Non-destructive detection method of dislocations based on confocal raman / photoluminescence spectroscopy

Abstract

The application provides a dislocation nondestructive detection method based on a confocal Raman / photoluminescence spectrum, and relates to the technical field of semiconductor material detection. In the detection process, no destructive treatment such as KOH corrosion is needed, the material integrity is ensured, the detection cost is reduced, and full sample inspection is realized. Through comprehensive spectrum scanning, the accidental and statistical deviation of the loss sampling detection is avoided, and the comprehensiveness and accuracy of the dislocation detection are improved. The confocal Raman / photoluminescence system is used to comprehensively scan the silicon carbide substrate material, the dislocation information is automatically judged based on the spectrum characteristics and a preset dislocation analysis model, a dislocation spatial distribution diagram and a dislocation detection report are generated, the problem that the accuracy of the loss detection method of the dislocation of the silicon carbide substrate and epitaxial material is low is fundamentally solved, the nondestructive detection of the dislocation of the silicon carbide substrate and epitaxial material is realized, and the dislocation detection accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of semiconductor material testing technology, and in particular to a non-destructive testing method for dislocations based on confocal Raman / photoluminescence spectroscopy. Background Technology

[0002] Silicon carbide (4H-SiC), as a representative of third-generation wide-bandgap semiconductor materials, has shown great application potential in high-temperature, high-frequency, high-voltage, and high-power devices due to its excellent properties such as high breakdown electric field, high thermal conductivity, and high electron saturation drift velocity. However, compared with mature silicon semiconductor technology, the high dislocation density in 4H-SiC materials, such as threading screw dislocations (TSD), basal plane dislocations (BPD), and threading edge dislocations (TED), is one of the key bottlenecks restricting the improvement of device performance, reliability, and yield. For example, BPD can transform into stacking faults during device operation, leading to increased on-resistance and performance degradation in bipolar devices.

[0003] Currently, the industry commonly uses a destructive testing method—high-temperature etching with molten KOH (or KOH + Na₂O₂) at 400℃~500℃—to detect dislocation density in 4H-SiC substrates and epitaxial materials. This method utilizes the difference in etching rate between dislocation outcrops and intact lattice regions to reveal characteristic etching pits under an optical microscope, which are then counted manually. During this process, the sample is permanently etched and cannot be used for epitaxial growth or device fabrication, resulting in material and cost waste. Because the sample is destroyed, it is impossible to trace and analyze defects at the same location in the substrate, epitaxial layer, and final device, making it difficult to accurately study the extension and transformation behavior of dislocations in subsequent processes and their specific impact on device performance. Destructive testing methods typically employ sampling inspection, and the identification and classification of etching pits heavily rely on operator experience, resulting in high subjectivity and low statistical efficiency and accuracy. Summary of the Invention

[0004] This invention provides a non-destructive testing method for dislocations based on confocal Raman / photoluminescence spectroscopy, which solves the problem of low accuracy of destructive testing methods for dislocations in silicon carbide substrates and epitaxial materials, realizes non-destructive testing of dislocations in silicon carbide substrates and epitaxial materials, and improves the accuracy of dislocation detection.

[0005] In a first aspect, the present invention provides a non-destructive dislocation detection method based on confocal Raman / photoluminescence spectroscopy. The method includes: performing a comprehensive scan of the silicon carbide substrate material to be tested using a confocal Raman / photoluminescence system to obtain the original spectra of each detection point on the silicon carbide substrate material; extracting features from the original spectra of each detection point to determine the spectral characteristics of each detection point; determining the dislocation analysis results of each detection point based on the spectral characteristics of each detection point and a preset dislocation analysis model, the dislocation analysis results including whether a dislocation has occurred and the type of dislocation; generating a dislocation spatial distribution map of the silicon carbide substrate material based on the dislocation spatial distribution map of the silicon carbide substrate material; and performing density calculation and statistics based on the dislocation spatial distribution map of the silicon carbide substrate material to determine a dislocation detection report for the silicon carbide substrate material.

[0006] In one possible implementation, the spectral features include the peak position of the characteristic peak, the full width at half maximum (FWHM) of the characteristic peak, the relative intensity of the characteristic peak, and the integral intensity within a specific energy range. Feature extraction is performed on the original spectra of each detection point to determine the spectral features of each detection point. This includes performing continuous first-order differential calculations on the original spectra to identify multiple candidate characteristic peaks; fitting a Gaussian-Lorentz mixture function to the multiple candidate characteristic peaks, and determining the energy value corresponding to the function's extremum point as the peak position of the characteristic peak; determining the FWHM of the characteristic peak using linear interpolation based on the peak position; selecting a reference characteristic peak based on the original spectrum; calculating the integral intensity of the reference characteristic peak and the integral intensity of the characteristic peak to be analyzed; determining the relative intensity of the characteristic peak based on the integral intensity of the reference characteristic peak and the integral intensity of the characteristic peak to be analyzed; performing integration based on the original spectrum and a preset integration interval, calculating the area between the original spectrum and the energy axis within the integration interval, and determining this as the integral intensity within a specific energy range; the integration interval is one or more broadened emission bands related to the dislocation type.

[0007] In one possible implementation, based on the spectral characteristics of each detection point and a pre-defined dislocation analysis model, the dislocation analysis results for each detection point are determined, including: determining the feature vector of each detection point based on the spectral characteristics of each detection point; outputting the probability of dislocation occurring at each detection point and the probability of each detection point belonging to each dislocation category based on the feature vector of each detection point and the dislocation analysis model; and determining the dislocation analysis results based on the probability of dislocation occurring at each detection point and the probability of each detection point belonging to each dislocation category.

[0008] In one possible implementation, before determining the dislocation analysis results for each detection point based on the spectral characteristics of each detection point and a pre-defined dislocation analysis model, the following steps are included: acquiring the original spectra of each detection point on multiple silicon carbide substrate materials; generating spectral characteristics of each detection point on each silicon carbide substrate material based on the original spectra of each detection point on each silicon carbide substrate material; using the melt etching method on multiple silicon carbide substrate materials respectively to identify and label the dislocation type of each spatial coordinate on each silicon carbide substrate material; constructing multiple training samples with the spectral characteristics of each detection point on each silicon carbide substrate material as input and the dislocation type of each spatial coordinate on each silicon carbide substrate material as output; and performing machine learning based on the multiple training samples to obtain the dislocation analysis model.

[0009] In one possible implementation, a spatial distribution map of dislocations in a silicon carbide substrate material is generated based on the dislocation analysis results of each detection point. This includes: associating the spatial coordinates of each detection point with the dislocation analysis results to construct a mapping dataset between spatial coordinates and dislocation categories; constructing a digital base map of the silicon carbide substrate material based on its size and shape information; and drawing each pixel on the digital base map based on the digital base map, the mapping dataset, and preset color mapping rules to obtain the spatial distribution map of dislocations in the silicon carbide substrate material.

[0010] In one possible implementation, based on the dislocation spatial distribution map of the silicon carbide substrate material, density calculation and statistics are performed to determine the dislocation detection report of the silicon carbide substrate material. This includes: classifying and statistically counting the number of pixels of each dislocation type based on the dislocation spatial distribution map of the silicon carbide substrate material; calculating the area of ​​the effective scanning area based on the physical dimensions of the scanning area in the dislocation spatial distribution map; calculating the density of each dislocation type based on the number of pixels of each dislocation type and the area of ​​the effective scanning area; calculating the total dislocation density of the silicon carbide substrate material based on the density of each dislocation type; and generating the dislocation detection report of the silicon carbide substrate material based on the basic information of the silicon carbide substrate material, the detection condition parameters of the full scan, the density of each dislocation type, the total dislocation density, and the dislocation spatial distribution map.

[0011] In one possible implementation, after calculating and statistically analyzing the dislocation spatial distribution map of the silicon carbide substrate material to determine the dislocation detection report, the process further includes: aligning and overlaying the dislocation spatial distribution map with the design layout of the target power device to obtain a fused image; using logical AND operations in image processing to identify and statistically analyze pixels of each dislocation type in the active region of the target power device; calculating the effective dislocation density in the active region of the target power device based on the pixels of each dislocation type in the active region; obtaining predicted values ​​of key electrical parameters and the pass probability of the device fabricated using the silicon carbide substrate material based on the effective dislocation density in the active region and a pre-established device performance testing model; the device performance testing model characterizes the quantitative relationship between the effective dislocation density in the active region and the key electrical parameters; the key electrical parameters include breakdown voltage, leakage current, and on-resistance; and generating a device performance prediction report for the silicon carbide substrate material based on the effective dislocation density in the active region, the predicted values ​​of the key electrical parameters, and the pass probability.

[0012] In one possible implementation, after performing density calculations and statistics based on the dislocation spatial distribution map of the silicon carbide substrate material to determine the dislocation detection report of the silicon carbide substrate material, the process further includes: performing spatial distribution statistical analysis on the dislocation spatial distribution map to identify defect aggregation patterns; the defect aggregation patterns include annular distribution, radial stripe distribution, localized cluster aggregation, or edge-dense distribution; based on the defect aggregation patterns and a preset defect pattern-process root cause knowledge base, matching is performed to query the root process parameters that cause the defect aggregation patterns, as well as the direction of process parameter adjustment; the root process parameters include crystal growth temperature field gradient, crucible rotation speed and pulling speed, substrate cutting angle, or etching conditions before epitaxial growth; the defect pattern-process root cause knowledge base represents the mapping relationship between different defect aggregation patterns and process parameters; based on the defect aggregation patterns, root process parameters, and the direction of process parameter adjustment, a process optimization suggestion report is generated.

[0013] Secondly, embodiments of the present invention provide a dislocation non-destructive testing device based on confocal Raman / photoluminescence spectroscopy. This non-destructive testing device includes: a data acquisition module and a data processing module. The data acquisition module is used to perform a comprehensive scan of the silicon carbide substrate material to be tested using a confocal Raman / photoluminescence system to obtain the original spectra of each detection point on the silicon carbide substrate material. The data processing module is used to extract features from the original spectra of each detection point to determine the spectral characteristics of each detection point; based on the spectral characteristics of each detection point and a preset dislocation analysis model, to determine the dislocation analysis results of each detection point, including whether a dislocation has occurred and the type of dislocation; based on the dislocation analysis results of each detection point, to generate a dislocation spatial distribution map of the silicon carbide substrate material; and based on the dislocation spatial distribution map of the silicon carbide substrate material, to perform density calculation and statistics to determine a dislocation detection report for the silicon carbide substrate material.

[0014] Thirdly, embodiments of the present invention provide an electronic device including a memory and a processor. The memory stores a computer program, and the processor is configured to call and run the computer program stored in the memory to perform the steps of the method as described in the first aspect and any possible implementation thereof.

[0015] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method as described in the first aspect and any possible implementation thereof.

[0016] This invention provides a non-destructive testing method for dislocations based on confocal Raman / photoluminescence spectroscopy. This method eliminates the need for destructive treatments such as KOH etching during the testing process, ensuring material integrity and subsequent usability, reducing testing costs, and enabling full sample inspection. Through fully automated, full-surface spectral scanning and analysis, this invention completely avoids the randomness and statistical biases of traditional destructive sampling testing, accurately capturing local defects and macroscopic distribution patterns, significantly improving the comprehensiveness and accuracy of dislocation detection. Furthermore, it transforms abstract defect information into intuitive spatial distribution maps and quantitative statistical reports of dislocations, providing unprecedentedly accurate and objective data support for process traceability and quality assessment, overcoming the limitations of traditional methods that rely on human experience, are highly subjective, and cannot perform spatial analysis. This invention employs a confocal Raman / photoluminescence system to perform a comprehensive spectral scan of silicon carbide substrate materials. Based on spectral characteristics and a preset dislocation analysis model, it automatically determines dislocation information, generates a spatial distribution map of dislocations and a dislocation detection report, fundamentally solving the problem of low accuracy in destructive detection methods for dislocations in silicon carbide substrates and epitaxial materials. This enables non-destructive detection of dislocations in silicon carbide substrates and epitaxial materials, improving the accuracy of dislocation detection. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic flowchart of a non-destructive testing method for dislocations based on confocal Raman / photoluminescence spectroscopy provided in an embodiment of the present invention;

[0019] Figure 2 This is a schematic diagram of the structure of a dislocation nondestructive testing device based on confocal Raman / photoluminescence spectroscopy provided in an embodiment of the present invention;

[0020] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0022] In the description of this invention, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" and "more than one" refer to two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0023] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner to facilitate understanding.

[0024] Furthermore, the terms "comprising" and "having," and any variations thereof, used in the description of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not limited to the steps or modules listed, but may optionally include other steps or modules not listed, or may optionally include other steps or modules inherent to such process, method, product, or device.

[0025] To make the objectives, technical solutions, and advantages of the present invention clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0026] like Figure 1 As shown, this embodiment of the invention provides a non-destructive testing method for dislocations based on confocal Raman / photoluminescence spectroscopy. The method includes steps S101-S105.

[0027] S101. The silicon carbide substrate material to be tested is scanned using a confocal Raman / photoluminescence system to obtain the original spectrum of each detection point of the silicon carbide substrate material.

[0028] For example, embodiments of the present invention can use an automated sample stage of a confocal Raman / photoluminescence system to drive the silicon carbide substrate material along a preset path, ensuring that the laser focus can cover the entire or a specified area of ​​the sample under test. During the scanning process, a constant or variable scanning step is used, with the step size set according to the detection resolution requirements, typically between 1 micrometer and 50 micrometers. At each detection point, the complete spectral signal containing Raman scattering and photoluminescence information generated by the micro-region under laser excitation is acquired and recorded, i.e., the raw spectrum.

[0029] S102. Perform background subtraction and feature extraction on the original spectra of each detection point to determine the spectral characteristics of each detection point.

[0030] In some embodiments, spectral features refer to physical parameters quantified from raw spectral data that characterize the microstructure and defect states of the material. This step processes the raw spectral data using computer algorithms to extract key information related to crystal defects such as dislocations, providing a basis for subsequent dislocation identification and classification.

[0031] In some embodiments, spectral features include the peak position of the characteristic peak, the full width at half maximum (FWHM) of the characteristic peak, the relative intensity of the characteristic peak, and the integrated intensity over a specific energy range.

[0032] As one possible implementation, step S102 can be specifically implemented as steps S1021-S1027.

[0033] S1021. Perform continuous first-order differential calculations on the original spectrum to identify multiple candidate characteristic peaks.

[0034] S1022. Fit the Gaussian-Lorentz mixture function to multiple candidate characteristic peaks, and determine the energy value corresponding to the extreme point of the function as the peak position of the characteristic peak.

[0035] For example, in this embodiment of the invention, the first-order differential calculation is performed on the original spectrum to find the zero-crossing point where the spectral intensity decreases from increasing to decreasing, thus initially locating candidate characteristic peaks. Subsequently, a Gaussian-Lorentz mixture function is fitted to the spectral interval where the candidate peak is located, and the energy value corresponding to the extreme point of the fitted function is precisely determined as the peak position of the characteristic peak.

[0036] S1023. Based on the peak position of the characteristic peak, the full width at half maximum (FWHM) of the characteristic peak is determined by linear interpolation.

[0037] For example, in this embodiment of the invention, based on the already determined precise peak position, two energy points corresponding to half the maximum intensity of the characteristic peak are found. If the data points are not exactly located at the half-width at half-maximum (WHM), linear interpolation is used to accurately determine the positions of these two points, and the energy difference between them is the WHM of the characteristic peak.

[0038] S1024. Select reference characteristic peaks based on the original spectrum.

[0039] S1025. Calculate the integral intensity of the reference characteristic peak and the integral intensity of the characteristic peak to be analyzed.

[0040] S1026. Determine the relative intensity of the characteristic peak based on the integral intensity of the reference characteristic peak and the integral intensity of the characteristic peak to be analyzed.

[0041] For example, in this embodiment of the invention, a reference characteristic peak with stable intensity (e.g., the intrinsic phonon peak of 4H-SiC) is selected from the original spectrum as an internal standard. The integrated intensities of the reference peak and the characteristic peak to be analyzed are calculated within a specific spectral range. Finally, the ratio of the integrated intensity of the characteristic peak to be analyzed to the integrated intensity of the reference peak is calculated, and this ratio is determined as the relative intensity of the characteristic peak. This method can effectively eliminate the influence of fluctuations in test conditions.

[0042] S1027. Based on the original spectrum and the preset integration interval, perform integration to calculate the area between the original spectrum and the energy axis within the integration interval, and determine it as the integrated intensity of the specific energy interval.

[0043] In some embodiments, the integration interval is one or more broadened emission bands associated with the dislocation type.

[0044] For example, in this embodiment of the invention, based on previous research or known literature, the energy range of photoluminescence characteristics related to the target dislocation type is determined, and this range is set as a preset integration interval. Numerical integration (e.g., using the trapezoidal rule) is performed on the original spectral data points within this interval, and the area enclosed by the spectral curve and the energy axis is calculated. This area value is the integrated intensity of the specific energy interval, reflecting the defect concentration or activity associated with that dislocation.

[0045] S103. Based on the spectral characteristics of each detection point and the preset dislocation analysis model, determine the dislocation analysis results of each detection point.

[0046] In some embodiments, the dislocation analysis model is a classifier based on a machine learning algorithm. This model is trained using a large amount of spectral data with known dislocation types and is able to establish a complex nonlinear mapping relationship between spectral features and dislocation categories. In application, the model receives the input spectral features, performs internal calculations, and outputs a judgment result indicating whether a dislocation exists at the detection point and its type.

[0047] In this embodiment of the application, the dislocation analysis results include whether a dislocation has occurred and the type of dislocation.

[0048] For example, dislocation types include screw dislocations, basal dislocations, and edge dislocations.

[0049] As one possible implementation, step S103 can be specifically implemented as steps S1031-S1033.

[0050] S1031. Based on the spectral characteristics of each detection point, determine the feature vector of each detection point.

[0051] In some embodiments, a feature vector refers to a digital sequence formed by combining multiple spectral features extracted from the original spectrum of a detection point in a preset order. For example, four key parameters—the peak position of the characteristic peak, the full width at half maximum (FWHM) of the characteristic peak, the relative intensity of the characteristic peak, and the integral intensity of the characteristic peak in a specific energy range—can be arranged in a fixed order to form a multidimensional feature vector representing the spectral characteristics of the detection point. This feature vector can comprehensively characterize the physical state of the material in the micro-region.

[0052] S1032. Based on the feature vectors of each detection point and the dislocation analysis model, output the probability of a dislocation occurring at each detection point and the probability of each detection point belonging to a dislocation category.

[0053] S1033. Based on the probability of dislocation occurring at each detection point and the probability of each detection point belonging to each dislocation category, determine the dislocation analysis results.

[0054] For example, in this embodiment of the invention, the final dislocation analysis result is determined based on the probability value output by the model and using a preset decision rule. First, the overall probability of a dislocation occurring is compared with a preset judgment threshold. If the probability is lower than the threshold, the detection point is ultimately determined to be without a dislocation. If the probability is higher than or equal to the threshold, the dislocation type determination stage is entered, where the category with the highest conditional probability among all types of dislocations is selected as the final dislocation type for the detection point. This dual judgment mechanism based on probability thresholds can effectively reduce misjudgments caused by noise interference and improve the accuracy and reliability of classification results.

[0055] S104. Based on the dislocation analysis results at each detection point, generate a spatial distribution map of dislocations in the silicon carbide substrate material.

[0056] For example, embodiments of the present invention can associate the two-dimensional spatial coordinates of each detection point with its corresponding dislocation analysis results (e.g., defect-free, screw dislocation, basal dislocation, edge dislocation). Subsequently, a digital image matching the actual size of the sample is generated in the computer, and pixels are drawn at the corresponding coordinate positions in the image according to a preset color scheme (e.g., using different colors to represent different dislocation types), ultimately forming an image that can intuitively reflect the spatial distribution of dislocations.

[0057] As one possible implementation, step S104 can be specifically implemented as steps S1041-S1043.

[0058] S1041. Correlate the spatial coordinates of each detection point with the dislocation analysis results to construct a mapping dataset between spatial coordinates and dislocation categories.

[0059] For example, in this embodiment of the invention, the two-dimensional coordinate information (X, Y) of each detection point on the silicon carbide substrate is automatically recorded, and the coordinates are associated with the dislocation category result corresponding to that point determined by the dislocation analysis model. The coordinate-category pairs of all detection points together constitute a structured mapping dataset, which fully records the spatial location information of each dislocation event on the sample surface.

[0060] S1042. Based on the size and shape information of the silicon carbide substrate material, construct a digital base map of the silicon carbide substrate material.

[0061] For example, a digital undermap refers to a two-dimensional digital image or matrix generated in a computer that perfectly corresponds to the actual size and shape of the silicon carbide substrate material under test. The undermap is created based on input sample size information (such as diameter and thickness) and shape information (such as circular, square, and orientation of locating edges / notches). This digital undermap serves as a spatial canvas to accurately represent the spatial distribution of dislocations.

[0062] S1043. Based on the digital base map, the mapping dataset, and the preset color mapping rules, draw each pixel on the digital base map to obtain the dislocation spatial distribution map of the silicon carbide substrate material.

[0063] For example, this embodiment of the invention pre-defines a color mapping rule, assigning a unique and easily distinguishable color identifier to each dislocation type. For instance, screw dislocations are represented by red, basal dislocations by blue, edge dislocations by green, and dislocation-free areas retain the background color. Then, the mapping dataset is read. For each data point in the dataset, its corresponding pixel position is located on the digital base map based on its coordinates. According to its dislocation category and the pre-define color mapping rule, the pixel is rendered into the specified color. By processing all data points, a colored spatial distribution map of dislocations is finally generated, which visually displays the specific distribution locations of different dislocation types on the sample surface.

[0064] S105. Based on the dislocation spatial distribution map of silicon carbide substrate material, perform density calculation and statistics to determine the dislocation detection report of silicon carbide substrate material.

[0065] For example, in this embodiment of the invention, the number of various types of dislocations is automatically counted based on the spatial distribution map of dislocations; combined with the known area of ​​the scanning region, the density of various types of dislocations and the total dislocation density are calculated; finally, the sample information, detection conditions, quantitative statistical results (dislocation density) and dislocation spatial distribution map are integrated to output a complete detection report.

[0066] As one possible implementation, step S105 can be specifically implemented as steps S1051-S1055.

[0067] S1051. Based on the spatial distribution map of dislocations on silicon carbide substrate material, classify and count the number of pixels for each dislocation type.

[0068] For example, in this embodiment of the invention, image processing technology is used to classify and count the pixels in the generated dislocation spatial distribution map. According to a preset color mapping rule, the number of all colored pixels representing each category, such as screw dislocations, basal dislocations, and edge dislocations, is identified and counted. This statistical process excludes background color pixels representing defect-free areas, thereby obtaining the frequency of occurrence of each type of dislocation within the scanned area.

[0069] S1052. Calculate the area of ​​the effective scanning region based on the physical dimensions of the scanning region in the dislocation spatial distribution map.

[0070] For example, the area of ​​the effective scan region refers to the area of ​​the actual physical region represented by the dislocation spatial distribution map. This area is calculated based on the physical dimensions of the scan region, determined by the stage movement accuracy and scan step parameters of the confocal Raman / photoluminescence system. Specifically, the effective scan region area is calculated by multiplying the number of detection points along the length of the scan region by the scan step, and then multiplying this by the number of detection points along the width by the scan step. This area provides an accurate area reference for subsequent dislocation density calculations.

[0071] S1053. Calculate the density of each error type based on the number of pixels of each error type and the area of ​​the effective scan area.

[0072] For example, dislocation density refers to the number of a specific type of dislocation per unit area. In this embodiment of the invention, the number of pixels of a certain type of dislocation obtained from the aforementioned steps can be multiplied by a preset calibration coefficient. This calibration coefficient is used to correct the statistical relationship between the number of pixels and the actual number of dislocations. Then, this product is divided by the area of ​​the effective scanned region to obtain the areal density of that type of dislocation. This calculation process is performed separately for all identified dislocation types.

[0073] S1054. Calculate the total dislocation density of the silicon carbide substrate material based on the density of each dislocation type.

[0074] For example, the total dislocation density refers to the total number of all types of dislocations per unit area. In this embodiment of the invention, the calculated dislocation densities of each type are arithmetically summed, or the sum of the calibrated pixel counts of all types of dislocations is divided by the effective scan area; both methods yield the total dislocation density value. This parameter provides a macroscopic evaluation index of the overall crystal quality of the material.

[0075] S1055. Based on the basic information of the silicon carbide substrate material, the detection condition parameters of the full scan, the density of each dislocation type, the total dislocation density, and the spatial distribution map of dislocations, generate a dislocation detection report for the silicon carbide substrate material.

[0076] For example, a dislocation detection report is a comprehensive document of detection results. This embodiment of the invention automatically integrates basic information about the silicon carbide substrate material, comprehensive scan detection condition parameters, calculated dislocation densities of various types, total dislocation density, and dislocation spatial distribution map, generating a structured document according to a preset report template. This report also includes metadata such as detection time, sample identification, and detection environmental conditions, forming a complete evaluation record of the sample's dislocation state.

[0077] This invention provides a non-destructive testing method for dislocations based on confocal Raman / photoluminescence spectroscopy. The method eliminates the need for destructive treatments such as KOH etching during testing, ensuring material integrity and subsequent usability, reducing testing costs, and enabling full sample inspection. Through fully automated, full-surface spectral scanning and analysis, this invention completely avoids the randomness and statistical biases of traditional destructive sampling testing, accurately capturing local defects and macroscopic distribution patterns, significantly improving the comprehensiveness and accuracy of dislocation detection. Furthermore, it transforms abstract defect information into intuitive spatial distribution maps and quantitative statistical reports of dislocations, providing unprecedentedly accurate and objective data support for process traceability and quality assessment, overcoming the limitations of traditional methods that rely on human experience, are highly subjective, and cannot perform spatial analysis. This invention employs a confocal Raman / photoluminescence system to perform a comprehensive spectral scan of silicon carbide substrate materials. Based on spectral characteristics and a preset dislocation analysis model, it automatically determines dislocation information, generates a spatial distribution map of dislocations and a dislocation detection report, fundamentally solving the problem of low accuracy in destructive detection methods for dislocations in silicon carbide substrates and epitaxial materials. This enables non-destructive detection of dislocations in silicon carbide substrates and epitaxial materials, improving the accuracy of dislocation detection.

[0078] Optionally, the dislocation non-destructive detection method based on confocal Raman / photoluminescence spectroscopy provided in this embodiment of the invention further includes steps S201-S205 before step S103.

[0079] S201. Obtain the original spectra of each detection point of multiple silicon carbide substrate materials.

[0080] For example, embodiments of the present invention require preparing a batch of representative silicon carbide substrate materials as standard samples. Using a confocal Raman / photoluminescence system, comprehensive spectral data is acquired for each standard sample according to a preset scanning path and step. This ensures that each sample obtains a raw spectral data set covering its specific region and containing a large number of detection points; this data constitutes the original input information for model training.

[0081] S202. Based on the original spectra of each detection point of each silicon carbide substrate material, generate the spectral characteristics of each detection point of each silicon carbide substrate material.

[0082] For example, in this embodiment of the invention, the raw spectra collected at each detection point of each standard sample are processed. The peak position, full width at half maximum (FWHM), relative intensity, and integral intensity of the characteristic peaks in each spectrum are accurately extracted, thereby transforming the massive amount of raw spectral data into a structured spectral feature dataset.

[0083] S203. For multiple silicon carbide substrate materials, the fusion etching method is used to identify and mark the dislocation type of each spatial coordinate on each silicon carbide substrate material.

[0084] For example, in this embodiment of the invention, a destructive testing method using molten KOH etching is employed on standard samples for which spectral acquisition has been completed. Under specific temperature and time conditions, the KOH etching solution selectively etches the sample surface, resulting in corrosion pits with specific morphologies at the outcrops of different dislocation types. Subsequently, by observation using an optical microscope or scanning electron microscope, the dislocation type corresponding to each corrosion pit and its precise spatial coordinates on the sample surface are accurately identified and recorded based on the geometric characteristics of the corrosion pits.

[0085] S204. Using the spectral characteristics of each detection point on each silicon carbide substrate as input and the dislocation type of each spatial coordinate on each silicon carbide substrate as output, construct multiple training samples.

[0086] For example, in this embodiment of the invention, spectral features are precisely matched with dislocation type labels obtained from destructive detection. Specifically, spectral features extracted from the same standard sample at the same spatial coordinate detection point are used as input data, and the dislocation type determined by the KOH etching method at that coordinate is used as the output label. The two are then bound together to form a training sample. By collecting all valid detection point data from all standard samples, a training sample set containing a large number of input-output correspondences is constructed for model training.

[0087] S205. Based on multiple training samples, machine learning is performed to obtain a dislocation analysis model.

[0088] For example, in this embodiment of the invention, the constructed training sample set is input into a selected machine learning algorithm. This algorithm, through an iterative optimization process, automatically learns the complex mapping relationship between spectral features and dislocation types, and adjusts the model's internal parameters to minimize the error between the predicted result and the true label. Commonly used algorithms include support vector machines, random forests, or neural networks. After training, the model possesses the ability to predict the dislocation type of an unknown sample based on its spectral characteristics, thus forming a dislocation analysis model.

[0089] Thus, by establishing a standard sample library and integrating high-confidence KOH corrosion calibration results, this invention provides accurate training data for machine learning models, fundamentally ensuring the accuracy and reliability of dislocation analysis models. This enables the model to learn the complex mapping relationship between spectral features and dislocation types, thereby achieving accurate and automatic identification of dislocation types in unknown samples in practical applications, significantly improving the objectivity and efficiency of nondestructive testing.

[0090] Optionally, the dislocation non-destructive detection method based on confocal Raman / photoluminescence spectroscopy provided in this embodiment of the invention further includes steps S301-S305 after step S105.

[0091] S301. Align and overlay the dislocation space distribution map with the design layout of the target power device to obtain a fused image.

[0092] For example, in this embodiment of the invention, a design layout file in GDSII or OASIS format for the target power device is obtained, and the boundary coordinate information of the active region is extracted from it. Subsequently, the coordinate system of the dislocation spatial distribution map is registered with the coordinate system of the design layout. This process is accomplished by identifying common alignment marks between the two or by implementing coordinate transformation through a specific algorithm, ultimately generating a precisely superimposed fused image, ensuring that the dislocation distribution and the active region pattern of the device correspond completely in space.

[0093] S302. Based on the fused image, the logical AND operation in image processing is used to identify and count the number of pixels of each fault type in the active region of the target power device.

[0094] For example, in this embodiment of the invention, based on the fused image with coordinate alignment, a logical AND operation from image processing is used for defect screening. Specifically, the boundary pattern of the active region of the device is used as a mask, and all pixels in the dislocation spatial distribution map are traversed. When a pixel representing a dislocation falls within the boundary of the active region mask, the pixel is determined to be a valid defect point and recorded; otherwise, it is excluded. Subsequently, the pixels captured within the mask are automatically classified and counted according to the dislocation type they represent.

[0095] S303. Based on the pixels of each dislocation type in the active region, calculate the effective dislocation density in the active region of the target power device.

[0096] For example, in this embodiment of the invention, the number of pixels identified for each type of dislocation is multiplied by a preset calibration coefficient to convert it into the corresponding equivalent dislocation number. Then, the actual physical area of ​​the active region mask of the device on the sample is calculated. Finally, the equivalent number of each type of dislocation is divided by the actual physical area of ​​its active region to obtain the effective dislocation density of each type of dislocation within the active region. This density value eliminates interference from defects in the non-active region and accurately reflects the concentration of key defects affecting device performance.

[0097] S304. Based on the effective dislocation density in the active region of the target power device and the pre-established device performance testing model, the predicted values ​​of key electrical parameters of the fabricated device corresponding to the silicon carbide substrate material and the pass probability are obtained.

[0098] In some embodiments, the device performance testing model characterizes the quantitative relationship between the effective dislocation density in the active region and key electrical parameters. Key electrical parameters include breakdown voltage, leakage current, and on-resistance.

[0099] For example, the device performance testing model is a quantitative relationship model built based on historical manufacturing data through regression analysis or machine learning training. Its input is the effective dislocation density in the active region, and its output is the predicted values ​​of the device's key electrical parameters. By inputting the effective dislocation density in the active region calculated for a specific substrate into this model, the model outputs predicted values ​​for parameters such as breakdown voltage, leakage current, and on-resistance of the device fabricated on that substrate. Simultaneously, the model calculates the predicted pass / fail probability of the device based on a comparison of the predicted parameters with the device datasheet.

[0100] S305. Based on the effective dislocation density, predicted values ​​of key electrical parameters, and pass probability in the active region, generate a device performance prediction report for silicon carbide substrate materials.

[0101] For example, the device performance prediction report includes the associated target power device model, the calculated effective dislocation density in the active region, the predicted values ​​of various key electrical parameters, the predicted device pass probability, and a probability-based risk level assessment (e.g., high risk, medium risk, low risk). This report provides direct data-driven decision-making support for substrate material selection and yield forecasting before device manufacturing.

[0102] Thus, this invention achieves a crucial leap from material-level quality assessment to device-level performance prediction by intelligently linking material defect distribution with device design layout. It can accurately quantify the effective dislocation density, which directly impacts device performance, and predict key electrical parameters and yield rates based on this. This enables chip manufacturers to accurately assess substrate quality and predict risks before tape-out, allowing for early screening of high-risk materials, thereby significantly reducing manufacturing costs, optimizing production processes, and improving the yield and reliability of the final product.

[0103] Optionally, the dislocation non-destructive detection method based on confocal Raman / photoluminescence spectroscopy provided in this embodiment of the invention further includes steps S401-S403 after step S105.

[0104] S401. Perform spatial distribution statistical analysis on the dislocation spatial distribution map to identify the defect clustering pattern.

[0105] In some embodiments, the defect clustering pattern includes a ring-shaped distribution, a radial stripe distribution, a localized clustering, or a dense edge distribution.

[0106] For example, embodiments of the present invention analyze the spatial distribution characteristics of dislocation spatial distribution maps using image processing algorithms. Density clustering algorithms are employed to identify the spatial clustering patterns of dislocation points, Fourier transform is used to analyze periodic distribution characteristics, and edge detection algorithms are combined to analyze the distribution characteristics of defects at the sample edges. Based on a preset pattern feature library, typical defect clustering patterns such as ring-shaped distributions, radial stripe distributions, localized clustering, or edge-dense distributions are automatically identified and classified.

[0107] S402. Based on the defect clustering pattern and the preset defect pattern-process root cause knowledge base, perform matching to query the root cause process parameters that lead to the defect clustering pattern, as well as the direction of process parameter adjustment.

[0108] In some embodiments, the root process parameters include the crystal growth temperature field gradient, crucible rotation speed and pulling speed, substrate cutting angle or etching conditions before epitaxial growth; the defect mode-process root knowledge base characterizes the mapping relationship between different defect aggregation modes and process parameters.

[0109] For example, the defect pattern-process root cause knowledge base is an expert system built upon a large amount of historical production data, storing the mapping relationship between specific defect patterns and process parameters. Identified defect cluster patterns are matched with feature patterns in the knowledge base based on similarity, and a preset matching algorithm identifies the most relevant process root causes. The knowledge base includes the correspondence between key process parameters such as crystal growth temperature gradient, crucible rotation speed and pulling speed, substrate cutting angle, and etching conditions before epitaxial growth and defect patterns.

[0110] For example, embodiments of the present invention can generate specific parameter adjustment suggestions based on process optimization experience rules stored in the knowledge base, including: for annular distribution defects, it is recommended to adjust the temperature field gradient distribution; for radial stripe distribution, it is recommended to optimize the matching relationship between crystal rotation rate and pulling speed; for local cluster aggregation, it is recommended to strengthen the substrate surface treatment process; for edge-dense distribution, it is recommended to optimize the etching conditions before epitaxial growth.

[0111] S403. Based on the defect clustering pattern, root cause process parameters, and the direction of process parameter adjustment, generate a process optimization suggestion report.

[0112] Thus, by establishing an intelligent correlation between defect distribution patterns and process parameters, this invention achieves closed-loop control from defect detection to process optimization. It can quickly and accurately pinpoint the root causes of problems in the production process and provide targeted adjustments to process parameters. This significantly shortens the process debugging cycle, improves the accuracy of problem solving, and realizes a shift from a passive detection to a proactive prevention-based quality control model, providing effective technical support for continuously improving product quality and process stability.

[0113] Optionally, the dislocation non-destructive testing method based on confocal Raman / photoluminescence spectroscopy provided in this embodiment of the invention further includes steps S501-S507.

[0114] S501. Periodically extract multiple samples from the samples identified as high-risk according to a preset ratio, perform destructive testing on the multiple samples using the melt corrosion method, and calibrate the dislocation type and spatial location of each sample.

[0115] In some embodiments, high-risk samples are silicon carbide substrate materials with dislocation density greater than a preset threshold or abnormal defect aggregation modes.

[0116] For example, abnormal defect aggregation patterns refer to specific spatial distribution patterns of defects that deviate significantly from normal random or expected distribution states, as identified through spatial distribution statistical analysis. For instance, macroscopic uneven distribution refers to the non-random, spatially concentrated nature of defects on the wafer surface, exhibiting specific patterns. This includes: Ring-shaped distribution: Defect density fluctuates in a ring-like ripple pattern along the wafer radius, usually related to thermal asymmetry, unstable melt convection, or mismatched rotation rates during crystal growth. Radial stripe distribution: Defects are distributed radially or fan-shaped along specific crystal orientations, originating from facet effects, thermal stress concentration, or improper substrate cutting angles during crystal growth. Edge-dense distribution: Defect density is significantly higher at the wafer edges than in the center, commonly seen in epitaxial growth due to deteriorated growth conditions caused by edge airflow and temperature boundary effects. Another example is localized micro-aggregation, which refers to anomalous regions with small scale but extremely high defect density against a macroscopically uniform background. This includes: Localized cluster-like aggregation: Defects appear tightly clustered in localized areas, resembling clusters or clumps. This typically points to localized quality issues within the substrate itself, such as: polymorphic inclusions: tiny regions of other crystal types, such as 3C-SiC, mixed into the 4H-SiC lattice; carbon inclusions or other second-phase particles; and surface damage or particulate contamination introduced during polishing or cleaning, which become nucleation sites for defects during epitaxial growth.

[0117] S502. For multiple samples, a confocal Raman / photoluminescence system is used to perform a comprehensive scan to obtain the original spectrum of each detection point for each sample.

[0118] S503. Based on the original spectra of multiple samples, feature extraction is performed to obtain the spectral features of each detection point of each sample.

[0119] For example, in this embodiment of the invention, samples with dislocation densities exceeding a set threshold or exhibiting specific abnormal defect aggregation patterns are automatically screened according to preset judgment rules and marked as high-risk samples. Subsequently, representative samples are randomly selected from the high-risk samples according to a preset sampling ratio. These samples are then subjected to destructive verification using the molten KOH etching method to accurately determine their dislocation types and spatial coordinates, establishing a high-confidence verification dataset.

[0120] S504. Based on the spectral characteristics of each detection point of each sample, as well as the dislocation type and spatial location of each sample, generate new training samples.

[0121] For example, samples that have already undergone KOH validation are re-scanned using a confocal Raman / photoluminescence system to obtain the original spectra. These new spectral data are then processed using the same feature extraction method as in the initial training phase to obtain the corresponding spectral features. These spectral features are then rigorously matched with the precise dislocation type and location information obtained from KOH validation to form new training samples.

[0122] S505. Based on the newly added training samples, amplify multiple training samples to obtain an amplified sample set.

[0123] S506. Based on the expanded sample set, the dislocation analysis model is incrementally trained to obtain the updated dislocation analysis model.

[0124] For example, in this embodiment of the invention, newly added training samples are merged with existing training samples to form an expanded sample set. An incremental learning algorithm is used to optimize and adjust the model parameters based on the original dislocation analysis model using the expanded sample set. This process maintains the model's original knowledge while focusing on learning feature patterns in the newly added samples, especially those defect features that were previously misjudged or difficult to identify.

[0125] S507. Based on the updated dislocation analysis model, perform non-destructive testing of dislocations on silicon carbide substrate materials.

[0126] For example, after incremental training is completed, the performance metrics of the updated model are evaluated using cross-validation. Once the model's classification accuracy and other key metrics on the test set reach the preset improvement standards, the new model version is deployed to the online detection system to replace the old model version for subsequent non-destructive testing of dislocations on all silicon carbide substrate materials.

[0127] For example, this embodiment of the invention achieves self-evolution of the detection system by establishing a continuous learning model optimization mechanism. It can continuously absorb new detection experience, correct misjudgment patterns, and adapt to new defect characteristics brought about by process changes, thereby continuously improving the accuracy and reliability of dislocation classification. This self-evolutionary capability ensures that the detection system can maintain optimal performance over a long period, effectively addressing the challenges brought about by improvements in production processes and material changes, and providing stable quality inspection assurance for industrial applications.

[0128] Optionally, the dislocation non-destructive testing method based on confocal Raman / photoluminescence spectroscopy provided in this embodiment of the invention further includes steps S601-S607.

[0129] S601. Based on the original spectra of each detection point of the silicon carbide substrate material, extract the peak position and full width at half maximum (FWHM) of the characteristic phonon modes of the silicon carbide substrate material.

[0130] In some embodiments, characteristic phonon modes originate from Raman scattering and are a direct reflection of the interaction between intrinsic lattice vibrations (phonons) and laser light; they characterize the integrity, symmetry, and stress of the material's lattice itself; they are signals of intrinsic or perfect lattice states; they serve as a precise scale for measuring lattice stress / strain; specifically, they refer to the inherent, known-position phonon peaks in 4H-SiC, most typically located at ~777 cm⁻¹. -1 The E2 (lateral optical) mode.

[0131] Characteristic peaks originate from photoluminescence; they are generated by electronic transitions at defect energy levels (such as dislocations and point defects) in crystals; they characterize the type, concentration, and state of defects in materials; they are signals of abnormality or imperfection; and they serve as direct fingerprints for identifying and classifying defects such as dislocations.

[0132] S602. Based on the peak position and full width at half maximum (FWHM) of the characteristic phonon modes of the silicon carbide substrate material, calculate the peak position offset of the characteristic phonon mode peak position relative to the stress-free standard reference value at each detection point.

[0133] For example, the characteristic phonon modes of 4H-SiC material were precisely located from the raw spectrum at each detection point, particularly those located at approximately 777 cm⁻¹. -1 The E2 transverse optical mode was determined. The peak position and full width at half maximum (FWHM) were accurately extracted by fitting the characteristic peak using a Gaussian-Lorentz mixture function. The offset of the characteristic phonon mode peak position from the stress-free standard reference value at each detection point was calculated; this offset directly reflects the change in lattice constant.

[0134] S603. Based on the preset Raman peak position offset-stress calibration curve and the peak position offset of each detection point, determine the local stress value of each detection point.

[0135] In some embodiments, the Raman peak offset-stress calibration curve is obtained by calibrating a standard sample using X-ray diffraction.

[0136] S604. Based on the spatial coordinates and local stress values ​​of each detection point, generate a two-dimensional stress distribution mapping map.

[0137] S605. Overlay the dislocation spatial distribution map and the two-dimensional stress distribution mapping map to generate a defect-stress integrated information map.

[0138] For example, in this embodiment of the invention, a Raman peak offset-stress calibration curve pre-calibrated by X-ray diffraction is used to convert the peak offset of each detection point into a quantitative stress value. Based on the spatial coordinates of all detection points and the corresponding stress values, a continuous two-dimensional stress distribution map is generated using a Kriging interpolation algorithm. This map displays the stress magnitude and distribution in pseudo-color.

[0139] S606. Based on the defect-stress integrated information map, perform image processing and statistics to determine the spatial coupling relationship between high dislocation density regions and high stress regions, as well as the spatial correlation coefficient between high dislocation density regions and high stress regions.

[0140] For example, in this embodiment of the invention, the stress distribution map and the dislocation spatial distribution map are spatially superimposed, and coordinate consistency is ensured through image registration technology. A spatial statistical analysis algorithm is used to calculate the spatial correlation coefficient between high dislocation density regions and high stress regions. A region growing algorithm is used to identify stress concentration regions and analyze their spatial correspondence with dislocation accumulation regions.

[0141] S607. Based on the spatial coupling relationship and spatial correlation coefficient between high dislocation density regions and high stress regions, assess the degree of dislocation-induced lattice strain and generate a long-term reliability risk level report for silicon carbide substrate materials.

[0142] In some embodiments, the long-term reliability risk level report includes the spatial coupling relationship and spatial correlation coefficient between high dislocation density regions and high stress regions, as well as the degree of dislocation-induced lattice strain.

[0143] For example, this embodiment of the invention establishes a quantitative relationship model between dislocation density, stress value, and device reliability based on the analysis results of spatial coupling. According to the degree of stress concentration and dislocation density, the risk level is divided into three levels: low, medium, and high. Combining historical failure data, the device lifetime and failure probability under different operating conditions are predicted.

[0144] Thus, this invention, by simultaneously achieving non-destructive mapping between dislocations and stress, establishes for the first time a direct correlation between defects and lattice strain at the material level; it can proactively identify material regions with high reliability risks, providing crucial data support for power device lifetime prediction and reliability design. This comprehensive analysis method not only assesses the current quality state of materials but also predicts the performance evolution of devices under long-term operating stress, providing an innovative solution for reliability-based material selection and device design.

[0145] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0146] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0147] Figure 2 This diagram illustrates a structural schematic of a dislocation nondestructive testing device based on confocal Raman / photoluminescence spectroscopy provided by an embodiment of the present invention. The nondestructive testing device 700 includes a data acquisition module 701 and a data processing module 702.

[0148] The data acquisition module 701 is used to perform a comprehensive scan of the silicon carbide substrate material to be tested using a confocal Raman / photoluminescence system to obtain the original spectrum of each detection point of the silicon carbide substrate material.

[0149] The data processing module 702 is used to extract features from the original spectra of each detection point to determine the spectral characteristics of each detection point; based on the spectral characteristics of each detection point and the preset dislocation analysis model, to determine the dislocation analysis results of each detection point, including whether a dislocation has occurred and the type of dislocation; based on the dislocation analysis results of each detection point, to generate a dislocation spatial distribution map of the silicon carbide substrate material; based on the dislocation spatial distribution map of the silicon carbide substrate material, to perform density calculation and statistics, and to determine the dislocation detection report of the silicon carbide substrate material.

[0150] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The electronic device 800 includes: a processor 801, a memory 802, and a computer program 803 stored in the memory 802 and executable on the processor 801. When the processor 801 executes the computer program 803, it implements the steps in the above-described method embodiments. Alternatively, when the processor 801 executes the computer program 803, it implements the functions of each module / unit in the above-described device embodiments.

[0151] For example, the computer program 803 may be divided into one or more modules / units, which are stored in the memory 802 and executed by the processor 801 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 803 in the electronic device 800.

[0152] The processor 801 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0153] The memory 802 can be an internal storage unit of the electronic device 800, such as a hard disk or memory of the electronic device 800. The memory 802 can also be an external storage device of the electronic device 800, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card (FC) equipped on the electronic device 800. Furthermore, the memory 802 can include both internal and external storage units of the electronic device 800. The memory 802 is used to store the computer program and other programs and data required by the terminal. The memory 802 can also be used to temporarily store data that has been output or will be output.

[0154] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A non-destructive testing method for dislocations based on confocal Raman and photoluminescence spectroscopy, characterized in that, include: The silicon carbide substrate material to be tested was scanned comprehensively using a confocal Raman and photoluminescence system to obtain the original spectra of each detection point of the silicon carbide substrate material; Feature extraction is performed on the original spectra of each detection point to determine the spectral characteristics of each detection point; Based on the spectral characteristics of each detection point and the preset dislocation analysis model, the dislocation analysis results of each detection point are determined. The dislocation analysis results include whether a dislocation has occurred and the type of dislocation. Based on the dislocation analysis results at each detection point, a dislocation spatial distribution map of the silicon carbide substrate material is generated. Based on the dislocation spatial distribution map of silicon carbide substrate material, density calculation and statistics are performed to determine the dislocation detection report of silicon carbide substrate material; The spectral features include the peak position of the characteristic peak, the full width at half maximum (FWHM) of the characteristic peak, the relative intensity of the characteristic peak, and the integrated intensity over a specific energy range. The step of extracting features from the original spectra of each detection point to determine the spectral characteristics of each detection point includes: performing continuous first-order differential calculations on the original spectra to identify multiple candidate feature peaks; fitting the multiple candidate feature peaks with a Gaussian-Lorentz mixture function to determine the energy value corresponding to the extreme point of the function as the peak position of the feature peak; determining the full width at half maximum (FWHM) of the feature peak based on the peak position of the feature peak using linear interpolation; selecting a reference feature peak based on the original spectrum; calculating the integral intensity of the reference feature peak and the integral intensity of the feature peak to be analyzed; and determining the relative intensity of the feature peak based on the integral intensity of the reference feature peak and the integral intensity of the feature peak to be analyzed; performing integration based on the original spectrum and a preset integration interval to calculate the area between the original spectrum and the energy axis within the integration interval, and determining it as the integral intensity of a specific energy interval; the integration interval is one or more broadened emission bands related to the dislocation type.

2. The non-destructive testing method for dislocations based on confocal Raman and photoluminescence spectroscopy according to claim 1, characterized in that, The determination of dislocation analysis results at each detection point based on the spectral characteristics of each detection point and a preset dislocation analysis model includes: Based on the spectral characteristics of each detection point, the feature vector of each detection point is determined; Based on the feature vectors of each detection point and the dislocation analysis model, the probability of a dislocation occurring at each detection point and the probability of each detection point belonging to each dislocation category are output. The dislocation analysis results are determined based on the probability of a dislocation occurring at each detection point and the probability that each detection point belongs to each dislocation category.

3. The non-destructive testing method for dislocations based on confocal Raman and photoluminescence spectroscopy according to claim 1, characterized in that, Before determining the dislocation analysis results for each detection point based on the spectral characteristics of each detection point and a preset dislocation analysis model, the process also includes: Obtain the original spectra at various detection points on multiple silicon carbide substrate materials; Based on the original spectra of each detection point of each silicon carbide substrate material, the spectral characteristics of each detection point of each silicon carbide substrate material are generated; The multiple silicon carbide substrate materials are subjected to melt etching method to identify and mark the dislocation type of each spatial coordinate on each silicon carbide substrate material; Multiple training samples are constructed by taking the spectral characteristics of each detection point of each silicon carbide substrate material as input and the dislocation type of each spatial coordinate on each silicon carbide substrate material as output. Based on the multiple training samples, machine learning is performed to obtain the dislocation analysis model.

4. The non-destructive testing method for dislocations based on confocal Raman and photoluminescence spectroscopy according to claim 1, characterized in that, The dislocation spatial distribution map of the silicon carbide substrate material is generated based on the dislocation analysis results at each detection point, including: Correlate the spatial coordinates of each detection point with the dislocation analysis results to construct a mapping dataset between spatial coordinates and dislocation categories; A digital base map of the silicon carbide substrate material is constructed based on its size and shape information. Based on the digital base map, the mapping dataset, and the preset color mapping rules, each pixel is drawn on the digital base map to obtain the dislocation spatial distribution map of the silicon carbide substrate material.

5. The non-destructive testing method for dislocations based on confocal Raman and photoluminescence spectroscopy according to claim 1, characterized in that, The dislocation spatial distribution map based on the silicon carbide substrate material is used to perform density calculations and statistics to determine the dislocation detection report of the silicon carbide substrate material, including: Based on the spatial distribution map of dislocations on silicon carbide substrates, the number of pixels for each dislocation type is classified and counted. Calculate the area of ​​the effective scanning region based on the physical dimensions of the scanning region in the dislocation space distribution map; The density of each error type is calculated based on the number of pixels of each error type and the area of ​​the effective scan region; Calculate the total dislocation density of the silicon carbide substrate material based on the density of each dislocation type; Based on the basic information of the silicon carbide substrate material, the detection condition parameters of the full scan, the density of each dislocation type, the total dislocation density, and the spatial distribution map of the dislocation, a dislocation detection report of the silicon carbide substrate material is generated.

6. The non-destructive testing method for dislocations based on confocal Raman and photoluminescence spectroscopy according to claim 1, characterized in that, After calculating and statistically analyzing the dislocation spatial distribution map based on the silicon carbide substrate material to determine the dislocation detection report for the silicon carbide substrate material, the following steps are also included: The dislocation space distribution map is aligned and superimposed with the design layout of the target power device to obtain a fused image; Based on the fused image, the logical AND operation in image processing is used to identify and count the number of each fault type pixel in the active region of the target power device. The effective dislocation density in the active region of the target power device is calculated based on the pixels of each dislocation type in the active region. Based on the effective dislocation density in the active region of the target power device and the pre-established device performance testing model, the predicted values ​​of key electrical parameters of the device fabricated with silicon carbide substrate material and the pass probability are obtained; the device performance testing model characterizes the quantitative relationship between the effective dislocation density in the active region and the key electrical parameters; the key electrical parameters include breakdown voltage, leakage current and on-resistance. Based on the effective dislocation density, predicted values ​​of key electrical parameters, and pass probability within the active region, a device performance prediction report for silicon carbide substrate material is generated.

7. The non-destructive testing method for dislocations based on confocal Raman and photoluminescence spectroscopy according to claim 1, characterized in that, After calculating and statistically analyzing the dislocation spatial distribution map based on the silicon carbide substrate material to determine the dislocation detection report for the silicon carbide substrate material, the following steps are also included: Spatial distribution statistical analysis is performed on the dislocation spatial distribution map to identify defect clustering patterns; the defect clustering patterns include ring-shaped distribution, radial stripe distribution, local clustering, or edge-dense distribution. Based on the aforementioned defect aggregation pattern and a preset defect pattern-process root cause knowledge base, a matching process is performed to query the root process parameters that cause the defect aggregation pattern, as well as the direction of process parameter adjustment. The root process parameters include the crystal growth temperature field gradient, crucible rotation speed and pulling speed, substrate cutting angle, or etching conditions before epitaxial growth. The defect pattern-process root cause knowledge base represents the mapping relationship between different defect aggregation patterns and process parameters. Based on the defect clustering pattern, the root cause process parameters, and the direction of process parameter adjustment, a process optimization suggestion report is generated.

8. The non-destructive testing method for dislocations based on confocal Raman and photoluminescence spectroscopy according to claim 1, characterized in that, The method further includes: Periodically, multiple samples are extracted from samples identified as high-risk according to a preset ratio, and destructive testing is performed on these multiple samples using the melt etching method to calibrate the dislocation type and spatial location of each sample; the high-risk samples are silicon carbide substrate materials with dislocation density greater than a preset threshold or abnormal defect aggregation mode. For the multiple samples, a comprehensive scan was performed using a confocal Raman and photoluminescence system to obtain the original spectra of each sample at each detection point; Based on the original spectra of multiple samples, feature extraction is performed to obtain the spectral features of each detection point of each sample; New training samples are generated based on the spectral characteristics of each detection point of each sample, as well as the dislocation type and spatial location of each sample. Based on the newly added training samples, multiple training samples are amplified to obtain an amplified sample set; Based on the amplified sample set, the dislocation analysis model is incrementally trained to obtain an updated dislocation analysis model. Based on the updated dislocation analysis model, non-destructive testing of dislocations in silicon carbide substrate materials is performed.

9. The non-destructive testing method for dislocations based on confocal Raman and photoluminescence spectroscopy according to claim 1, characterized in that, The method further includes: Based on the original spectra of each detection point of the silicon carbide substrate material, the peak positions and full width at half maximum (FWHM) of the characteristic phonon modes of the silicon carbide substrate material are extracted. Based on the peak position and full width at half maximum (FWHM) of the characteristic phonon modes of silicon carbide substrate material, the peak position offset of the characteristic phonon mode at each detection point relative to the stress-free standard reference value is calculated. Based on the preset Raman peak position offset-stress calibration curve and the peak position offset of each detection point, the local stress value of each detection point is determined; the Raman peak position offset-stress calibration curve is obtained by calibrating the standard sample by X-ray diffraction. A two-dimensional stress distribution mapping map is generated based on the spatial coordinates and local stress values ​​of each detection point; The dislocation spatial distribution map and the two-dimensional stress distribution mapping map are superimposed to generate a defect-stress integrated information map; Based on the defect-stress integrated information map, image processing and statistics are performed to determine the spatial coupling relationship between the high dislocation density region and the high stress region, as well as the spatial correlation coefficient between the high dislocation density region and the high stress region. Based on the spatial coupling relationship and spatial correlation coefficient between high dislocation density regions and high stress regions, the degree of dislocation-induced lattice strain is assessed, and a long-term reliability risk level report for silicon carbide substrate materials is generated. The long-term reliability risk level report includes the spatial coupling relationship and spatial correlation coefficient between high dislocation density regions and high stress regions, as well as the degree of dislocation-induced lattice strain.

Images