Electron microscope, aberration correction method, and imaging method

Abstract

To provide an aberration-correctable electron microscope that is hardly affected by a drift of a sample.SOLUTION: An electron microscope includes an electron optical system and a control part which controls the electron optical system, and the control part performs processing to find a standard deviation of a luminance distribution of an electron microscopic image, processing to find an optimum value of a parameter of the electron optical system which maximizes the standard deviation through Gaussian process regression, and processing to capture the electron microscopic image with the optimum value of the parameter, and then determines the value of the parameter by repeating the processing to find the standard deviation, the processing to find the optimum value, and the processing to capture the electron microscopic image.SELECTED DRAWING: Figure 11

Description

[Technical field]

[0001] The present invention relates to an electron microscope, an aberration correction method, and an imaging method. [Background technology]

[0002] In electron microscopes such as transmission electron microscopes, scanning transmission electron microscopes, and scanning electron microscopes, aberration correction is an important technique for obtaining high-resolution images.

[0003] For example, Patent Document 1 discloses a method for correcting the focus and astigmatism of an electron lens, in which focus correction and astigmatism correction are performed based on the image sharpness of an electron microscope image. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2005-108567 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, when correcting aberrations based on the shape of an image, such as the image sharpness of an electron microscope image, as in the aberration correction method disclosed in Patent Document 1, if the image is deformed due to drift of the sample, the focus and astigmatism cannot be accurately corrected.

[0006] Furthermore, in a method such as the method disclosed in Patent Document 1, in which a parameter such as a focal position is swept to search for an optimal condition where an index such as image sharpness has an extreme value, when there are multiple parameters to be adjusted, the index must be calculated for all combinations of the parameters, and correction takes a lot of time. For example, when sweeping three parameters under five conditions, the number of measurements and calculations is 5 x 5 x 5 = 625. Increasing the number of measurements not only increases the time required, but also increases the amount of electron beam irradiation to the sample, which increases the damage to the sample caused by electron beam irradiation, making it difficult to perform good microscope observation. [Means for solving the problem]

[0007] One aspect of the electron microscope according to the present invention is An electron optical system; A control unit for controlling the electron optical system; Including, The control unit is A process of calculating a standard deviation of the brightness distribution of the electron microscope image; A process of obtaining optimal values ​​of parameters of the electron optical system that maximize the standard deviation by Gaussian process regression; a process of capturing the electron microscope image with the parameter values ​​set to the optimal values; Do the following: The process of determining the standard deviation, the process of determining the optimum value, and the process of taking the electron microscope image are repeated to determine the value of the parameter.

[0008] In such an electron microscope, the control unit determines the optimal values ​​of the parameters of the electron optical system using the standard deviation of the brightness distribution of the electron microscope image as an index, so that the influence of sample drift can be reduced compared to, for example, a case where aberration is corrected based on the shape of the image in the electron microscope image. Therefore, in such an electron microscope, aberration can be corrected more accurately.

[0009] Furthermore, with such an electron microscope, the optimal values ​​for parameters can be found by combining only a very small number of parameters, which not only shortens the time required for correction, making it more efficient, but also reduces damage to the sample.

[0010] One aspect of the aberration correction method according to the present invention includes: 1. A method for correcting aberrations in an electron microscope including an electron optical system, comprising: determining a standard deviation of the brightness distribution of the electron microscope image; determining, by Gaussian process regression, optimal values ​​of parameters of the electron optical system that maximize the standard deviation; taking the electron microscope image with the parameter values ​​set to the optimal values; Including, The step of determining the standard deviation, the step of determining the optimum value, and the step of taking an electron microscope image are repeated to determine the value of the parameter.

[0011] In this aberration correction method, the standard deviation of the brightness distribution of the atomic microscope image is used as an index to find the optimal values ​​of the parameters of the electron optical system, so the influence of sample drift can be reduced compared to, for example, a case where aberration is corrected based on the shape of the image in the electron microscope image. In addition, the use of Gaussian process regression makes it possible to significantly reduce the number of iterations. Therefore, this aberration correction method can correct aberrations more accurately and quickly.

[0012] One aspect of the imaging method according to the present invention includes: correcting the aberration of the electron optical system by the aberration correction method; a step of capturing an electron microscope image having a number of pixels greater than the number of pixels of the electron microscope image captured in the step of capturing the electron microscope image; Includes.

[0013] With this imaging method, it is possible to obtain an electron microscope image with a large number of pixels in a state in which the aberrations have been accurately corrected by the aberration correction method, and further, with this imaging method, it is possible to shorten the imaging time of the electron microscope image used for aberration correction.

[0014] One aspect of the imaging method according to the present invention includes: correcting the aberration of the electron optical system by the aberration correction method; capturing an electron microscope image at a magnification higher than the magnification of the electron microscope image captured in the step of capturing the electron microscope image; Includes.

[0015] In this imaging method, an electron microscope image with a high imaging magnification can be obtained in a state in which the aberration has been accurately corrected by the aberration correction method. Furthermore, in this imaging method, damage to a sample caused by irradiation with an electron beam can be reduced when capturing an electron microscope image used for aberration correction. [Brief description of the drawings]

[0016] [Figure 1] FIG. 1 is a diagram showing the configuration of an electron microscope according to a first embodiment. [Diagram 2] 4 is a flowchart showing an example of an aberration correction method in the electron microscope according to the first embodiment. [Diagram 3] 1 is a table showing conditions for correction of low-order aberrations. [Figure 4] Fitting results using the first Gaussian process regression. [Diagram 5] Second Gaussian process regression fitting results. [Figure 6] Fitting results using the third Gaussian process regression. [Figure 7] Fitting results using the fourth Gaussian process regression. [Figure 8] Fitting results using Gaussian process regression. [Figure 9] Fitting results using Gaussian process regression. [Figure 10]11 is a flowchart showing an example of a process for correcting low-order aberrations by a control unit. [Figure 11] 10 is a flowchart showing an example of a process for adjusting an objective lens of a control unit. [Figure 12] 10 is a flowchart showing an example of a process for adjusting an astigmatism corrector in a control unit. [Figure 13] 10 is a flowchart showing an example of a process for adjusting an axial coma aberration corrector in a control unit. [Figure 14] FIG. 1 is a diagram for explaining the hill climbing method. [Figure 15] FIG. 1 is a diagram for explaining the hill climbing method. [Figure 16] FIG. 1 is a diagram for explaining the hill climbing method. [Figure 17] 11 is a flowchart showing an example of a process for correcting low-order aberrations by a control unit. [Figure 18] 10 is a flowchart showing a modified example of the process of correcting low-order aberrations by the control unit. [Figure 19] 10 is a flowchart showing a modified example of the process of adjusting the objective lens of the control unit. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0017] Preferred embodiments of the present invention will be described in detail below with reference to the drawings. Note that the embodiments described below do not unduly limit the contents of the present invention described in the claims. In addition, not all of the configurations described below are necessarily essential components of the present invention.

[0018] 1. First embodiment 1.1. Electron microscope First, an electron microscope according to a first embodiment will be described with reference to the drawings. Fig. 1 is a diagram showing the configuration of an electron microscope 100 according to the first embodiment.

[0019] As shown in FIG. 1, the electron microscope 100 includes an electron source 10, an irradiation optical system 20, a sample holder 32, an imaging optical system 40, an imaging device 50, an annular detector 52, a detector 54, a control unit 60, an operation unit 70, a display unit 72, and a memory unit 74.

[0020] The electron microscope 100 can be switched between a TEM mode and a STEM mode. In the TEM mode, a sample image (TEM image) and an electron diffraction pattern are formed by electrons transmitted through or diffracted by the sample S, and these can be imaged by the imaging device 50. In the STEM mode, a STEM image (bright-field STEM image) can be acquired by scanning the sample S with a finely focused electron beam and detecting the electrons transmitted through the sample S at each scanning position with the detector 54. Also, a STEM image (dark-field STEM image) can be acquired by scanning the sample S with a finely focused electron beam and detecting the electrons scattered by the sample S at each scanning position with the annular detector 52.

[0021] The electron source 10 generates an electron beam. The electron source 10 is, for example, an electron gun that accelerates electrons emitted from a cathode by an anode to emit an electron beam.

[0022] The projection optical system 20 irradiates the electron beam emitted from the electron source 10 onto the sample S. The projection optical system 20 includes, for example, a condenser lens 22, an aperture 24, a two-fold astigmatism corrector 26, and an axial coma aberration corrector 28.

[0023] The condenser lens 22 focuses the electron beam emitted from the electron source 10 and irradiates the sample S. In the electron microscope 100, the electron beam is narrowed to form an electron probe by the forward magnetic field of the condenser lens 22 and the objective lens 42. That is, in the STEM mode, the irradiation optical system 20 includes a lens created by the forward magnetic field of the objective lens 42.

[0024] The two-fold astigmatism corrector 26 corrects the two-fold astigmatism of the condenser lens 22. The astigmatism corrector 26 includes, for example, an X stigma coil for correcting distortion of the electron beam in the X direction (first direction) and a Y stigma coil for correcting distortion of the electron beam in the Y direction (second direction). The X and Y directions are two axes that are orthogonal to each other and perpendicular to the optical axis OA of the electron optical system.

[0025] The axial coma aberration corrector 28 corrects the axial coma aberration by two-dimensionally deflecting the electron beam incident on the sample S. The axial coma aberration corrector 28 includes, for example, an X alignment coil for correcting the tilt of the electron beam in the X direction (third direction) and a Y alignment coil for correcting the tilt of the electron beam in the Y direction (fourth direction).

[0026] Here, the first direction in which the X stigmer coil corrects distortion and the third direction in which the X alignment coil tilts the electron beam are the same direction (X direction), but the first direction in which the X stigmer coil corrects distortion and the third direction in which the X alignment coil tilts the electron beam may be different. Similarly, here, the second direction in which the Y stigmer coil corrects distortion and the fourth direction in which the Y alignment coil tilts the electron beam are the same direction (Y direction), but the second direction in which the Y stigmer coil corrects distortion and the fourth direction in which the Y alignment coil tilts the electron beam may be different.

[0027] In the electron microscope 100, an electron probe is formed by narrowing an electron beam using the forward magnetic field of the condenser lens 22 and the objective lens 42, and the electron probe is two-dimensionally deflected by a scanning deflector (not shown). This allows the sample S to be scanned with the electron probe.

[0028] The aperture 24 (condenser aperture) cuts off a part of the electron beam irradiated to the sample S. For example, the aperture 24 has a plurality of aperture holes with different diameters. By switching between these aperture holes, the aperture angle and the irradiation amount of the electron beam can be adjusted.

[0029] The sample holder 32 holds the sample S. The tip of the sample holder 32 supports the sample S. The sample holder 32 is attached to the microscope body via a goniometer 34. The goniometer 34 has a drive mechanism that moves and tilts the sample holder 32.

[0030] The imaging optical system 40 forms an image of the electrons transmitted through the sample S. The imaging optical system 40 includes an objective lens 42, an intermediate lens 44, and a projection lens 46.

[0031] In the TEM mode, the imaging optical system 40 forms an electron diffraction pattern and a specimen image (TEM image) with electrons (transmitted electrons) that have passed through the specimen S. For example, by focusing the intermediate lens 44 on the specimen image (TEM image) created by the objective lens 42, a specimen image can be created on the object plane of the projection lens 46. This allows the specimen image to be captured by the imaging device 50. Also, by changing the focal length of the intermediate lens 44 and focusing on the electron diffraction pattern created by the objective lens 42, an electron diffraction pattern can be created on the object plane of the projection lens 46. This allows the electron diffraction pattern to be captured by the imaging device 50.

[0032] In the STEM mode, the imaging optical system 40 guides electrons transmitted through the sample S to the detector 54 and guides electrons scattered by the sample S to the annular detector 52 .

[0033] The illumination optical system 20 and the imaging optical system 40 constitute the electron optical system of the electron microscope 100 .

[0034] The imaging device 50 captures the sample image and the electron diffraction pattern formed by the imaging optical system 40. The imaging device 50 is, for example, a digital camera such as a CCD (Charge Coupled Device) camera. The image data of the TEM image and the electron diffraction pattern captured by the imaging device 50 is The data is sent to the control unit 60 and stored in the memory unit 74 .

[0035] The annular detector 52 detects electrons scattered by the sample S. The annular detector 52 may be, for example, a detector that detects electrons inelastically scattered at a high angle by the sample S. This makes it possible to acquire a high-angle annular dark-field STEM image (HAADF-STEM image).

[0036] The detector 54 detects the electrons that have transmitted through the sample S. By detecting the electrons that have transmitted through the sample S with the detector 54, a bright-field STEM image can be obtained.

[0037] The operation unit 70 allows the user to input operation information and outputs the input operation information to the control unit 60. The functions of the operation unit 70 can be realized by hardware such as a keyboard, a mouse, buttons, a touch panel, and a touch pad.

[0038] The display unit 72 displays an image generated by the control unit 60. For example, a captured sample image or an electron diffraction pattern is displayed on the display unit 72. The function of the display unit 72 can be realized by an LCD (Liquid Crystal Display), a CRT (Cathode Ray Tube), a touch panel that also functions as the operation unit 70, or the like.

[0039] The storage unit 74 stores programs and various data for causing a computer to function as the control unit 60. The storage unit 74 also functions as a work area for the control unit 60. The functions of the storage unit 74 can be realized by a RAM (Random Access Memory), a hard disk, or the like.

[0040] The control unit 60 controls each part of the electron microscope 100, such as the electron source 10, the irradiation optical system 20, the sample holder 32, the goniometer 34, the imaging optical system 40, the image pickup device 50, the annular detector 52, and the detector 54. The functions of the control unit 60 can be realized by executing programs using various processors, such as a CPU (Central Processing Unit) and a DSP (Digital Signal Processor).

[0041] 1.2. Aberration correction method 1.2.1. Overview of aberration correction methods Low-order aberrations such as defocus, two-fold astigmatism, and axial coma tend to fluctuate, while higher-order aberrations tend to be more stable. Therefore, once high-order aberrations have been adjusted, there is almost no need to readjust them until the end of the observation or analysis. However, low-order aberrations require readjustment in a short period of time. Therefore, it is desirable to be able to correct low-order aberrations using the actual sample to be observed, without using a dedicated sample for aberration correction.

[0042] As a method for correcting aberration, a method of correcting aberration using a Ronchigram is known. To obtain a Ronchigram, it is better to use an amorphous region of a sample, and an aperture with a larger diameter is required than the aperture for imaging an atomic image. Therefore, a Ronchigram is obtained in an amorphous region using an aperture with a larger diameter, and after correcting the aberration, the aperture is changed from the large diameter to the small diameter, the field of view is moved to the region to be imaged, and the sample is imaged. Since the conditions for obtaining a Ronchigram are different from the imaging conditions, it is desirable to correct low-order aberrations again when imaging.

[0043] In the aberration correction method in the first embodiment, low-order aberrations are corrected using the standard deviation of the luminance distribution of the atomic image as an index. In other words, the index is whether or not a bright atomic image is captured in a dark vacuum region without blurring and with clear contrast. Therefore, in the aberration correction method in the first embodiment, low-order aberrations can be corrected using the sample that is actually observed, without using a dedicated sample. Furthermore, the standard deviation of the luminance distribution of the atomic image does not depend on the shape of the image in the image or the type of residual aberration. Therefore, by using the standard deviation of the luminance distribution of the atomic image as an index for evaluating aberrations, This makes it possible to reduce the influence of sample drift, as compared to, for example, a case in which aberrations are corrected based on the shape of the image.

[0044] The standard deviation of the brightness distribution of the atomic image can be obtained, for example, by extracting the brightness of each pixel of the atomic image and calculating the standard deviation from the brightness of each extracted pixel. The atomic image used for the aberration correction is, for example, an annular dark-field STEM image (ADF-STEM image) acquired using the annular detector 52. The atomic image used for the aberration correction may be, for example, an HAADF-STEM image.

[0045] The parameters of the optical elements adjusted for the correction of low-order aberrations are the excitation of the objective lens 42, the excitation of the X stigma coil (first excitation) and the excitation of the Y stigma coil (second excitation) of the two-fold astigmatism corrector 26, and the excitation of the X alignment coil (third excitation) and the excitation of the Y alignment coil (fourth excitation) of the axial coma aberration corrector 28.

[0046] Hereinafter, the current setting value for exciting objective lens 42 is referred to as the OL value. Also, the current setting value for exciting the X stigma coil of two-fold astigmatism corrector 26 is referred to as the X stigma value, and the current setting value for exciting the Y stigma coil of two-fold astigmatism corrector 26 is referred to as the Y stigma value. Also, the current setting value for exciting the X alignment coil of axial coma aberration corrector 28 is referred to as the X alignment value, and the current setting value for exciting the Y alignment coil of axial coma aberration corrector 28 is referred to as the Y alignment value.

[0047] In addition, when the irradiation optical system 20 has two stages of alignment coils that tilt the electron beam, i.e., a first stage alignment coil and a second stage alignment coil, the on-axis coma aberration is corrected using the second stage alignment coil, which can adjust the tilt of the irradiation axis of the electron beam.

[0048] In the aberration correction method in the first embodiment, the low-order aberrations are corrected by finding the optimal values ​​of these five parameters. The optimal parameter values ​​are values ​​that maximize the standard deviation of the brightness distribution of the atomic image. The optimal parameter values ​​are found using Gaussian process regression.

[0049] If we were to search a five-dimensional space with each of the above five parameters as an axis, the large number of dimensions would result in many combinations of conditions at the edges of the search area that are less likely to produce the optimal value (the curse of dimensionality), making the search inefficient.

[0050] Therefore, we divide the five parameters into multiple groups to reduce the dimension, and search for the optimal value in each group. The parameters are classified by considering the independence between the parameters. Here, we will show two classification methods.

[0051] One of the two classification methods is classification according to the order of the aberration. If low-order aberrations remain, higher-order aberrations are drowned out by the lower-order aberrations and are less likely to appear. Therefore, classification is performed in ascending order of order. Defocus and two-fold astigmatism have the same order but different symmetries, and defocus has a greater effect on higher-order aberrations. Therefore, the optimum values ​​are searched for in the order of defocus, two-fold astigmatism, and on-axis coma aberration. For example, the space is divided into three groups: a one-dimensional space of only the OL value, a two-dimensional space of the X stigma value and the Y stigma value, and a two-dimensional space of the X alignment value and the Y alignment value, and the optimum values ​​are searched for in this order. It is also possible to divide the space into two groups: a three-dimensional space of the OL value, the X stigma value, and the Y stigma value, and a two-dimensional space of the X alignment value and the Y alignment value, and the optimum values ​​are searched for in this order.

[0052] The other of the two classification methods is classification that considers the correlation between parameters. Parameter X and parameter Y are said to be correlated if the peak position obtained by sweeping in the X direction is closer to that of Y. It means that it varies depending on the value.

[0053] If the OL value is deviated, the beam shape may resemble the atomic arrangement under specific conditions where the X stigma value and Y stigma value are deviated from the optimal value, and the standard deviation may become large. This is because <110> This can occur when there are places where the atomic arrangement is very close, such as a dumbbell structure obtained by irradiating the electron beam from the direction, or when the irradiation optical system 20 does not have an aberration corrector and the diameter of the electron probe is as large as the atomic spacing. In this case, it can be said that there is a correlation between the OL value, the X stigma value, and the Y stigma value. On the other hand, in the case of the X alignment value and the Y alignment value, the correlation with the OL value is smaller than that in the case of the stigma.

[0054] Therefore, the optimum value may be searched for by dividing the space into two groups: the three-dimensional space of OL value, X stigma value, and Y stigma value, and the two-dimensional space of X alignment value and Y alignment value.The optimum value may also be searched for by dividing the space into three groups: the two-dimensional space of OL value and X stigma value, the two-dimensional space of OL value and Y stigma value, and the two-dimensional space of X alignment value and Y alignment value.

[0055] Whether classification by the order of aberration or by the correlation between parameters is more appropriate depends on the type of sample and the configuration of the instrument. Therefore, which classification method to use is determined based on the type of sample and the configuration of the instrument.

[0056] As described above, the number of searches can be reduced by classifying the parameters and then finding the optimal parameter values ​​using Gaussian process regression.

[0057] 1.2.2. Aberration correction process FIG. 2 is a flowchart showing an example of a method for correcting aberrations in the electron microscope 100.

[0058] <Preparation> First, before correcting the low-order aberrations, the high-order aberrations are corrected and the low-order aberrations are roughly adjusted. The high-order aberrations and the low-order aberrations may be corrected using, for example, a Ronchigram or a sample image. Alternatively, the high-order aberrations and the low-order aberrations may be corrected using a dedicated sample.

[0059] For the correction of higher-order aberrations and the coarse adjustment of lower-order aberrations, known aberration correction methods can be used, such as the Segmental Ronchigram Autocorrelation function Method (SRAM method) using a Ronchigram or a deconvolution method using a sample image of a dedicated sample in which gold particles are scattered on amorphous carbon. When a dedicated sample is used, the observation sample is introduced into the electron microscope 100 after the correction of higher-order aberrations and the coarse adjustment of lower-order aberrations. The orientation of the sample (the orientation of the sample relative to the direction of incidence of the electron beam) is adjusted in the region to be observed of the introduced observation sample. When the SRAM method is used to correct higher-order aberrations and the coarse adjustment of lower-order aberrations, the aperture for Ronchigram is replaced with the aperture for imaging, and the field of view is moved to enable imaging of atomic images.

[0060] <Setting conditions S10> In order to correct low-order aberrations, the type of optical element, the search range of the parameters of each optical element, the imaging conditions for the atomic image, the function form of the Gaussian process regression, and the hyperparameters are set.

[0061] Fig. 3 is a table showing the conditions for correcting low-order aberrations. The table in Fig. 3 lists the types of optical elements, the search ranges for the setting values ​​of each optical element, the imaging conditions, the function form of Gaussian process regression, and the conditions for hyperparameters.

[0062] The optical elements to be adjusted are the objective lens 42 (OL), the two-fold astigmatism corrector 26 (CL Stig), and the second-stage X-alignment coil and Y-alignment coil (CL2 Align) of the axial coma aberration corrector 28. Here, the optimum values ​​are searched for in the order of defocus, two-fold astigmatism, and axial coma aberration. That is, the optimum values ​​are searched for by dividing into three groups: a one-dimensional space of only the OL value, a two-dimensional space of the X-stigma value and the Y-stigma value, and a two-dimensional space of the X-alignment value and the Y-alignment value. As described above, the method of dividing into groups is not limited to this.

[0063] The imaging conditions for the STEM image used for aberration correction are set as imaging magnification, image size, and irradiation time. A certain number of pixels is required for the image size in order to calculate the standard deviation from the brightness distribution, and it is desirable for the image size to be 32 pixels x 32 pixels = 1024 pixels or more. The more pixels there are, the more precisely the standard deviation can be calculated, but since imaging takes time, it is desirable for the image size to be 512 pixels x 512 pixels or less.

[0064] Also, to obtain an image with atomic resolution, the image resolution must be smaller than the atomic interval. In general, the atomic interval is about several hundred pm (picometers), so a high magnification is required to make the image resolution about a fraction of that, about 100 pm or less. However, if the magnification is too high, fewer atoms are captured in the image, and even a slight difference in the shooting position changes the number of atoms that fit in the image, causing the standard deviation of the brightness distribution to fluctuate. Therefore, it is desirable to capture a certain number of atoms within the set number of pixels, and it is desirable to capture about several tens of atoms, or about four periods of crystal lattice. The shooting conditions must satisfy these constraints, and although the specific conditions vary depending on the model, the lower limit of the imaging magnification is generally 40M (mega) times, and the upper limit depends on the number of pixels, but is about 100M (mega) times or less in the case of 32 pixels x 32 pixels. In the example shown in Figure 3, the imaging magnification is set to 80M times. The image size is 64 pixels x 64 pixels. The irradiation time of the electron beam per pixel is, for example, about 10 μs or less. In the example shown in Figure 3, the electron beam irradiation time per pixel is set to 8 μs. Note that the appropriate imaging conditions for the atomic image used for aberration correction depend on the performance of the instrument and the atomic arrangement of the sample, and can therefore be changed as appropriate.

[0065] In the example shown in FIG. 3, the search range of the setting value of each optical element is expressed in bits. In this embodiment, prior to the aberration correction by this method, aberration correction using a Ronchigram is performed, so the amount of aberration remaining at this stage is small. Therefore, the search range can be set narrow. In the example shown in FIG. 3, the search range of the OL value is -32 bits or more and +32 bits or less, which corresponds to a focus deviation amount of -20 nm or more and +20 nm or less. In the example shown in FIG. 3, the search range of the X stigma value and the Y stigma value is -1024 bits or more and +1024 bits or less, and the search range of the X alignment value and the Y alignment value is -32 bits or more and +32 bits or less. This corresponds to approximately ±100 nm and ±1000 nm, respectively, in terms of aberration. Note that the appropriate search range depends on the state and stability of the device and the performance of the aberration corrector, and can be changed as appropriate.

[0066] In the example shown in Figure 3, the function form of the Gaussian process regression is set to a normalized Gaussian (normalized Gaussian + error) that includes error. The normalized Gaussian is the same type as the Gaussian kernel, but it is a kernel function multiplied by hyperparameters (coefficients) so that the shape of the fitting function changes relatively when the search range or the fluctuation range of the standard deviation changes. The hyperparameters θ1, θ2, and θ3 will be described later.

[0067] The normalized Gaussian K including error (Gaussian kernel K including error) is expressed by the following equation (1).

[0068]

number

[0069] where θ1, θ2, and θ3 are hyperparameters, and x i ,x j is the input variable vector, and δ ij is the Kronecker delta.

[0070] As described above, the optical elements are adjusted in ascending order of aberration order. Therefore, the adjustments are performed in the order of the objective lens 42, the two-fold astigmatism corrector 26, and the axial coma aberration corrector 28. The order in which the optical elements are adjusted can be changed as appropriate depending on the sample and the device.

[0071] <Adjusting the objective lens S20> First, the electron optical system is set to the initial state, and the standard deviation of the brightness distribution of the atomic image is measured. Specifically, first, an atomic image of the observation sample is captured in the electron microscope 100. The image is captured under the imaging conditions shown in FIG. 3. That is, the image is captured at an imaging magnification of 80M times, an image size of 64 pixels x 64 pixels, and an electron beam irradiation time of 8 μs per pixel. This makes it possible to obtain an atomic image of the observation sample. Next, the standard deviation of the brightness distribution is calculated from the captured atomic image. The above steps allow the standard deviation of the brightness distribution of the atomic image to be measured.

[0072] Next, the OL value of the objective lens 42 that maximizes the standard deviation is obtained by Gaussian process regression.

[0073] Specifically, first, the measurement result of the standard deviation of the brightness distribution of the atomic image captured with the OL value as the initial value is plotted on a graph with the OL value on the horizontal axis and the standard deviation of the brightness distribution of the atomic image on the vertical axis. Next, fitting is performed with a predetermined function and hyperparameters on the graph on which the measurement points showing the standard deviation when the OL value is the initial value are plotted. Here, the above formula (1) is used as the predetermined function. In addition, θ1=1, θ2=1, and θ3=10 are used as the hyperparameters.

[0074] Figure 4 shows the fitting results by the first Gaussian process regression. In Figure 4, the fitting function F of the fitting result is shown by a solid line, and the +1σ confidence interval and the -1σ confidence interval are shown by dashed lines. The graph shown in Figure 4 plots a measurement point P0 indicating the measurement result of the standard deviation of the brightness distribution of an atomic image captured with the OL value as the initial value. The initial value of the OL value is, for example, the center value of the search range.

[0075] As shown in Fig. 4, a fitting function F can be obtained as a prediction model for predicting the relationship between the OL value and the standard deviation by Gaussian process regression. Furthermore, the reliability of the OL value predicted by the prediction model (fitting function F) can be obtained by Gaussian process regression.

[0076] The fitting function F is expressed as a straight line parallel to the horizontal axis that passes through the measurement point P0, which indicates the standard deviation when the OL value is the initial value. The two dashed lines indicate the range where the probability is ±1σ, that is, the 1σ confidence interval. In other words, this indicates that an error will occur in the fitting result within the range indicated by the two dashed lines.

[0077] In this way, in Gaussian process regression, not only is the fitting function F obtained, but also the reliability of the fitting result is obtained. In other words, in Gaussian process regression, not only is a prediction model obtained, but also the reliability of the prediction result predicted by the prediction model is obtained. Therefore, in order to improve the accuracy of the fitting function F in Gaussian process regression, Therefore, the search position can be efficiently determined, and the peak can be efficiently searched for.

[0078] Here, the hyperparameters will be described. θ1 is a parameter that represents the maximum value of the confidence interval. The larger θ1 is, the larger the confidence interval becomes overall. For example, in FIG. 4, the larger θ1 is, the greater the distance between the solid line representing the fitting function F and the dashed line representing the confidence interval becomes.

[0079] θ2 is a parameter that specifies how sharply the confidence interval and the fitting function F change. The larger θ2 is, the sharper the change in the confidence interval and the fitting function F is allowed. θ3 is a parameter that represents the magnitude of the noise (error) in the search data. In other words, θ3 is a parameter that represents the size of the confidence interval at the measurement point. For example, in FIG. 4, when θ3 is set to zero, the two dashed lines that represent the fitting function F and the confidence interval pass through the measurement point P0. In other words, when θ3 is set to zero, the confidence interval becomes zero at the measurement position.

[0080] As shown in FIG. 4, at the position where the OL value is the initial value, that is, the measurement position where the standard deviation is actually measured, the confidence level is high and the confidence interval is small. That is, at the measurement position, the uncertainty of the fitting function F is small. The confidence interval becomes larger as the distance from the measurement position increases. That is, the uncertainty of the fitting function F becomes larger as the distance from the measurement position increases. Therefore, although the fitting function F is a straight line parallel to the horizontal axis, it can be seen that the positions at both ends of the search range are likely to show the maximum value of the standard deviation when the confidence interval is included. In this way, the next search position is determined using the UCB (Upper Confidence Bound) algorithm, which determines the next search position based on the confidence interval. For example, the position where the upper limit value of the confidence interval (confidence interval of +1σ) is maximum is likely to have the maximum standard deviation. Therefore, the position where the upper limit value of the confidence interval is maximum is set as the optimal value of the OL value, that is, the next search position.

[0081] 4, the OL value of the objective lens 42 that is likely to maximize the standard deviation, i.e., the optimal value of the OL value, is the position at both ends of the search range. Therefore, one of the values ​​at both ends of the search range, i.e., the lower limit value or the upper limit value of the search range, is set as the optimal value of the OL value. The found optimal value becomes the next search position for improving the accuracy of the fitting function F.

[0082] In this way, once the next search position is determined, the standard deviation of the luminance distribution of the atomic image is measured at the next search position. That is, the standard deviation of the luminance distribution of the atomic image is measured with the OL value set to the optimum value. The measurement of the standard deviation with the OL value set to the optimum value is performed in the same manner as the measurement of the standard deviation with the OL value set to the initial value described above, except that the OL value is set to the optimum value. Here, a case will be described where the standard deviation is measured with the optimum OL value set to the lower limit value of the search range.

[0083] Next, the OL value of the objective lens 42 that maximizes the standard deviation is obtained by Gaussian process regression. At this time, θ1=1, θ2=1, and θ3=10 are used as hyperparameters, similarly to the first fitting shown in Fig. 4. Here, the values ​​of the hyperparameters θ1, θ2, and θ3 are fixed, but the hyperparameters may be optimized using a known algorithm each time the fitting is repeated.

[0084] Figure 5 shows the fitting results by the second Gaussian process regression. The graph shown in Figure 5 plots measurement point P0 and measurement point P1, which shows the measurement results of the standard deviation of the brightness distribution of an atomic image captured with the OL value set as the lower limit of the search range. Note that the dashed lines indicating the -1σ confidence interval are omitted in Figure 5.

[0085] As shown in Fig. 5, by plotting measurement point P1 in addition to measurement point P0, The fitting function F has a slope, and the confidence interval also changes. In this case, the optimal value of the OL value at which the standard deviation is maximum is the upper limit value of the search range of the OL value, as shown in FIG. 5. Therefore, the upper limit value of the search range of the OL value is set as the optimal value. In other words, the upper limit value of the search range of the OL value is set as the next search position.

[0086] In this way, when the next search position is determined, the standard deviation of the brightness distribution of the atomic image is measured at the next search position in the same way. That is, the standard deviation of the brightness distribution of the atomic image is measured with the OL value set to the optimum value (upper limit of the search range). Next, the OL value of the objective lens 42 that maximizes the standard deviation is found by Gaussian process regression.

[0087] Figure 6 shows the fitting results by the third Gaussian process regression. The graph shown in Figure 6 plots measurement point P0, measurement point P1, and measurement point P2, which shows the measurement result of the standard deviation of the brightness distribution of an atomic image captured with the OL value as the upper limit value of the search range.

[0088] As shown in Fig. 6, by plotting measurement point P2 in addition to measurement points P0 and P1, the fitting function F becomes a parabola. The optimum value of the OL value is found from the fitting result shown in Fig. 6. In the example shown in Fig. 6, taking into consideration the fitting function F and the confidence interval, the optimum value of the OL value is set to the position where OL value = A. For example, the position where the dashed line of +1σ indicating the upper limit of the confidence interval peaks is set to the optimum value of the OL value.

[0089] Fig. 7 shows the fitting results by the fourth Gaussian process regression. The graph shown in Fig. 7 plots measurement point P0, measurement point P1, measurement point P2, and measurement point P3, which shows the measurement result of the standard deviation of the brightness distribution of an atomic image captured with an OL value of OL=A.

[0090] 7, the optimum value obtained as a fitting result by the fourth Gaussian process regression coincides with the optimum value (OL value=A) obtained as a fitting result by the third Gaussian process regression. Therefore, it is determined that the optimum value has converged, and the OL value of the objective lens 42 is determined to be OL value=A.

[0091] Here, in FIG. 7, there is an area where the confidence interval is large. Also, even in the vicinity of the peak position (OL value=A) where the standard deviation is maximum, the confidence interval is wide. A large confidence interval means that the fitting accuracy is not high. However, the objective here is to search for a peak (maximum value). Therefore, even if the fitting accuracy is not high, the objective can be achieved as long as the peak position can be confident. As shown in FIG. 7, the fact that the obtained optimum value matches the previous optimum value means that the possibility of a peak being obtained at another position is low. Therefore, by converging the optimum value, the OL value of the objective lens 42 where the standard deviation is maximum can be determined. The objective lens 42 can be adjusted by the above steps.

[0092] The condition for ending the adjustment of the objective lens 42 is not limited to when the optimum value has converged. For example, the adjustment of the objective lens 42 may be ended when the width of the confidence interval becomes smaller than a preset size in the entire search range.

[0093] Furthermore, if the converged optimal value is generally located at the edge of the search range, it is highly likely that the true optimal value is outside the search range, so by moving the search range - for example, if the optimal value is the minimum value of the range, the search range can be shifted to the smaller side and similar adjustments can be made to arrive at the true optimal value.

[0094] <Adjustment of astigmatism corrector S30> Next, the two-fold astigmatism corrector 26 is adjusted. The standard deviation of the luminance distribution of the atomic image is measured using the value determined in adjusting the lens 42 in S20 and setting the other electron optical systems to their initial states. The measurement of the standard deviation is performed in the same manner as the measurement of the standard deviation in adjusting the objective lens 42 in S20 described above.

[0095] Next, the X stigma value and the Y stigma value of the two-fold astigmatism corrector 26 that maximize the standard deviation are determined by Gaussian process regression.

[0096] Here, in the adjustment of the two-time astigmatism corrector 26, the X stigma value and the Y stigma value are adjusted simultaneously. That is, the two-dimensional space of the X stigma value and the Y stigma value is searched. Therefore, the fitting result by the Gaussian process regression becomes a three-dimensional graph.

[0097] Next, the measurement results of the standard deviation of the brightness distribution of the atomic image captured with the X stigma value and the Y stigma value as the initial values ​​are plotted on a three-dimensional graph and fitting is performed. The fitting is performed in the same manner as the search for the optimal value of the OL value described above, except that it is a three-dimensional graph. As shown in Figure 3, θ1=1, θ2=1, and θ3=10 are used as hyper parameters.

[0098] Figure 8 shows the fitting results by Gaussian process regression. The x1 axis of the graph in Figure 8 is the X-stigma value, the x2 axis is the Y-stigma value, and the y axis is the standard deviation of the brightness distribution of the atomic image. Figure 8 also shows the fitting function F and the confidence interval of ±1σ.

[0099] As shown in Fig. 8, the X stigma value and Y stigma value that are likely to maximize the standard deviation, i.e., the optimal values ​​of the X stigma value and Y stigma value, are the four corner positions of the search range where the confidence interval takes the upper limit. Therefore, the search positions for the X stigma value and the Y stigma value are set to the four corner positions of the search range, and the standard deviation of the brightness distribution of each atomic image is measured.

[0100] Figure 9 shows the fitting results by Gaussian process regression. The graph in Figure 9 plots a measurement point showing the measurement results of the standard deviation of the brightness distribution of atomic images captured with the X stigma value and the Y stigma value as the initial values, and four measurement points showing the measurement results of the standard deviation of the brightness distribution of atomic images captured with the X stigma value and the Y stigma value at each of the four corner positions.

[0101] From the fitting results shown in Fig. 9, the optimal values ​​of the X stigma value and the Y stigma value can be obtained. In this manner, the standard deviation of the brightness distribution of the atomic image is measured with the X stigma value and the Y stigma value set as the optimal values, the optimal values ​​of the X stigma value and the Y stigma value that maximize the standard deviation are obtained by Gaussian process regression, and the atomic image is captured with the X stigma value and the Y stigma value set as the optimal values, and the optimal values ​​are converged upon by repeating this process. For example, if the obtained optimal value matches the previous optimal value, it is determined that the optimal value has converged.

[0102] In this manner, it is possible to determine the X stigma value and the Y stigma value of the two-fold astigmatism corrector 26 that maximizes the brightness distribution of the atomic image. Through the above steps, the two-fold astigmatism corrector 26 can be adjusted.

[0103] <Adjustment of the axial coma aberration compensator S40> Next, the axial coma aberration corrector 28 is adjusted. In adjusting the axial coma aberration corrector 28, an X alignment value and a Y alignment value are determined. The adjustment of the axial coma aberration corrector 28 is similar to the adjustment of the two-fold astigmatism corrector 26 described above, except that the parameters are the X alignment value and the Y alignment value, and therefore a description thereof will be omitted.

[0104] Through the above steps, low-order aberrations can be corrected.

[0105] 1.2.3. Processing of the control section <Processing to correct low-order aberrations> In the electron microscope 100, the control unit 60 performs processing to correct low-order aberrations. Fig. 10 is a flowchart showing an example of processing by the control unit 60 to correct low-order aberrations.

[0106] The control unit 60 judges whether or not the user has instructed to start the correction of low-order aberrations (S100). Although not shown in the figure, when an aberration correction start button is pressed or when an aberration correction start button on a GUI (Graphical User Interface) is operated, it is judged that the user has instructed to start the aberration correction.

[0107] When it is determined that the user has issued an instruction to start the aberration correction (Yes in S100), the control unit 60 adjusts the objective lens 42 (S110).

[0108] After adjusting the objective lens 42, the control unit 60 adjusts the two-fold astigmatism corrector 26 (S120). After adjusting the two-fold astigmatism corrector 26, the control unit 60 adjusts the on-axis coma aberration corrector 28 (S130). After adjusting the on-axis coma aberration corrector 28, the control unit 60 ends the process of correcting low-order aberrations.

[0109] <Process to adjust the objective lens> FIG. 11 is a flowchart showing an example of a process S110 for adjusting the objective lens 42 by the control unit 60. As shown in FIG.

[0110] The control unit 60 sets the OL value of the objective lens 42 to an initial value (S111), and captures an atomic image of the observation sample S (S112).

[0111] The control unit 60 determines the standard deviation of the brightness distribution of the captured atomic image (S113). Based on the determined standard deviation, the control unit 60 determines a fitting function F (prediction model) that predicts the relationship between the OL value and the standard deviation by Gaussian process regression, and a confidence interval (reliability of the prediction result) (S114).

[0112] Next, the control unit 60 obtains the optimum value of the OL value based on the fitting function F and the confidence interval (S115). The control unit 60 determines the OL value that is likely to maximize the standard deviation as the optimum value based on the fitting function F and the confidence interval, as shown in Figures 4, 5, and 6. The control unit 60 determines, for example, the position where the upper limit value of the confidence interval is maximized as the optimum value of the OL value.

[0113] The control unit 60 determines whether the obtained optimum value has converged (S116). For example, the control unit 60 determines that the optimum value has converged when the obtained optimum value matches the previously obtained optimum value (or the initial value).

[0114] If the control unit 60 determines that the found optimal value has not converged (No in S116), it takes an atomic image (S117) with the OL value set as the optimal value found in step S115. After taking an atomic image (after step S117), the control unit 60 returns to step S113, finds the standard deviation of the brightness distribution of the atomic image (S113), finds a fitting function F and a confidence interval that predict the relationship between the OL value and the standard deviation by Gaussian process regression (S114), and finds the optimal value of the OL value based on the fitting function F and the confidence interval (S115).

[0115] The control unit 60 repeats process S117 of capturing an atomic image with the OL value set to the optimal value, process S113 of calculating the standard deviation of the brightness distribution of the atomic image, process S114 of calculating the fitting function F and the confidence interval, and process S115 of calculating the optimal value of the OL value, until it determines in process S116 that the optimal value has converged.

[0116] When the control unit 60 determines that the optimal value has converged (Yes in S116), it ends the process of adjusting the objective lens 42. As a result of the process of adjusting the objective lens 42, the OL value of the objective lens 42 can be set to a value that maximizes the standard deviation of the brightness distribution of the atomic image.

[0117] <Processing for adjusting the astigmatism corrector> FIG. 12 is a flowchart showing an example of a process S120 for adjusting the two-fold astigmatism corrector 26 of the control unit 60. As shown in FIG.

[0118] The control unit 60 sets the X stigma value of the two-fold astigmatism corrector 26 to an initial value, and sets the Y stigma value to an initial value (S121), and captures an atomic image of the sample S (S122).

[0119] The control unit 60 obtains the standard deviation of the brightness distribution of the captured atomic image (S123). Based on the obtained standard deviation, the control unit 60 obtains a fitting function F (prediction model) that predicts the relationship between the X stigma value and the Y stigma value and the standard deviation by Gaussian process regression, and a confidence interval (reliability of the prediction result) (S124).

[0120] Next, the control unit 60 determines the optimal value of the X stigma value and the optimal value of the Y stigma value based on the fitting function F and the confidence interval (S125). As shown in Figures 8 and 9, the control unit 60 determines the optimal value of the X stigma value and the Y stigma value that are likely to maximize the standard deviation, taking into account the fitting function F and the confidence interval.

[0121] The control unit 60 determines whether the obtained optimum value has converged (S126). For example, the control unit 60 determines that the optimum value has converged when the obtained optimum value matches the previously obtained optimum value (or the initial value).

[0122] When the control unit 60 determines that the obtained optimal values ​​have not converged (No in S126), it takes an atomic image with the X stigma value and the Y stigma value set to the optimal values ​​obtained in step S125 (S127).

[0123] After capturing the atomic image (after step S127), the control unit 60 returns to step S123, calculates the standard deviation of the brightness distribution of the atomic image (S123), calculates a fitting function F and a confidence interval for predicting the relationship between the X-stigma value and the Y-stigma value and the standard deviation by Gaussian process regression (S124), and calculates optimal values ​​of the X-stigma value and the Y-stigma value based on the fitting function F and the confidence interval (S125).

[0124] The control unit 60 repeats process S127 of capturing an atomic image with the X stigma value and the Y stigma value as optimal values, process S123 of calculating the standard deviation of the brightness distribution of the atomic image, process S124 of calculating the fitting function F and the confidence interval, and process S125 of calculating the optimal values ​​of the X stigma value and the Y stigma value, until it determines in process S126 that the optimal values ​​have converged.

[0125] When the control unit 60 determines that the optimal values ​​have converged (Yes in S126), it ends the process of adjusting the two-fold astigmatism corrector 26. As a result of the process of adjusting the two-fold astigmatism corrector 26, it is possible to set the X stigma value and the Y stigma value of the two-fold astigmatism corrector 26 to values ​​that maximize the standard deviation of the brightness distribution of the atomic image.

[0126] <Processing for adjusting the axial coma aberration corrector> FIG. 13 is a flowchart showing an example of a process S130 for adjusting the axial coma aberration corrector 28 of the control unit 60.

[0127] The control unit 60 sets the X alignment value of the axial coma aberration corrector 28 to an initial value, and sets the Y alignment value to an initial value (S131), and captures an atomic image of the sample S (S132).

[0128] The control unit 60 obtains the standard deviation of the brightness distribution of the captured atomic image (S133). Based on the obtained standard deviation, the control unit 60 obtains a fitting function F (prediction model) that predicts the relationship between the X alignment value and the Y alignment value and the standard deviation by Gaussian process regression, and a confidence interval (reliability of the prediction result) (S134).

[0129] Next, the control unit 60 determines the optimum value of the X alignment value and the optimum value of the Y alignment value based on the fitting function F and the confidence interval (S135). The control unit 60 determines the X alignment value and the Y alignment value that are likely to maximize the standard deviation as the optimum value, taking into account the fitting function F and the confidence interval.

[0130] The control unit 60 determines whether the obtained optimum value has converged (S136). For example, the control unit 60 determines that the optimum value has converged when the obtained optimum value matches the previously obtained optimum value (or the initial value).

[0131] If the control unit 60 determines that the obtained optimal values ​​have not converged (No in S136), it takes an atomic image with the X alignment value and the Y alignment value set to the optimal values ​​obtained in step S135 (S137).

[0132] After capturing the atomic image (after step S137), the control unit 60 returns to step S133, determines the standard deviation of the brightness distribution of the atomic image (S133), determines a fitting function F and a confidence interval for predicting the relationship between the X alignment value and the Y alignment value and the standard deviation by Gaussian process regression (S134), and determines optimal values ​​of the X alignment value and the Y alignment value based on the fitting function F and the confidence interval (S135).

[0133] The control unit 60 repeats process S137 of capturing an atomic image with the X alignment value and the Y alignment value set to the optimal values, process S133 of determining the standard deviation of the brightness distribution of the atomic image, process S134 of determining the fitting function F and the confidence interval, and process S135 of determining the optimal values ​​of the X alignment value and the Y alignment value, until it determines in process S136 that the optimal values ​​have converged.

[0134] When the control unit 60 determines that the optimal values ​​have converged (Yes in S136), it ends the process of adjusting the axial coma aberration corrector 28. As a result of the process of adjusting the axial coma aberration corrector 28, it is possible to set the X alignment value and the Y alignment value of the axial coma aberration corrector 28 to values ​​that maximize the standard deviation of the brightness distribution of the atomic image.

[0135] 1.3. Imaging method 1.3.1. Electron microscope imaging After correcting the low-order aberrations of the electron optical system by the above-mentioned aberration correction method shown in FIG. 2, a STEM image of the observation sample S is captured. At this time, the number of pixels of the STEM image is made larger than the number of pixels of the atomic image captured for the aberration correction. In this way, with the imaging method in the electron microscope 100, the image size of the STEM image when the observation sample S is imaged can be made larger than the image size of the atomic image captured for the aberration correction. In other words, Therefore, the image size of the atomic image captured for this purpose can be made smaller than the image size when capturing a STEM image of the observation sample S. As a result, the time required to capture the atomic image used for aberration correction can be shortened.

[0136] For example, under the imaging conditions for aberration correction shown in FIG. 3, the image size is 64 pixels×64 pixels, but under the imaging conditions for imaging the observation sample S, the image size is made larger than 512 pixels×512 pixels.

[0137] 1.3.2. Processing of the control section The control unit 60 captures a STEM image of the sample S after the process of correcting low-order aberrations shown in Fig. 10 is completed, that is, after the adjustment process S130 of the axial coma aberration corrector 28 is completed. At this time, the control unit 60 sets the number of pixels of the STEM image to be larger than the number of pixels of the atomic images in the process S117 for capturing an atomic image in Fig. 11, the process S127 for capturing an atomic image in Fig. 12, and the process S137 for capturing an atomic image in Fig. 13. After capturing the STEM image, the control unit 60 ends the process.

[0138] 1.4. Effects The electron microscope 100 includes an electron optical system having an illumination optical system 20, and a control unit 60 that controls the electron optical system. The control unit 60 performs a process of calculating a standard deviation of the brightness distribution of the electron microscope image, a process of calculating optimal values ​​of parameters of the electron optical system that maximize the standard deviation of the brightness distribution of the electron microscope image by Gaussian process regression, and a process of capturing an atomic microscope image with the parameter values ​​set to the optimal values, and determines the parameter values ​​by repeating the process of calculating the standard deviation, the process of calculating the optimal values, and the process of capturing an electron microscope image.

[0139] In addition, in the process of determining the optimal value, the control unit 60 determines a prediction model that predicts the relationship between the parameter and the standard deviation by Gaussian process regression based on the standard deviation of the brightness distribution of the electron microscope image, and the reliability of the parameter value predicted by the prediction model, and determines the optimal value of the parameter based on the prediction model and the reliability.

[0140] In this way, the control unit 60 uses the standard deviation of the brightness distribution of the electron microscope image as an index to find the optimal value of the parameter for correcting the aberration. Since the standard deviation of the brightness distribution of the image does not depend on the shape of the image, the optimal value can be found even for a low-resolution image. Therefore, in the electron microscope 100, the electron microscope image can be captured in a short time, and the aberration can be corrected in a short time. Furthermore, since the standard deviation of the brightness distribution of the image does not depend on the shape of the image, even if the image is deformed due to the drift of the sample, the effect is small. Therefore, in the electron microscope 100, the effect of the drift of the sample can be reduced compared to, for example, a case where the aberration is corrected based on the shape of the image. Therefore, in the electron microscope 100, the aberration can be corrected accurately.

[0141] Furthermore, the control unit 60 uses Gaussian process regression to determine the optimum values ​​of the parameters of the electron optical system that maximize the standard deviation of the brightness distribution of the electron microscope image. Therefore, the electron microscope 100 can efficiently determine the optimum parameter values. The reason for this will be explained below in comparison with other methods.

[0142] Known methods for searching for peaks similar to Gaussian process regression include hill climbing, steepest descent, and conjugate gradient methods.

[0143] 14, 15, and 16 are diagrams for explaining the hill-climbing method.

[0144] The hill climbing method is an algorithm that starts from an initial value, searches the surroundings in a predetermined direction and with a predetermined step size, and if a better condition is found, moves to the better condition. However, However, in the hill climbing method, the accuracy is determined by the step width. Therefore, in the hill climbing method, there is a trade-off between accuracy and speed (number of trials). Furthermore, as shown in FIG. 14, if the step width is large, there is a high possibility that a peak including a maximum value cannot be detected. In the example shown in FIG. 14, the step width is large, so point P is detected as a peak even though it is not a peak. Furthermore, as shown in FIG. 15, if the initial value is not set appropriately, if the step width is small, it is easy to be trapped by a local peak. In the example shown in FIG. 15, if the initial value is set appropriately, point Pa can be detected as the maximum value, but if the initial value is not set appropriately, point Pb, which is a local peak, is detected as the maximum value. Therefore, in the hill climbing method, a technique is used to search with multiple step widths or various initial values. However, the number of trials is large.

[0145] Furthermore, as shown in Fig. 16, the hill-climbing method has a problem that it is difficult to converge if the search direction is not appropriate. In Fig. 16, the search direction is indicated by an arrow, and the detected peak positions are indicated by points. Furthermore, the points indicating the peak positions are given numbers indicating the order in which they were detected.

[0146] As described above, the hill-climbing method requires a large number of trials, and is therefore not suitable for aberration correction.

[0147] The steepest descent method and the conjugate gradient method are methods that address the problems of the hill climbing method. In the steepest descent method and the conjugate gradient method, the gradient at the search position is used to determine the search direction and step size. This makes it possible to improve the speed (number of trials) and to eliminate the difficulty in convergence caused by the search direction.

[0148] However, in the steepest descent method and the conjugate gradient method, in order to calculate the gradient, it is necessary to image not only the search position (x, y) but also the neighboring points (x+dx, y) and (x, y+dy). In addition, there are cases where the method is trapped at a local peak or a position where the gradient is zero.

[0149] Furthermore, the presence of noise is an obstacle to searching for peaks. The standard deviation is affected by luminance noise. When luminance noise is present, it is difficult to use the hill climbing method because of the occurrence of local peaks. In addition, when luminance noise is present, it is difficult to use the steepest descent method and the conjugate gradient method because the gradient calculation becomes inaccurate.

[0150] The luminance noise can be reduced by extending the electron beam irradiation time or by enlarging the imaging range. However, extending the electron beam irradiation time or enlarging the imaging range results in a longer imaging time, which is not suitable for aberration correction.

[0151] In contrast, Gaussian process regression performs fitting over the entire specified search range. Furthermore, Gaussian process regression can calculate the reliability of the fitting result. Therefore, it is less susceptible to local peaks and noise.

[0152] Generally, when there are few measurement points, the accuracy of the fitting is low, and the maximum value cannot be obtained based on the fitting result. However, Gaussian process regression can calculate the reliability of the fitting result, that is, the uncertainty of the fitting result. Therefore, if the upper limit value of the confidence interval at a certain position does not exceed the maximum value of the fitting result, it can be determined that the maximum value is not obtained at that position. Conversely, if the upper limit value of the confidence interval at a certain position exceeds the maximum value of the fitting result, there is a possibility that the maximum value is obtained at that position.

[0153] Therefore, in Gaussian process regression, as described above, the maximum value and its position can be efficiently searched for by preferentially measuring the position where the upper limit value of the confidence interval is maximum. If the position where the upper limit between the two is maximum has already been measured, re-measuring that position will not change the fitting result or the confidence interval. Therefore, the results can be considered to have converged, and the maximum value and its position can be determined.

[0154] In this way, Gaussian process regression searches for the maximum value in the entire search range, so it is not trapped by local peaks. Also, since the measurement positions are not divided and selected, it is possible to search in increments of the minimum unit of the control width, regardless of the size of the search range. Furthermore, Gaussian process regression is resistant to errors, because it can calculate fitting and confidence intervals by incorporating measurement error (noise).

[0155] In Gaussian process regression, a kernel function is set and hyperparameters used for that function are specified. The kernel function specifies how the confidence interval expands as it moves away from the measurement point. There are various kernel functions, such as linear and Gaussian. The magnitude of the noise to be considered is expressed in constant terms. The function form can be selected appropriately.

[0156] The coefficients used in the kernel function, i.e., the hyperparameters, are directly linked to the number of trials. In the first embodiment, the case where the hyperparameters are fixed has been described, but the hyperparameters may be optimized for each search. The hyperparameters can be optimized using a known algorithm.

[0157] In Gaussian process regression, unknown values ​​are assumed to be zero. Therefore, if the standard deviation at the initial value is very large, a very large and steep peak will exist within the search range. In this case, the problem occurs that the position of the upper limit of the confidence interval (peak position) does not change from the initial value. Therefore, instead of using the standard deviation itself as an index, it is also possible to use the index obtained by subtracting the average value of the measured standard deviation, or to use the difference between the measured standard deviation and the standard deviation at the initial value. This makes it possible to prevent the problem of erroneously determining that the initial value is the peak position.

[0158] In the electron microscope 100, the control unit 60 sets the parameter as the excitation (OL value) of the objective lens 42, and repeats the above-mentioned process of finding the standard deviation, the process of finding the optimum value, and the process of capturing an electron microscope image to determine the OL value and adjust the objective lens 42. In addition, the control unit 60 sets the parameter as the excitation (X stigma value) for correcting the distortion of the electron beam in the X direction in the two-fold astigmatism corrector 26 and the excitation (Y stigma value) for correcting the distortion of the electron beam in the Y direction in the two-fold astigmatism corrector 26, and repeats the above-mentioned process of finding the standard deviation, the process of finding the optimum value, and the process of capturing an electron microscope image to determine the X stigma value and the Y stigma value and adjust the two-fold astigmatism corrector 26. Furthermore, control unit 60 repeats the process of obtaining the standard deviation, the process of obtaining the optimum value, and the process of capturing an electron microscope image as parameters, which are excitation for correcting the tilt of the electron beam in the X direction in axial coma aberration corrector 28 (X alignment value) and excitation for correcting the distortion of the electron beam in the Y direction (Y alignment value), to determine the X alignment value and the Y alignment value, and adjusts axial coma aberration corrector 28.

[0159] In this way, the control unit 60 adjusts the parameters by dividing them into three groups: a group of OL values, a group of X stigma values ​​and Y stigma values, and a group of X alignment values ​​and Y alignment values, so that aberrations can be corrected efficiently.

[0160] In the correction of low-order aberrations, five parameters, OL value, X stigma value, Y stigma value, X alignment value, and Y alignment value, are adjusted. Therefore, the optimal values ​​of the parameters can be found by searching a five-dimensional space. However, if the dimension is increased, the search becomes more wasteful due to the so-called curse of dimensionality.

[0161] Therefore, five parameters are divided into groups for searching and adjustment. In the correction of low-order aberrations, the OL value and the stigma value (X stigma value and Y stigma value) are highly independent. In addition, the OL value and the alignment value (X alignment value and Y alignment value) are highly independent. In addition, the stigma value and the alignment value are highly independent. This means that the results of optimizing the X stigma value and the Y stigma value are almost independent of the OL value and the alignment value.

[0162] On the other hand, the X stigma value and the Y stigma value have low independence, so it is better to search and adjust them together. Similarly, the X alignment value and the Y alignment value have low independence, so it is better to search and adjust them together.

[0163] Therefore, search and adjustment are performed by dividing into three groups: OL value, X stigma value and Y stigma value, and X alignment value and Y alignment value. By dividing into three groups in this way, it is possible to reduce waste of search due to the curse of dimensionality and to efficiently correct aberration.

[0164] Note that since the OL value is one-dimensional, it may be included in the group of X stigma values ​​and Y stigma values, or it may be included in the group of X alignment values ​​and Y alignment values.

[0165] In the electron microscope 100, the control unit 60 determines the parameter values, and then performs processing to capture an electron microscope image having a larger number of pixels than the number of pixels of the electron microscope image captured in the processing to capture the electron microscope image. Therefore, it is possible to obtain an electron microscope image with a larger number of pixels, and also to shorten the time required to capture the electron microscope image used for aberration correction.

[0166] 2. Second embodiment 2.1. Electron microscope Next, an electron microscope according to a second embodiment will be described. The configuration of the electron microscope according to the second embodiment is similar to that of the electron microscope 100 according to the first embodiment, and a description thereof will be omitted. Hereinafter, in the electron microscope 100 according to the second embodiment, members having the same functions as those of the constituent members of the electron microscope 100 according to the first embodiment will be given the same reference numerals, and detailed descriptions thereof will be omitted.

[0167] 2.2. Aberration correction method The electron microscope 100 according to the second embodiment differs from the electron microscope 100 according to the first embodiment described above in that the control unit 60 performs a process of determining whether a predetermined time has elapsed since adjusting the objective lens 42, a process of determining whether a predetermined time has elapsed since adjusting the two-fold astigmatism corrector 26, and a process of determining whether a predetermined time has elapsed since adjusting the axial coma aberration corrector 28. Below, the points that differ from the example of the electron microscope 100 according to the first embodiment described above will be described, and a description of the similar points will be omitted.

[0168] FIG. 17 is a flowchart showing an example of processing for correcting low-order aberrations in the control unit 60 of the electron microscope 100 according to the second embodiment.

[0169] The control unit 60 determines whether or not the user has instructed to start the correction of low-order aberrations (S100). When the control unit 60 determines that the user has instructed to start the aberration correction (Yes in S100), the control unit 60 determines whether or not the elapsed time tO from the end of the adjustment process S110 of the objective lens 42 is greater than the elapsed time TO at which the objective lens 42 needs to be readjusted (S210).

[0170] The optimum OL value of the objective lens 42 changes over time. The elapsed time TO is set to a time at which the change in the optimum OL value falls within an allowable range. The control unit 60 starts measuring the elapsed time tO at the timing when the adjustment process S110 of the objective lens 42 shown in Fig. 11 is completed. If the adjustment process S110 of the objective lens 42 has never been performed, the control unit 60 sets the elapsed time tO to a value larger than TO, for example, infinity.

[0171] When the control unit 60 determines that the elapsed time t is greater than the elapsed time TO (Yes in S210), it performs an adjustment process for the objective lens 42 shown in Fig. 11 (S110). After performing the adjustment process for the objective lens 42 (after S110), the control unit 60 returns to the process S210 and determines whether the elapsed time t is greater than the elapsed time TO (S210).

[0172] If the control unit 60 determines that the elapsed time tO is not greater than the elapsed time TO (No in S210), it determines whether the elapsed time tS since the end of the adjustment process S120 of the two-fold astigmatism corrector 26 is greater than the elapsed time TS at which readjustment of the two-fold astigmatism corrector 26 is required (S220).

[0173] The optimal values ​​of the X stigma value and the Y stigma value of the two-fold astigmatism corrector 26 change over time. The elapsed time TS is set to a time at which the change in the optimal value of the X stigma value and the change in the optimal value of the Y stigma value fall within an allowable range. The control unit 60 starts measuring the elapsed time tS at the timing when the adjustment process S120 of the two-fold astigmatism corrector 26 shown in FIG. 12 is completed. Note that, if the adjustment process S120 of the two-fold astigmatism corrector 26 has never been performed, the control unit 60 sets the elapsed time tS to a value larger than TS, for example, infinity.

[0174] When the control unit 60 determines that the elapsed time tS is greater than the elapsed time TS (Yes in S220), it performs the adjustment process of the two-time astigmatism corrector 26 shown in Fig. 12 (S120). After performing the adjustment process of the two-time astigmatism corrector 26 (after S120), the control unit 60 returns to the process S210 and determines whether the elapsed time tO is greater than the elapsed time TO (S210).

[0175] If the control unit 60 determines that the elapsed time tS is not greater than the elapsed time TS (No in S220), it determines whether the elapsed time tA since the end of the adjustment process S130 of the axial coma aberration compensator 28 is greater than the elapsed time TA at which readjustment of the axial coma aberration compensator 28 is required (S230).

[0176] The optimal values ​​of the X alignment value and the Y alignment value of the axial coma aberration corrector 28 change over time. The elapsed time TA is set to a time at which the change in the optimal value of the X alignment value and the change in the optimal value of the Y alignment value fall within an allowable range. The control unit 60 starts measuring the elapsed time tA at the timing when the adjustment process S130 of the axial coma aberration corrector 28 shown in FIG. 13 is completed. Note that if the adjustment process S130 of the axial coma aberration corrector 28 has never been performed, the control unit 60 sets the elapsed time tA to a value larger than TA, for example, infinity.

[0177] When the control unit 60 determines that the elapsed time tA is greater than the elapsed time TA (Yes in S230), it performs an adjustment process of the axial coma aberration corrector 28 shown in Fig. 13 (S130). After performing the adjustment process of the axial coma aberration corrector 28 (after S130), the control unit 60 returns to the process S210 and determines whether the elapsed time tO is greater than the elapsed time TO (S210).

[0178] When the control unit 60 determines that the elapsed time tA is not greater than the elapsed time TA (No in S230), it ends the process of correcting the low-order aberrations.

[0179] 2.3. Effects In the electron microscope 100 according to the second embodiment, the control unit 60 performs a process of determining whether or not a predetermined time has elapsed since performing a process of adjusting the objective lens 42, and, if it is determined that the predetermined time has elapsed, a process of again adjusting the objective lens 42. Therefore, in the electron microscope 100 according to the second embodiment, it is possible to reduce the influence of changes over time in the optimal value of the OL value of the objective lens 42, and to accurately correct low-order aberrations.

[0180] Furthermore, in the electron microscope 100 according to the second embodiment, the control unit 60 performs a process of determining whether or not a predetermined time has elapsed since performing a process of adjusting the two-fold astigmatism corrector 26, and, if it is determined that the predetermined time has elapsed, a process of again adjusting the two-fold astigmatism corrector 26. Therefore, in the electron microscope 100 according to the second embodiment, it is possible to reduce the influence of changes over time in the optimal values ​​of the X stigma value and the Y stigma value of the two-fold astigmatism corrector 26, and to accurately correct low-order aberrations.

[0181] Furthermore, in the electron microscope 100 according to the second embodiment, the control unit 60 performs a process of determining whether or not a predetermined time has elapsed since performing a process of adjusting the axial coma aberration corrector 28, and, if it is determined that the predetermined time has elapsed, a process of again adjusting the axial coma aberration corrector 28. Therefore, in the electron microscope 100 according to the second embodiment, it is possible to reduce the influence of changes over time in the optimal values ​​of the X alignment value and the Y alignment value of the axial coma aberration corrector 28, and to accurately correct low-order aberrations.

[0182] 2.4. Variations 18 is a flowchart showing a modified example of the process of correcting low-order aberrations by the control unit 60. Below, differences from the example of the electron microscope 100 according to the first embodiment described above will be described, and a description of similarities will be omitted.

[0183] The optimum value of the OL value of the objective lens 42 varies more greatly due to changes over time than the optimum values ​​of the X stigma value and the Y stigma value of the two-fold astigmatism corrector 26 and the optimum values ​​of the X alignment value and the Y alignment value of the axial coma aberration corrector 28. Therefore, as shown in FIG. 18, the control unit 60 performs an adjustment process S110 of the objective lens 42 and an adjustment process S120 of the two-fold astigmatism corrector 26, and then performs an adjustment process S1100 of the objective lens 42. The adjustment process S1100 of the objective lens 42 is performed in the same manner as the adjustment process S110 of the objective lens 42. This can reduce the influence of changes over time in the optimum value of the OL value, and can accurately correct low-order aberrations.

[0184] After performing the adjustment process S1100 for the objective lens 42, the control unit 60 performs the adjustment process S130 for the axial coma aberration corrector 28.

[0185] 3. Third embodiment 3.1. Electron microscope Next, an electron microscope according to a third embodiment will be described. The configuration of the electron microscope according to the third embodiment is similar to that of the electron microscope 100 according to the first embodiment, and a description thereof will be omitted. Hereinafter, in the electron microscope 100 according to the third embodiment, members having the same functions as those of the constituent members of the electron microscope 100 according to the first embodiment will be given the same reference numerals, and detailed descriptions thereof will be omitted.

[0186] 3.2. Aberration correction method In the first embodiment described above, an atomic image is captured at an imaging magnification of about 80M, and low-order aberrations are corrected using the standard deviation of the brightness distribution of the atomic image as an index. In contrast, in the third embodiment, the imaging magnification is set to a relatively low magnification of about 10M. In this case, even if the size of the electron probe is sufficiently small, the scanning interval becomes large, so the atomic image cannot be clearly seen and moire patterns can be seen.

[0187] Moiré in a STEM image is a periodic pattern that occurs when the scanning pattern when scanning a sample with an electron beam and the atomic arrangement pattern overlap. Therefore, the brightness and darkness of the moiré are equal to the brightness and darkness of an image in which the atomic image is clearly visible, and the same results are obtained when low-order aberrations are corrected using the standard deviation of the brightness distribution of the moiré as an index.

[0188] 3.3. Imaging method The control unit 60 performs processing to correct the above-mentioned low-order aberrations shown in Fig. 10 using the standard deviation of the luminance distribution of the moiré as an index. At this time, the control unit 60 performs imaging at a relatively low magnification of about 10M times in processing S117 for imaging atomic images in Fig. 11, processing S127 for imaging atomic images in Fig. 12, and processing S137 for imaging atomic images in Fig. 13. This makes it possible to correct low-order aberrations using the standard deviation of the luminance distribution of the moiré as an index.

[0189] Moreover, the control unit 60 captures a STEM image of the sample S after completing the process of correcting the low-order aberrations shown in FIG. 10 described above. At this time, the imaging magnification is set to be higher than the imaging magnifications in the above-mentioned processes S117, S127, and S137. In this way, in the imaging method in the electron microscope 100, the imaging magnification when imaging the observation sample S can be made higher than the imaging magnification when imaging an atomic image for aberration correction. In other words, the imaging magnification when imaging an atomic image for aberration correction can be made lower than the imaging magnification when imaging a STEM image of the observation sample S. Therefore, a STEM image with a high imaging magnification can be obtained in a state where the aberrations are corrected, and damage to the sample S caused by irradiation with an electron beam can be reduced in imaging an atomic image for aberration correction.

[0190] 4. Fourth embodiment 4.1. Electron microscope Next, an electron microscope according to a fourth embodiment will be described. The configuration of the electron microscope according to the fourth embodiment is similar to that of the electron microscope 100 according to the first embodiment, and a description thereof will be omitted. Hereinafter, in the electron microscope 100 according to the fourth embodiment, members having the same functions as those of the constituent members of the electron microscope 100 according to the first embodiment will be given the same reference numerals, and detailed descriptions thereof will be omitted.

[0191] 4.2. Aberration correction method FIG. 19 is a flowchart showing a modified example of the process S110 for adjusting the objective lens 42 by the control unit 60. In FIG.

[0192] As shown in FIG. 19, the control unit 60 performs a process S1170 of changing the area where the atomic image is captured in the process S110 of adjusting the objective lens 42. For example, after the process S116 of determining whether the optimal value has converged or not and before the process S117 of capturing the atomic image, the control unit 60 performs a process S1170 of changing the area where the atomic image is captured. As a result, the area where the atomic image is captured is changed each time the process S113 of calculating the standard deviation, the process S114 of calculating the fitting function F and the confidence interval, the process S115 of calculating the optimal value of the OL value, and the process S117 of capturing the atomic image are repeated. For example, if the axial coma aberration corrector 28 has two alignment coils, the area where the atomic image is captured can be changed by using the alignment coil of the first stage. The area where the atomic image is captured may be changed by moving the sample S with the goniometer 34.

[0193] Here, since the atomic image is a periodic pattern, the standard deviation of the brightness distribution of the atomic image is The aberration correction is independent of the area where the atomic image is captured. Therefore, even if the area where the atomic image is captured is changed, it does not affect the aberration correction. By changing the area where the atomic image is captured each time the process is repeated, damage to the sample S caused by irradiation with the electron beam can be reduced.

[0194] As long as steps S113 to S117 are repeated, the timing of performing step S1170 for changing the region in which the atomic image is captured is not particularly limited.

[0195] The control unit 60 also performs a process of changing the region in which an atomic image is captured in the adjustment process S120 of the two-fold astigmatism corrector 26 shown in Fig. 12. The control unit 60 also performs a process of changing the region in which an atomic image is captured in the adjustment process S130 of the on-axis coma aberration corrector 28 shown in Fig. 13.

[0196] The above-described embodiment and modifications are merely examples, and the present invention is not limited to these. For example, the embodiments and modifications can be appropriately combined.

[0197] The present invention is not limited to the above-described embodiment, and various modifications are possible. For example, the present invention includes a configuration that is substantially the same as the configuration described in the embodiment. A substantially identical configuration is, for example, a configuration that has the same function, method, and result, or a configuration that has the same purpose and effect. The present invention also includes a configuration in which non-essential parts of the configuration described in the embodiment are replaced. The present invention also includes a configuration that has the same action and effect as the configuration described in the embodiment, or a configuration that can achieve the same purpose. The present invention also includes a configuration in which publicly known technology is added to the configuration described in the embodiment. [Explanation of symbols]

[0198] 10...electron source, 20...irradiation optical system, 22...condenser lens, 24...diaphragm, 26...two-fold astigmatism corrector, 28...axial coma aberration corrector, 32...sample holder, 34...goniometer, 40...imaging optical system, 42...objective lens, 44...intermediate lens, 46...projection lens, 50...imaging device, 52...annular detector, 54...detector, 60...control unit, 70...operation unit, 72...display unit, 74...storage unit, 100...electron microscope

Claims

1. An electron optical system; A control unit for controlling the electron optical system; Including, The control unit is A process of calculating a standard deviation of the brightness distribution of the electron microscope image; A process of obtaining optimal values ​​of parameters of the electron optical system that maximize the standard deviation by Gaussian process regression; a process of capturing the electron microscope image with the parameter values ​​set to the optimal values; Do the following: The process of determining the standard deviation, the process of determining the optimum value, and the process of capturing the electron microscope image are repeated to determine the value of the parameter.

2. In claim 1, The control unit, in the process of obtaining the optimal value, determining a prediction model for predicting a relationship between the parameter and the standard deviation by the Gaussian process regression and a reliability of the value of the parameter predicted by the prediction model based on the standard deviation; The electron microscope determines the optimum value based on the prediction model and the confidence level.

3. In claim 1, The electron optical system includes: An objective lens; a two-fold astigmatism corrector for correcting distortion of the electron beam; an axial coma aberration corrector for correcting the inclination of the electron beam; Including, a process of repeating the process of obtaining the standard deviation, the process of obtaining the optimum value, and the process of capturing the electron microscope image, using the parameter as the excitation of the objective lens, to determine the excitation value of the objective lens and adjust the objective lens; a process of repeating the steps of determining the standard deviation, determining the optimum value, and capturing the electron microscope image, with the parameters being a first excitation for correcting distortion of the electron beam in a first direction in the two-fold astigmatism corrector and a second excitation for correcting distortion of the electron beam in a second direction different from the first direction, thereby determining the value of the first excitation and the value of the second excitation, and adjusting the two-fold astigmatism corrector; a process of repeating the process of obtaining the standard deviation, the process of obtaining the optimum value, and the process of capturing the electron microscope image, with the parameters being a third excitation for correcting the inclination of the electron beam in a third direction in the axial coma aberration corrector and a fourth excitation for correcting the distortion of the electron beam in a fourth direction different from the third direction, thereby determining the value of the third excitation and the value of the fourth excitation, and adjusting the axial coma aberration corrector; This is done using an electron microscope.

4. In claim 3, The control unit is A process of determining whether a predetermined time has elapsed since performing a process of adjusting the objective lens; a process of adjusting the objective lens again when it is determined that the predetermined time has elapsed; This is done using an electron microscope.

5. In claim 1, The control unit changes the area in which the electron microscope image is captured each time the process of determining the standard deviation, the process of determining the optimum value, and the process of capturing the electron microscope image are repeated.

6. In any one of claims 1 to 5, The control unit, after determining the value of the parameter, performs a process of capturing an electron microscope image having a number of pixels greater than the number of pixels of the electron microscope image captured in the process of capturing the electron microscope image, in an electron microscope.

7. In any one of claims 1 to 5, The control unit, after determining the value of the parameter, performs a process of capturing an electron microscope image at an imaging magnification higher than the imaging magnification of the electron microscope image captured in the process of capturing the electron microscope image, in an electron microscope.

8. 1. A method for correcting aberrations in an electron microscope including an electron optical system, comprising: determining a standard deviation of the brightness distribution of the electron microscope image; determining, by Gaussian process regression, optimal values ​​of parameters of the electron optical system that maximize the standard deviation; taking the electron microscope image with the parameter values ​​set to the optimal values; Including, a method for correcting an aberration, the step of determining the standard deviation, the step of determining the optimum value, and the step of capturing an electron microscope image are repeated to determine the value of the parameter.

9. In claim 8, In the step of determining the optimal value, determining a prediction model for predicting a relationship between the parameter and the standard deviation by the Gaussian process regression and a reliability of the value of the parameter predicted by the prediction model based on the standard deviation; determining the optimum value based on the prediction model and the reliability.

10. In claim 8, The electron optical system includes: An objective lens; a two-fold astigmatism corrector for correcting distortion of the electron beam; an axial coma aberration corrector for correcting the inclination of the electron beam; Including, a step of repeating the step of determining the standard deviation, the step of determining the optimum value, and the step of capturing the electron microscope image, with the parameter being the excitation of the objective lens, thereby determining the excitation value of the objective lens and adjusting the objective lens; a step of repeating the step of calculating the standard deviation, the step of calculating the optimum value, and the step of capturing the electron microscope image, with the parameters being a first excitation for correcting distortion of the electron beam in a first direction in the two-fold astigmatism corrector and a second excitation for correcting distortion of the electron beam in a second direction different from the first direction, thereby determining the value of the first excitation and the value of the second excitation, and adjusting the two-fold astigmatism corrector; a step of repeating the step of calculating the standard deviation, the step of calculating the optimum value, and the step of capturing the electron microscope image, with the parameters being a third excitation for correcting a tilt of the electron beam in a third direction in the axial coma aberration corrector and a fourth excitation for correcting a distortion of the electron beam in a fourth direction different from the third direction, thereby determining a value of the third excitation and a value of the fourth excitation, and adjusting the axial coma aberration corrector; The aberration correction method includes:

11. In claim 10, determining whether a predetermined time has elapsed since performing the step of adjusting the objective lens; adjusting the objective lens again when it is determined that the predetermined time has elapsed; The aberration correction method includes:

12. In claim 8, an area for capturing the electron microscope image is changed each time the steps of determining the standard deviation, determining the optimum value, and capturing the electron microscope image are repeated.

13. correcting the aberration of the electron optical system by the aberration correcting method according to any one of claims 8 to 12; a step of capturing an electron microscope image having a number of pixels greater than the number of pixels of the electron microscope image captured in the step of capturing the electron microscope image; An imaging method comprising:

14. correcting the aberration of the electron optical system by the aberration correcting method according to any one of claims 8 to 12; capturing an electron microscope image at a magnification higher than the magnification of the electron microscope image captured in the step of capturing the electron microscope image; An imaging method comprising:

Images