Method and system for automatic crystallographic axis alignment

Abstract

Methods and systems for automatic alignment of crystallographic axes. Automatic alignment of a sample's crystallographic axes with a charged particle beam is achieved based on a diffraction pattern of the sample. A region corresponding to a Laue circle is segmented using a trained network. The sample is aligned with the charged particle beam by tilting the sample with a crystallographic axis tilt determined based on the segmented region.

Description

TECHNICAL FIELD

[0001] The present specification generally relates to methods and systems for aligning a sample with an incident beam, and more specifically, to methods and systems for automatically aligning a crystallographic axis of a crystalline sample with a charged particle beam. BACKGROUND

[0002] For high-resolution charged particle beam microscopes, to image a crystalline sample with high precision, the charged particle beam must be aligned with the crystallographic axis of the crystalline sample. If the crystallographic axis of the sample is not aligned, for example when the crystallographic axis is not oriented parallel to the incident beam, measurements of nanoscale features on the sample can be inaccurate. The process of aligning the crystal structure of a sample with the incident beam is referred to as crystallographic axis alignment.

[0003] One method of crystallographic axis alignment is based on a diffraction pattern formed by a parallel beam. When a parallel charged particle beam passes through a thin crystalline sample, the charged particles interfere with each other and form a diffraction pattern on the back focal plane of an objective lens located below the sample. The diffraction pattern consists of a plurality of bright diffraction spots. The diffraction spots belonging to the zeroth Laue circle can be determined, and the crystallographic axis misalignment can be determined based on the positions of the center of the zeroth Laue circle and the center of the direct beam. However, the applicants have recognized that under certain conditions, the diffraction spots of the zeroth Laue circle are not easily identifiable. As one example, when the sample is curved or bent, the diffraction spots of different crystallographic axis orientations can be mixed in the diffraction pattern. As another example, under illumination by a convergent beam, the diffraction spots become discs and can be elongated and overlap with each other and / or the direct beam. SUMMARY

[0004] In one embodiment, a method can be used to align the crystallographic axis of a sample, the method comprising directing a charged particle beam toward the sample; acquiring a diffraction pattern of the sample; segmenting, using a trained network, regions of the diffraction pattern corresponding to Laue circles; determining a crystallographic axis tilt based on the segmented regions; and tilting the sample based on the determined crystallographic axis tilt. In this way, the crystallographic axis of a curved sample can be automatically aligned based on a diffraction pattern acquired by a convergent beam.

[0005] It should be appreciated that the above Summary is provided to introduce some concepts in a simplified form, not to identify key or essential features of the claimed subject matter. The scope of the claimed subject matter is defined only by the claims that follow the detailed description, and the claimed subject matter encompasses many alternatives, modifications and equivalents. Furthermore, the claimed subject matter is not limited to implementations that solve any or all of the disadvantages of the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0006] Figure 1 A system for imaging a sample is shown in accordance with some embodiments.

[0007] Figure 2A and Figure 2B Illustrates the relationship between the incident beam and the sample.

[0008] Figure 3 is a high level flowchart of a method for zone axis alignment.

[0009] Figure 4 Illustrates a coordinate system for adjusting the orientation of a sample.

[0010] Figure 5A is an example diffraction pattern acquired by a convergent beam.

[0011] Figure 5B Illustrates the output of a trained network.

[0012] Figure 5C Illustrates the positions of the Laue circle center and the direct beam center.

[0013] Figure 5D is a diffraction pattern of a sample with the zone axis aligned with the incident beam.

[0014] Figure 6 is a flowchart of a method for training a network.

[0015] Figure 7A Illustrates an example annotated diffraction pattern.

[0016] Figure 7B Illustrates an example output of a network.

[0017] Throughout the several views of the drawings, like reference numerals refer to corresponding parts. DETAILED DESCRIPTION

[0018] The following description relates to systems and methods for aligning the zone axis of a sample with a charged particle beam. To perform high resolution imaging, for example, high resolution scanning transmission electron microscope (STEM) imaging using an imaging system of Figure 1 The zone axis of a crystalline sample must be parallel to the incident beam for high resolution imaging. In a STEM, charged particles including transmitted charged particles, scattered electrons, and secondary electrons are detected via a detector on one side of the sample in response to charged particles impinging on the other side of the sample.

[0019] A sample imaged in transmission mode can be curved due to reduced sample thickness or poor mechanical support. As one example, a sample can have a curvature greater than 5 degrees over a small region of 2 pm 2 Figure 2A ​As shown, because the sample 202 is not flat under the illumination of the parallel beam 201, diffraction spots from different crystal orientations can be mixed in the diffraction pattern. Thus, it is difficult to align the charged particle beam with one of the multiple crystal orientations based on the diffraction pattern. In contrast, as shown, the diffraction pattern acquired by the convergent beam 203 is formed by a sample with a uniform crystal orientation. However, the diffraction pattern formed by the convergent beam is more complex than the diffraction pattern formed by the parallel beam. Unlike the spots in the parallel diffraction pattern that resemble high intensity diffraction spots, the diffraction spots formed by the convergent beam diffraction pattern are disc-like, elongated, and overlap with each other and the direct beam. Figure 2B

[0020] Figure 3 A method for crystallographic axis alignment based on a diffraction pattern is shown. The diffraction pattern can be acquired using a convergent beam or a parallel beam. A trained network is used to segment a first region in the diffraction pattern corresponding to a Laue circle and a second region corresponding to a direct beam from the diffraction pattern. Based on the output of the trained network, a quality factor representing the quality of the segmentation is generated. In one instance, the quality factor is generated based on the shape of the segmented region corresponding to the Laue circle. In another instance, the quality factor is generated based on the relative position of the two segmented regions. Crystallographic axis tilts can be determined based on the position of the segmented regions. For example, based on the relative position of the center of the first segmented region and the center of the second segmented region, the angles of the crystallographic axis tilts in two orthogonal tilt directions are derived. The sample orientation can be adjusted based on the crystallographic axis tilts to align the crystallographic axis of the sample with the incident beam. The sample orientation can be further adjusted until the crystallographic axis tilts converge to a threshold crystallographic axis tilt. Figure 4 A coordinate system for adjusting the sample orientation is shown. Figure 5A to Figure 5B Exemplary results of crystallographic axis alignment are shown.

[0021] A method for generating a trained network is shown in Figure 6 The network can be trained with a plurality of diffraction patterns acquired by tilting a reference sample at a plurality of tilt angles. The diffraction patterns of the reference sample can be acquired under the same beam conditions as used for imaging. The regions of the Laue circle and the direct beam in the plurality of diffraction patterns are automatically annotated based on the known tilt angles. The regions of the Laue circle and the direct beam can each be a circle. The plurality of diffraction patterns and the annotated diffraction patterns are used to train the network. The network outputs two segmented regions corresponding to the Laue circle and the direct beam in the diffraction patterns. The parameters of the network can be adjusted based on the similarity between the segmented Laue circle and the annotated Laue circle. Figure 7A to Figure 7B Exemplary annotated diffraction patterns and network outputs are shown.

[0022] Referring to Figure 1 ​STEM system 100 includes an electron source 10 that emits charged particles, such as an electron beam 11, toward a focusing column 12. The electron beam can generate high-energy electrons, i.e., electrons having typical energies of about 10 keV and 1,000 keV. In some embodiments, the focusing column 12 can include one or more of a condenser lens 121, an aperture 122, a scan coil 123, and an upper objective lens 124. The focusing column 12 focuses the electrons from the electron source 10 into a small spot on a sample 14. Different locations of the sample can be scanned by adjusting the electron beam direction through the scan coil 123. For example, by operating the scan coil 123, the incident beam 112 (as shown in dashed line) can be deflected or scanned to focus on different locations of the sample 14. The sample 14 can be thin enough to not obstruct a majority of the electrons in the electron beam 11.

[0023] A principal axis 110 of the imaging system can be a central axis of the electron beam emitted from the electron source 10. The principal axis 110 can also be a central axis of the condenser lens 121. When the incident beam is not deflected or scanned (i.e., incident beam 112), the incident beam can be focused at a location where the principal axis 110 intersects the sample 14.

[0024] The sample 14 can be held by a sample holder 13. The sample holder 13 can adjust the sample orientation by tilting and / or translating the sample. As an example, Figure 4 A coordinate system for adjusting the sample orientation is shown. In Figure 4 the incident beam 112 can be focused on the sample 14 along the principal axis 110 of the imaging system. The z-axis can be parallel to the principal axis 110. The x-y plane can be a plane that is perpendicular to the z-axis. The sample 14 can be tilted relative to the principal axis 110 by rotating about the x-axis or about the y-axis. For example, the direction of rotation about the x-axis can be an alpha tilt direction 1001, and the direction of rotation about the y-axis can be a beta tilt direction 1002. The sample holder can also translate or deflect the sample 14 along any of the x-axis, the y-axis, and the z-axis. In some embodiments, the sample 14 can be rotated about the z-axis.

[0025] Returning to Figure 1 Electrons 101 that pass through the sample 14 can enter a projector 116. In one embodiment, the projector 116 can be a separate part from the focusing column. In another embodiment, the projector 116 can be an extension of the lens field from the lenses in the focusing column 12.

[0026] The controller 30 can adjust the projector 116 so that the direct electrons passing through the sample impinge on the disc-shaped bright-field detector 115, while the diffracted or scattered electrons that are more strongly deflected by the sample are detected by one or more of the high-angle annular dark-field (HAADF) detector 18 and the annular dark-field (ADF) detector 19. The signals from the HAADF and ADF detectors can be amplified by amplifiers 20 and 21, respectively. The signal from the bright-field detector 115 can be amplified by amplifier 22. The signals from the amplifiers 20, 21, and 22 can be sent to the image processor 24, which can form an image of the sample 14 from the detected electrons. The HAADF detector 18, the ADF detector 19, and the bright-field detector 115 can be scintillator-photomultiplier detectors, solid-state PIN detectors, or metal plates. The STEM system 100 can simultaneously detect signals from one or more of the ADF detector, the ADF detector, and the HAADF detector.

[0027] The axial region of the sample 14 can be aligned with the incident beam 112 based on a diffraction pattern of the sample 14 obtained when the sample is illuminated with the incident beam 112. In one embodiment, the diffraction pattern can be obtained via the camera 142 by obtaining the diffraction pattern formed on the phosphor screen 141. The phosphor screen 141 can be inserted between the projector 116 and the bright-field detector 115 during the zone axis alignment. For example, the phosphor screen 141 can be positioned between the HAADF detector 18 and the ADF detector 19. The HAADF detector can be retracted to acquire the diffraction pattern. In another embodiment, the diffraction pattern on the phosphor screen can be captured via the camera 143 positioned downstream of the bright-field detector 115. The camera 143 can be a CCD or CMOS camera. In some embodiments, the diffraction pattern can be acquired by a pixelated detector. The pixelated detector can also be used to detect one or more of the bright-field, ADF, and HAADF images. The acquired diffraction pattern can be sent to the controller 30 for determining the zone axis tilt.

[0028] The controller 30 can control the operation of the STEM system 100 manually in response to operator instructions or automatically in accordance with computer readable instructions stored in the non-transitory memory 32. The controller 30 can be configured to execute the computer readable instructions and control the various components of the STEM system 100 in order to implement any of the methods described herein. For example, the controller can adjust the beam position on the sample by operating the scanning coils 123. The controller can adjust the cross-section of the incident beam by adjusting one or more apertures and / or lenses in the focusing column 12. The controller can adjust the sample orientation relative to the incident beam by tilting the sample holder 13. The controller can shift the sample relative to the incident beam by translating the sample holder 13. The controller 30 can be further coupled to the display 31 to display notifications and / or images of the sample. The controller 30 can receive user input from the user input device 33. The user input device 33 can include a keyboard, mouse, or touch screen.

[0029] Although a STEM system is described by way of example, it should be understood that the present technology can be used to align a zone axis with a collimated incident beam. The present technology can also be useful when applied to sample alignment in other charged particle beam microscope systems such as transmission electron microscope (TEM) systems, scanning electron microscope (SEM) systems, and dual beam microscope systems. The current discussion of STEM imaging is provided as an example of only one suitable imaging modality.

[0030] Figure 3 A method 300 for aligning a zone axis of a sample with an incident beam in an imaging system of a STEM system such as Figure 1 The sample can be curved or bent. In one example, the sample has a curvature equal to or greater than 0.5 degrees in a 2um 2 In another example, the sample has a curvature equal to or greater than 5 degrees in a 2um 2 Based on the convergent beam diffraction pattern, the zone axis at a particular sample position can be automatically aligned with the incident beam (e.g. the principal axis 110 of the STEM system 100). Figure 1

[0031] At 302, the condition of the imaging system is checked. Checking the system condition can include checking one or more of whether the system is operational, whether a suitable sample is inserted, and whether the required system settings are in place. Operating the system can include, but is not limited to, opening the column valve, operating the electron source, and functioning system vacuum. The system settings include, but are not limited to, the required aperture size, condenser lens current, beam position, camera length, electron potential, and beam current. Step 302 also includes acquiring a low resolution, large field of view (FOV) sample image. The sample image can be a STEM image. The sample image can be used to locate a region of interest (ROI) for high resolution imaging. For example, a large FOV STEM image is acquired at 5,000X magnification.​

[0032] At 304, an initial sample image of the ROI is acquired. The initial sample image has a higher resolution and a smaller FOV than the sample image acquired at 302. The initial sample image can be acquired by the HAADF detector.

[0033] At 306, a diffraction pattern of the sample is acquired. The diffraction pattern can be acquired by focusing the charged particle beam at a point within the ROI. For example, the diffraction pattern is acquired by focusing the charged particle beam at the center of the ROI and receiving the charged particles with the detector in transmission mode. During the diffraction pattern acquisition, the HAADF detector is retracted. The diffraction pattern can be acquired with the camera 143 of the Figure 1

[0034] At 308, the trained network receives the diffraction pattern and outputs two segmented regions of the diffraction pattern. Figure 5A An exemplary diffraction pattern formed by a convergent charged particle beam is shown. The direct beam 503 is a bright circle produced by the incident beam that is transmitted directly through the sample to hit the detector without scattering. The diffraction spots 504 of the Laue circle overlap with the direct beam. Figure 5B The output of the trained network is shown. The output of the trained network is an image with the same size (or pixels) as the diffraction pattern. The output image contains two segmented regions 501 and 502. The first segmented region 501 corresponds to the Laue circle and the second segmented region 502 corresponds to the direct beam.

[0035] At 310, a quality factor representing the quality of the output of the trained network (or the segmentation quality of the trained network) is compared to a threshold quality factor. In one example, the quality factor is determined based on the shape of the segmented region corresponding to the Laue circle. If the shape of the segmented Laue circle in the diffraction pattern is closer to a circle, the quality factor is higher. In another example, the quality factor is lower if the two segmented regions do not intersect. In yet another example, the quality factor can be determined based on the degree of overlap between the two segmented regions. The higher the degree of overlap, the higher the quality factor. If the quality factor is greater than a predetermined threshold quality factor, the method 300 proceeds to step 310, if not, the operator can be notified at 322.

[0036] At 326, the diffraction pattern can optionally be used to update the parameters of the trained network at 326. Step 326 also includes determining the validity of the diffraction pattern. Non-crystalline materials can result in an invalid diffraction pattern, in which case, zone axis alignment cannot be performed. If the diffraction pattern is valid, the diffraction pattern can be annotated and used to update and retrain the trained network.

[0037] ​At 312, a crystallographic axis tilt is determined based on the network output. The crystallographic axis tilt includes tilt angles with respect to two orthogonal axes (e.g., a and b tilt angles with respect to the x and y axes of Figure 4 ). The crystallographic axis tilt can be determined based on the locations of the centers of the two segmentation regions in the trained network output. In one example, the center of each segmentation region can be the geometric center of the segmentation region. In another example, one or more of the segmentation regions can be fitted with a circle. The center of the segmentation region is the center of the fitted circle. In yet another example, the trained network is used to only segment the region corresponding to the Laue circle, but not the region corresponding to the direct beam. The direct beam location can be at a fixed known location, in which case only the segmentation region of the Laue circle is needed.

[0038] Figure 5C The center 505 of the segmentation region 502 corresponding to the direct beam is shown, and the center 506 of the segmentation region 501 of the Laue circle corresponding to the network output of Figure 5B is shown. These centers are shown with respect to the diffraction pattern of Figure 5A . The distance 507 between the centers of the two segmentation regions increases with the degree of misalignment between the crystallographic axis and the incident beam axis. The crystallographic axis tilt increases with the distance 507. The distance 507 on each of the two orthogonal axes (e.g., the x and y axes of Figure 4 ) increases with the increase in the crystallographic axis tilt angle about the axis.

[0039] At 314, the crystallographic axis tilt determined at 312 is compared to a threshold crystallographic axis tilt to assess the degree of misalignment. In one example, the crystallographic axis tilt in the a direction and the b direction can be compared to threshold a and b region tilt thresholds, respectively. The a and b crystallographic axis tilt thresholds can be 0.5 degrees. If the crystallographic axis tilt is less than the threshold crystallographic axis tilt, the crystallographic axis is aligned, and the method 300 proceeds to 324 to acquire a high resolution sample image. If not, the sample is tilted at 316 to align the crystallographic axis of the sample with the incident beam.

[0040] In another example, instead of the crystallographic axis tilt, the degree of overlap between the direct beam and the Laue circle is compared to a threshold overlap to assess the degree of misalignment. The method can proceed to 324 in response to the degree of overlap being greater than the threshold overlap.

[0041] At 316, the sample orientation is adjusted based on the crystallographic axis tilt. For example, the sample is tilted by the crystallographic axis tilt with respect to the two orthogonal axes. After the sample is tilted, the sample height and / or the focus of the charged particle beam on the sample surface are adjusted.

[0042] Figure 5DA diffraction pattern of the sample is shown after adjusting the sample orientation based on the goniometer axis tilt. The diffraction points of Laue circles converge to and overlap with the direct beam. The center of the Laue circles and the direct beam also overlap with each other.

[0043] At 318, a second sample image is acquired after tilting the sample, and the sample is shifted based on a comparison between the second sample image and the initial sample image. The sample is shifted by operating a sample holder of the sample holder 13, for example Figure 1 The second sample image is acquired with the same parameters as the initial sample image acquired at 304. The second image can be acquired by the HAADF detector after the HAADF detector is inserted into the beam path. The sample is shifted in the x-y plane shown at Figure 4 based on a displacement between the second sample image and the initial sample image. The sample is shifted to compensate for sample drift caused by tilting the sample and mechanical shift caused by tilting. If the system has a concentric holder in both orthogonal directions and the sample drift is minimal, step 318 can be skipped. The concentric holder maintains the concentricity of the sample and does not cause sample drift / shift during sample tilting.

[0044] At 320, a third sample image can be optionally acquired, and the charged particle beam is shifted based on a comparison between the third sample image and the initial sample image. The beam shift can be achieved by operating the scan coils 123. Figure 1 The third sample image is acquired with the same parameters as the initial sample image acquired at 304. The sample is shifted in the x-y plane shown at Figure 4 based on a displacement between the third sample image and the initial sample image. Compared to shifting the sample using the sample holder at 318, the beam shift can achieve a higher resolution shift between the sample and the incident beam. If the sample holder has sufficient shift precision, this step is not needed.

[0045] If the goniometer axis tilt determined at 314 is less than a threshold goniometer axis tilt, a high resolution image of the sample is acquired at 324. The resolution of the acquired sample image is higher than the resolution of the sample images acquired with goniometer axis alignment or sample drift compensation (e.g., the sample images acquired at 304, 318, and 320).

[0046] In some examples, the trained network can only segment the region corresponding to the Laue circles. The center of the Laue circles can be determined based on the segmented region. The center of the direct beam can be predetermined, for example, the center of the diffraction pattern. The sample can be tilted based on the positions of the Laue circle center and the direct beam center in the diffraction pattern.

[0047] In this way, the crystallographic axis of a curved or bent sample is automatically aligned with the incident beam based on the convergent beam diffraction pattern of the sample. Since the convergent beam can have a nanometer scale beam profile (or beam cross section at the sample surface), the crystal orientation can be aligned at a small selected area. Using the trained network, the crystallographic axis tilt is automatically determined based on the location of the segmented direct beam and segmented Laue circle in the diffraction pattern. The quality of the network output is evaluated prior to tilting the sample to ensure the accuracy of the crystallographic axis tilt estimation. Although the crystallographic axis alignment using convergent beam diffraction patterns is provided as an example herein, the method 300 can also implement the crystallographic axis automatic alignment using parallel beam diffraction patterns.

[0048] Figure 6 A method 600 for training a network with one or more reference samples is shown. The network can include a single machine learning network or multiple machine learning networks working in combination. The individual machine learning networks can correspond to a CAN, ANN, GAN, FCN, U-NET, YOLO, Mask R-CNN, or any other type of machine learning network capable of image segmentation. For example, the network can include a fully convolutional neural network. In accordance with the present disclosure, the network is trained with a plurality of diffraction patterns of the one or more reference samples and annotated diffraction patterns.

[0049] The reference sample can be flat or have a curvature less than Figure 3 The curvature of the imaged sample. A flat reference sample can ensure that the acquired diffraction pattern represents the specified tilt. Otherwise, due to the sample curvature, the actual tilt can not match the specified tilt if the sample shifts when the holder is tilted. If sample drift compensation is used, as described in step 608, the need for flatness is reduced. The sample material can or can not be the same material as the imaged sample. The reference sample and the imaged sample are crystalline of the same lattice type. For example, the reference sample and the imaged sample can be silicon near the 110 crystallographic axis. It does not matter whether the silicon has been processed by lithography, etching, doping, or other semiconductor manufacturing processes. Support for other lattice types can be achieved by including these lattice types during training. The thickness of the reference sample can be different or the same as the imaged sample if the charged particle beam can be transmitted through the material.

[0050] At 602, the crystallographic axis of the reference sample is aligned with the incident beam axis. The crystallographic axis of the reference sample can be aligned manually.

[0051] At 604, an initial image of the reference sample is acquired. The reference sample image can be a STEM image used for compensating sample drift.

[0052] At 606, the reference sample is tilted within a predetermined tilt range for a predetermined step size, resulting in a diffraction pattern of the tilted reference sample. For example, the tilt step size can be 1 degree, and the tilt range in both the α-rotation direction and the β-rotation direction can be -5 degrees to 5 degrees. In another example, the tilt step size can be varied based on the total sample tilt angle. The tilt step size can be reduced at smaller tilt angles. This can be compared with... Figure 3 The same system configuration or parameters are used during zone axis alignment to obtain diffraction patterns. For example, with... Figure 3 The same beam convergence angle in 304 is used to obtain the diffraction pattern.

[0053] At position 608, a second reference sample image is acquired, and the reference sample is offset by comparing the second reference sample image with the initial reference sample image. Similar to... Figure 3 In step 314, at 606, the sample is offset in the xy plane to compensate for sample drift during sample tilting. Step 608 may further include offsetting the incident beam based on a comparison of the second reference sample image and the initial reference sample image.

[0054] At 610, method 600 checks whether a complete tilt sequence has been acquired. If a complete tilt sequence has been acquired, the method moves to 612. Otherwise, at 606, the sample is tilted further according to a predetermined step size.

[0055] At position 612, each diffraction pattern in the tilt sequence is annotated based on the known zone axis tilt corresponding to the diffraction pattern. Annotating the diffraction pattern involves masking the regions corresponding to the direct beam and the Laue circle within the diffraction pattern. The region corresponding to the direct beam can be annotated based on the known beam position in the diffraction pattern and the known beam size proportional to the beam convergence angle. The region corresponding to the Laue circle can be annotated based on the known zone axis tilt angle.

[0056] In one instance, such as Figure 7A The annotated diffraction pattern shows two circles representing the direct beam and the Laue circle. Circle 701 corresponds to the direct beam, and circle 702 corresponds to the Laue circle. The center position and radius of circle 702 are determined based on the known sample zone axis tilt from the incident beam and the beam convergence angle.

[0057] Return to Figure 6 At position 614, both the tilted sequence and the annotated tilted sequence are used to train the network. The network receives the diffraction pattern as input and outputs two segmented regions corresponding to the straight beam and the Laue circle. Figure 7B An exemplary output of the network is shown. The output is an image containing a first segmented region 704 corresponding to the Laue circle and a second segmented region 703 corresponding to the direct beam. This can be based on... Figure 7A Annotated diffraction patterns andFigure 7B The similarity of the network output to the network output in the training set can be used to adjust the parameters of the network. For example, the similarity can be computed based on the cross-entropy of the two images. Training can be completed until the difference between the annotated diffraction pattern and the network output is less than a predetermined threshold level or when the difference no longer improves. The parameters of the trained network can be saved in the non-transitory memory of the system for use in the determination of the crystallographic orientation Figure 3 Method 300) of the crystallographic orientation.

[0058] In some examples, the network is trained with multiple reference samples. The reference samples can have different thicknesses or different crystal orientations. They can be different materials or lattice types. In some examples, the network is trained on tilt series collected at different points of the reference sample.

[0059] In some examples, the Laue circles used for network training and crystallographic orientation determination are zeroth order Laue circles. In other examples, the Laue circles can include higher order Laue circles, such as a combination of zeroth order Laue circles and first order Laue circles.

[0060] In some examples, the network is trained to segment only the regions corresponding to Laue circles and not the regions corresponding to straight beams.

[0061] In this way, a trained network for segmenting Laue circles and straight beams is generated based on multiple diffraction patterns of one or more reference samples. Annotations for each diffraction pattern are automatically generated based on the known crystallographic tilt angles and known beam convergence angles of the diffraction patterns.

Claims

1. A method for laue axis alignment, comprising: directing a charged particle beam at a sample; acquiring a diffraction pattern of the sample; acquiring a diffraction pattern tilt sequence comprising a plurality of diffraction patterns; annotating the diffraction pattern tilt sequence by annotating a direct beam and a Laue circle in each diffraction pattern of the diffraction pattern tilt sequence; training a network using the acquired diffraction pattern tilt sequence and the annotated diffraction pattern tilt sequence; segmenting a first region of the diffraction pattern corresponding to the Laue circle and segmenting a second region of the diffraction pattern corresponding to the direct beam using the trained network; and determining a laue axis tilt based on relative positions of the first region and the second region.

2. The method of claim 1, further comprising: tilting the sample based on the determined laue axis.

3. The method of claim 1, further comprising: determining a shape of the segmented regions; and in response to the shape of the segmented regions being circular, determining the laue axis tilt based on the segmented regions.

4. The method of claim 1, wherein, the charged particle beam is a convergent charged particle beam.

5. The method of claim 1, wherein, the charged particle beam is a parallel charged particle beam.

6. The method of claim 1, further comprising: training the network with a plurality of diffraction patterns acquired with the charged particle beam.

7. The method of claim 1 or 6, further comprising: training the network with a plurality of diffraction patterns of a second sample, the second sample having a smaller curvature than the sample.

8. The method of claim 1, wherein, training the network with the acquired plurality of diffraction patterns and a plurality of annotated diffraction patterns comprises: inputting the plurality of diffraction patterns to the network; and adjusting parameters of the network by comparing an output of the network to the plurality of annotated diffraction patterns.

9. A method for laue axis alignment, comprising: directing a convergent charged particle beam at a sample; acquiring a diffraction pattern of the sample; acquiring a diffraction pattern tilt sequence comprising a plurality of diffraction patterns; annotating the diffraction pattern tilt sequence by annotating a direct beam and a Laue circle in each diffraction pattern of the diffraction pattern tilt sequence; training a network using the acquired diffraction pattern tilt sequence and the annotated diffraction pattern tilt sequence; segmenting a first region of the diffraction pattern corresponding to the Laue circle and segmenting a second region of the diffraction pattern corresponding to the direct beam using the trained network; and determining a laue axis tilt based on relative positions of the first region and the second region.

10. The method of claim 9, wherein, the annotated direct beam and the annotated Laue circle are circles.

11. The method of any of claims 9-10, further comprising: determining a quality factor based on the first region and the second region; and determining the laue axis tilt based on the first region and the second region comprises determining the laue axis tilt in response to the quality factor.

12. The method of claim 11, determining the quality factor based on the first region and the second region comprises determining the quality factor based on an amount of overlap between the first region and the second region.

13. The method of claim 11, determining the quality factor based on the first region and the second region comprises determining the quality factor based on a shape of the first region.

14. The method of claim 9, further comprising, in response to the determined crystallographic axis tilt being greater than a threshold crystallographic axis tilt, acquiring a second diffraction pattern of the sample and tilting the sample using a second crystallographic axis tilt determined based on the second diffraction pattern using the trained network.

15. A system for imaging a sample, comprising: a source to generate a charged particle beam; a sample holder to tilt the sample; a detector; and a controller having instructions stored in non-transitory memory, the controller configured to: direct the charged particle beam toward the sample; acquire a diffraction pattern of the sample; acquire a diffraction pattern tilt sequence comprising a plurality of diffraction patterns; annotate the diffraction pattern tilt sequence by annotating each diffraction pattern of the diffraction pattern tilt sequence with a direct beam and a Laue circle; train a network using the acquired diffraction pattern tilt sequence and the annotated diffraction pattern tilt sequence; segment, using the trained network, a first region of the diffraction pattern corresponding to the Laue circle and a second region of the diffraction pattern corresponding to the direct beam; and determine a crystallographic axis tilt based on relative positions of the first region and the second region. the controller is further configured to determine a quality factor based on the segmented regions and tilt the sample in response to the quality factor. determining the crystallographic axis tilt based on the segmented regions includes determining a position of a Laue circle center in the diffraction pattern based on the segmented regions and determining the crystallographic axis tilt based on the position of the Laue circle center and a position of a direct beam center in the diffraction pattern.

16. The system of claim 15, wherein, the charged particle beam is a convergent charged particle beam, and wherein the controller is further configured to tilt the sample based on the determined crystallographic axis and acquire an image of the sample by scanning the convergent charged particle beam over the sample after tilting the sample.

17. The system of claim 15, wherein, ​ 18. The system of any one of claims 15 to 17, wherein, ​

Images