Multi-mode detection method for small and medium-sized stacking fault defects of silicon carbide substrate

Abstract

The invention belongs to the technical field of silicon carbide defect detection, and particularly relates to a multi-mode detection method for small and medium-sized stacking fault defects of a silicon carbide substrate, which comprises the following steps of: performing photoluminescence imaging on a silicon carbide substrate to be detected to obtain a candidate area of the small-sized stacking fault defects; performing microscopic photoluminescence spectrum two-dimensional mapping scanning on the candidate area to obtain a spectrum intensity mapping image in the candidate area; constructing a multi-modal matching model, performing spatial registration on a spectral intensity mapping image and a photoluminescence image, and determining a single small stacking fault defect by calculating a weighted pixel overlapping rate; microscopic photoluminescence spectrum two-dimensional mapping scanning is carried out on a single small stacking fault defect, and the morphology and size of the defect are determined. The method has the advantages of high sensitivity and high resolution, is suitable for rapid detection and online quality control of defects of the large-size silicon carbide substrate, and has a good industrial application prospect.

Description

Technical Field

[0001] The present invention belongs to the technical field of silicon carbide defect detection, and in particular relates to a multimodal detection method for small stacking fault defects in a silicon carbide substrate. Background Art

[0002] Wide-bandgap semiconductor silicon carbide (SiC) is widely used in high-voltage, high-frequency, and high-temperature power devices due to its outstanding physical properties, including high breakdown field strength, high thermal conductivity, and excellent chemical stability. 4H silicon carbide (4H-SiC), the mainstream epitaxial substrate material, has a crystal quality that directly impacts device performance. However, during the fabrication process of 4H-SiC substrates, various crystal defects, such as stacking faults (SFs), micropipes, and dislocations, are inevitably introduced. These defects affect the quality of subsequent epitaxial growth, the electrical performance, and long-term stability of the device. Small stacking faults, due to their tiny size, are difficult to accurately identify using current detection technologies. However, these defects can develop into more serious problems in subsequent processing steps, impacting device reliability. Therefore, the development of highly sensitive and reliable defect detection technologies is crucial.

[0003] Currently, methods used for crystal defect detection, such as optical microscopy, X-ray diffraction (XRD), scanning electron microscopy (SEM), transmission electron microscopy (TEM), and chemical etching, while somewhat effective in identifying large defects, have significant limitations: Optical microscopy and XRD have insufficient spatial resolution, making it difficult to capture microscopic defect details; SEM and TEM inspections require complex sample pretreatment, resulting in cumbersome procedures, high equipment costs, and low inspection efficiency; and chemical etching is a destructive method that damages the substrate and cannot meet the requirements of subsequent epitaxial growth or device fabrication. These limitations make it difficult to meet the demand for efficient, non-destructive, and accurate inspection of small stacking fault defects.

[0004] In recent years, photoluminescence (PL) technology has gained widespread application in the study of wide-bandgap semiconductor defects due to its non-contact, non-destructive, and highly sensitive nature. PL imaging can rapidly identify luminescence non-uniformities on the substrate surface, assisting in analyzing defect distribution. However, conventional PL imaging suffers from low spatial resolution, making it difficult to clearly discern microscopic stacking faults with small size and low luminescence contrast. Summary of the Invention

[0005] The purpose of the present invention is to provide a multimodal detection method for small stacking fault defects in silicon carbide substrates, thereby overcoming the shortcomings of the existing technology. By introducing a matching model based on affine transformation, the precise registration and fusion of PL imaging and microscopic PL spectral mapping are achieved, which greatly improves the consistency of the multimodal data in spatial position. It can quickly and reliably identify small stacking fault defects in 4H-SiC single crystal substrates under non-destructive conditions. It has high spatial resolution and simple operation, and is suitable for industrial silicon carbide quality control.

[0006] In order to achieve the above object, the technical solution of the present invention is: The present invention provides a method for detecting stacking fault defects in silicon carbide, comprising the following steps: (1) Photoluminescence imaging is performed on the silicon carbide substrate to obtain candidate areas for small stacking fault defects; (2) performing a two-dimensional mapping scan of the microscopic photoluminescence spectrum of the candidate area to obtain a spectral intensity mapping image within the candidate area; (3) A multimodal matching model was constructed to spatially register the spectral intensity mapping image and the photoluminescence image, and a single small stacking fault defect was confirmed by calculating the pixel overlap ratio; (4) Perform two-dimensional mapping scanning of microscopic photoluminescence spectra on a single small stacking fault defect to determine the defect morphology and size.

[0007] The present invention uses an ultraviolet excitation light source to perform large-area photoluminescence (PL) imaging of the silicon carbide substrate to be tested at room temperature, identifies linear areas with localized attenuation of luminescence intensity as candidate areas for small stacking fault defects, performs two-dimensional mapping scanning of the microscopic PL spectrum on the candidate areas, and obtains their two-dimensional PL intensity distribution maps. By introducing a multimodal matching model based on two-dimensional affine transformation, the PL imaging image and the two-dimensional mapping image of the microscopic PL spectrum are spatially aligned, and the alignment accuracy is evaluated by calculating the pixel overlap ratio to achieve precise positioning of single small defects. Finally, based on the two-dimensional PL intensity distribution map and characteristic peak position information at a smaller scanning step size, the morphological characteristics and size information of the single small stacking fault defect are determined, thereby realizing multimodal nondestructive detection of small stacking fault defects in silicon carbide substrates.

[0008] In some other embodiments, in step (1), the silicon carbide substrate is a 4H-SiC substrate, and the small stacking fault defect is a continuous linear defect with an aspect ratio greater than 5:1 and a length in the range of 0.2-1 mm; To further improve the accuracy of determining small stacking fault defects, transmission electron microscopy can be used to verify the microstructure of the candidate areas of small stacking fault defects initially identified by photoluminescence imaging. By observing the atomic-scale morphology of the defects, the accuracy of the photoluminescence detection results can be verified.

[0009] In some other embodiments, in step (1), the light source used for the photoluminescence imaging is an ultraviolet laser light source, and a red light high-pass filter is configured in the signal receiving path. Specifically, the wavelength of the ultraviolet laser light source is 313 nm or 320 nm; the cutoff wavelength of the red light high-pass filter is 660 nm or 700 nm (only red light signals with a wavelength ≥660 nm or ≥700 nm are allowed to pass).

[0010] In some other embodiments, in step (2), the light source for the two-dimensional mapping scan of the microscopic photoluminescence spectrum is an ultraviolet excitation light source, the spatial resolution is greater than 1 μm, the luminescence spectrum covers a range of 360-600 nm, the scanning step is set to ≤5 μm, and the integration time is ≥0.01 s. Specifically, the characteristic luminescence wavelength is 426 nm.

[0011] In some other embodiments, in step (3), the multimodal image matching model is a two-dimensional affine transformation multimodal matching model, and the least squares method is used to estimate the model parameters.

[0012] Specifically, the endpoints of large-scale rod-shaped SFs are preferred as feature points. If the sample has no large-scale defects, the laser markings on the edge of the wafer can be manually marked as feature points. This is because the endpoints of large-scale rod-shaped SFs are clearly visible in both PL imaging and micro-PL imaging systems. The coordinates of each pair of feature points in the two images are recorded, where the pixel coordinates of the PL imaging image are ( ), the pixel coordinates of the micro PL spectrum image are ( ), = 1, 2, 3; substitute the pixel coordinates of the feature points into the multimodal matching model of the two-dimensional affine transformation, achieve pixel-level spatial alignment through feature point calibration, and solve the vector p using the least squares method to complete the spatial registration.

[0013] The mathematical expression of the multimodal matching model of two-dimensional affine transformation is: ,

[0014] in,( ) is the pixel coordinate of the PL imaging image, ( ) is the pixel coordinate of the microscopic PL spectrum mapping image, =1, 2, 3; , , ， is the linear transformation coefficient, is the translation coefficient.

[0015] The mathematical expression of the least squares method for solving vector p is:

[0016] in, is the parameter vector to be determined, A and It consists of the coordinates of the feature points of the source image and the target image respectively.

[0017] The obtained optimal parameters are used to perform geometric transformation on the PL imaging image so that it can achieve one-to-one correspondence with the microscopic PL spectrum mapping image in spatial position.

[0018] In some other embodiments, in step (3), the pixel overlap ratio is calculated as follows: aligning the photoluminescence imaging image and the microscopic photoluminescence spectrum mapping image based on a binarization process to obtain a corresponding binary pixel matrix and calculate the number of intersection pixels; and calculating the pixel overlap ratio based on the total number of pixels in the candidate area and the number of intersection pixels.

[0019] Specifically, the binarization process uses an adaptive threshold method (the PL imaging threshold is 70% of the average intensity of the defect-free area, and the microscopic PL threshold is three times the noise value of the 426 nm peak intensity) to mark defective pixels as 1 and the remaining pixels as 0 to align the two images.

[0020] After the two images are aligned, compare the two images pixel by pixel to get the number of intersection pixels N overlap :

[0021] in, The serial number of a single pixel in the candidate area (corresponding to each pixel selected one by one, ensuring that all pixels are covered without duplication or omission); The first The binarization result of pixels (1 is a defective pixel, 0 is a non-defective pixel), The first The binarization result of pixels, I imaging and I spectra The binary pixel matrices of the PL imaging image and the microscopic PL spectrum mapping image are respectively multiplied and summed. In essence, the total number of pixels that are judged as defects in both images is counted.

[0022] The total number of pixels in the candidate area is recorded as N candidate , then the pixel overlap rate Calculate using the following formula:

[0023] in, is the total number of intersection pixels in both images that are judged as defects, N candidate is the total number of pixels in the candidate region.

[0024] like R If the probability of defect detection is ≥85%, it can be preliminarily confirmed that the candidate area contains small SF defects.

[0025] In some other embodiments, in step (3), accurate identification of a single small stacking fault defect can also be achieved by calculating a weighted pixel overlap ratio, specifically including: assigning a weight to the pixel point according to its position in the defect area, and obtaining the weighted pixel overlap ratio through weighted calculation.

[0026] Specifically, considering the uneven contribution of different factors to the overlap rate, such as image resolution difference, optical distortion, noise, background change and defect morphology difference, a weighted pixel overlap rate is further proposed. Calculation method:

[0027] in, It is also the serial number of a single pixel point in the candidate area (corresponding to each pixel "selected one by one", ensuring that all pixels are covered without duplication or omission). =1 means that the pixel is judged as a defect in both images, otherwise it is 0 (i.e. It is Pixel matching results, 1 for match, 0 for mismatch); For the The weight value of each pixel is set (the weight of the defect center pixel is set to 1.0, the weight of the edge pixel is set to 0.5, and the weight of the non-defect area pixel is 0).

[0028] If the calculated If the value is ≥85%, it can be further confirmed that the area is a single small stacking fault defect.

[0029] In step (4), the shape of the single small stacking fault defect is an irregular polygon; The scanning step length of the microscopic photoluminescence spectrum two-dimensional mapping scan is ≤2 μm, and the luminescence spectrum covers a range of 360-600 nm.

[0030] Beneficial effects of the present invention: (1) The detection method provided by the present invention, based on PL imaging and microscopic PL spectral fusion analysis, combined with a two-dimensional affine transformation matching model, can identify small stacking fault defects in 4H-SiC substrates at room temperature without loss and with high resolution. This method spatially locates the defects by identifying the local attenuation area of ​​luminescence intensity and accurately extracts the optical characteristics of the defects (central wavelength, number, morphology, size) using two-dimensional spectral scanning. Compared with traditional technologies, the present invention has the significant advantages of non-contact, high sensitivity, and high spatial resolution, and is particularly good at rapid screening of submillimeter defects. This method can be directly applied to industrial large-scale silicon carbide wafers, adapted to online quality control processes, and serve defect classification, yield prediction, and substrate process optimization, significantly improving detection efficiency and reliability.

[0031] (2) The present invention uses an ultraviolet excitation light source to perform large-area photoluminescence (PL) imaging on the silicon carbide substrate to be tested at room temperature, identifying linear areas with locally attenuated luminescence intensity as candidate areas for small stacking fault defects. A two-dimensional mapping scan of the microscopic PL spectrum is performed on the candidate areas to obtain their two-dimensional PL intensity distribution maps. The PL imaging results are combined to accurately locate individual small defects, extract the luminescence intensity and luminescence peak position, and confirm the defect characteristics with a central wavelength of approximately 426 nm. Based on the two-dimensional PL intensity distribution map and peak position characteristics with a smaller scanning step, the morphological characteristics and size information of individual small stacking fault defects are determined, achieving accurate identification. This method has the advantages of high sensitivity and high resolution, is suitable for rapid detection and online quality control of defects in large-scale silicon carbide substrates, and has good industrial application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0033] Figure 1 Schematic diagram of the composition of the PL imaging system used in Example 1 of the present invention; Figure 2 This is a distribution diagram of small stacking fault defects detected by the PL imaging system in Example 1 of the present invention; Figure 3 is a PL imaging result diagram of any candidate area in Example 1 of the present invention; Figure 4 Typical PL spectra of the small SF region and the SF-free region in Example 1 of the present invention; Figure 5 2D high-resolution microscopic PL spectrum mapping diagrams of any three typical small SF defects in Example 1 of the present invention, wherein the three typical small SF defect images are marked as A, B and C respectively. DETAILED DESCRIPTION

[0034] Those skilled in the art will appreciate that the following examples are intended only to illustrate the present invention and should not be construed as limiting the scope of the present invention. Specific conditions are not specified in the examples, and the experiments were performed according to conventional conditions or conditions recommended by the manufacturer.

[0035] Terminology Notes: Photoluminescence: PL for short, is the process in which a material, after being excited by an external light source (usually ultraviolet light), absorbs the energy of photons, produces electron transitions, and then the electrons recombine with holes and release energy in the form of emitted photons.

[0036] Stacking fault: abbreviated as SF, is a common two-dimensional lattice defect in crystal structure. It refers to the abnormal atomic arrangement that locally deviates from the regular order in the normal stacking sequence of crystal atoms along a specific crystal plane (such as a close-packed plane). This abnormality is only limited to a thin layer (usually a few atomic layers thick) and does not change the overall structure type of the crystal.

[0037] Traditional PL imaging is limited by the optical diffraction limit and has difficulty distinguishing small stacking faults, resulting in missed detection of tiny defects. While microscopic PL spectroscopy can break the diffraction limit, its point-by-point scanning mode makes full-area detection of large-sized wafers too time-consuming and unable to meet the industrial online efficiency requirements. At the same time, there are inherent contradictions between the two technologies: (1) The difference in field of view and coordinate system between PL imaging (fast, low resolution) and microscopic PL spectroscopy (slow, high resolution) makes spatial registration difficult due to sample tilt and optical distortion, affecting the accuracy of defect positioning; (2) The lack of a physical correlation model between luminescence intensity decay (PL imaging feature) and specific spectral peak position / linewidth (microscopic PL feature) hinders the joint analysis of multimodal defects; (3) The weak luminescence decay of small stacking faults is easily overwhelmed by substrate background noise (impurity luminescence, surface scattering), significantly reducing the signal-to-noise ratio.

[0038] The present invention provides a multimodal detection method for small stacking fault defects in silicon carbide substrates, which can be applied to multimodal nondestructive detection of small stacking fault defects in silicon carbide substrates, and specifically includes the following steps: (1) Photoluminescence imaging is performed on the silicon carbide substrate to obtain candidate areas for small stacking fault defects; (2) performing a two-dimensional mapping scan of the microscopic photoluminescence spectrum of the candidate area to obtain a spectral intensity mapping image within the candidate area; (3) A multimodal matching model was constructed to spatially register the spectral intensity mapping image and the photoluminescence image, and a single small stacking fault defect was confirmed by calculating the pixel overlap ratio; (4) Perform two-dimensional mapping scanning of microscopic photoluminescence spectra on a single small stacking fault defect to determine the defect morphology and size.

[0039] The solution of the present invention is described below in conjunction with specific embodiments: Example 1 This embodiment provides a multimodal detection method for small stacking fault defects in silicon carbide substrates, which specifically includes the following steps: S1. Perform photoluminescence imaging on the silicon carbide substrate to obtain candidate regions of small stacking fault defects; S2, performing a two-dimensional mapping scan of the microscopic photoluminescence spectrum of the candidate area to obtain a spectral intensity mapping image within the candidate area; S3. Construct a multimodal matching model to spatially register the spectral intensity mapping image and the photoluminescence image, and confirm a single small stacking fault defect by calculating the pixel overlap ratio; S4. Perform two-dimensional mapping scanning of a single small stacking fault defect using microscopic photoluminescence spectrum to determine the defect morphology and size.

[0040] In step S1, a photoluminescence (PL) imaging system is used to perform large-area imaging of the entire silicon carbide substrate under room temperature. Relying on the image processing function of the photoluminescence (PL) imaging system, the imaging image is analyzed to obtain candidate areas for small stacking fault defects.

[0041] The components of the PL imaging system used are as follows: Figure 1 As shown in the figure, it includes an imaging camera, a red high-pass filter, a substrate placement platform, and a UV laser light source that illuminates the substrate placement platform. The UV laser light source has a wavelength of 313 nm or 320 nm; the red high-pass filter has a cutoff wavelength of 660 nm or 700 nm (only red light signals with wavelengths ≥660 nm or ≥700 nm are allowed to pass).

[0042] The PL imaging system works as follows: a UV laser illuminates the silicon carbide substrate under test, exciting the silicon carbide material and generating a PL signal. This PL signal passes through a red high-pass filter placed in the signal receiving path, effectively removing short-wavelength interference signals from the background, thereby improving the image contrast and signal-to-noise ratio. The filtered PL signal is ultimately received and recorded by an imaging camera, forming a PL image that characterizes the defect.

[0043] Specifically, the process of obtaining candidate regions for small stacking fault defects is as follows: (1) Under room temperature conditions, a PL imaging system is used to perform large-area imaging of the entire silicon carbide substrate to be tested. The PL imaging system analyzes and processes the large-area imaging based on its own image processing function, that is, the luminous intensity of each area of ​​the substrate is intuitively reflected through the difference in light and dark - the area with stronger luminescence appears bright, and the area with locally attenuated luminescence intensity appears dim.

[0044] The system uses the luminous intensity of defect-free areas on the substrate surface as a benchmark and automatically marks areas that significantly deviate from this benchmark (too bright or too dark) as areas of abnormal luminous intensity, which serve as preliminary screening defects. By analyzing and processing large-area imaging, dim areas are initially screened.

[0045] (2) For the dark areas initially screened out, linear areas with an aspect ratio of >5:1, a length in the range of 0.2-1 mm, and continuous without obvious breakpoints are screened out; at the same time, non-linear interferences such as point-like dark spots formed by surface contamination and irregular block dark areas caused by scratches are automatically excluded, and continuous dark lines that are highly matched with the local luminescence attenuation characteristics caused by small stacking fault lattice defects are retained. By performing multiple imaging verifications on the same substrate, if the imaging results of the above-mentioned linear areas are consistent, it can be confirmed that they are candidate areas for small stacking fault defects, and the system marks and records their specific distribution positions on the substrate. The distribution map of small stacking fault defects detected by the PL imaging system is shown in the figure below. Figure 2 shown.

[0046] (3) The candidate regions of small stacking fault defects obtained by the PL imaging system were further verified by spherical aberration-corrected scanning transmission electron microscopy, and this type of defect was attributed to stacking faults.

[0047] Will Figure 2 Microscale linear stacking faults detected in the image are defined as small stacking faults (SFs). In the defect distribution map obtained from PL imaging, the location of each detected SF is marked with a black dot to facilitate subsequent analysis and screening. Furthermore, local regions containing these SFs are preliminarily identified as candidate regions.

[0048] In the present invention, Figure 2 Only the distribution of small stacking fault defects involved is given, which does not mean that other defects will not be identified and classified. Figure 3 In the present invention, the silicon carbide substrate to be tested is a 4H-SiC substrate material, and there is no particular restriction on its specific source, and commercially available products well known to those skilled in the art can be selected.

[0049] In step S2, a two-dimensional microscopic PL spectrum mapping scan is performed on the candidate area determined in S1 to obtain a spectral intensity mapping image within the area. The specific process is as follows: the silicon carbide substrate to be tested is fixed on the precision displacement platform of the microscopic PL spectrometer, and the candidate area is observed and accurately positioned using an optical microscope. The scanning range is set (covering the candidate area and its surrounding normal range) to ensure that the target defect and the surrounding reference area are completely included.

[0050] More specifically, after starting the microscopic PL spectrometer scan, the precision displacement stage moves the silicon carbide substrate under test point by point within the XY plane at a set step size. At each test point, the excitation light is focused on that point, emitting a PL signal. The spectral detector simultaneously captures the complete PL spectrum at that point. After the scan is complete, the PL spectra of the defect-free region are compared with the PL spectra of the candidate region, confirming that the characteristic emission wavelength of the defect region is 426 nm. The spectral peak intensity at and near 426 nm is integrated for an integration time of ≥0.01 s. The integration result is correlated with the spatial coordinates of each test point to generate a spectral intensity map image that visually reflects the defect emission intensity distribution, as shown in Figure 4. This indicates that the characteristic emission wavelength of the small SF is 426 nm, providing fundamental data support for subsequent small SF defect identification.

[0051] The microscopic PL spectrometer used is equipped with a 320 nm ultraviolet excitation light source, with a spatial scanning resolution better than 1 μm and a luminescence spectrum detection range of 360-600 nm (covering the main defect-related luminescence peaks in SiC); the scanning step size is set to ≤5 μm.

[0052] In step S3, due to differences in resolution and field of view between PL imaging and microscopic PL imaging techniques (PL imaging is low-resolution over a large area, while microscopic PL is high-resolution over a small area), direct comparison will result in spatial offset. To accurately locate the defect, the spectral intensity mapping image obtained in S2 must be spatially registered with the PL imaging result obtained in S1. Geometric transformations are performed using a multimodal matching model based on a two-dimensional affine transformation to achieve spatial registration of the spectral intensity mapping image and the photoluminescence imaging. Individual small stacking fault defects are then further confirmed by calculating pixel overlap or weighted pixel overlap. This specifically includes the following steps: (1) The endpoints of large-scale rod-shaped SFs are preferably selected as feature points. If the sample has no large-scale defects, the laser marking on the edge of the wafer can be manually marked as feature points. This is because the endpoints of large-scale rod-shaped SFs are clearly visible in both PL imaging and microscopic PL imaging systems.

[0053] (2) Record the pixel coordinates of each pair of feature points in the two images, where the pixel coordinates of the PL imaging image are ( ), the pixel coordinates of the micro PL spectrum image are ( ), = 1, 2, 3; substitute the pixel coordinates of the feature points into the multimodal matching model of the two-dimensional affine transformation, achieve pixel-level spatial alignment through feature point calibration, and solve the vector p using the least squares method to complete the spatial registration.

[0054] The mathematical expression of the multimodal matching model of two-dimensional affine transformation is: ,

[0055] in,( ) is the pixel coordinate of the PL imaging image, ( ) is the pixel coordinate of the microscopic PL spectrum mapping image, =1, 2, 3; , , ， is the linear transformation coefficient, is the translation coefficient.

[0056] The mathematical expression of the least squares method for solving vector p is:

[0057] in, is the parameter vector to be determined, A and It consists of the coordinates of the feature points of the source image and the target image respectively.

[0058] The obtained vector p is used to perform geometric transformation on the PL imaging image so that it can achieve one-to-one correspondence with the microscopic PL spectrum mapping image in terms of spatial position.

[0059] (3) After the spatial registration is completed, the pixel overlap ratio is introduced to quantitatively evaluate the registration accuracy of the two images. The specific process is as follows: Binarization: Adaptive thresholding was used (the PL imaging threshold was 70% of the average intensity of the defect-free area, and the microscopic PL threshold was three times the noise value of the 426 nm peak intensity). Defective pixels were marked as 1 and the remaining pixels were marked as 0 to align the two images.

[0060] After the two images are aligned, compare the two images pixel by pixel to get the number of intersection pixels N overlap :

[0061] in, The serial number of a single pixel in the candidate area (corresponding to each pixel selected one by one, ensuring that all pixels are covered without duplication or omission); The first The binarization result of pixels (1 is a defective pixel, 0 is a non-defective pixel), The first The binarization result of pixels, I imaging and I spectraThe binary pixel matrices of the PL imaging image and the microscopic PL spectrum mapping image are respectively multiplied and summed. In essence, the total number of pixels that are judged as defects in both images is counted.

[0062] The total number of pixels in the candidate area is recorded as N candidate , then the pixel overlap rate Calculate using the following formula:

[0063] in, is the total number of intersection pixels in both images that are judged as defects, N candidate is the total number of pixels in the candidate region.

[0064] like R If the percentage is ≥85%, the candidate area can be preliminarily confirmed to contain small SF defects.

[0065] Considering the uneven contribution of different factors to the overlap rate, such as image resolution difference, optical distortion, noise, background change and defect morphology difference, a weighted pixel overlap rate is further proposed. Calculation method:

[0066] in, It is also the serial number of a single pixel point in the candidate area (corresponding to each pixel "selected one by one", ensuring that all pixels are covered without duplication or omission). =1 means that the pixel is judged as a defect in both images, otherwise it is 0 (i.e. M i It is Pixel matching results, 1 for match, 0 for mismatch); For the The weight value of each pixel is set (the weight of the defect center pixel is set to 1.0, the weight of the edge pixel is set to 0.5, and the weight of the non-defect area pixel is 0).

[0067] If the calculated If the wavelength is ≥85%, it can be further confirmed that the area is a single small stacking fault defect with a central wavelength of approximately 426 nm, and its spatial position on the substrate is recorded.

[0068] Through the above method, the PL imaging characteristics and microscopic spectral characteristics of each small SF defect can be obtained simultaneously, achieving precise spatial positioning and confirming that the central wavelength of the defect is approximately 426 nm.

[0069] After registering the candidate regions of the 4H-SiC substrate to be tested in this embodiment, R=89% is calculated. =91%, meeting the accuracy requirement. At this point, the dark line region in the PL image completely overlaps with the 426 nm strong peak region in the microscopic PL. The deviation mainly comes from edge distortion of the PL image (deviation ≤ 2 pixels at 10 mm from the center), which can be effectively corrected through weighted calculation.

[0070] Through the above method, the PL imaging characteristics and microscopic spectral characteristics of each small SF defect can be obtained simultaneously, achieving precise spatial positioning and confirming that the central wavelength of the defect is approximately 426nm.

[0071] In step S4, a microscopic PL spectrometer is used to perform high-precision point-by-point scanning with a smaller scanning step size for the single small stacking fault defect area located in S3. The microscopic PL spectrum intensity mapping image of the micro area is obtained by integrating the spectral peak intensity of 426 nm and its surrounding bands. The small stacking fault appears as a bright contrast in the image, based on which its true morphology can be observed and its length, width and other dimensional parameters can be quantified through image calibration.

[0072] Specifically, a microscopic PL two-dimensional mapping scan is further performed on a single small SF defect in any three regions. The microscopic PL spectrometer used is equipped with a 320 nm ultraviolet excitation light source, with a spatial scanning resolution better than 1 μm and a luminescence spectrum detection range of 360-600 nm (covering the main defect-related luminescence peaks in SiC). The scanning step is set to ≤ 2 μm, and the PL spectrum intensity in the 426±10 nm band is integrated to obtain the cumulative light intensity signal in this band, i.e., the microscopic PL two-dimensional mapping scan result, as shown in Fig. Figure 5 As shown in the figure, the "CCD cts" marked in the figure is the charge-coupled device photon count, which represents the photon signal count detected and quantified by the CCD sensor at the corresponding point of the "specific spatial position" and the "specific wavelength in the 426±10nm band" during the scanning process. The value directly reflects the relative strength of the PL luminescence intensity at that position and wavelength. Figure 5 The numbers in Figure A are 609.3, -98.5, Figure 5 The numbers 407.3, -102.8 and Figure 5 The numbers 371.5 and -102 in Figure C are the relative PL luminescence intensity values ​​(unit: CCD cts) in the 426±10nm band of the corresponding micro-area position: the positive values ​​609.3, 407.3, and 371.5 represent the luminescence signal intensity of the core area of ​​the small stacking fault defect in the three sub-images A, B, and C, respectively. The higher the value, the more significant the defect characteristic luminescence at that position; the negative values ​​-98.5, -102.8, and -102 represent the luminescence signal intensity of the non-defect area around the defect in the corresponding sub-image, respectively. The difference between the two types of values ​​clearly distinguishes the luminescence characteristics of the defect area from the non-defect area.

[0073] like Figure 5 Figures A, B, and C in the figure show that a single small SF defect appears as a polygonal morphology with varying sizes and irregular boundaries in the 2D PL mapping scan results. This morphological information serves as an important basis for substrate quality assessment and is significantly beneficial for improving the efficiency and accuracy of defect identification.

[0074] In summary, the detection method provided in this embodiment is based on the joint analysis of PL imaging and microscopic PL spectroscopy, and introduces a two-dimensional affine transformation matching model. It can achieve non-destructive, high-resolution detection of small stacking fault defects in 4H-SiC substrates at room temperature.

[0075] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A multimodal detection method for small stacking fault defects in silicon carbide substrates, characterized in that: The following steps are involved: (1) Photoluminescence imaging is performed on the silicon carbide substrate to obtain candidate areas for small and medium-sized stacking fault defects; (2) performing a two-dimensional mapping scan of the microscopic photoluminescence spectrum of the candidate area to obtain a spectral intensity mapping image; (3) A multimodal matching model was constructed to spatially register the spectral intensity mapping image and the photoluminescence image, and a single small stacking fault defect was confirmed by calculating the pixel overlap ratio; (4) Perform two-dimensional mapping scanning of microscopic photoluminescence spectra on a single small stacking fault defect to determine the defect morphology and size.

2. The multimodal detection method for small stacking fault defects in silicon carbide substrates according to claim 1, characterized in that: In step (1), the silicon carbide substrate is a 4H-SiC substrate, and the small and medium-sized stacking fault defects are continuous linear defects with no breakpoints and an aspect ratio greater than 5:1 and a length in the range of 0.2-1 mm.

3. The multimodal detection method for small stacking fault defects in silicon carbide substrates according to claim 1, characterized in that: In step (1), the light source used for the photoluminescence imaging is an ultraviolet laser light source, and a red light high-pass filter is configured in the signal receiving path.

4. The multimodal detection method for small stacking fault defects in silicon carbide substrates according to claim 1, characterized in that: In step (2), the light source for the two-dimensional mapping scan of the microscopic photoluminescence spectrum is an ultraviolet excitation light source, and the spatial resolution is >1 μm.

5. The multimodal detection method for small stacking fault defects in silicon carbide substrates according to claim 1, characterized in that: In step (2), the luminescence spectrum of the two-dimensional mapping scan of the microscopic photoluminescence spectrum covers a range of 360-600 nm, the scanning step is ≤5 μm, and the integration time is ≥0.01 s.

6. The multimodal detection method for small stacking fault defects in silicon carbide substrates according to claim 1, characterized in that: In step (3), the multimodal image matching model is a two-dimensional affine transformation model; it also includes performing geometric transformation on the photoluminescence imaging based on the least squares method, and performing spatial position matching on the spectral intensity mapping image and the photoluminescence imaging.

7. The multimodal detection method for small stacking fault defects in silicon carbide substrates according to claim 1, characterized in that: In step (3), the pixel overlap ratio is calculated as follows: the photoluminescence imaging image and the microscopic photoluminescence spectrum mapping image are binarized and aligned to obtain the number of intersection pixels, and the pixel overlap ratio is calculated in combination with the total number of pixels in the candidate area.

8. The multimodal detection method for small stacking fault defects in silicon carbide substrates according to claim 1, characterized in that: Step (3) also includes confirming a single small stacking fault defect by calculating a weighted pixel overlap ratio; the weighted pixel overlap ratio is calculated as follows: different weights are assigned according to the spatial position of the pixel point in the defect area, and after weighted summation of the matched pixels, the weighted pixel overlap ratio is calculated in combination with the total weight.

9. The multimodal detection method for small stacking fault defects in silicon carbide substrates according to claim 1, characterized in that: In step (4), the scanning step length of the two-dimensional mapping scan of the microscopic photoluminescence spectrum is ≤2 μm, and the luminescence spectrum coverage range is 360-600 nm.

10. The multimodal detection method for small stacking fault defects in silicon carbide substrates according to claim 1, characterized in that: In step (4), the shape of the single small stacking fault defect is an irregular polygon.

Images