Defect detection method and defect detection apparatus

Abstract

This method improves the accuracy of defect detection when determining the defect state of a semiconductor substrate using XRT measurements, even when there is a brightness distribution in the band of X-rays used to generate the XRT image or when a brightness gradient is generated due to data merging. [Solution] A method for determining the defect state of a semiconductor substrate using XRT measurement, wherein a band measurement is performed by scanning a specific region S of the semiconductor substrate while moving a band-shaped X-ray extending in a first direction along a second direction substantially perpendicular to the first direction, and this is repeated while making step movements in the first direction to scan the entire specific region S, the data from each band measurement are merged to output an XRT image of the entire specific region S, an image of a unit region of a predetermined size is extracted from the XRT image, brightness gradient correction is performed on the image of the unit region to remove the brightness distribution of the band-shaped X-ray and the brightness gradient caused by merging, and the defect state in the semiconductor substrate is determined using the corrected image of the unit region.

Description

Technical Field

[0001] The present invention relates to a method and an apparatus (defect determination method and defect determination apparatus) for determining the defect state of a semiconductor substrate using X-ray topography (XRT) measurement.

Background Art

[0002] Currently, power devices, which are semiconductor devices that convert and control high-voltage and large-current electricity, are used in all electrical equipment. Power devices are also important devices for achieving carbon neutrality due to their characteristics. For example, they are used in power conversion equipment that converts the power sent from a power plant into a voltage that can be used in homes, and in the power control units of hybrid vehicles. Almost all of the currently commercialized power devices use Si substrates, but due to the physical properties of Si as a material, the limits of device performance are approaching. Therefore, currently, wide-bandgap semiconductors such as SiC and GaN have been found as new materials for power devices.

[0003] However, since the crystal growth of semiconductors, which are wide-bandgap semiconductors, is very difficult, single-crystal substrates with a low defect density and a large diameter like Si have not yet been realized. In particular, when there are linear crystal defects called dislocations among crystal defects, dislocations cause disorder in the atomic arrangement in the epitaxial growth process controlled at the atomic level, and have a great negative impact on device characteristics. Therefore, in the development of a growth technology for wide-bandgap semiconductor single-crystal substrates with a low dislocation density, it is beneficial to obtain distribution information such as the type and density of dislocations, and an evaluation technology for obtaining such information is desired. In this regard, in recent years, a method for evaluating defects in a multilayer substrate in which an SOI substrate or a single-crystal substrate is directly bonded using XRT measurement has been proposed (see Patent Document 1). It is said that by adopting such a method, highly sensitive measurement of small defects is possible even after a plurality of substrates are bonded together.

Prior Art Documents

Patent Documents

[0004] [Patent Document 1] Japanese Patent Publication No. 2022-609934 [Overview of the project] [Problems that the invention aims to solve]

[0005] Incidentally, when performing XRT measurements, a common method involves scanning a specific region of a semiconductor substrate by moving a band of X-rays extending in a predetermined direction (X direction) along a direction approximately perpendicular to the X direction (Y direction), and repeating this band measurement while stepping in the X direction to scan the entire specific region. The data from each band measurement are then merged to output an XRT image of the entire specific region. However, with this method, if a brightness distribution exists along the X direction of the band of X-rays, or if a brightness gradient is generated due to data merging, the accuracy of the XRT image may decrease, potentially hindering accurate defect detection. This problem has not been solved by conventional techniques such as those described in Patent Document 1.

[0006] This invention has been made in view of the above circumstances, and aims to improve the accuracy of defect detection when determining the defect state of a semiconductor substrate using XRT measurement, even when a brightness distribution exists in the band of X-rays used to generate the XRT image or when a brightness gradient is generated due to data merging. [Means for solving the problem]

[0007] A first aspect of the present invention for achieving the above objective is a method for determining the defect state of a semiconductor substrate using X-ray topography (XRT) measurement, comprising: an image output step of scanning a specific region of a semiconductor substrate by scanning a band measurement while moving a band of X-rays extending in a first direction relative to the semiconductor substrate along a second direction substantially perpendicular to the first direction, while performing step movements of the band of X-rays relative to the semiconductor substrate in the first direction, and merging the data of each band measurement to output an XRT image of the entire specific region; an image extraction step of extracting an image of a unit region of a predetermined size from the XRT image obtained by the image output step; an image correction step of performing brightness gradient correction to remove the brightness distribution of the band of X-rays and the brightness gradient caused by merging from the image of the unit region obtained by the image extraction step; and a defect determination step of determining the defect state in the semiconductor substrate using the image of the unit region corrected by the image correction step.

[0008] A second aspect of the present invention is a defect determination apparatus for determining the defect state of a semiconductor substrate using X-ray topography (XRT) measurement, comprising: an X-ray irradiation unit that irradiates a band of X-rays extending in a first direction; an image detection unit configured to scan the entire specific region by repeatedly performing band measurements, which scan a specific region of the semiconductor substrate while moving the band of X-rays relative to the semiconductor substrate along a second direction substantially perpendicular to the first direction, while performing step movements of the band of X-rays relative to the semiconductor substrate in the first direction; an image output unit that merges the data of each band measurement obtained by the image detection unit and outputs an XRT image of the entire specific region; an image extraction unit that extracts an image of a unit region of a predetermined size from the XRT image obtained by the image output unit; an image correction unit that performs brightness gradient correction to remove the brightness distribution of the band of X-rays and the brightness gradient caused by merging from the image of the unit region obtained by the image extraction unit; and a defect determination unit that determines the defect state in the semiconductor substrate using the image of the unit region corrected by the image correction unit.

[0009] By employing this method and apparatus, band measurements are performed by scanning a specific region of a semiconductor substrate while moving a band of X-rays extending in the first direction (X direction) along the second direction (Y direction). These measurements are repeated while stepping in the first direction (X direction) to scan the entire specific region. The data from each band measurement are merged to output an XRT image of the entire specific region. A unit image of a predetermined size is extracted from this XRT image. Brightness gradient correction is performed on this unit image to remove the brightness distribution of the band of X-rays and the brightness gradient caused by merging. This corrected unit image can then be used to determine the defect state within the semiconductor substrate. Therefore, even when the brightness distribution along the first direction of the band of X-rays and the brightness gradient caused by merging are significant, the defect state can be determined using the corrected unit image from which these have been removed, thereby improving the accuracy of defect determination (for example, detecting the dislocation state within a specific region of the semiconductor substrate (measurement of dislocation number, calculation of dislocation density, etc.) with relatively high accuracy). Furthermore, because correction can be performed to remove the brightness distribution of the band of X-rays and the brightness gradient in the unit image caused by merging, the length of the band of X-rays in the first direction can be increased. Therefore, there is also the advantage that the number of step movements of the band X-ray can be reduced, and the measurement time can be shortened.

[0010] In the defect detection method (defect detection apparatus) according to the present invention, dislocations occurring within a specific region of a semiconductor substrate can be identified as defects. In such cases, the image correction step of the defect detection method according to the present invention can use a filter that passes through the spatial frequency band related to the image of the dislocation but blocks the spatial frequency band related to the background image containing noise different from the dislocation, and the image correction unit of the defect detection apparatus according to the present invention can have a filter that passes through the spatial frequency band related to the image of the dislocation but blocks the spatial frequency band related to the background image containing noise different from the dislocation.

[0011] In the defect determination method according to the present invention, in the image correction step, one dislocation image is extracted from the XRT image obtained by the image output step, the dislocation image is Fourier transformed to generate a power spectrum, a specific line profile is extracted from the power spectrum, the spatial frequency at which the area of ​​the peak right shoulder or peak left shoulder of the line profile is 70% is set as the maximum spatial frequency, the same process is performed on other dislocation images to obtain the maximum spatial frequency, and the average value of these maximum spatial frequencies can be used for filtering. Furthermore, in the defect determination apparatus according to the present invention, the image correction unit extracts one dislocation image from the XRT image obtained by the image output unit, the dislocation image is Fourier transformed to generate a power spectrum, a specific line profile is extracted from the power spectrum, the spatial frequency at which the area of ​​the peak right shoulder or peak left shoulder of the line profile is 70% is set as the maximum spatial frequency, the same process is performed on other dislocation images to obtain the maximum spatial frequency, and filtering can be performed using the average value of these maximum spatial frequencies.

[0012] The defect determination method according to the present invention further includes: a first dislocation number measurement step in which the brightness of each pixel of an image in a unit region corrected for brightness gradient in the image correction step is converted into a histogram, the peaks of the histogram are separated into a dislocation peak caused by dislocations and a first background peak caused by the background, a binarized image is generated with the position 3σ from the position of the first background peak as the first threshold, and only dislocations are extracted from the binarized image to measure the first dislocation number; and a second dislocation number measurement step in which the image is divided into search window sizes calculated based on the first dislocation number, multiple search windows without dislocations are selected from the divided images, the peak of the histogram generated by summing the histograms which are the brightness distributions of the images of the selected search windows is taken as the second background peak, peak fitting is performed by applying the second background peak in place of the first background peak, a binarized image is generated with the position 3σ from the position of the second background peak as the second threshold, a contour detection image is obtained by extracting only dislocations from the binarized image, and the second dislocation number is measured from the contour detection image. Furthermore, the defect detection apparatus according to the present invention may further include: a first dislocation number measurement unit that converts the brightness of each pixel of an image in a unit region corrected for brightness gradient by an image correction unit into a histogram, separates the peaks of the histogram into a dislocation peak caused by dislocations and a first background peak caused by the background, generates a binarized image with the position 3σ from the position of the first background peak as the first threshold, extracts only dislocations from the binarized image to measure the first dislocation number; and a second dislocation number measurement step that divides the image with a search window size calculated based on the first dislocation number, selects multiple search windows without dislocations from the divided images, adds up the histograms which are the brightness distributions of the images in the selected search windows and uses the peak of the histogram generated as the second background peak, applies the second background peak in place of the first background peak to perform peak fitting, generates a binarized image with the position 3σ from the position of the second background peak as the second threshold, obtains a contour detection image with only dislocations extracted from the binarized image, and measures the second dislocation number from the contour detection image.

[0013] In the defect determination method according to the present invention, the defect determination step can measure the number of dislocations across the entire surface of a specific region of a semiconductor substrate by repeatedly performing the image extraction step, image correction step, first dislocation number measurement step, and second dislocation number measurement step for each unit region image. Furthermore, in the defect determination apparatus according to the present invention, the defect determination unit can measure the number of dislocations across the entire surface of a specific region of a semiconductor substrate by repeatedly performing the processing by the image extraction unit, image correction unit, first dislocation number measurement unit, and second dislocation number measurement unit for each unit region image.

[0014] In the defect determination method according to the present invention, the image correction step can remove brightness gradients caused by the warping of the semiconductor substrate and / or the brightness distribution in the first direction of the band X-rays. Furthermore, in the defect determination apparatus according to the present invention, the image correction unit can remove brightness gradients caused by the warping of the semiconductor substrate and / or the brightness distribution in the first direction of the band X-rays. [Effects of the Invention]

[0015] According to the present invention, when determining the defect state of a semiconductor substrate using XRT measurement, it is possible to improve the accuracy of defect determination even when a brightness distribution exists in the band of X-rays used for XRT image generation or when a brightness gradient is generated due to data merging. [Brief explanation of the drawing]

[0016] [Figure 1] This is a block diagram illustrating the configuration of a defect detection device according to an embodiment of the present invention. [Figure 2] This diagram illustrates the overview of XRT measurement. [Figure 3] This figure illustrates the XRT image obtained by merging the measurement data from each band. [Figure 4] This is a flowchart illustrating a defect determination method according to an embodiment of the present invention. [Figure 5] This figure shows an example of a unit image (original image) of an XRT image. [Figure 6]It is a diagram showing the power spectrum of the original image shown in FIG. 5. [Figure 7] It is a diagram showing the power spectrum after mask processing. [Figure 8] It is a diagram showing the unit image corresponding to the power spectrum after mask processing. [Figure 9] It is a diagram for explaining the luminance gradient correction for the unit image after mask processing. [Figure 10] It is for explaining the normalization of the histogram. (A) is a diagram showing an example of specifying the peak position and the half-value width of the histogram, (B) is a diagram showing the set peak position, and (C) is a diagram showing the set half-value width. [Figure 11] It is a diagram for explaining the low-pass filter processing. [Figure 12] It is a diagram showing the histogram after low-pass filter processing. [Figure 13] It is a diagram for explaining the threshold setting of peak separation and binarization. [Figure 14] It is a diagram showing the image in which the dislocation is extracted using the set first threshold. [Figure 15] It is a diagram showing the state in which the image is divided and displayed in the search window determined based on the first dislocation number. [Figure 16] It is a diagram showing the luminance histogram of the search window. [Figure 17] It is a diagram showing the state in which the area on the right side from the peak position is folded back symmetrically with respect to the Y-axis, overlapped with the area on the left side from the peak position, and then the absolute value of the difference in the Y direction is taken. [Figure 18] It is a diagram showing the state in which 20 images are specified from the side where the absolute value of the difference in the Y direction is small. [Figure 19] It is a diagram for confirming that the 20 images specified in FIG. 18 do not contain dislocations. [Figure 20] It is a diagram showing one luminance histogram by overlapping the luminance histograms of the 20 images shown in FIGS. 18 and 19. [Figure 21] It is a diagram showing the image in which the dislocation is extracted using the set second threshold. [Figure 22] This figure shows the number of dislocations and dislocation density per unit region obtained for a specific region of a semiconductor substrate. [Figure 23] This figure shows a color map of the dislocation density in the region shown in Figure 22. [Modes for carrying out the invention]

[0017] Embodiments of the present invention will be described below with reference to the drawings. Note that the following embodiments are merely preferred examples, and the scope of application of the present invention is not limited to these.

[0018] First, the configuration of the defect detection device 1 according to this embodiment will be described using Figure 1, etc. The defect detection device 1 is a device for determining the defect state of a semiconductor substrate using X-ray topography (XRT) measurement, and includes a sample stage (not shown) for placing the semiconductor substrate to be measured, an X-ray irradiation unit 10, an image detection unit 20, an image output unit 30, an image extraction unit 40, an image correction unit 50, a first dislocation number measurement unit 60, a second dislocation number measurement unit 70, a defect detection unit 80, etc.

[0019] In this embodiment, "X-ray topography" refers to a method of observing the distribution and shape of defects within a crystal as a two-dimensional image (also called an X-ray topographic image or XRT image) using the X-ray diffraction phenomenon. In this embodiment, "defect" includes "dislocations" (linear crystal defects contained in the crystal) that occur within a specific region of the semiconductor substrate (for example, a square region with sides of 50 mm). In this embodiment, the type of semiconductor substrate to be measured is not particularly limited, but the defect determination method according to this embodiment is effective in improving the accuracy of dislocation measurements in SiC substrates, AlN substrates, and GaN substrates.

[0020] As shown in Figure 2, the X-ray irradiation unit 10 functions to irradiate a band of X-rays extending in the X direction (first direction), and its configuration is not particularly limited as long as it performs this function.

[0021] As shown in Figure 2, the image detection unit 20 is configured to scan the entire specific region S of the semiconductor substrate by repeatedly performing band measurements, which involves moving the band-shaped X-rays irradiated from the X-ray irradiation unit 10 relative to the semiconductor substrate along the Y direction (second direction) which is approximately perpendicular to the X direction, while performing step movements of the band-shaped X-rays relative to the semiconductor substrate in the X direction. The image detection unit 20 enables mapping of the dislocation distribution using XRT images by detecting the increase or decrease in diffracted X-ray intensity caused by the presence or absence of dislocations at each position within the X-ray irradiation region. The image detection unit 20 includes an X-ray sensor as shown in Figure 2 and a control unit (not shown) that drives and controls the X-ray sensor. Note that during band measurements and step movements, the band-shaped X-rays may be moved relative to the semiconductor substrate, or the semiconductor substrate (sample stage) may be moved relative to the band-shaped X-rays.

[0022] As shown in Figure 3, the image output unit 30 functions to merge the data of each band measurement obtained by the image detection unit 20 and output an XRT image of the entire specific region S, and the image extraction unit 40 functions to extract an image of a predetermined size unit region from the XRT image obtained by the image output unit 30. The image output unit 30 and the image extraction unit 40 may consist of a computer configured to execute a computer program for image output and extraction based on user operation, a user interface that accepts user input, a monitor that visualizes the XRT image, etc.

[0023] The image correction unit 50 functions to perform brightness gradient correction to remove brightness gradients (caused by semiconductor substrate warping, brightness distribution in the first direction of band X-rays, data merging of each band measurement, etc.) from the image of a unit region obtained by the image extraction unit 40. The image correction unit 50 in this embodiment has a filter that passes through the spatial frequency band related to the dislocation image but does not pass through the spatial frequency band related to the background image containing noise different from the dislocation. The image correction unit 50 in this embodiment also functions to extract one dislocation image from the XRT image obtained by the image output unit 30, perform a Fourier transform on the dislocation image to generate a power spectrum, extract a specific line profile from the power spectrum, set the spatial frequency at which the area of ​​the peak right shoulder or peak left shoulder of the line profile is 70% as the maximum spatial frequency, perform the same process on other dislocation images to obtain the maximum spatial frequency, and perform filtering using the average value of these maximum spatial frequencies. The image correction unit 50 may be configured by a computer or the like configured to execute an image correction program based on user operation.

[0024] The first dislocation number measurement unit 60 converts the brightness of each pixel in the unit region image corrected for brightness gradient by the image correction unit 50 into a histogram, separates the peaks of the histogram into dislocation peaks caused by dislocations and first background peaks caused by the background, generates a binarized image with the position 3σ from the position of the first background peak as the first threshold, and extracts only the dislocations from the binarized image to measure the first dislocation number. The first dislocation number measurement unit 60 may be configured as a computer or the like configured to execute a program for measuring the first dislocation number based on user operation.

[0025] The second dislocation number measurement unit 70 divides the image into search window sizes calculated based on the first dislocation number, selects multiple search windows without dislocations from the divided images, adds up the histograms of the brightness distribution of the selected search windows, sets the peak of the histogram to be the second background peak, applies the second background peak in place of the first background peak to perform peak fitting, generates a binarized image with the position 3σ from the position of the second background peak as the second threshold, obtains a contour detection image by extracting only dislocations from the binarized image, and measures the second dislocation number from the contour detection image. The second dislocation number measurement unit 70 may be configured as a computer or the like configured to execute a program for measuring the second dislocation number based on user operation.

[0026] The defect determination unit 80 is configured to determine the defect state (dislocation state) in the semiconductor substrate using an image of a unit region corrected by the image correction unit 50. In this embodiment, the defect determination unit 80 functions to measure the number of dislocations across the entire surface of a specific region S of the semiconductor substrate by repeatedly performing processing by the image extraction unit 40, the image correction unit 50, the first dislocation number measurement unit 60, and the second dislocation number measurement unit 70 for each unit region image. The defect determination unit 80 may be configured by a computer or the like that is configured to execute a program for determining the dislocation state based on user operation.

[0027] Next, a defect determination method according to an embodiment of the present invention will be described using the flowchart in Figure 4 and the like.

[0028] The defect determination method according to this embodiment is a method for determining the defect state of a semiconductor substrate using XRT measurement, and as shown in Figure 4, includes an image output step S1, an image extraction step S2, an image correction step S3, a first dislocation number measurement step S4, a second dislocation number measurement step S5, and a defect determination step S6.

[0029] Image output step S1 is a process in which a band measurement is performed by scanning a specific region S of a semiconductor substrate while moving a band-shaped X-ray extending in the X direction (first direction) relative to the semiconductor substrate along the Y direction (second direction) which is substantially perpendicular to the X direction, while repeating this band measurement while making step movements relative to the semiconductor substrate in the first direction, thereby scanning the entire specific region S, merging the data from each band measurement, and outputting an XRT image of the entire specific region. In this embodiment, image output step S1 is performed using the sample stage, X-ray irradiation unit 10, image detection unit 20, and image output unit 30 of the defect determination apparatus 1 according to this embodiment.

[0030] In the image output process S1, each axis can be adjusted to satisfy the diffraction conditions on the crystal plane of the semiconductor substrate to be measured. The angles of the axes to be adjusted are ω, which is the angle between the X-ray irradiation surface and the incident X-ray, and 2θ, which is the angle between the incident X-ray and the optical axis of the detector. The X-ray irradiation unit 10 and the image detection unit 20 are scanned on a concentric circle centered on the X-ray irradiation point, within the plane containing the X-ray irradiation unit 10, the image detection unit 20, and the crystal, to identify the position where the intensity of the diffracted light is maximum. In addition, ω scanning can be performed at multiple locations on the semiconductor substrate surface to check the degree of wafer warping.

[0031] Figure 2 shows the measurement setup of this embodiment. The image detection unit 20 is set directly above the semiconductor substrate, and the position of the image detection unit 20 is adjusted so that the diffracted X-ray intensity is maximized, and the bandwidth of the diffracted X-ray beam taken up by the image detection unit 20 is determined. Next, by repeatedly performing band measurements that traverse scan within a specific region S with a band of irradiated X-rays, the entire specific region S is scanned, and as shown in Figure 3, the band data is merged and output as a single image. Note that if the semiconductor substrate has a large curvature, the ω value that satisfies the diffraction condition changes along the curvature direction, so by performing XRT measurements on the entire surface of the sample with multiple ω values ​​and finally performing image synthesis, an image in which dislocations appear across the entire surface of the sample can be obtained.

[0032] Image extraction step S2 is a step of extracting an image of a predetermined unit region from the XRT image obtained in image output step S1. In image extraction step S2, it is preferable to extract a region from the entire surface of a specific region S of the semiconductor substrate in which there is little overlap between dislocations and each dislocation can be visually identified. There are no particular restrictions on the region to be extracted as long as the above conditions are satisfied, but for example, as shown in Figure 5, a square region with sides of 2.7 mm near the center of the specific region S can be extracted as a unit region.

[0033] Image correction step S3 is a step in which brightness gradient correction is performed to remove the brightness distribution of band-shaped X-rays and the brightness gradient caused by merging from the unit region image (hereinafter referred to as the "original image") obtained by the image extraction step S2. As shown in Figure 5, the original image contains not only horizontal stripes (horizontal bands) at the time of each band data merging, but also brightness gradients in the background in the X and Y directions caused by substrate warping and divergence of incident X-rays. Therefore, in image correction step S3, brightness gradient correction is performed using a filter that passes through the spatial frequency band related to the dislocation image but does not pass through the spatial frequency band related to the background image which contains noise different from the dislocation.

[0034] First, in the image correction step S3, the power spectrum of the original image is masked. Figure 6 shows the power spectrum of the image before masking, Figure 7 shows the power spectrum of the image after masking, and Figure 8 shows the image after masking. Comparing the unprocessed image (original image) in Figure 5 with the processed image in Figure 8, it can be seen that the transverse bands and luminance gradients caused by the luminance distribution in the X direction of band-like X-rays resulting from data merging have been removed. Next, in the image correction step S3, in order to remove the low-spatial-frequency background gradient that could not be removed by the power spectrum masking, a luminance profile in the X·Y direction as shown in Figure 9 is taken, and the luminance gradient is corrected.

[0035] Furthermore, in the image correction process S3, the luminance gradient corrected image is converted into a histogram, and the histogram is normalized. As shown in Figure 10, the luminance of each pixel in the luminance gradient corrected unit region image is converted into a histogram, and the peak position and half-width of this histogram are identified. For example, in Figure 10(A), the peak position is 81, the right side of the half-width is 9, and the right side of the half-width is 8. Next, as shown in Figures 10(B) and 10(C), the histogram is normalized by setting the peak position and half-width to specific values. Subsequently, the luminance values ​​of the original image are converted to the values ​​set by normalization. Then, as shown in Figure 11, for the image whose histogram has been normalized, the spatial frequency components contained in the image are expressed as a power spectrum using a Fast Fourier Transform, and then a low-pass filter is used to remove high spatial frequency components contained in noise while retaining low spatial frequency components contained in dislocation images, thereby obtaining an image that emphasizes dislocations. Furthermore, filters are not limited to low-pass filters; other filters, such as bilateral filters that can remove noise and emphasize dislocations, can also be used. Other filters include bilateral filters, Gaussian filters, and median filters.

[0036] Here, we describe the design method for the low-pass filter. From the XRT image obtained in the image output process S1, one dislocation image is extracted, for example, by cropping it to a 64-pixel square. The dislocation image is Fourier transformed to generate a power spectrum. A specific line profile is extracted from this power spectrum, and the spatial frequency at which the area of ​​the right shoulder of the peak of this line profile is 70% is set as the maximum spatial frequency. The same process is performed on other dislocation images to obtain the maximum spatial frequencies, and the average of these maximum spatial frequencies is used as the filter parameter. Note that the maximum spatial frequency is not limited to the spatial frequency at which the area of ​​the right shoulder of the peak is 70%. For example, the spatial frequency at which the area of ​​the left shoulder of the peak is 70% can be set as the maximum spatial frequency, or the spatial frequency at which the area of ​​the right (or left) shoulder of the peak is 50-60% can be set as the maximum spatial frequency. In the image correction process S3, by using the low-pass filter obtained in this way, an image with noise removed and dislocations emphasized can be obtained, as shown in Figure 11.

[0037] The first dislocation number measurement step S4 involves converting the brightness of each pixel in the image of a unit region obtained by applying brightness gradient correction and low-pass filtering in the image correction step S3 into a histogram, separating the peaks of this histogram into dislocation peaks caused by dislocations and first background peaks caused by the background, generating a binarized image with the position 3σ from the position of the first background peak as the first threshold, and extracting only the dislocations from the binarized image to measure the first dislocation number. It is also possible to calculate the dislocation density (first dislocation density) from the measured first dislocation number.

[0038] In the first dislocation number measurement step S4, as shown in Figure 12, a luminance histogram is obtained again from the image obtained by applying a low-pass filter, and then a one-component peak fitting is performed on the strong peaks in this luminance histogram that originate from the background other than dislocations. Next, since a fitting residual originating from dislocations is generated at the left shoulder of the peaks in the original histogram, peak fitting is performed again on that region. By performing two-component peak fitting using the peak positions and standard deviations obtained from these two peak fittings as fitting parameters, the peaks are separated into a first background peak originating from the background and a dislocation peak originating from the dislocations, as shown in Figure 13. Then, the original image is binarized with a threshold (first threshold) set at a position 3σ from the position of the first background peak, and contour detection is performed on the obtained binarized image to obtain an image in which dislocations have been extracted, as shown in Figure 14, and the first dislocation number is measured. In this embodiment, an example is shown where the first threshold is set at a position 3σ from the position of the first background peak, but it is also possible to set the first threshold at a position 1σ or 5σ from the position of the first background peak.

[0039] The second dislocation number measurement step S5 involves dividing the image into search window sizes (e.g., 20 pixels) calculated based on the first dislocation number, selecting multiple search windows without dislocations from the divided images, summing the histograms of the brightness distribution of the selected search windows to generate a histogram peak which is used as the second background peak, applying the second background peak in place of the first background peak to perform peak fitting, generating a binarized image with the position 3σ from the second background peak as the second threshold, obtaining a contour detection image by extracting only dislocations from the binarized image, and measuring the second dislocation number from the contour detection image. It is also possible to calculate the dislocation density (second dislocation density) from the measured second dislocation number.

[0040] In the second dislocation number measurement step S5, the number of pixels in the search window is determined based on the first dislocation number, as shown in Figure 15. At this time, the larger the first dislocation number, the smaller the size of the search window is determined. Using the search window determined in this way, the image within the search window is trimmed and histogrammed as shown in Figure 16, and then separated into a region to the left and a region to the right of the peak position. Next, as shown in Figure 17, the region to the right (or left) of the peak position is folded symmetrically along the Y axis and superimposed with the region to the left (or right) of the peak position, and the absolute value of the difference in the Y direction is taken. If the absolute value of the difference is small, the histogram is close to a Gaussian distribution, and is judged to be the background region. The same process is performed for each search window and a list is created in ascending order of the absolute values ​​of the differences. Subsequently, as shown in Figure 18, an arbitrary number of corresponding images are taken from those with small absolute values ​​of differences, and as shown in Figure 19, it is confirmed that each image does not contain dislocations, and as shown in Figure 20, the luminance histograms of each image are superimposed and combined into a single luminance histogram, which is the luminance histogram of the background component, and this peak is taken as the second background peak.

[0041] Then, as shown in Figure 20, the original image is binarized using a threshold (second threshold) set at a position 3σ from the second background peak, and contour detection is performed on the resulting binarized image to obtain an image with dislocations extracted, as shown in Figure 21, and the second number of dislocations is measured. In this embodiment, an example is shown where the second threshold is set at a position 3σ from the second background peak, but it is also possible to set the second threshold at a position 1σ or 5σ from the second background peak.

[0042] The defect determination step S6 is a process that measures the number of dislocations across the entire surface of a specific region S of the semiconductor substrate by repeating the image extraction step S2, image correction step S3, first dislocation number measurement step S4, and second dislocation number measurement step S5 for each image of a unit region, and determines the defect state within the semiconductor substrate. Figure 22 shows the number of dislocations and dislocation density for each 5.4 mm square region obtained for a 50 mm square region of the semiconductor substrate, and Figure 23 is a color map display of the dislocation density of the same region. By referring to these figures, the user can accurately obtain the distribution of defects within the semiconductor substrate and accurately determine the defect state within the semiconductor substrate.

[0043] In the defect detection apparatus 1 according to the embodiment described above, band measurement is performed by scanning a specific region S of a semiconductor substrate while moving a band-shaped X-ray extending in the X direction (first direction) along the Y direction (second direction). This is repeated while performing step movements in the X direction to scan the entire specific region S. The data from each band measurement are merged to output an XRT image of the entire specific region S. An image of a predetermined size unit region (unit image) is extracted from this XRT image. Brightness gradient correction is performed on this unit image to remove the brightness distribution of the band-shaped X-ray and the brightness gradient caused by merging. The defect state in the semiconductor substrate can then be determined using this corrected unit image. Therefore, even when the brightness distribution along the X direction of the band-shaped X-ray and the brightness gradient caused by merging are significant, the defect state can be determined using the corrected unit image from which these have been removed, thereby improving the accuracy of defect detection (for example, detecting the dislocation state in the specific region S of the semiconductor substrate (measurement of the number of dislocations, calculation of dislocation density, etc.) with relatively high accuracy). Furthermore, since correction can be performed to remove the brightness distribution of the band-shaped X-ray and the brightness gradient in the unit image caused by merging, the length of the band-shaped X-ray in the X direction can be increased. Therefore, there is also the advantage that the number of step movements of the band X-ray can be reduced, and the measurement time can be shortened. [Examples]

[0044] Next, embodiments of the present invention will be described. However, the present invention is not limited to these embodiments.

[0045] In this embodiment, a silicon carbide substrate with a diameter of 4 inches and a 4-degree off-angle at <11-20> was prepared as the semiconductor substrate to be measured. A Rigaku SmartLab (copper target, tungsten filament, tube voltage / tube current (45 kV / 200 mA)) was used as the defect detection device 1. The crystal plane to be measured was (0-1110), and the number of measurements was 2 (1st time: ω=30.62°, 2nd time: ω=30.59°). The specific region S to be measured was a square region with sides of 50 mm, the longitudinal limiting slit (a slit that limits the irradiation width in the X direction (first direction) of the X-ray irradiation unit 10) was 15 mm, the incident slit (a slit that limits the irradiation width in the Y direction (second direction) of the X-ray irradiation unit 10) was 0.2 mm, the beam cutting range was 200 pixels, and the scan speed was 15 mm per second.

[0046] In the image output process S1, first, each axis was moved to satisfy the diffraction conditions (0-1110). Then, the X-ray irradiation unit 10 and the image detection unit 20 were scanned in a concentric circle within the plane containing the X-ray irradiation unit 10, the image detection unit 20, and the crystal, centered on the X-ray irradiation point, to identify the position where the intensity of the diffracted light was maximum. In addition, an ω scan was performed at 9 points on the semiconductor substrate surface to check the degree of substrate warping, and a warp of 0.03° was confirmed at both ends of the substrate. The image detection unit 20 was set directly above the semiconductor substrate, and the position of the image detection unit 20 was adjusted to maximize the diffracted X-ray intensity, and the beam width of the diffracted X-rays captured by the image detection unit 20 was determined. Next, band measurements were repeatedly performed by traversing the specific region S with band-shaped X-rays to scan the entire specific region S, and the band data was merged to output a single image. Furthermore, if the substrate has a large degree of warping, the ω value that satisfies the diffraction conditions changes along the direction of warping. Therefore, XRT measurements were performed on the entire surface of a specific region S using multiple ω values, and finally, the images were combined to obtain an image in which dislocations appear across the entire sample surface.

[0047] In the image extraction step S2, a region (unit region) was extracted from the entire surface of a specific region S on the semiconductor substrate, where there was little overlap between dislocations and each dislocation could be visually identified. In this embodiment, a square region with sides of 2.7 mm was extracted as the unit region.

[0048] In the image correction process S3, first, the power spectrum of the original image was masked in both the X and Y directions to allow the DC component to pass through, thereby removing frequency components parallel to the X and Y directions from the image (see Figures 5 to 8). Next, since a background gradient of low spatial frequencies that could not be removed by the power spectrum masking remained, a luminance profile was taken in the X and Y directions to remove it, and the luminance gradient was corrected (see Figure 9). Subsequently, the image with the luminance gradient corrected was converted into a histogram, the peak position and the left and right sides of the full width at half maximum (FWHM) of this histogram were identified, and the histogram was normalized by setting the peak position and the left and right sides of the FWHM to arbitrary values ​​(see Figure 10). In this embodiment, as shown in Figures 10(B) and 10(C), the peak position was set to 128, and both the left and right sides of the FWHM were set to 12.

[0049] Furthermore, in this embodiment, a low-pass filter shown in Figure 11 was designed and used. For the above image, after normalizing the histogram, the spatial frequency components contained in the image were represented by a power spectrum using a Fast Fourier Transform. Then, a low-pass filter was used to remove high spatial frequency components containing a lot of noise while retaining low spatial frequency components containing many dislocations, thereby obtaining an image with emphasized dislocations. Note that the parameter for the frequency range to be passed by the low-pass filter can be changed depending on the sample. In this embodiment, approximately 0.1 to 0.3 pixels were used. -1 The parameters were set so that the filter would pass through spatial frequency components lower than the value of [value].

[0050] In the first dislocation number measurement step S4, as shown in Figure 12, a luminance histogram was obtained again from the image obtained by applying a low-pass filter, and then a one-component peak fitting was performed on the strong peaks in this luminance histogram that originated from the background other than dislocations. Next, since a fitting residual originating from dislocations was generated at the left shoulder of the peaks in the original histogram, peak fitting was performed again on that region. By performing a two-component peak fitting using the peak positions and standard deviations obtained from these two peak fittings as fitting parameters, the peaks were separated into a first background peak originating from the background and a dislocation peak originating from the dislocations, as shown in Figure 13. Then, the original image was binarized with a position 3σ from the position of the first background peak as the first threshold, and contour detection was performed on the obtained binarized image to obtain an image in which dislocations were extracted, as shown in Figure 14, and the first dislocation number was measured. In this embodiment, the first dislocation number was 156.

[0051] In the second dislocation number measurement step S5, the number of pixels in the search window was determined based on the first dislocation number, as shown in Figure 15 (in this embodiment, one side of the search window was set to 20 pixels). Using the search window thus determined, the image within the search window was cropped and histogrammed as shown in Figure 16, and then separated into a region to the left and a region to the right of the peak position. Next, as shown in Figure 17, the region to the right of the peak position was folded symmetrically along the Y axis and superimposed with the region to the left of the peak position, and the absolute value of the difference in the Y direction was taken. If the absolute value of the difference was small, the histogram was close to a Gaussian distribution, and was judged to be the background region. The same process was performed for each search window, and a list was created sorted in ascending order of the absolute values ​​of the differences. Subsequently, as shown in Figure 18, an arbitrary number of corresponding images were taken from those with the smallest absolute values ​​of the differences, and it was confirmed that no dislocations were contained in each image, as shown in Figure 19 (in this embodiment, as shown in Figures 18 and 19, 20 images were taken from those with the smallest absolute values ​​of the differences). Subsequently, as shown in Figure 20, the luminance histograms of each image were superimposed and combined into a single luminance histogram, which was then used as the background component luminance histogram, and this peak was designated as the second background peak.

[0052] Then, as shown in Figure 20, the original image was binarized using a second threshold of 3σ from the second background peak, and contour detection was performed on the resulting binarized image to obtain an image with dislocations extracted, as shown in Figure 21, and the second dislocation count was measured. In this embodiment, the second dislocation count was 143.

[0053] In defect detection step S6, by repeatedly applying the automatic dislocation detection algorithm to the image of a unit region (a square region with sides of 2.7 mm), it became possible to extend the automatic analysis to a square region with sides of approximately 2 inches. Figure 22 shows the number of dislocations and dislocation density for each 5.4 mm square region obtained for a 50 mm square region of a semiconductor (SiC) substrate, and Figure 23 shows the result of the color map display of the dislocation density.

[0054] The present invention is not limited to the embodiments described above, and any modifications made to these embodiments by those skilled in the art are also included within the scope of the present invention, as long as they retain the features of the present invention. In other words, the elements of the embodiments and their arrangement, materials, conditions, shapes, sizes, etc., are not limited to those exemplified and can be modified as appropriate. Furthermore, the elements of the embodiments can be combined to the extent that it is technically possible, and any combinations thereof are also included within the scope of the present invention, as long as they retain the features of the present invention. [Industrial applicability]

[0055] This invention makes it possible to effectively remove noise from images obtained using X-ray topography (XRT images) through a specific process, thereby enabling accurate observation of dislocation states in wide-bandgap semiconductors, particularly hexagonal semiconductor single-crystal substrates such as SiC, AlN, and GaN substrates. This contributes to improving the quality of wide-bandgap semiconductors and can also be used for quality assurance. [Explanation of symbols]

[0056] 1…Defect detection device 10...X-ray irradiation section 20...Image detection unit 30...Image output unit 40...Image extraction unit 50…Image correction unit 60...First dislocation number measurement unit 70...Second dislocation number measurement unit 80...Defect detection unit S…Specific area S1...Image output process S2…Image extraction process S3...Image correction process S4...First dislocation number measurement process S5... Second dislocation number measurement process S6…Defect determination process

Claims

1. A method for determining the defect state of a semiconductor substrate using X-ray topography (XRT) measurement, An image output process involves scanning a specific region of the semiconductor substrate by moving a band-shaped X-ray beam extending in a first direction relative to the semiconductor substrate along a second direction substantially perpendicular to the first direction, repeating this band measurement while making step movements relative to the semiconductor substrate in the first direction, scanning the entire specific region, merging the data from each band measurement, and outputting an XRT image of the entire specific region. An image extraction step is performed to extract an image of a predetermined unit region from the XRT image obtained by the image output step, An image correction step is performed to remove the brightness gradient from the image of the unit region obtained by the image extraction step, in order to remove the brightness distribution of the band-shaped X-rays and the brightness gradient caused by merging. A defect determination step, which determines the defect state in the semiconductor substrate using the image of the unit region corrected by the image correction step, A defect detection method, including the following.

2. The defect is a dislocation that occurred within the specific region of the semiconductor substrate. The defect determination method according to claim 1, wherein the image correction step uses a filter that passes through the spatial frequency band related to the image of the dislocation but does not pass through the spatial frequency band related to the background image containing noise different from the dislocation.

3. The defect determination method according to claim 2, wherein the image correction step involves extracting one dislocation image from the XRT image obtained in the image output step, performing a Fourier transform on the dislocation image to generate a power spectrum, extracting a specific line profile from the power spectrum, setting the spatial frequency at which the area of ​​the peak right shoulder or peak left shoulder of the line profile is 70% as the maximum spatial frequency, performing the same process on other dislocation images to obtain the maximum spatial frequencies, and using the average value of these maximum spatial frequencies to perform the filter processing.

4. A first dislocation number measurement step is performed, in which the brightness of each pixel of the unit region image corrected for brightness gradient in the above image correction step is converted into a histogram, the peaks of the histogram are separated into a dislocation peak caused by the dislocation and a first background peak caused by the background, a binarized image is generated with the position of 3σ from the position of the first background peak as the first threshold, and only the dislocations are extracted from the binarized image to measure the first dislocation number, The process involves dividing the image into search window sizes calculated based on the first dislocation number, selecting multiple search windows from the divided images in which no dislocations exist, summing the histograms of the brightness distributions of the selected search windows to generate a histogram whose peak is the second background peak, applying the second background peak in place of the first background peak to perform peak fitting, generating a binarized image with the position of 3σ from the position of the second background peak as the second threshold, obtaining a contour detection image from which only the dislocations are extracted from the binarized image, and measuring the second dislocation number from the contour detection image. The defect determination method according to claim 2 or 3, further comprising:

5. The defect determination method according to claim 4, wherein the defect determination step involves repeating the image extraction step, the image correction step, the first dislocation number measurement step, and the second dislocation number measurement step for each of the images of the unit region, thereby measuring the number of dislocations across the entire specific region of the semiconductor substrate.

6. The defect determination method according to claim 1, wherein the image correction step removes the warping of the semiconductor substrate and / or the brightness gradient caused by the brightness distribution of the band-shaped X-rays in the first direction.

7. An apparatus for determining the defect state of a semiconductor substrate using X-ray topography (XRT) measurement, An X-ray irradiation unit that irradiates a band of X-rays extending in the first direction, An image detection unit is configured to scan the entire specific region by repeatedly performing a band measurement, which involves moving the band-shaped X-ray relative to the semiconductor substrate along a second direction substantially perpendicular to the first direction, while scanning a specific region of the semiconductor substrate, and performing this band measurement while making a step movement relative to the semiconductor substrate in the first direction. An image output unit merges the data from each band measurement obtained by the image detection unit and outputs an XRT image of the entire specific region. An image extraction unit extracts an image of a predetermined size unit region from the XRT image obtained by the image output unit, An image correction unit performs brightness gradient correction to remove the brightness distribution and brightness gradient caused by the merging of the band-shaped X-rays from the image of the unit region obtained by the image extraction unit, A defect determination unit that determines the defect state in the semiconductor substrate using the image of the unit region corrected by the image correction unit, A defect detection device equipped with the following features.

8. The defect is a dislocation that occurred within the specific region of the semiconductor substrate. The defect determination device according to claim 7, wherein the image correction unit has a filter that allows spatial frequency bands related to the image of the dislocation to pass through, but does not allow spatial frequency bands related to the background image containing noise different from the dislocation to pass through.

9. The defect determination apparatus according to claim 8, wherein the image correction unit extracts one dislocation image from the XRT image obtained by the image output unit, performs a Fourier transform on the dislocation image to generate a power spectrum, extracts a specific line profile from the power spectrum, sets the spatial frequency at which the area of ​​the peak right shoulder or peak left shoulder of the line profile is 70% as the maximum spatial frequency, performs the same process on other dislocation images to obtain the maximum spatial frequencies, and uses the average value of these maximum spatial frequencies to perform the filter processing.

10. A first dislocation number measurement unit converts the brightness of each pixel in the unit region image corrected for brightness gradient by the image correction unit into a histogram, separates the peaks of the histogram into a dislocation peak caused by the dislocation and a first background peak caused by the background, generates a binarized image with the position of 3σ from the position of the first background peak as the first threshold, and extracts only the dislocations from the binarized image to measure the first dislocation number. A second dislocation number measurement unit divides the image into search window sizes calculated based on the first dislocation number, selects multiple search windows from the divided images in which no dislocations exist, adds up the histograms of the brightness distribution of the selected search windows and sets the peak of the histogram to be the second background peak, applies the second background peak in place of the first background peak to perform peak fitting, generates a binarized image with the position of 3σ from the position of the second background peak as the second threshold, obtains a contour detection image from the binarized image in which only the dislocations are extracted, and measures the second dislocation number from the contour detection image. A defect determination device according to claim 8 or 9, further comprising:

11. The defect determination device according to claim 10, wherein the defect determination unit measures the number of dislocations across the entire specific region of the semiconductor substrate by repeatedly performing the processing by the image extraction unit, the image correction unit, the first dislocation number measurement unit, and the second dislocation number measurement unit for each of the images of the unit region.

12. The defect determination apparatus according to claim 7, wherein the image correction unit removes the brightness gradient caused by the warping of the semiconductor substrate and / or the brightness distribution of the band-shaped X-rays in the first direction.

Images