wafer surface. The interaction products form a plurality of primary charged particle beams, which are focused by a common objective lens and guided by a projection imaging system of a multi-beam detection system to a detector disposed on a detector plane. The detector comprises a plurality of detection regions, each containing a plurality of detection pixels, and detects the intensity distribution of each of the plurality of primary charged particle beams, obtaining, for example, an image patch of 100 µm × 100 µm. Prior art multi-beam charged particle microscopes include a series of electrostatic and magnetic elements. At least some of these electrostatic and magnetic elements are adjustable to adjust the focal position and astigmatism of the plurality of primary charged particle beams. Prior art multi-beam charged particle microscopes include at least one intersecting plane of primary or secondary charged particles. Prior art multi-beam charged particle microscopes include a detection system that facilitates adjustment. Prior art multi-beam charged particle microscopes include at least one deflector scanner for collectively scanning a plurality of single-beam charged particle beams over a region of the sample surface to obtain an image patch of the sample surface. Further details of multi-beam charged particle microscopes and methods of operating multi-beam charged particle microscopes are described in PCT / EP2021 / 061216, filed April 29, 2021, which is incorporated herein by reference.

[0006] However, in charged particle microscopy used for wafer inspection, it is desirable to maintain stable imaging conditions to enable imaging with high reliability and repeatability. Throughput depends on several parameters, such as stage speed and realignment of new measurement points, as well as the measurement area per acquisition time, which is determined by dwell time, resolution, and the number of small beams. Furthermore, for multi-beam charged particle microscopy, time-consuming image post-processing is required; for example, the signals generated by the multi-beam charged particle microscopy detection system must be digitally corrected before image patches from multiple image subfields can be stitched together.

[0007] Multiple primary charged particle beams can deviate from regular grating positions within a grating configuration (e.g., a hexagonal grating configuration). Furthermore, these multiple primary charged particle beams deviate from regular grating positions during grating scanning operations within planar segments, and the resolution of a multi-beam charged particle detection system may vary and depend on the individual scanning position of each of the multiple primary charged particle beams. Using multiple primary charged particle beams, each beam is incident at a different angle onto the intersection volume of a common scanning deflector, and each beam is deflected to a different exit angle, and each beam traverses the intersection volume of the common scanning deflector along different paths. Therefore, each beam experiences a different distortion mode during the scanning operation. Prior art single-beam dynamic correctors are not suitable for mitigating distortions caused by any scanning of multiple primary charged particle beams. Patent US20090001267 A1 illustrates the calibration of a primary beam layout or static grating pattern configuration in a multi-beam charged particle system comprising five primary charged particle mini-beams. Three causes of grating pattern deviation are illustrated here: rotation of the primary beam layout, magnification or reduction of the primary beam layout, and offset of the entire primary beam layout. Therefore, US20090001267 A1 considers the fundamental first-order distortion (rotation, magnification, global offset, or displacement) of a static primary beam grating pattern formed by the static focal points of a plurality of primary mini-beams. Furthermore, US20090001267 A1 includes calibrating the first-order characteristics, deflection width, and deflection direction of the focusing grating scanner to scan the plurality of primary mini-beams with a focusing grating. Components for compensating for these fundamental errors in the primary beam layout have been discussed here. No solution is provided for higher-order distortions of the static grating pattern, such as third-order distortion. Even after calibrating the primary beam layout and selectively calibrating the secondary electron beam path, scanning distortion is introduced during scanning in each individual primary beam, which cannot be resolved by calibrating the static grating pattern of multiple primary beams.

[0008] Typically, basic first-order image distortions (rotation, magnification, and global offset or displacement) are corrected in today's high-tech multibeam charged particle microscopes. However, with the increasing demand in metrology for higher-precision measurements using MSEMs, higher-order distortions derived from scan processing are becoming increasingly important and must be properly considered.

[0009] PCT patent application PCT / EP2021 / 066255, filed on June 16, 2021, discloses a method for minimizing distortion differences caused by scanning between multiple primary charged particle beams. The disclosure of that patent application is incorporated herein by reference in its entirety. This international patent application employs a method to minimize scanning-induced distortion by improving the configuration of the grating scanner itself. However, such improved grating scanning devices are typically only implemented in newly developed multi-beam charged particle microscopes. Nevertheless, when using existing microscopes, there is a need for higher precision, particularly in handling quantitative metrological detection tasks, such as determining the feature dimensions of integrated semiconductor structures.

[0010] PCT patent WO 2021 / 239380 A1 (corresponding to PCT / EP2021 / 061216 above) discloses a multi-beam charged particle detection system and a method for using the multi-beam charged particle detection system for wafer inspection, which offers high throughput, high resolution, and high reliability. The method and the multi-beam charged particle detection system are configured to extract a set of control signals from data from a plurality of sensors to control the multi-beam charged particle detection system and thereby maintain imaging specifications including wafer stage movement during wafer inspection operations. WO 2021 / 139380 A1 does not address the problem of time-consuming image post-processing. Furthermore, WO 2021 / 139380 A1 does not address distortion caused by scanning, nor any specific problems arising from distortion caused by scanning. Summary of the Invention

[0011] Therefore, one object of the present invention is to provide an alternative solution for correcting scanning-induced distortions in images acquired using a multi-beam charged particle microscope. In particular, this solution should be applicable to accurately determining the feature dimensions of integrated semiconductor structures.

[0012] Unlike the hardware / physical method used in PCT / EP2021 / 066255, this invention employs an algorithmic approach. According to a first embodiment of the invention, distortion caused by scanning is corrected during image post-processing. Distortion correction is performed based on existing scanned distorted images, for example, using a computer (PC). Nevertheless, this correction is neither time-consuming nor energy-intensive, but rather provides an elegant solution for specific detection tasks. According to a second embodiment of the invention, distortion correction is performed during image post-processing. This is performed via specially configured or programmed hardware components of the MSEM. Therefore, this MSEM is an MSEM with integrated distortion correction. Furthermore, the first and second embodiments can be combined with each other.

[0013] According to the first state, the present invention relates to a method for determining the distortion-corrected location of features in an image composed of one or more image patches, each image patch being composed of multiple image subfields, each image subfield being imaged by a coherent small beam of a multi-beam charged particle microscope, the method comprising the following steps: a) Provide multiple vector distortion maps for each image subfield, each vector distortion map characterizing the position-related distortion of each pixel in the relevant image subfield; b) Identify a feature within the image; c) Extract the geometric properties of this feature; d) Determine the corresponding image subfield containing the captured geometric feature; e) Determine one or more locations of the captured geometric feature within the corresponding image subfield; and f) Based on the vector distortion mapping of the corresponding image subfield, correct one or more locations of the geometric characteristics captured in the image, thereby establishing distortion-corrected image data.

[0014] Typically, an image contains a plurality of image patches; however, the method also applies if the image contains only one image "patch". In any case, the image patch contains a plurality of image subfields, each of which has been imaged or has been imaged with a related small beam using a multi-beam particle microscope.

[0015] This method is particularly suitable for correcting distortions caused by scanning, and it provides a high-precision correction. A key feature of this invention is that it provides an individual vector distortion map for each image subfield, because scanning-induced distortion is typically domain-specific, which is why scanning-induced distortion cannot be compensated for simultaneously across all subfields using a conventional spotlight scanner (see above). The vector distortion map is not necessarily provided as a "map". The term "map" should only indicate that the distortion is a vector and that the vector is position-dependent. Therefore, the vector distortion map is, in principle, a vector field.

[0016] To describe the position of the distortion vector within the image subfield, the internal coordinates of the image subfield (generally referred to as p, q in this patent application) are used. Furthermore, the internal coordinates must be connected to the global coordinate system (generally referred to as x, y in this patent application). The position of each subfield, labeled with an index nm relative to the global coordinate system, can be, for example, the midpoint position of each subfield (p0, q0) in the global coordinate system (xnm, ynm).

[0017] The vector distortion mapping for each subfield can be predetermined, and therefore the vector distortion mapping for each small beam can also be predetermined. The determination method will be described in more detail below. Typically, the vector distortion mapping remains valid across multiple imaging procedures. Therefore, compared to patent WO 2021 / 239380 A1, this invention is particularly suitable for correcting regular or persistent distortions, especially distortions caused by regular scanning. However, the vector distortion mapping according to the invention can also be updated periodically. This also allows for the correction of more unpredictable or irregular distortions during image post-processing.

[0018] Method steps b) identifying features of interest in the image and c) extracting the geometric properties of the features can be performed individually or in combination. In principle, the features of interest can be of any type and shape. Examples of features of interest when studying semiconductor structures include HAR structures (high aspect ratio structures, also known as light pillars, holes, or contact channels) or other features.

[0019] The geometric properties of a feature can be, for example, the outline of the feature, or only a part of that outline, such as an edge or corner. In principle, such pixels can also represent features. According to one specific embodiment, the geometric properties of the feature are at least one of the following: outline, edge, corner, point, line, circle, ellipse, center, diameter, radius, distance.

[0020] Image data typically contains information of interest to be measured, such as the center or edge location, size, area, or volume of an object of interest, or the distance or gap between several objects of interest. Further image data may include attributes such as line edge roughness, the angle between two lines, and the radius.

[0021] Such feature extraction is well-known in image processing. For an example of contour extraction, see Li Huanliang, “Image Contour Extraction Method Based on Computer Technology”, 4th National Conference on Electrical, Electronic and Computer Engineering (NCEECE 2015), 1185-1189 (2016).

[0022] According to one specific embodiment, capturing geometric features involves generating a binary image. Images taken with a multibeam particle microscope are typically grayscale images, representing the intensity of detected secondary particles. Such images contain a considerable amount of data. In contrast, binary images that only show, for example, outlines contain relatively little data.

[0023] According to embodiments of the present invention, distortion correction is performed only on a portion of the entire image, more precisely, regarding the geometric properties of the captured features, such as the captured contours. This makes distortion correction much faster than conventional distortion correction according to the prior art, where distortion correction is performed on each pixel of the grayscale image. Furthermore, the distortion correction according to the present invention requires fewer energy resources.

[0024] The distortion correction includes the following steps: d) determining the corresponding image subfield containing the geometric features of the captured features; e) determining one or more locations of the captured geometric features of the features within the determined corresponding image subfield; and f) correcting one or more locations of the captured geometric features in the image based on the vector distortion mapping of the corresponding image subfield, thereby establishing distortion-corrected image data.

[0025] In order to correct the captured geometric features with relevant image distortion mapping, it is necessary to determine the corresponding image subfield. The corresponding image subfield can be indicated, for example, in the image relay data, or it can be determined based on the location of the data in memory or image data archive.

[0026] The captured geometric features of the feature are determined at one or more locations within the determined corresponding image subfield, because the distortion correction depends on the one or more locations.

[0027] According to one specific embodiment, correcting one or more locations of captured geometric features in an image based on vector distortion mapping of the corresponding image subfield includes determining a distortion vector for at least one location of the captured geometric feature. If, for example, the center of a feature (the location of the feature) is the geometric feature, then determining a distortion vector for only the center location is sufficient. If the geometric feature is, for example, an edge or line, then the edge or line is described by a plurality of locations, and therefore it is necessary to determine a plurality of relative distortion vectors for each of the plurality of locations. Similar considerations apply to geometric features of other shapes.

[0028] According to one specific embodiment, each of the complex vector distortion maps is described by a polynomial expansion of a vector polynomial. Therefore, in principle, the correlated distortion vector at any location or pixel within an image subfield can be computed. Alternatively, each of the complex vector distortion maps can be described by a two-dimensional lookup table. Other representations of the vector distortion "maps" are also possible in principle.

[0029] For example, a vector polynomial can be calculated as follows: , Where (dp, dq) represents the distortion vector. In one example, the summation is calculated only for lower-order terms, such as up to the third order. For instance, some terms in the summation might be related to specific types of corrections, such as scaling, rotation, shearing, trapezoidal distortion, or deformation.

[0030] According to one specific embodiment, method steps b) to f) are repeatedly performed for a plurality of features. It should be noted that method step a is not necessarily repeated.

[0031] According to one specific embodiment, other regions in the image that do not contain any features of interest are not subject to distortion correction. This significantly reduces computation and saves resources.

[0032] According to one embodiment, the geometric characteristics of features of interest are extracted from the entire image. In one instance, feature extraction results in a relatively small amount of data in a binary image. According to a further embodiment, feature extraction results in determining at least one location of a geometric characteristic, such as a center, point, edge, contour, or line.

[0033] According to one specific embodiment, correcting one or more locations in an image whose geometric features have been captured is based on a vector distortion mapping of a corresponding image subfield, including converting at least one pixel of the image into a distortion-corrected image based on the distortion vector. This is because distortion correction does not necessarily lead to a positional shift of all pixels. Instead, for example, one pixel may be shifted across two, three, or four pixels (interpolation).

[0034] According to one specific embodiment, correcting one or more locations in an image with captured geometric features based on vector distortion of a corresponding image subfield includes transforming the image location to a distortion-corrected location based on a distortion vector polygon. The vector distortion polynomial is described by a vector polynomial expansion of the vector distortion mapping of the subfield in subfield coordinates (p,q), global coordinates (x,y), or both sets of coordinates.

[0035] According to one specific embodiment, the captured geometric feature of a characteristic extends across a plurality of image subfields and is thus divided into a plurality of relative parts. In this case, individual correction is performed on one or more locations of each part of the captured geometric feature based on the relevant individual vector distortion mapping in the corresponding image subfield of the relative part. Here, the principle is also based on the vector distortion mapping of each part of the geometric feature relative to the image subfield to which that part belongs. This division of the feature into parts and relative partial distortion correction allows for more precise metrological applications.

[0036] According to one specific embodiment, the method further includes at least one of the following steps: Determine the dimensions of the semiconductor device structure from the distortion-corrected image data; Identify the region of the semiconductor device structure within the distortion-corrected image data; In the distortion-corrected image data, the positions of multiple regular objects in the semiconductor device, especially the HAR structure, are determined; Determine the line edge roughness in the distortion-corrected image data; and / or Determine the overlap error between different features within the semiconductor device in the distortion-corrected image data.

[0037] In each case, the determination / measurement steps are performed based on the distortion-corrected image data, which may be represented, for example, as a set of positional data or binary images. This improves the accuracy of the determination or measurement.

[0038] According to one specific embodiment, the method further includes the following steps: Provide test samples with precisely known, and especially repeating, patterns that define the target mesh; The test sample was imaged using a multibeam charged particle microscope, the obtained images were analyzed, and the actual grid was determined based on the analysis. Determine the positional deviation between the actual grid and the target grid; and Based on this positional deviation, a vector distortion mapping for each image subfield is obtained.

[0039] The determination of vector distortion maps or vector distortion fields from imaging-calibrated test samples is, in principle, known in the art. The accuracy of the obtained vector distortion maps depends largely on the manufacturing precision of the patterns on the test samples and the measurement precision during the analysis of the test samples.

[0040] According to one specific embodiment, the method further includes moving a test sample from a first position to a second position relative to a multi-beam charged particle microscope, and imaging the test sample within both the first and second positions. Preferably, the stage is moved to move, for example, about half an image subfield. This method step is particularly helpful in improving accuracy when imaging high-frequency structures / patterns that are statistically distributed on the sample.

[0041] According to one specific embodiment, determining the positional deviation between the actual grid and the target grid involves a two-step determination. In the first step, the offset, rotation, and magnification of each image subfield are compensated. In the second step, the remaining, and particularly higher-order, distortions are determined. These latter distortions may be scan-induced distortions. Therefore, scan-induced distortions can be clearly distinguished from other distortions.

[0042] According to one specific embodiment, the method further includes updating the vector distortion map. The update may be performed, for example, at regular time intervals, or upon user request, or whenever the configuration or operating parameters of the multi-beam charged particle microscope change.

[0043] According to a second aspect of the present invention, the present invention relates to a method for correcting distortion in an image composed of one or more image patches, each image patch being composed of a plurality of image subfields, each image subfield being imaged by a coherent small beam of a multi-beam charged particle microscope, the method comprising the following steps: g) Provide a plurality of vector distortion maps for each image subfield, each vector distortion map characterizing the position-related distortion of each pixel in the relevant image subfield; h) For each pixel in the image: determine the corresponding image subfield containing that pixel; and i) For each pixel in the image: based on the vector distortion mapping of the corresponding image subfield, convert the pixel in the image into at least one pixel in the distortion-corrected image.

[0044] The definitions of the terms used above are identical to those used in the description or definition of the first aspect of the invention. According to the second aspect of the invention, distortion correction is performed not only on the captured features but also on the entire distorted image. This can be performed, for example, using a personal computer (PC), after imaging with the multibeam particle microscope.

[0045] According to a third aspect of the invention, the invention relates to a computer program product comprising program code for performing the methods described in any of the specific embodiments described above with respect to the first and second aspects of the invention. This program code may be subdivided into one or more parts. For example, it is suitable to provide individual code for controlling a multi-beam particle microscope in one part, while another part contains multiple routines for distortion correction. Such distortion correction may, for example, be performed on a personal computer.

[0046] According to a fourth aspect of the present invention, the present invention relates to a multi-beam charged particle microscope having a controller configured to perform the methods described above in various specific embodiments.

[0047] According to a fifth aspect of the invention, correction of scanning-induced distortion is performed during image post-processing. This means that correction is performed before the digitized image data is written to image memory, which can be implemented as a parallel-access memory. For example, an FPGA ("Field-Programmable Gate Array") is configured or programmed to perform spatially correlated distortion correction on pixels describing image subfields. To achieve relative distortion correction, a filtering operation is implemented through appropriate hardware design / programming, using a spatially varied filter kernel that takes into account spatially varying distortion within the image subfield, for example, by referring to a vector distortion map determined for each image subfield as described above. To account for the spatial variation of the filter kernel, a kernel generation unit is applied, which calculates the relative filter kernel individually and preferably "on the fly" for each segment of the image subfield. Distortion correction must be performed in parallel for all small beam data streams, but it must be numerically adapted individually to the image subfield / small beam (imaging channel) in question.

[0048] More specifically, the present invention relates to a multi-beam charged particle microscope, comprising: At least one first focused beam grating scanner is used for focused scanning of a complex J primary charged particle small beams above a complex J image subfields; A detection unit includes a detector for detecting a plurality of J secondary electron microbeams, each of the J secondary electron microbeams corresponding to one of the J image subfields; and A control unit (800, 820) includes: A scanning control unit is connected to the first focused beam grating scanner and configured to use the first focused beam grating scanner to control the grating scanning operation of the plurality of J primary charged particle beams during use; A kernel generation unit configured to generate a spatial variation filter kernel for spatial variation distortion correction of the image subfield; and An image data acquisition unit, the operation of which is synchronized with the operation of the detector, the scanning control unit, and the kernel generation unit, wherein the image data acquisition unit includes, for each of the J image subfields: - A comparison digitizer for converting an analog data stream received from a detector into a digital data stream describing an image subfield; - A hardware filter unit configured to receive the digital data stream and configured to perform convolution of the image subfield segment with a spatial variation filter kernel, thereby generating a distortion-corrected data stream; and - An image memory configured to store the distortion-corrected data stream as a 2D representation of the image subfield.

[0049] The fifth characteristic feature of the present invention is the hardware filter unit and the kernel generation unit. The hardware filter unit is configured to receive the digital data stream and to perform convolution of the image subfield segment with the spatially varied filter kernel during use, thereby generating a distortion-corrected data stream, which is achieved for the first time in multi-beam charged particle microscopy. Since the distortion correction within the image subfield is not constant but varies within the image subfield, the filter kernel used must also vary spatially. To take into account this spatial dependence, the kernel generation unit is applied, which allows the spatially varied filter kernel to be calculated / determined for each segment of the image subfield currently filtered within the hardware filter unit.

[0050] Furthermore, it must be considered that for multiple small beams, there are a relatively large number of imaging channels. Therefore, distortion correction must be performed individually for each imaging channel, or in other words, for each J image subfield. Thus, the image data acquisition unit includes an analog-to-logarithmic converter, a hardware filter unit, and image memory for each of these imaging channels and therefore for each of these J image subfields.

[0051] As mentioned above, distortion correction in image post-processing typically incurs enormous computational costs. However, by performing image distortion correction on each image subfield using hardware filters, computational costs and energy requirements can be significantly reduced. This hardware filtering effect is achieved with minimal time delay during data generation before the data stream is stored in image memory. The kernel generation unit can compute the spatial variation filter kernel for spatial variation distortion correction of each image subfield "on the fly," and the computational cost of generating such filter kernels is quite moderate.

[0052] Of course, the operation of different parts of a multi-beam charged particle microscope must be synchronized, for example, by applying clock signals and counting units. Those familiar with this technique know the possible implementation methods.

[0053] According to a specific embodiment of the present invention, the hardware filter unit includes: A grid configuration of filter elements, each filter element comprising a first register for storing pixel values ​​and a second register for storing coefficients generated by the kernel generation unit, wherein the pixel values ​​stored in the first register represent segments of the image subfield; A plurality of multiplication blocks are configured to multiply the pixel values ​​stored in a first register with the corresponding coefficients stored in a second register; and Multiple addition blocks are configured to sum the results of multiplication. As described above, the hardware filter unit is configured to perform convolution of segments of the image subfield with the spatial variation filter kernel during use. Mathematically, the convolution between two matrices can be described as the sum of the products calculated from the items in the matrices. Applied to this invention, the first register stores a first matrix entry (pixel values ​​of the image subfield segment), and the second matrix entry corresponds to the coefficients generated by the kernel generation unit. To perform the necessary multiplication of the inner terms of the two matrices, a plurality of multiplication blocks are provided. Similarly, to perform the necessary summation of the products, a plurality of summation blocks are provided.

[0054] The term grid configuration should indicate the intrinsic relationship / context between pixel values ​​and coefficients. Grid configuration logically corresponds to a matrix representation.

[0055] Typically, filtering is a neighborhood operation. This means that a filter unit operates only on a segment of an image subfield, not the entire image subfield. Therefore, according to one embodiment, the hardware filter unit includes a plurality of shift registers configured to implement a grid configuration of the filter unit and to maintain the order of data in the data stream as it passes through the hardware filter unit. These measures ensure the implementation of the grid configuration of micro-image subfield segments, and thus the implementation of pixels within the image subfield that are in the neighborhood of the image pixel to be distorted. Shift registers typically have a predetermined size, such as 512 bits, 1024 bits, 2048 bits, or 4096 bits. Thus, shift registers can store a relatively large number of pixels. However, the grid configuration size of the filter elements is typically much smaller. Typically, an image segment may contain, for example, 11×11 filter elements, 21×21 filter elements, or 31×31 filter elements. If the grid configuration of the filter elements has a general size A×A, then a plurality of A-shift registers can be applied, wherein the first A-term in such shift registers belongs to the representation of an image subfield segment, and wherein the remaining terms in such shift registers can fill the remaining pixels of a column (or bar) of the image subfield. Thus, essentially, the size of the shift register limits the number of pixels in a column (or bar) of the image subfield.

[0056] According to one embodiment of the invention, the grid configuration size of the filter elements is adapted to correct at least ten times the distortion of the pixel size of the image subfield. This means that the grid configuration size of the filter elements is at least 20×20, or more precisely, 21×21 items. It should be noted that the number of filter elements in a column or column is typically chosen to be odd, because then the filter kernel can be represented in a symmetrical manner with a unique center. However, mathematically, the grid configuration size of the filter kernel can also be even. Furthermore, the pixel size can be the same or different in different scanning directions.

[0057] For example, the pixel size in an image subfield can be 2 nm. Then, by applying a 20×20 or 21×21 filter kernel, distortion of about 20 nm can be corrected.

[0058] Typically, the grid configuration size of a filter element determines the maximum distortion that can be corrected, which is approximately half the grid configuration size / dimension multiplied by the relative dimension or the pixel size in the direction.

[0059] According to one embodiment, the size of the grid configuration corresponds to the size of the filter kernel. Therefore, the number of multiplications that must be performed is equal to the number of filter elements. However, the number of subsequent necessary multiplications increases quadratically with the number of pixels in a column or bar. Therefore, the computational load increases, and the number of logic units also increases, since the hardware filter units are implemented in hardware. Therefore, it is preferable to reduce the number of logic units. According to one embodiment of the invention, the size of the predetermined kernel window is equal to or smaller than the grid configuration size of the filter elements. Herein, it is necessary to consider performing filtering according to the invention for the purpose of distortion correction. Distortion correction can be understood as pixel offset. This means that even if a full-size kernel filter is fully convolved with the pixel values ​​stored in the first register of the filter element, many multiplications will not affect the result. In other words, for example, moving a pixel typically results in the pixel being "distributed" across four other pixels. Therefore, the kernel window reflects a portion of the filter kernel, where filter kernel entries affect the result. Other multiplications that can theoretically be performed in a full convolution have no effect and can therefore be omitted. This saves logic units, and more precisely, it saves multiplication and addition blocks. Of course, the exact location of the kernel window within the entire filter kernel must be considered. Therefore, according to a specific embodiment of the invention, the kernel generation unit is configured to determine the position of the kernel window relative to the filter element grid configuration during use.

[0060] According to one embodiment, the hardware filter unit further includes a plurality of switching components configured to combine terms and filter elements with multiplication block logic based on the position of the kernel window during use. Therefore, to reduce the number of multiplication blocks and addition blocks, the number of switching components (e.g., multiplexers) must be increased. However, this is easier to implement.

[0061] According to a specific embodiment of the present invention, the kernel generation unit is configured to determine the spatial variation filter kernel based on a vector distortion mapping of spatial variation distortion in a characteristic image subfield. For details describing the vector distortion mapping, please refer to the definitions and explanations given in the first to fourth embodiments of the present invention.

[0062] According to one specific embodiment, the vector distortion mapping is described by a polynomial expansion in a vector polynomial. Alternatively, the vector distortion mapping is described by a multidimensional lookup table.

[0063] According to one specific embodiment, the kernel generation unit is configured to determine the filter kernel based on a function f that represents the pixels. In other words, in addition to the distortion subject itself, the filter kernel also considers the "shape" of the pixels. Possible functions describing pixels could be, for example, the Rect2D function describing rectangular pixels; this corresponds to linear or bilinear filters. Since a pixel may be blurred along the scan direction, a possible function f could also be a function Rect(p,q) with different degrees of blurring in different scan directions p and q.

[0064] Optionally, the function f describing the pixel can also have the shape of the pixel beam focus, such as a Gaussian function, an isotropic function, a cubic function, a sinc function, a ventilation pattern, etc., where the filter is truncated at a certain low value. Furthermore, according to one example, the filter should conserve energy; therefore, a higher-order, truncated filter kernel should be positively normalized to a sum of weights equal to 1. Optionally, the positive normalization can be performed at a later stage rather than directly within the filter; those skilled in the art will understand the advantages and disadvantages of this specific implementation.

[0065] It is important to note that pixels at the boundaries of image subfields will be unavailable. However, this effect is well-known in filtering processes during image post-processing. To address this, scaling down is necessary depending on the filter kernel size. However, this does not pose any problem, as overlap between adjacent image subfields is typically achieved in multibeam charged particle microscopy.

[0066] According to one embodiment, the image data acquisition unit further includes a counter configured to indicate the local coordinates (p, q) of pixels within a filtered image subfield during use. This is related to both synchronization purposes and determining the distortion caused by independent spatially correlated scans within the image subfield.

[0067] According to one embodiment, the image data acquisition unit further includes an averaging unit implemented in the data streaming direction after the analog-to-digital converter and before the hardware filter unit. The averaging unit can be applied to increase the signal-to-noise ratio. Possible implementations are described in PCT patent application WO 2021 / 156198 A1, which is incorporated herein by reference in its entirety.

[0068] According to one specific embodiment, the image data acquisition unit further includes an additional hardware filter unit configured to perform additional filtering operations during use, particularly low-pass filtering, morphological operations, and / or deconvolution with a point spread function. Of course, the image data acquisition unit may also include a plurality of further hardware filter units. The principle applied here is that filtering operations can also be implemented by specially configured hardware, without needing to force filtering operations during image post-processing.

[0069] According to one specific embodiment, the hardware filter unit includes a field-programmable gate array (FPGA) or an application-specific integrated circuit (ASIC).

[0070] According to one specific embodiment, the hardware filter unit includes a series of FIFOs. It can implement shift registers as described above to achieve a grid configuration of filter elements.

[0071] According to one embodiment, the FIFO is implemented as a block RAM. According to another embodiment, the FIFO may be implemented as a LUT (lookup table) or an externally connected SRAM / DRAM (static or dynamic random access memory). It should be noted that there are typically pre-built IP blocks from the relative chip manufacturer to instantiate the hardware.

[0072] The specific embodiments of the fifth state of the present invention can be partially or completely combined with each other, as long as no technical contradictions arise.

[0073] Of course, other implementations and configurations of the hardware filter unit are also possible.

[0074] According to the sixth aspect of the present invention, the present invention relates to a system comprising: A multi-beam charged particle microscope, as described above in several specific embodiments; and An image post-processing unit is provided for distortion correction of image data. This image post-processing unit can be provided in addition to a multibeam charged particle microscope. For example, it may include an additional personal computer. However, alternatively, the image post-processing unit may be included within the multibeam charged particle microscope. The image post-processing unit can be configured to perform distortion correction through image post-processing as described above with respect to the first aspect of the invention. Importantly, according to this specific embodiment of the invention, two different types of distortion correction can be combined with each other. A first distortion correction can be performed in image pre-processing (implemented as data streaming processing), followed by a second distortion correction in image post-processing, preferably performed only on the geometric characteristics of the captured targeted features. Reference is made explicitly to the description of the first aspect of the invention in this regard.

[0075] Similarly, different versions of the present invention can be combined in whole or in part, as long as no technical contradictions occur. The definition of one version of the present invention also applies to other versions of the present invention.

[0076] According to one example, in the first step, distortion caused by scanning that occurs according to a second embodiment of the invention can be corrected (where distortion correction is performed during image preprocessing), and then in the second step, another or still existing distortion can be corrected according to the first embodiment of the invention (where distortion caused by scanning is corrected during image postprocessing). Simple Explanation of the Diagram

[0077] The invention will be fully understood with reference to the accompanying drawings: Figure 1 illustrates a multi-beam charged particle microscope according to a specific embodiment; Figure 2 illustrates the coordinates of the first detection region and the second detection region, which contain the first and second image patches; Figure 3 illustrates the static distortion shift of a plurality of single-stage charged particle beams; Figure 4a illustrates the scanning deflection diagram at the scanning deflector of the small-axis beam; Figure 4b illustrates the scanning deflection at the scanning deflector, and the distortion caused by scanning an off-axis small beam with a propagation angle β; Figure 5 illustrates the telecentric aberration pattern caused by scanning an off-axis small beam with a propagation angle β; Figure 6 illustrates the distortion caused by a typical scan of a single small beam during scanning on an image subfield with image subfield coordinates (p, q); Figure 7 illustrates distortion correction commonly used in image processing; Figure 8 illustrates grayscale image distortion correction and subsequent feature extraction; Figure 9 illustrates the feature extraction and subsequent distortion correction according to the present invention; Figure 10 is a flowchart of the method for determining the distortion-corrected position of a feature according to the present invention; Figure 11 illustrates the determination of vector distortion mapping based on the target mesh; Figure 12 illustrates the determination of the distortion vector; Figure 13 illustrates the determination of grid points; Figure 14 illustrates dimensional measurements based on distortion-corrected image data; Figure 15 illustrates the statistical evaluation of the position of regular objects based on distortion-corrected image data; Figure 16 illustrates the image data acquisition unit and related units or modules; Figure 17 illustrates a hardware filter unit; Figure 18 illustrates the convolution of an image subfield segment with a filter kernel; Figure 19 illustrates an excerpt of filter elements and related components; Figure 20 illustrates a hardware filter unit with a 3×3 filter kernel window; and Figure 21 illustrates a hardware filter unit with a 2×2 filter kernel window. Implementation

[0078] In the exemplary embodiments described below, components that are similar in function and structure are represented by similar or identical reference numerals whenever possible.

[0079] Figure 1 is a schematic diagram illustrating the basic features and functions of a multi-beam charged particle microscope 1 according to some specific embodiments of the present invention. It should be noted that the symbols used in the figure do not represent the physical configuration of the illustrated components, but have been selected to symbolize their respective functions. The system type shown is a multi-beam scanning electron microscope (MSEM or Multi-SEM), which uses a plurality of primary electron beamlets 3 on the surface of an object 7, such as a wafer with its top surface 25 located in the object plane 101 of the objective lens 102, to generate a plurality of primary charged particle beams 5. For simplicity, only five primary charged particle beamlets 3 and five primary charged particle beams 5 are shown. The characteristics and functions of the multi-beam charged particle microscope 1 can be realized using electrons or other types of primary charged particles (e.g., ions, especially helium ions).

[0080] The multi-beam charged particle microscope 1 includes an object illumination unit 100 and a detection unit 200; and a beam splitter unit 400 for separating the primary charged particle beam path 11 from the primary charged particle beam path 13. The object illumination unit 100 includes a charged particle multi-beam generator 300 for generating a plurality of primary charged particle small beams 3 and adjusting them to focus the plurality of primary charged particle small beams 3 on the object plane 101, wherein the platform 500 positions the surface 25 of the wafer 7.

[0081] A charged particle multi-beam generator 300 generates a plurality of primary charged particle small beams 311 within an intermediate image surface 321, which is typically a spherically curved surface to compensate for the field curvature of the object illumination unit 100. The charged particle multi-beam generator 300 includes a source 301 of primary charged particles (e.g., electrons). The primary charged particle source 301 emits a diverging primary charged particle beam 309, which is collimated by at least one collimating lens 303 to form a collimated beam. The collimating lens 303 typically consists of one or more electrostatic or magnetic lenses, or a combination of electrostatic and magnetic lenses. The collimated primary charged particle beam is incident on a primary multi-beam forming unit 305. The primary multi-beam forming unit 305 essentially includes a first porous plate 306.1 irradiated by the primary charged particle beam 309. The first porous plate 306.1 includes a plurality of apertures in the grating configuration for generating a plurality of primary charged particle beams 3, which are generated by passing a collimated primary charged particle beam 309 through the plurality of apertures. The primary multi-beam forming unit 305 includes at least two additional porous plates 306.2 and 306.3, located downstream of the first porous plate 306.1 relative to the direction of electron motion in the electron beam 309. For example, the second porous plate 306.2 functions as a microlens array and is preferably configured with a defined potential to adjust the focusing position of the plurality of primary beams 3 within the intermediate image surface 321. The third porous plate configuration 306.3 (not shown) includes individual electrostatic elements for each of the plurality of apertures to influence each of the plurality of beams individually. The porous plate configuration 306.3 consists of one or more porous plates with electrostatic elements, such as circular electrodes, multipole electrodes, or a series of multipole electrodes for microlenses to form a static deflector array, microlens array, or stigmator array. The primary multi-beam forming unit 305 is composed of adjacent first electrostatic field lenses 307, and together with the second field lens 308 and the second porous plate 306.2, focuses a plurality of primary charged particle beams 3 into or near the intermediate image plane 321.

[0082] Within or near the intermediate image plane 321, a static beam control aperture plate 390 is configured with a plurality of apertures having electrostatic elements (e.g., deflectors) to manipulate each of the plurality of charged particle microbeams 3 individually. The apertures of the beam control aperture plate 390 are configured to have a larger diameter to allow the plurality of primary charged particle microbeams 3 to pass through, even if the focal point of the primary charged particle microbeam 3 deviates from the intermediate image plane or its designed position. In one example, the beam control aperture plate 390 may also be formed as a single aperture element.

[0083] A plurality of focal points of the primary charged particle beams 3 passing through the intermediate image plane 321 are imaged in the image plane 101 by the field lens group 103 and the objective lens 102, positioning the surface 25 of the object 7 in the image plane. The object illumination system 100 further includes a beam-focusing grating scanner 110 near the first beam intersection point 108, so that the plurality of charged particle beams 3 can be deflected in a direction perpendicular to the beam propagation direction or the optical axis 105 of the objective lens 102. In the example of FIG1, the optical axis 105 is parallel to the z-direction. The objective lens 102 and the beam-focusing grating scanner 110 are centered on the optical axis 105 of the multi-beam charged particle microscope 1, which is perpendicular to the wafer surface 25. The wafer surface 25, configured in the image plane 101, is then scanned by the beam-focusing grating scanner 110. Thus, a plurality of primary charged particle beams 3, forming multiple beam points 5 in a grating configuration, are simultaneously scanned on the wafer surface 25. In one example, the grating configuration of the focal points 5 of a plurality of primary charged particle small beams 3 is a hexagonal grating of approximately one hundred or more primary charged particle small beams 3. The focal points 5 have a distance of approximately 6 µm to 15 µm and a diameter less than 5 nm, for example 3 nm, 2 nm or even smaller. In one example, the focal point size is approximately 2 nm, and the distance between two adjacent focal points is 8 µm. At each scan position of each of the plurality of focal points 5, a plurality of secondary electrons are generated to form a plurality of secondary electron small beams 9 with the same grating configuration as the focal point 5. The intensity of the primary charged particles generated at each focal point 5 depends on the intensity of the impacting primary charged particle small beam 3, the illuminated relative point, and the material composition and morphology of the object 7 below the focal point 5. The primary charged particle small beams 9 are accelerated by the electrostatic field generated by the sample charging unit 503 and collected by the objective lens 102, and guided by the beam splitter 400 to the detection unit 200. The detection unit 200 images the secondary electron beam 9 onto the image sensor 207 to form a plurality of secondary charged particle beam dots 15 therein. The detector includes multiple detector pixels or individual detectors. For each of the plurality of secondary charged particle beam dots 15, the intensity is detected, and the material composition of the wafer surface 25 is detected at high resolution over a large image patch with high throughput. For example, for a 10×10 small beam grating with an 8 µm pitch, an image scan using a beamforming grating scanner 110 with an image resolution of, for example, 2 nm or less, produces an image patch of approximately 88 µm × 88 µm. For example, the image patch is sampled at half the beam dot size, so that for each small beam, each image row has 8000 pixels, resulting in a digital dataset of 6.4 billion pixels for the image patch produced by 100 small beams. The control unit 800 collects image data. Details of image data collection and processing using, for example, parallel processing are described in German Patent Application 102019000470.1 and US Patent Application 9,536,702, which are incorporated herein by reference.

[0084] A plurality of secondary electron beams 9 pass through a focusing grating scanner 110, are deflected by the focusing grating scanner 110, and are guided by a beam splitter unit 400 to follow the secondary particle beam path 11 of the detection unit 200. The plurality of secondary electron beams 9 travel in opposite directions to the primary charged particle beam 3, and the beam splitter unit 400 is configured to separate the secondary particle beam path 11 from the primary particle beam path 13, typically by means of a combination of magnetic or electromagnetic fields. Optionally, an additional magnetic correction element 420 is present in either the primary or secondary particle beam path. The projection system 205 further includes at least one focusing grating scanner 222 connected to the projection system control unit 820, or more generally to the imaging control module 820. The control unit 800 is configured to compensate for residual positional differences in the plurality of focal points 15 of the plurality of secondary electron beams 9, such that the positions of the plurality of focal points 15 remain constant on the image sensor 207.

[0085] The projection system 205 of the detection unit 200 includes additional electrostatic or magnetic lenses 208, 209, 210 and a plurality of secondary electron microbeams 9 at a second intersection point 212, wherein an aperture 214 is located. In one example, the aperture 214 further includes a detector (not shown) connected to the projection system control unit 820. The projection system control unit 820 is further connected to at least one electrostatic lens 206 and a third deflection unit 218. The projection system 205 further includes at least one multi-aperture modifier having apertures and electrodes for respectively influencing each of the plurality of secondary electron microbeams 9; and a selective additional active element 216, such as a multipole element connected to the control unit 800.

[0086] The image sensor 207 is composed of an array of sensing areas, the pattern of which is compatible with the grating configuration of the secondary electron microbeams 9 focused onto the image sensor 207 by the projection lens 205. This allows for the detection of each individual secondary electron microbeam independently of other secondary electron microbeams incident on the image sensor 207. A plurality of electrical signals are established and converted into digital image data, which is then processed by the control unit 800. During image scanning, the control unit 800 is configured to trigger the image sensor 207 to detect multiple real-time resolution intensity signals from the plurality of secondary electron microbeams 9 at predetermined time intervals, and the digital image of the image patch is accumulated and stitched together from all scan positions of the plurality of primary charged particle microbeams 3.

[0087] The image sensor 207 shown in Figure 1 can be an array of electron sensitivity detectors, such as a CMOS or CCD sensor. This array of electron sensitivity detectors may include electron-to-photon conversion units, such as scintillator elements or an array of scintillator elements. In one example, the image sensor 207 may be configured as an electron-to-photon conversion unit or scintillator plate disposed in the focal plane of a plurality of secondary electron particle image points 15. In this example, the image sensor 207 may further include a relay optics system for imaging and guiding photons generated by the electron-to-photon conversion unit at the secondary charged particle image points 15 on a dedicated photon detection element such as a plurality of photomultiplier tubes or avalanche photodiodes (not shown). Such an image sensor is disclosed in US 9,536,702, which has been referenced above. In one example, the relay optics system further includes a beam splitter for separating and guiding light to a first slow-light detector and a second fast-light detector. The second fast light detector, for example, is composed of a photodiode array such as an avalanche photodiode, which is fast enough to resolve the image signals of the plurality of secondary electron beams 9 based on the scanning speed of the plurality of primary charged particle beams 3. The first slow light detector is preferably a CMOS or CCD sensor, which provides high-resolution sensor data signals to monitor the focal point 15 or the plurality of secondary electron beams 9 and control the operation of the multi-beam charged particle microscope.

[0088] In one example, the primary charged particle source is implemented as an electron source 301 having an emitter tip and a pickup electrode. When using primary charged particles other than electrons, such as helium ions, the configuration of the primary charged particle source 301 may differ from that shown. The primary charged particle source 301, the active porous plate configurations 306.1…306.3, and the beam control porous plate 390 are controlled by a primary small beam control module 830, which is connected to the control unit 800.

[0089] During image patch acquisition by scanning a plurality of primary charged particle beams 3, it is preferable not to move the platform 500, and after acquiring an image patch, move the platform 500 to the next image patch to be acquired. In an alternative embodiment, the platform 500 moves continuously in a second direction while acquiring images by scanning a plurality of primary charged particle beams 3 in a first direction using a focusing grating scanner 110. The platform movement and platform position are monitored and controlled by sensors known in the industry, such as laser interferometers, grating interferometers, confocal microlens arrays, or similar instruments.

[0090] Figure 2 illustrates in more detail the method of inspecting a wafer by acquiring image patches. The wafer, with its wafer surface 25, is placed in the focusing plane of a plurality of primary charged particle beams 3, and positioned at the center 21.1 of a first image patch 17.1. Predefined positions of image patches 17.1...k correspond to detection areas on the wafer used for semiconductor feature detection. This application is not limited to the wafer surface 25, but also, for example, applies to lithography masks used in semiconductor manufacturing. Therefore, the term "wafer" should not be limited to semiconductor wafers, but includes any object used in or manufactured during semiconductor manufacturing.

[0091] From a standard file format inspection file, predefined positions of the first inspection area 33 and the second inspection area 35 are loaded. The predetermined first inspection area 33 is divided into multiple image blocks, such as the first image block 17.1 and the second image block 17.2, and the first center position 21.1 of the first image block 17.1 is aligned below the optical axis 105 of the multi-beam charged particle microscope 1 for the first image acquisition step of this inspection task. The first center position 21.1 of the first image block is selected as the origin of the first local wafer coordinate system for acquiring the first image block 17.1. The method of aligning the wafer 7 to register the wafer surface 25 and generating a local coordinate system for wafer coordinates is well known in the art.

[0092] A plurality of primary small beams 3 are distributed in a regular grating configuration in each image patch 17.1...k and are scanned by a grating scanning mechanism to produce a digital image of the image patch. In this example, the plurality of primary charged particle small beams 3 are configured in a rectangular grating configuration, with N beam points 5.11, 5.12 to 5.1N in the first row and beam points 5.11 to 5.MN in the Mth row. For simplicity, only M = five times N = five beam points are shown, but the number of beam points J = M times N can be larger, for example J = 61 small beams, or about 100 small beams or more, and the plurality of beam points 5.11 to 5.MN can have different grating configurations, such as hexagonal or circular gratings.

[0093] Each of the charged particle mini-beams scans through wafer surface 25, as exemplified by a charged particle mini-beam with focal points 5.11 and 5.MN and scan paths 27.11 and 27.MN. For example, scanning of each of the multiple charged particles is performed by moving back and forth along scan paths 27.11...27.MN, and the multi-beam scanning deflector system 110 causes each focal point 5.11...5.MN of each charged particle mini-beam to move together in the x-direction from the starting position of an image subfield line, which in this example is, for example, the leftmost image point of image subfield 31.MN. Then, each focal point 5.11...5.MN is focused and scanned by concentrating the charged particle mini-beams 3 to the correct position, and then the focused raster scanner 110 moves each of the multiple charged particle mini-beams in parallel to the starting position of the next line in each subfield 31.11...31.MN. The movement back to the starting position of the next scan line is called a flyback. A plurality of primary charged particle beams 3 follow in most parallel scan paths 27.11 to 27.MN, thereby simultaneously acquiring multiple scan images of each subfield 31.11 to 31.MN. For image acquisition, as described above, a plurality of secondary electrons are emitted at the focal point 5.11 to 5.MN, generating a plurality of secondary electron beams 9. These beams 9 are collected by the objective lens 102, passed through a beamforming scanner 110, guided to the detection unit 200, and detected by the image sensor 207. The sequential data stream of each of the plurality of secondary electron beams 9 is synchronously transformed with the scan paths 27.11…27.MN within the multiple 2D datasets, thereby forming digital image data for each subfield. Finally, the digital images of multiple image subfields are stitched together by the image stitching unit to form the digital image of the first image patch 17.1. Each image subfield is configured to have a small overlapping area with the adjacent image subfield, such as the overlapping area 39 of subfield 31.mn and subfield 31.m(n+1).

[0094] Next, the requirements or specifications for wafer inspection tasks will be explained. For high-throughput wafer inspection, the image acquisition time (including the time required for image post-processing) for each image patch 17.1...k must be fast. On the other hand, strict image quality specifications must be maintained, such as image resolution, image accuracy, and repeatability. For example, the image resolution requirement is typically 2 nm or less, with high repeatability. Image accuracy, also known as image fidelity, refers to, for example, the edge position of a component. The absolute positional accuracy of the component is typically determined by high absolute accuracy. Typically, the positional accuracy requirement is about 50% or even lower than the resolution requirement. For example, measurement tasks require absolute accuracy of semiconductor component dimensions, less than 1 nm, less than 0.3 nm, or even less than 0.1 nm. Therefore, the lateral positional accuracy of each focus 5 of a plurality of primary charged particle beams 3 must be less than 1 nm, for example, less than 0.3 nm or even less than 0.1 nm. Under high image repeatability, it should be understood that repeated image acquisition of the same area produces first and second repeated digital images, and the difference between the first and second repeated digital images is below a predetermined threshold. For example, the difference in image distortion between the first and second repeated digital images must be less than 1 nm, such as 0.3 nm, or even more preferably less than 0.1 nm, and the difference in image contrast must be less than 10%. In this way, similar image results can be obtained even through repeated imaging operations. This is important for, for example, image capture and comparison of similar semiconductor structures in different wafer dies, or for comparing the obtained image with a representative image obtained from image simulation of CAD data, a database, or a reference image.

[0095] One of the requirements or specifications for wafer inspection tasks is throughput. The area measured per acquisition time is determined by dwell time, pixel size, and the number of small beams. Typical examples of dwell time are between 2 ns and 800 ns. Therefore, the pixel rate at the fast image sensor 207 is in the range of 1.25 MHz to 500 MHz, and approximately 15 to 20 image patches or frames can be acquired per minute. For 100 small beams in a high-resolution mode with a pixel size of 0.5 nm, a typical example of throughput is approximately 0.045 sqmm / min, and with a larger number of small beams, such as 10,000 small beams and a dwell time of 25 ns, the throughput may exceed 7 sqmm / min. However, in existing systems, the requirements for digital image processing greatly limit throughput; for example, digital compensation for scanning distortion in existing technology is very time-consuming and therefore undesirable.

[0096] The imaging performance of the multi-beam charged particle microscope 1 is limited by the design of the electrostatic or magnetic elements of the object illumination unit 100, as well as higher-order aberrations and manufacturing tolerances, such as those of the primary multi-beam forming unit 305. Imaging performance is limited by aberrations, such as distortion of the plurality of charged particle microbeams, focusing aberrations, telecentricity, and astigmatism. Figure 3 illustrates a typical static distortion aberration of the plurality of primary charged particle microbeams 3 in the image plane 101. The plurality of primary charged particle microbeams 3 are focused in the image plane to form a plurality of primary charged particle beam points 5 (three are listed) in a grating configuration, which in this example is a hexagonal grating. In an ideal system, with the beam-focusing grating scanner 110 off, each beam point 5 is formed at the center position 29.mn (see Figure 2) of the corresponding image subfield 31.mn (index m is used for row number and n is column number). However, in practical systems, the beam points 5 are formed at slightly off-center positions, deviating from the ideal position on the ideal grating, as shown by the static distortion vector in Figure 3. For the exemplary case of the main beam point 141, the deviation from the ideal position on the hexagonal grating is described by the distortion vector 143. Given the lateral difference [dx, dy] from the ideal position, the maximum absolute value of the distortion vector can be in the range of several nm, for example, above 1 nm, 2 nm, or even above 5 nm. Typically, the static distortion vector of a real system is measured and compensated by an array of static deflection elements, such as any active porous plate configuration 306.2. Furthermore, as described in German Patent Application No. 102020206739.2, filed May 28, 2020, which is incorporated herein by reference, drift or dynamic changes in static distortion are considered and compensated. Aberration control and compensation are achieved through a monitoring or detection system and a control loop capable of driving the compensator multiple times during image scanning, thereby compensating for aberrations in the multi-beam charged particle microscope 1.

[0097] However, the imaging performance of charged particle microscopes is limited not only by the design aberrations and drift aberrations of the electrostatic or magnetic elements of the object illumination unit 100, but also, particularly, by the beam-focusing grating scanner 110. Deflection scanning systems and their characteristics for single-beam microscopes have been extensively studied. However, for multi-beam microscopes, conventional deflection scanning systems used to scan multiple small beams of charged particles exhibit inherent characteristics. The beam path through the deflection scanner in Figure 4 illustrates these inherent characteristics in more detail.

[0098] Figure 4a illustrates the beam path of a single primary charged particle beam through a prior art beam-focusing grating scanner 110 with deflection electrodes 153.1 and 153.2 and a voltage source. For simplicity, only the deflection scanner electrodes used for grating scanning deflection in the first direction are illustrated. During use, a scanning deflection voltage difference VSp(t) is applied, and an electrostatic field is formed by the equipotential line 155 between the deflection electrodes 153.1 and 153.2. An axially charged particle beam 150a corresponding to image patch 31.c having an image patch center 29.c coinciding with the optical axis 105 is deflected by the electrostatic field and passes through the intersection volume 189 between the deflection electrodes 153.1 and 153.2 along the actual beam path 151f. The beam trajectory can be approximated by a single virtual deflection at a virtual pivot point 159 by the first-order beam paths 150a and 150f. A small beam of charged particles traveling along path 150z is focused in object plane 101 by objective lens 102, as shown in the lower part of Figure 4a. The subfield coordinates are given in relative coordinates (p, q) with respect to the center point 29.c of subfield 31.c.

[0099] For the maximum deflection at the maximum subfield point at coordinate pf, a maximum voltage difference VSP max is applied, and for the deflection of the incident small beam 150a at a distance pz to the subfield point, a relative voltage VSP is applied, and the incident small beam 150a is deflected by a deflection angle α in the direction of the beam path 150z. The nonlinearity of the deflector is compensated by determining the functional correlation between the deflection angle α and the deflector voltage difference VSP. By calibrating the functional correlation VSP(sin(α)), a near-ideal scanner for a single primary charged particle small beam is achieved, with a single virtual pivot point 159 for deflection scanning of a single charged particle small beam. It should be noted that the lateral displacement (p, q) of the beam point position in the image plane is proportional to the focal length f of the objective lens 102 multiplied by sin(α). For example, for a zonal field point, pz = f sin(αz). With respect to small angles α, the function sin(α) is usually approximated by α. As will be described in more detail below, although distortion caused by scanning with a single-beam microscope can be minimized, aberrations caused by other scanning, such as astigmatism, defocus, hairline aberration, or spherical aberration, reduce the resolution of charged particle microscopy as the field size increases. Furthermore, the deviation from the virtual pivot point 159 becomes increasingly significant as the field size increases.

[0100] In a multi-beam system, multiple charged particle beams are scanned in parallel using the same deflection scanning gas and the same voltage difference, based on the functional correlation VSp(sin(α)). In Figure 4b, the intersection point 108 of the multiple primary charged particle beams coincides with the virtual pivot point 159 of the axial primary beam 150a, and each charged particle beam passes through the electrostatic field at a different angle. The path 157a of the charged particle beam, exemplified by the incident angle β, corresponds to a subfield 31.o with the center of the image subfield 29.o. This angle β is related to the distance X from the center coordinate 29.o to the optical axis 105 by sin(β) = X / f, and the focal length of the objective lens 102 is f. With the condenser grating scanner 110 turned off (VSp(t) = 0V), the beam passes through path 157a and is focused by the objective lens 102 onto the center point 29.o of the subfield 31.o. However, if a voltage difference is applied, although the deflection scanner is approximately ideal for the axial small beam as shown in Figure 4a, it is not ideal for the field small beam at the incident angle β. Due to the finite thickness of the deflection field, the path length through the electrostatic field is different for each incident small beam at different incident angles β, and the actual beam paths 157z and 157f deviate from the first-order ideal beam paths 163z and 163f. This is explained with respect to the beam paths of the two sub-field points at coordinates pz and pf, and the actual beam paths 157z and 157f. The angles of the actual beam paths 157z and 157f deviate from the angles of the ideal beam paths 163z and 163f, and each beam is virtually deflected at different virtual pivot points 161z and 161f, deviating from the beam intersection point 108. For example, if a voltage VSp(sin(α 0)) is applied, the path 157a of the single charged particle beam deflects by an angle α1 instead of an angle α0, and along the beam path 157z with a virtual pivot point 161z. Therefore, the charged particle beam point is distorted due to the local distortion vector dpz.

[0101] The deviation of the deflection angle increases with the increase of the incident angle β, and the distortion caused by the scanning generated by the beam grating scanner 110 also increases.

[0102] Differences in deflection angle α cause scanning-induced distortion, and differences in the position of the virtual pivot point are the cause of telecentric aberrations in scanning. Figure 5 illustrates a simplified system 171 in front of the scanning beam grating scanner 110, from which a plurality of primary charged particles are incident on the first beam grating scanner 110. The plurality of charged particle beams are shown by two beams, including an axial charged particle beam 3.0 and an off-axis beam 3.1, which pass through the intersection volume 189 of the beam grating scanner 110 and are focused by the objective lens 102 to form a plurality of focal points, shown by focal points 5.0 and 5.1 on the surface 25 of the wafer 7. When the beam grating scanner 110 is in the off state and no voltage difference VSp is applied to the deflection electrode 153, the beam points 5.0 and 5.1 are located at the center points 29.0 and 29.1 of the respective image subfields. If a voltage difference VSp(sin(α 0)) has been applied, the small beam 3.0 follows the ideal path 150 and deflects to the strip field point Z 0. In the linear representation of Figure 5, the small beam 3.0 appears to deflect at the beam intersection point 108 corresponding to the virtual pivot point 159 of Figure 4a. Therefore, the small beam 3.0 illuminates the wafer surface 25 at the same angle of incidence as the central position 29.0. The off-axis small beam 3.1 has deflected to the relative strip field point Z 1 of the corresponding image subfield. The off-axis small beam 3.1 appears to deflect along the representative beam path 157 at the virtual pivot point 161, deviating from the beam intersection point 108. Therefore, the telecentric angle of the small beam 3.1 at the scan position with respect to the strip field point Z 1 deviates from the telecentric angle at the central field 29.1, corresponding to the telecentric aberration caused by the scan of the small beam 3.1 in addition to the aforementioned distortion. In a third embodiment of the present invention, the second multi-beam scanning correction system 602 reduces telecentric aberrations caused by scanning.

[0103] The deviation of the focal position at the scanning position of each of the plurality of charged particle beams 3 is described by the scanning distortion vector field (also called vector distortion mapping) of each image subfield 31.11 to 31.MN. Figure 6 illustrates the scanning distortion in an example of image subfield 31.15 (see Figure 7). Throughout the specification, image subfield coordinates (p, q) relative to the respective center of each image subfield 31.mn are used, and the scanning distortion is described by the scanning distortion vector [dp, dq] as a function of the image subfield coordinates (p, q) of each individual image subfield 31.mn. The center position (p, q) of each image subfield = (0, 0) is described in (x, y) coordinates relative to the optical axis 105. The center coordinates of each image can be distorted from a predetermined ideal grating configuration by a static offset (dx, dy) as a function of the (x, y) coordinates, as shown in Figure 3. Static distortion is typically compensated by a static porous plate 306.2 and is not considered in the scan distortion vector [dp, dq]. Since the scan distortion differs in each image subfield 31.11…31.MN, it is generally described by a scan distortion vector [dp, dq] = [dp, dq](p, q; xij, yij) based on four coordinates. These four coordinates consist of the local image subfield coordinates (p, q) and the discrete center coordinates (xij, yij) of the image subfield.

[0104] Figure 6 shows the scan distortion vector [dp, dq] on image subfield 31.15. In this example, the maximum scan distortion is located at the maximum image subfield coordinates p = q = 6µm, with a scan distortion vector [dp, dq] = [2.7nm, -1.6nm]. The length of the maximum scan distortion vector in this image subfield is 3.5 nm. Typical maximum scan distortion aberrations in image subfields range from 1 nm to 4 nm, but can even exceed 5 nm.

[0105] Figure 7 illustrates a common method for distortion correction in image processing. Such image distortion correction is well-known in the art. Image distortion correction is then performed in post-processing. The correction of distortion can be described as pixel displacement with a position-dependent displacement vector, since distortion varies from pixel to pixel. The position-dependent displacement vector can be mathematically described by matrix-vector multiplication. Furthermore, it must be considered that distortion is not typically given in the form of all pixels. In other words, in addition to simple displacement, pixel value interpolation must also be performed. These facts are schematically shown in Figure 7: pixel 700 is displaced due to distortion, and the resulting pixel position is indicated by the reference symbol 700'. The value of pixel 700 is set to 1. Due to the displacement, the value or intensity 1 must be distributed across four pixels in the distortion-corrected image: relative pixels with intensities / values ​​I1, I2, I3, and I4.

[0106] If image processing is used to correct distortion across the entire image, it is numerically costly: for each original pixel in the distorted image, a multiplication operation with an n×m matrix must be performed, in addition to interpolation operations. For example, an image from a multi-beam charged particle microscope contains 10 G pixels. Therefore, distortion correction requires four operations per pixel plus interpolation, resulting in at least 40 billion operations—a massive number.

[0107] However, in metrology, what truly matters is the accurate location of image details. According to embodiments of the invention, the locations of image details are determined in the original, still distorted image, and then these locations are corrected for distortion. For example, if the goal is to determine the location of a HAR (high aspect ratio) structure in a semiconductor sample, the numerical computation can be reduced by approximately 100,000 times (assuming a 100 × 100 µm² image field contains 10 G pixels, and the approximate diameter of the HAR structure is approximately 100 nanometers, with a spacing of approximately 300 nanometers).

[0108] According to an embodiment of the invention, a distortion represented by a vector distortion map 730 is determined for each image subfield 31.mn, because the distortion of each image subfield 31.mn is different and varies within each image subfield 31.mn. The generation of the vector distortion map is known in itself. The distortion in each image subfield 31.mn can be described, for example, by a polynomial expansion in a vector polynomial. This is known in principle, for example, from measurements of a calibration object. Furthermore, the object or test sample can be moved between a first and a second measurement, and the distortion can be determined based on the difference between the two measurements. These measurements can also be repeated. Thus, the distortion can be determined. The distortion, and more precisely the vector distortion map 730 and / or its representation as a polynomial expansion in a vector polynomial, can be stored in memory with respect to each image subfield. It can also be updated at predetermined time intervals.

[0109] Figures 8 and 9 illustrate distortion correction according to conventional image processing (Figure 8) and, on the other hand, according to an embodiment of the present invention (Figure 9). More specifically, Figure 8A describes a grayscale image 702. The grayscale image 702 can, in principle, be a complete image, a single image block, or even just a subfield of an image; this makes no difference in explaining the principles. The grayscale image 702 comprises three specific features 701a, 701b, and 701c. In principle, these features 701a, 701b, and 701c can be distorted, with distortion illustratively shown for the curved feature 701c. Distortion correction is performed on the original grayscale image 702 according to the prior art, wherein distortion correction is performed on each pixel of the grayscale image. The result is shown in Figure 8B. Feature 701c is no longer distorted, and feature 701c is no longer curved. In the next step, the contours of features 701a, 701b, and 701c are extracted from the grayscale image 702, generating a binary image 710 as shown in Figure 8C. Based on the contours in the binary image 710, precision measurement or metrological applications can be performed. Note that for illustrative and distinguishing purposes, the grayscale image 702 includes a dotted background and the binary image 710 includes a white background.

[0110] Please refer to Figure 9 for an illustration of the correction process according to the present invention. The original situation described in Figure 9A is the same. However, firstly, all targeted features are identified and extracted. Figure 9B illustrates a binary image 710 containing only the contours of features 701a, 701b, and 701c. These contours are still distorted. However, compared to grayscale images according to the prior art, the amount of data in the binary image is significantly reduced. Then, in the next step, distortion correction is performed on the contours of features 701a, 701b, and 701c. In this text, since the distortion is of the nature of scanning-induced distortion, distortion correction is performed individually for each image subfield, and the distortion correction for each pixel in each image subfield 31.mn is position-dependent.

[0111] Examplely, Figure 9 shows a simplified method for improving the correction of distortion caused by scanning. According to another example of a method for correcting distortion caused by scanning, at least the locations of targeted features 701a, 701b, and 701c are extracted from the uncorrected digital image, and distortion correction is applied only to the locations of the targeted features 701a, 701b, and 701c through, for example, a polynomial expansion of a vector distortion map. Therefore, distortion correction is not limited to the pixel raster of the digital image.

[0112] Therefore, more generally, the diagram shown in Figure 9B can be interpreted as a visualization of connecting line segments, which consist of a set of non-integer positions or non-integer coordinates of one of the targeted features 701a, 701b, and 701c obtained from features extracted from the grayscale image 702. Similarly, Figure 9C can be interpreted as a visualization of connecting line segments of the non-integer positions or non-integer coordinates of the targeted distortion correction features 701a, 701b, and 701c.

[0113] Figure 10 illustrates a flowchart of a method for determining the distortion-corrected location of feature 701 in an image composed of one or more image patches, wherein each image patch consists of a plurality of image subfields 31.mn, and each image subfield 31.mn is imaged by a coherent small beam of a multi-beam charged particle microscope. In the first method step S1, a plurality of vector distortion maps 730 are provided for each image subfield 31.mn. Each vector distortion map 730 characterizes the position-related distortion of each pixel in the relevant image subfield 31.mn. Furthermore, as explained in the general part of this document, the term "map" must be interpreted broadly. It should refer to providing a vector field with distortion vectors for each image subfield 31.mn. For example, each of the plurality of vector distortion maps 730 can be described by a polynomial expansion in a vector polynomial. The specific distortion at position p, q in the image subfield 31.mn can then be calculated from the polynomial expansion. Optionally, each of the complex vector distortion maps 730 can be described by a two-dimensional lookup table. Other representations are possible in principle.

[0114] In method step S2, feature 701 of interest is identified in the image. In method step S3, the geometric properties of feature 701 are extracted. Method steps S2 and S3 can be performed individually, or they can be combined. In principle, the geometric properties of feature 701 of interest can be of any type or shape. The geometric properties of feature 701 can be, for example, the outline of feature 701, or only a part of the outline, such as an edge or corner. It can also be the center of feature 701. Examples of geometric properties of feature 701 can be at least one of the following: outline, edge, corner, point, line, circle, ellipse, center, diameter, radius, distance. Other geometric properties and irregular forms are also possible. Geometric properties can further include attributes such as line edge roughness, angle between two lines, or area or volume.

[0115] In the next step S4, the corresponding image subfield 31.mn containing the geometric characteristics of the captured feature 701 is determined. In step S5, one or more positions of the captured geometric characteristics of feature 701 within the determined corresponding image subfield 31.mn are determined. Whether only one position or multiple positions are determined depends on the nature of the captured geometric characteristics. By determining the corresponding image subfield 31.mn and the positions of one or more pixels within it, distortion vectors 715 (or multiple distortion vectors 715) can be explicitly assigned for the correction performed in method step S6: According to method step S6, the positions of the captured geometric features in the image are corrected based on the vector distortion mapping 730 of the corresponding image subfield 31.mn, thereby establishing distortion-corrected image data. Method steps S2 to S6 can be repeated for multiple features 701.

[0116] Then, in method S7, the program may terminate or execute one or more metrological applications or measurements: examples include determining the structural dimensions of a semiconductor device in a distortion-corrected image; determining the structural area of ​​the semiconductor device in the distortion-corrected image; determining the positions of a plurality of regular objects in the semiconductor device, particularly the HAR structure, in the distortion-corrected image; determining the line edge roughness in the distortion-corrected image data; and / or determining the overlap error between different features within the semiconductor device in the distortion-corrected image. These example applications will be described in detail below.

[0117] The captured geometric feature of feature 701 can extend over a plurality of image subfields 31.mn, and is therefore divided into a plurality of relative parts. In this case, based on the correlation individual vector distortion mapping 730 in the corresponding image subfields 31.mn of the relative parts, one or more locations of each part of the captured geometric feature are individually corrected. This significantly improves the accuracy of the measurement processing because the distortion caused by the scan is not necessarily a smooth function on the subfield boundary 725.

[0118] Figure 11 is a diagram illustrating the determination of vector distortion mapping 730 based on target mesh 711. Figure 11A shows a test sample with a repeating pattern of structure 712 having a precisely known target mesh defined in this instance. In the present case, target mesh 711 contains a plurality of circles. However, other target meshes 711 may be selected, such as those containing squares or a combination of squares and circles. Ideally, the target mesh is a perfect mesh with nominal spacing between a plurality of structures 712 arranged in a regular pattern. The test sample is then imaged using a multibeam charged particle microscope 1, and the obtained images are analyzed to determine the actual mesh 720 based on the analysis. Target mesh 711 and actual mesh 720 are different from each other. The difference is described with respect to the center 713 of structure 712 and is indicated in Figure 11B by means of distortion vector 715. The field of distortion vector 715 is an example of vector distortion mapping 730 used for distortion correction.

[0119] Figure 12 illustrates the determination of the distortion vector 715. Vector 717, defined within an internal coordinate system with coordinates (p, q), points to the center 713 of the structure 712 of the ideal target grid 711. However, when determining the actual grid 714, this center 713 is imaged at the location of the actual grid 714, which can be described by vector 716 according to the internal coordinates (p, q) of the image subfield. Subtracting vector 717 from vector 716 yields the distortion vector 715. Note that the distortion vector 715 can be defined as the vector pointing from the center 713 of the target grid 711 to the actual measurement center of the actual grid 714. However, in principle, the distortion vector 715 can also be defined as the inversion of the currently described vector. By definition, the distortion vector 715 itself, or its reciprocal, is used to correct one or more locations of the captured geometric features in the image subfield 31.mn.

[0120] Figure 13 illustrates the determination of grid points in the actual grid 720. The target grid 711 contains a plurality of regular and highly accurate known structures 712. These structures 712 have ideal profiles. In the described example, the structure 712 is circular. When the test sample is imaged, several single profile locations 721 are determined. Due to the point-symmetric geometry of the structure 712 in the current case, a connecting line 722 connecting the two edge locations on opposite sides of the structure 712 can be defined. Reference symbol 723 indicates the region containing the midpoint 724 of the line containing the center 713. The center 713 is used to define the grid location. The average location of these midpoints 724 can be taken as the center of the actual structure, i.e., the center of the structure relative to the actual grid 720. The standard deviation of the midpoint locations 724 is a measure of the accuracy or reliability with which the center 713 of the feature can be determined. If this deviation is too large, the structure can be excluded from further processing.

[0121] Figure 14 illustrates a dimensional measurement diagram based on distortion-corrected geometric data. Figure 14A exemplarily shows two image subfields 31.mn and 31.m(n+1) and their corresponding vector distortion mapping 730, which contains the distortion vector domain 715. In a conventional single-beam charged particle microscope with a single image field, distortion is a slowly varying continuous function over the single image field, and its effect on dimensional measurement is negligible. However, in a multi-beam charged particle microscope with multiple image subfields 31.mn, such as subfields 31.mn and 31.m(n+1), the total distortion is a discontinuity function at the subfield boundary 725. Therefore, large differences in the discontinuous distortion function can worsen the dimensional measurement of feature 701 extending over the two image subfields 31.mn and 31.m(n+1). According to an embodiment of the present invention, the two parts 726 and 727 of feature 701 are respectively distorted based on the vector distortion mapping 730 of each image subfield 31.mn and 31.m(n+1). More specifically, the geometric characteristics of the features extracted from the image are distances dv, more precisely, two positions (p1;q) and (p2;q), where the value of q is the same, and therefore will not be further explained. However, the coordinates (p1;q) are determined for the image subfield 31.mn, while the coordinates (p2;q) are determined for the image subfield 31.m(n+1). The position of (p1:q) is corrected based on the vector distortion mapping 730 of the image subfield 31.mn, and the position of (p2;q) is corrected based on the vector distortion mapping 730 of the image subfield 31.m(n+1). The respective distortion vectors vp1 and vp2 are also shown in FIG14B. Therefore, the distance dv is distorted to distance d.

[0122] Figure 14 illustrates the case when compensating for the static distortion of a plurality of primary small beams. Therefore, the vector distortion map 730 at the center position of the corresponding image subfields 31.mn, 31.m(n+1) shows no distortion or an offset of the distortion vector. However, it is also possible that, depending on the distortion caused by the scanning of the image subfields 31.mn, 31.m(n+1), each of the vector distortion maps 730 contains an additional offset distortion vector caused by the static distortion of the multi-beam charged particle system 1. The offset of each distortion vector in each image subfield can be different, as shown in the example in Figure 3.

[0123] Figure 15 is a diagram illustrating the statistical evaluation of the position of regular objects based on image data after distortion correction. Figure 16A depicts multiple HAR features, where reference symbols 80.1 and 80.2 respectively label the first HAR structure and the second HAR feature. These HAR features 80.1 and 80.2 can be identified, for example, through pattern recognition, which is well-known in the art. For example, pattern recognition can be assisted by machine learning. The geometric properties of HAR features 80.1 and 80.2 are the center positions of HAR features 80.1 and 80.2, respectively. The center position of each HAR structure 80 is extracted and its location is determined. Further, it is determined which image subfield 31.mn the center position of HAR structure 80 belongs to: in the current case, the center of HAR structure 80.1 belongs to image subfield 31.mn and the center of HAR structure 80.2 belongs to image subfield 31.m(n+1). Then, the center positions of HAR structures 80.1 and 80.2 are corrected based on the relative vector distortion mapping 730 of the corresponding image subfields 31.mn and 31.m(n+1), respectively. The corrected center positions can then be analyzed, and compared, for example, with the design center positions 96 of multiple HAR structures, and the deviations 97 from the design center positions 96 are analyzed. Furthermore, in the example shown in Figure 15, it is important to first perform all feature extraction and position or measurement in the still distorted binary image. Afterwards, distortion correction is performed in a position-dependent manner and relative to the relevant image subfields 31.mn and 31.m(n+1).

[0124] Besides the specific applications described in Figures 14 and 15, many other applications of the invention are possible. One of these is the determination of LER (Line Edge Roughness) across subfield boundaries 725. Distortion discontinuities across subfield boundaries 725 can create discontinuities within the line itself. A possible solution according to the invention is essentially to first extract the line, divide it into portions belonging to different image subfields, apply distortion correction to each portion of the line, and then determine the line edge roughness.

[0125] The positional deviation between feature 701 in the first layer and feature 701' in the second layer is called the overlap error. The overlap error can be determined at features 701 and 701', which are generated in different lithography steps or in different layers. Again, according to an embodiment of the present invention, features 701 and 701' are first extracted. Then, distortion correction is applied to features 701 and 701'. This invention is particularly important when features 701 and 701' are within different image subfields 31.mn.

[0126] The general objective of embodiments of the present invention is to reduce or avoid distortion compensation during image post-processing of 2D image data. As described above, distortion compensation during post-processing of 2D image data requires storing the source image data and calculating the distortion-corrected target image data. According to the improved distortion correction method provided above, distortion correction is performed in a clustered manner with respect to captured parameters such as edges or center locations, rather than on full-size 2D image data. Therefore, computational load and power consumption are reduced by at least one order of magnitude, and even up to five orders of magnitude. According to another embodiment of the invention, the computational load and power consumption required for post-processing are further reduced. In this embodiment, the digital image data stream received from image sensor 207 is directly written to image memory 814, thereby reducing or compensating for distortion aberrations during data stream processing. Therefore, at least a major portion of the distortion in each subfield 31.mn can be compensated during stream processing.

[0127] Figure 16 is a diagram of the image data acquisition unit and related units or modules. For ease of explanation, only one image channel is described; the other image channels are not shown in Figure 16. In this case, the number of image channels corresponds to the number of J small beams used for imaging with the multi-beam charged particle microscope 1.

[0128] In one example, the image sensor 207 includes a plurality of J photodiodes corresponding to a plurality of J secondary electron beams. Each of the J photodiodes, such as an avalanche photodiode (APD), is connected to a separate analog-to-logarithmic converter. The image sensor may further include an electron-to-photon converter, such as that described in German Patent DE 102018007455 B4, which is incorporated herein by reference in its entirety.

[0129] An analog-to-digital converter 811 converts an analog data stream into a digital data stream. After conversion to a digital data stream, the data is provided to an averaging unit 815; however, the averaging unit 815 may be omitted. In principle, pixel averaging or line averaging can be performed; for more detailed information, please refer to PCT patent WO 2021 / 156198 A1, which is incorporated herein by reference in its entirety.

[0130] The image data acquisition unit includes a hardware filter unit 813 for each of the J image subfields. This hardware filter unit 813 is configured to receive a digital data stream and, during use of the multi-beam charged particle microscope 1, to perform convolution of a segment of the image subfield 31.mn with the spatial variation filter kernel 910, thereby producing a distortion-corrected data stream. The details of this distortion correction will be described in more detail below.

[0131] The data image acquisition unit 810 further includes an image memory 814, which is configured to store the distortion-corrected data stream as a 2D representation of the image subfield 31.mn.

[0132] In the described example, the image acquisition unit 810 is part of the imaging control module 820, which further includes a scan control unit 930. In this example, the scan control unit 930 is configured to control the strafe grating scanner 110 and the strafe grating scanner 220. Further control mechanisms for the scan control unit 930 can also be implemented within the multi-beam charged particle microscope 1, which is not shown in Figure 16.

[0133] In principle, the overall control of the multi-beam charged particle microscope 1 comprises different units or modules. However, it must be remembered that different modules belonging to the control can be selected and implemented in different ways; therefore, the structure described in Figure 16 is merely an example. In addition to the imaging control module 820, a control unit 800 is also provided. An image memory 814 is connected for parallel readout to the control unit 800, which is configured to read out a plurality of J digital images corresponding to J image subfields 31.11 to 31.mn. The image stitching unit 817 of the control unit 800 is configured to stitch the J digital image subfields into a digital image file corresponding to an image patch (e.g., image patch 17.k). The image stitching unit 817 is connected to an image data processor and an output 818, which is configured to retrieve information from the digital image file and to write the digital image file to memory or to provide information from the digital image file to a display.

[0134] Note that the modules and processing illustrated in Figure 16 are precisely synchronized, which is achieved by providing an appropriate clock signal (not further shown in Figure 16). Furthermore, since the hardware filter unit 813 is configured to perform convolution of image subfield segments with the spatial variation filter kernel 910, the counting unit 816 is implemented within the control unit 800, which provides input to the kernel generation unit 812, which provides the filter kernel data to the hardware filter unit 813. Again, it is emphasized that one filter kernel 910 is calculated for each imaging channel; however, for ease of illustration, these plurality of imaging channels are not further illustrated in Figure 16.

[0135] The imaging control module 820 of the multi-beam charged particle microscope 1 may include a plurality of L image data acquisition units 810.n, which include at least a first image data acquisition unit 810.1 and a second image data acquisition unit 810.2 configured in parallel. Each of the plurality of image data acquisition units 810.n may be configured to receive sensor data from the image sensor 207, corresponding to a subset of S small beams and a plurality of J primary charged particle small beams, and to generate a subset of S streams of digital image data values ​​of a plurality of J digital image data value streams. The number of S sub-beams belonging to each of the L image data acquisition units 810.n may be the same, and S×L = J. The number of S is, for example, between 6 and 10, for example, S = 8. The number L of parallel image data acquisition units 810.n may be, for example, 10 to 100 or greater, depending on the number J of primary charged particle small beams. By using the modular concept of the imaging control module 820, the number of small charged particle beams J in the multi-beam charged particle microscope 1 can be increased by adding parallel image data acquisition units 810.n.

[0136] Figure 17 is a diagram of the hardware filter unit 813. The arrows in Figure 17 indicate data input to the hardware filter unit 813. In the described embodiment, the hardware filter unit 813 comprises a grid configuration 900 with 5×5 filter elements 901. The grid configuration 900 of the filter elements 901 should reflect or be equivalent to a segment representation of the image subfield 31.mn. Therefore, the order and arrangement of the data within the grid configuration 900 are quite important to ensure this relationship or equivalence. In an exemplary embodiment, the hardware filter unit 813 is implemented by a series of FIFOs 906. This series of FIFOs 906 ensures that the order of data entering the hardware filter unit 813 is maintained. Furthermore, the FIFOs 906 ensure that data correctly jumps from the first column or first row of the image subfield 31.mn to the second column or second row, etc. Therefore, when the filter element 901 is filled with pixel values ​​step by step and the sequence of pixel values ​​is passed through the filter unit 813, the pixel value items in the grid configuration 900 can correspond to segments of the image subfield 31.mn for distortion correction.

[0137] As described above, the hardware filter unit 813 is configured to perform convolution of segment 32 of the image subfield 31.mn with the spatial variation filter kernel 910. In other words, the value or coefficients of the filter kernel 910 must be calculated individually for the filtering of a specific filtered segment 32. Each filter element 901 within the described grid configuration 900 contains two items: the pixel value itself and the coefficients generated by the kernel generation unit. To perform convolution, multiplication of the entries within the filter element 901 must be performed. The results of the multiplication must then be summed, as shown by the line connecting filter element 901 and box 905 in Figure 17. The performed filtering operations (multiplication and summation) result in a time delay that remains constant throughout the filtering process of the entire image subfield 31.mn. The distorted data stream (data input) is converted into a distortion-corrected data stream (data output).

[0138] Figure 18 is a diagram illustrating the convolution of segment 32 of image subfield 31.mn with filter kernel 910. Segment 32 of image subfield 32.mn and filter kernel 910 are described as a grid configuration of filter element 901, and the size of filter kernel 910 is the same in this case. Here, a 5×5 implementation is described. On the left side of Figure 18A, the uncorrected pixel value or intensity I is described in the first register 902. In filter kernel 910, a plurality of coefficients 903 generated by kernel generation unit 812 are stored in the second register 903.

[0139] Figure 18B shows the mathematical equivalence to the situation shown in Figure 18A: describing two matrices that must be convolved. The result is a double sum of the products of some matrix terms with other terms. Generally, it must be noted that different terms of the matrix must be multiplied by each other; for example, it is not usually the terms I11 and K11 that must be multiplied by each other. This is only the case for symmetric filter kernels. However, there is still a fixed scheme based on which different terms must be multiplied. This scheme can also be implemented by the relative hardware representation of filter kernel 910 (the flipping of columns and bars in the kernel).

[0140] Figure 19 is a schematic diagram of the filter element 901 and related elements. More specifically, according to the described specific embodiment, each filter element 901 includes a first register 902 temporarily storing pixel values ​​and a second register 903 temporarily storing coefficients generated by the kernel generation unit 812. Furthermore, filter element 901 includes a multiplication block 904 for multiplying the pixel values ​​stored in the first register 902 with the corresponding coefficients stored in the second register 903. Note that the multiplication block 904 is not necessarily part of the filter element 901 itself; it can also be implemented separately. After performing the multiplication with the multiplication block 904, the relative result is provided to the addition block 905. Figure 19 shows only two filter elements 901 and one addition block 905; it should be noted that more filter elements 901 and multiple addition blocks 905 are typically provided for successful distortion correction. The arrows in Figure 19 indicate data flow. In addition, the entries in the second temporary register 903 are provided by the kernel generation unit 812 (not shown in Figure 19).

[0141] According to a more general embodiment, the hardware filter unit 813 may include a grid configuration 900 of filter elements 901, each filter element 901 including a first register 902 for storing pixel values ​​and a second register 903 for storing coefficients generated by the kernel generation unit 812, wherein the pixel values ​​temporarily stored in the first register 902 represent segments of the image subfield 31.mn. The hardware filter unit 813 may further include a plurality of multiplication blocks 904 for multiplying the pixel values ​​stored in the first register 902 with relative coefficients stored in the second register 903. The hardware filter unit 813 may further include a plurality of addition blocks 905 configured to sum the multiplication results. According to this more general formula, the number of multiplication blocks is not necessarily the same as the number of filter elements 901, but may be reduced.

[0142] The latter case is illustrated in Figure 20: Figure 20 is a diagram of a hardware filter unit 813 with a 3×3 filter kernel window. Therefore, the filter kernel window (3×3) is smaller than the grid configuration 900 (5×5). Importantly, the filtering process according to the invention is performed for a specific purpose, namely distortion correction. Distortion correction can be interpreted as pixel offset. This means that even if a full convolution is performed between the full-size kernel filter 910 and the pixel values ​​stored in the first register 902 of the filter element 901, many multiplications will not affect the distortion correction; therefore, more precisely, they will not affect the resulting summation. Thus, if all filter elements 901 are considered in the convolution, the result (summation) is indistinguishable. Instead, it is important to select the relevant filter element 901 for the calculation process. This selection can be made by choosing a suitable kernel window 907. Of course, the exact position of the filter kernel window 907 within the grid 900 is not arbitrary. The position of the kernel window 907 can be determined by the kernel generation unit 812, particularly "on the fly." If this specific embodiment variant is chosen, it is not necessary to provide a multiplication block for each filter element 901. This reduces the number of logic units within the hardware filter unit 813. However, since the position of the kernel window 907 is not fixed for each segment 32 of the image subfield 31.mn, the possibility of performing different multiplications must be guaranteed. Therefore, a plurality of switching components must be provided, configured to logically combine terms and filter elements 901 with the multiplication block 904 based on the position of the kernel window 907 during use.

[0143] According to one embodiment, the kernel generation unit 812 is configured to determine the spatial variation filter kernel 910 based on a vector distortion map 730 that characterizes spatial variation distortion in the image subfield 31.mn. According to one embodiment, the vector distortion map 730 is described by a polynomial expansion in a vector polynomial. Alternatively, the vector distortion map 730 is described by a multidimensional lookup table. Furthermore, the kernel generation unit 812 may be configured to determine the filter kernel 910 based on a function f that represents the pixels. Possible functions f describing pixels may be, for example, the Rect2D function describing rectangular pixels. Alternatively, the beam focal shape of the pixel may be used as the function f, such as a Gaussian function, an isotropic function, a cubic function, a sinc function, an airy pattern, etc., with the filter truncated at a low-order value. Furthermore, the filter should be energy conserved; therefore, the higher-order, truncated filter kernel 910 should be positively normalized to a weight sum equal to 1.

[0144] As explained in Figure 7 regarding this case, pixel 700 is "distributed" across four pixels 700' in the distortion-corrected image. Therefore, a kernel window 907 of size 2×2 can be applied.

[0145] Figure 21 is a diagram of a hardware filter unit 813 with only a 2×2 filter kernel window 907. The diagram shown in Figure 21 corresponds to the displacement shown in Figure 7 in this case.

[0146] Specific embodiments of the present invention reduce or eliminate distortion compensation during image post-processing of 2D image data. Therefore, distortion correction for each pixel in a massive 2D image containing gigapixels and requiring substantial image memory is unnecessary. Instead, distortion correction is performed, for example, on reduced clustering of extraction parameters such as edge or center locations, rather than on the full-size 2D image data. According to a further example, distortion for each subfield 31.mn is compensated during streaming processing of the data stream from image sensor 207. Streaming of analog data from image sensor 207 is necessary regardless, and the additional distortion compensation during streaming requires only minimal additional computational power and a reduced amount of additional memory. With this invention, computational load and power consumption are reduced by at least an order of magnitude, and even up to five orders of magnitude. These two methods and configurations can also be combined. In one example, it is advantageous to compensate for the first part of the vector distortion polynomial of each image subfield 31.mn through streaming processing, and to compensate for the second part of the vector distortion polynomial via distortion correction at the reduced-bundle of the extracted parameters or geometric characteristics. For example, the linear part of the distortion polynomial is compensated during streaming processing, and higher-order distortion is compensated by distortion correction at the reduced-bundle of the extracted parameters. Thus, the additional computational cost of calculating higher-order vector polynomials during streaming processing is reduced. In general, embodiments of the present invention allow distortion correction of a multi-beam charged particle microscope 1 with reduced computational power and reduced energy consumption. Therefore, the present invention enables detection or metrology tasks in semiconductor processes with high efficiency, reduced computational workload, and reduced energy consumption.

[0147] It should be noted that the specific embodiments of the present invention described with reference to the accompanying drawings are not intended to limit the present invention. The drawings only show possible implementations of the present invention.

[0148] The following describes further examples of the present invention, which may be combined with other specific embodiments and examples as described above.

[0149] Example 1. A method for determining the distortion-corrected location of features in an image composed of one or more image patches, each image patch consisting of multiple image subfields, each image subfield being imaged by a correlated small beam of a multi-beam charged particle microscope, the method comprising the following steps: a) Provide multiple vector distortion maps for each image subfield, each vector distortion map characterizing the position-related distortion of each pixel in the relevant image subfield; b) Identify features of interest within the image; c) Extract the geometric properties of this feature; d) Determine the corresponding image subfield containing the captured geometric feature; e) Determine one or more locations of the captured geometric feature within the corresponding image subfield; and f) Based on the vector distortion mapping of the corresponding image subfield, correct one or more locations of the geometric characteristics captured in the image, thereby establishing distortion-corrected image data.

[0150] Example 2. The method as described in Example 1, wherein steps b) to f) of the method are repeated for a plurality of features.

[0151] Example 3. The method described in any of the preceding examples, wherein other regions in the image that do not contain any features of interest are not subject to distortion correction.

[0152] Example 4. The method as described in any of the preceding examples, wherein the geometric properties of the feature are at least one of the following: contour, edge, corner, point, line, circle, ellipse, center, diameter, radius, distance.

[0153] Example 5. The method described in any of the preceding examples, wherein capturing geometric features involves generating a binary image.

[0154] Example 6. The method as described in any of the preceding examples, One of the features, the captured geometric property, extends across a plurality of image subfields and is therefore divided into individual plurality of parts, and The distortion mapping of relevant individual vectors in the corresponding image subfield of individual parts is used to individually correct one or more locations of each part of the captured geometric characteristic.

[0155] Example 7. The method described in any of the preceding examples, wherein geometric properties of the features of interest are extracted over the entire image.

[0156] Example 8. The method as described in any of the preceding examples, wherein the step of correcting one or more locations of the geometric feature captured in the image based on the vector distortion mapping of the corresponding image subfield includes determining the distortion vector of at least one location of the captured geometric feature.

[0157] Example 9. The method as described in any of the preceding examples, wherein the step of correcting one or more locations of the geometric feature captured in the image based on the vector distortion mapping of the corresponding image subfield includes converting the pixels of the image into at least one pixel of the distortion-corrected image based on the distortion vector.

[0158] Example 10. The method as described in any of the preceding examples, wherein each of the complex vector distortion maps is described by a polynomial expansion in a vector polynomial.

[0159] Example 11. The method as described in any of Examples 1 to 9, wherein each of the plurality of vector distortion maps is described by a 2D lookup table.

[0160] Example 12. The method as described in any of the preceding examples further comprises at least one of the following steps: Determine the dimensions of the semiconductor device structure from the distortion-corrected image data; Identify the region of the semiconductor device structure within the distortion-corrected image data; In the distortion-corrected image data, the positions of multiple regular objects in the semiconductor device, especially the HAR structure, are determined; Determine the line edge roughness in the distortion-corrected image data; and / or Determine the overlap error between different features within the semiconductor device in the distortion-corrected image data.

[0161] Example 13. The method as described in any of the preceding examples further includes the following steps: Provide test samples with precisely known, and especially repeating, patterns that define the target mesh; The test sample was imaged using a multibeam charged particle microscope, the obtained images were analyzed, and the actual grid was determined based on the analysis. Determine the positional deviation between the actual grid and the target grid; and Based on this positional deviation, a vector distortion mapping for each image subfield is obtained.

[0162] Example 14. The method as described in any of the preceding examples further includes moving a test sample from a first position to a second position relative to a multi-beam charged particle microscope, and imaging the test sample in the first position and the second position.

[0163] Example 15. The method as described in any of Examples 13 to 14, wherein determining the positional deviation comprises two steps, wherein in the first step, the offset of each image subfield, the rotation of each image subfield, and the magnification of each subfield are compensated, and wherein in the second step, the remaining higher-order distortions are determined.

[0164] Example 16. The method as described in any of the preceding examples further includes the following steps: Update the vector distortion mapping.

[0165] Example 17. The method as described in any of the preceding examples further includes the following steps: Image distortion is corrected by streaming data during image preprocessing.

[0166] Example 18. A method for correcting distortion in an image composed of one or more image patches, each image patch consisting of multiple image subfields, each image subfield being imaged by a correlated small beam from a multi-beam charged particle microscope, the method comprising the following steps: g) Provide a plurality of vector distortion maps for each image subfield, each vector distortion map characterizing the position-related distortion of each pixel in the relevant image subfield; h) For each pixel in the image: determine the corresponding image subfield containing that pixel; and i) For each pixel in the image: based on the vector distortion mapping of the corresponding image subfield, convert the pixel in the image into at least one pixel in the distortion-corrected image.

[0167] Example 19. A computer program product comprising program code for performing the methods described in any one of Examples 1 to 18 above.

[0168] Example 20. A multi-beam charged particle microscope having a controller configured to perform the method as described in any one of Examples 1 to 18.

[0169] 1: Multibeam charged particle microscope 3: A single charged particle beam 5: Primary charged particle beam focus / beam point / focal point 7: Objects / Wafer 9: Secondary Electron Small Beam 11: Secondary electron beam path 13: Primary beam path 15: Image point / focal point of secondary charged particles 17: Image tiles 19: Overlapping areas of image tiles 21: Central position 25: Wafer surface / surface 27: Scanning path of a single small beam 29: Center of the image subfield 31: Image Subfield 32: Excerpt 33: First detection site 35: Second detection site 39: Overlapping region of subfield 31 80.1: HAR Structure 80.2: HAR structure 96: The design center position of the HAR structure 97: Deviation from the design center position of the HAR structure 100: Object Illumination Unit 101: Object plane or image plane 102: Objective lens 103: Field Lens Group 105: Optical Axis 108: First beam intersection 110: Focused beam grating scanner 141: Principal Bundle Point 143: Distortion Vector 150: Charged Particle Small Beam 151f: Real Beam Trajectory 153: Deflection Electrode 155: Equipotential line 157: Path 159: Virtual Pivot Point 161: Virtual pivot point 163: First-order beam path 171: System 189: Intersection Volume 200: Detection Unit 205: Projection System 206: Electrostatic Lens 207: Image Sensor 208: Electrostatic or magnetic lens 209: Electrostatic or magnetic lens 210: Electrostatic or magnetic lens 212: Second intersection 214: Aperture 216: Active Components 218: Third Deflection Unit 220: Focused beam grating scanner 222: Second Deflection System 300: Charged Particle Multibeam Generator 301: Primary charged particle source 303: Collimating Lens 305: Single-beam forming unit 306: Perforated plate 307: First Field Lens 308: Second Field Lens 309: A single charged particle beam 311: Primary Electron Beam Point 321: Intermediate Image Plane 390: Light Beam Controlled Perforated Plate 400: Beam splitter unit 420: Magnetic components 500: Platform 503: Sample Charged Unit 700 pixels 701, 701': Features 702: Grayscale image 710: Binary Image 711: Target Grid 712: Structure 713: Center 714: Actual Mesh 715: Distortion Vector 716: Vector 717: Vector 720: Actual Mesh 721: Single contour position 722: Connecting line at the upper edges of opposite sides of the connecting structure 723: The region containing the midpoint of the line at the center of the structure 724: Midpoint of the line 725: Subfield Boundary 726: The first part of the feature 727: The second part of the feature 730: Vector Distortion Mapping 800: Control Unit 810: Image Data Acquisition Unit 811: Analog-to-logarithmic converter 812: Kernel Generation Unit 813: Hardware Filter Unit 814: Image Memory 815: Average Unit 816: Counting Unit 817: Image stitching unit 818: Image Processing and Output 820: Projection System Control Module / Imaging Control Module 830: Primary Beam Path Control Module 900: Grid Configuration 901: Filter element 902: First temporary register 903: Second temporary register 904: Multiplication Block 905: Addition block 906: Shift register 907: Kernel Window 910: Filter kernel 930: Scanning Control Unit S1: Provide a complex number of vector distortion maps for each image subfield. S2: Identify features of interest within the image. S3: Extract the geometric properties of this feature. S4: Determine the corresponding image subfield containing the geometric properties of the captured features. S5: Determine one or more locations of the captured geometric feature within the determined corresponding image subfield. S6: Based on the vector distortion mapping of the corresponding image subfield, correct one or more locations in the image where geometric features have been extracted, thereby establishing distortion-corrected image data. S7: End or other steps dv: Distance within distorted image d: Distance within the image after distortion correction vp1: First part of the distortion vector vp1: Distortion vector, second part p: Internal coordinates of the image subfield q: Internal coordinates of the image subfield x: Global coordinates y: Global coordinates

Claims

1. A method for determining the distortion correction location of a feature in an image composed of one or more image patches, each image patch being composed of a plurality of image subfields, each of the image subfields being imaged by a correlated small beam of a multi-beam charged particle microscope, the method comprising the following steps: a) providing a plurality of vector distortion maps for each of the image subfields, each of the vector distortion maps characterizing the position-related distortion of each pixel of the correlated image subfield; b) identifying a feature of interest in the image; c) extracting the geometric characteristics of the feature of interest; d) determining a corresponding image subfield containing the extracted geometric characteristics; e) determining one or more locations of the extracted geometric characteristics within the corresponding image subfield; and f) correcting the one or more locations of the extracted geometric characteristics in the image based on the vector distortion maps of the corresponding image subfields, thereby establishing distortion-corrected image data.

2. The method as described in claim 1, wherein steps b) to f) of the method are repeatedly performed for a plurality of features of interest.

3. The method as described in claim 1, wherein other regions in the image that do not contain any of the features of interest are not subject to distortion correction.

4. The method as described in claim 1 or 3, wherein the geometric properties of the feature of interest are at least one of the following: contour, edge, corner, point, line, circle, ellipse, center, diameter, radius, distance.

5. The method as described in claim 1 or 3, wherein capturing the geometric feature includes generating a binary image.

6. The method as described in claim 1 or 3, wherein the captured geometric feature of the feature of interest extends over the plurality of image subfields and is thus divided into individual plurality of parts, and wherein, based on the relevant individual vector distortion mapping in the corresponding image subfield of the individual parts, one or more locations of each part of the captured geometric feature are individually corrected.

7. The method as described in claim 1 or 3, wherein the geometric properties of the feature of interest are extracted over the entire image.

8. The method as described in claim 1 or 3, wherein the step of correcting one or more locations of the geometric feature captured in the image based on a vector distortion mapping of a corresponding image subfield includes determining a distortion vector for at least one location of the captured geometric feature.

9. The method as described in claim 1 or 3, wherein the step of correcting one or more locations of the geometric feature captured in the image based on a vector distortion mapping of a corresponding image subfield includes converting a pixel of the image into at least one pixel of the distortion-corrected image based on a distortion vector.

10. The method as described in claim 1 or 3, wherein each of the plurality of vector distortion maps is described by a polynomial expansion in a vector polynomial.

11. The method as described in claim 1 or 3, wherein each of the plurality of vector distortion maps is described by a 2D lookup table.

12. The method as described in claim 1 or 3, further comprising at least one of the following steps: determining the dimensions of a semiconductor device structure in the distortion-corrected image data; determining a region of a semiconductor device structure in the distortion-corrected image data; determining the positions of a plurality of regular objects in the semiconductor device, particularly HAR structures, in the distortion-corrected image data; determining the line edge roughness in the distortion-corrected image data; and / or determining the overlap error between different features within the semiconductor device in the distortion-corrected image data.

13. The method as described in claim 1, further comprising the steps of: providing a test sample having a precisely known and particularly repetitive pattern that defines a target mesh; imaging the test sample using a multibeam charged particle microscope, analyzing the obtained image, and determining the actual mesh based on the analysis; determining the positional deviation between the actual mesh and the target mesh; and obtaining the vector distortion mapping for each image subfield based on the positional deviation.

14. The method as described in claim 13, further comprising moving the test sample relative to the multi-beam charged particle microscope from a first position to a second position, and imaging the test sample in the first position and the second position.

15. The method as described in any of claims 13 to 14, wherein determining the positional deviation comprises two steps, wherein in the first step, offset of each image subfield, rotation of each image subfield, and magnification of each subfield are compensated, and wherein in the second step, the remaining higher-order distortions are determined.

16. The method as described in request item 1 or 3 further comprises the step of: updating the vector distortion mapping.

17. The method as described in claim 1 or 3 further comprises the step of: correcting distortion in the image by means of streaming the data during image preprocessing.

18. A computer program product comprising program code for performing the methods described in claim 1 or 3 above.

19. A multi-beam charged particle microscope having a controller configured to perform the method as described in claim 1 or 3.

Images