Semitransparent SiC sheet defect detection system and method based on multi-light-source cooperation

Abstract

The invention discloses a semitransparent SiC wafer defect detection system and method based on multi-light-source cooperation, and relates to the technical field of product defect analysis. Light sources are divided into a template light source and a non-template light source, SiC wafer features detected by the light sources are called, a template light source database is established, the limiting conditions of the template light sources for available light sources are analyzed based on the template light source database, and the defect detection result of the semitransparent SiC wafer is obtained. The method comprises the following steps: establishing a limiting characteristic range database according to a limiting condition, when receiving a defect detection task for a current batch of SiC chips, judging a matching relationship between a template light source and the current batch of SiC chips, obtaining a to-be-selected light source of the current batch of SiC chips according to the matching relationship, selecting a detection light source by considering a historical defect misjudgment rate, and when the to-be-selected light source of the current batch of SiC chips does not exist, selecting the detection light source according to the historical defect misjudgment rate. According to the method, the optimal light source is dynamically selected based on the actual detection effect, and a reliable basis is provided for subsequent quality control and processing.

Description

Technical Field

[0001] This invention relates to the field of product defect analysis technology, specifically to a defect detection system and method for semi-transparent SiC wafers based on multi-light source collaboration. Background Technology

[0002] Translucent SiC wafers, as key functional materials, are widely used in various precision electronics and industrial applications. Their surface and internal defects directly affect product performance and lifespan, making defect detection an indispensable and crucial step in the production process. With the increasing demands for product quality from related industries, higher standards are being set for the accuracy and reliability of SiC wafer defect detection. Detection methods must be able to accurately identify various potential defects to provide effective support for quality control. However, the material properties of translucent SiC wafers make defect detection susceptible to the influence of light sources. Different SiC wafers have different requirements for detection light sources, making the selection of a suitable light source a core challenge for improving detection effectiveness. Traditional SiC wafer defect detection often relies on a single fixed light source or manual experience in selecting the light source, lacking a systematic mechanism for screening the effectiveness of light sources. These methods do not fully consider the characteristic differences between different batches of SiC wafers, making it difficult to match the optimal detection light source for a specific batch, often resulting in mismatches between the light source and the material being tested. While some testing solutions attempt to use multiple light sources, they lack a scientific system for evaluating and matching these sources. This makes it difficult to accurately determine the effectiveness of light sources for detecting SiC wafers with different characteristics, leading to issues such as misjudgments and missed detections during the testing process. This affects the reliability of the test results and increases the operational complexity and uncertainty of the testing process. Existing testing technologies lack a systematic support system based on historical data, lacking long-term accumulation and analysis of light source testing effects. It is impossible to screen effective light sources and establish adaptation standards based on historical testing data. The lack of correlation analysis between SiC wafer characteristics and light source characteristics during the testing process makes it difficult to form standardized feature matching rules, resulting in insufficient adaptability and scalability of the testing methods. When dealing with batches of SiC wafers with different characteristics, frequent adjustments to testing parameters or reliance on manual re-selection of light sources are required. This not only reduces testing efficiency but also makes it difficult to guarantee the stability of testing quality, failing to meet the needs of efficient and accurate defect detection in large-scale production. Summary of the Invention

[0003] The purpose of this invention is to provide a defect detection system and method for semi-transparent SiC wafers based on multi-light source collaboration, so as to solve the problems mentioned in the background art.

[0004] To address the aforementioned technical problems, this invention provides the following technical solution: a defect detection method for semi-transparent SiC wafers based on multi-source collaborative methods, comprising the following steps: S1. Based on historical data, light sources are divided into template light sources and non-template light sources; S2. After preprocessing and normalizing the light source data, the light source characteristics are obtained. The SiC wafer characteristics detected by the light source are called to establish a template light source database. S3. Analyze the constraints of template light sources on available light sources based on the template light source database, and establish a database of constraint feature ranges based on the constraints. S4. When receiving the defect detection task of the current batch of SiC wafers, determine the matching relationship between the template light source and the current batch of SiC wafers; S5. When there are candidate light sources for the current batch of SiC wafers, obtain the candidate light sources for the current batch of SiC wafers according to the matching relationship, and select the detection light source considering the historical defect misjudgment rate. S6. If there is no candidate light source for the current batch of SiC wafers, select the detection light source by considering the matching degree between the characteristics of the current batch of SiC wafers and the template light source.

[0005] Furthermore, in step S1, in historical data, when using a light source to detect defects in a batch of SiC wafers, the number of defects α detected by the light source is recorded. The actual number of defects β of the SiC wafers is obtained through microscopy, thus obtaining the defect rate α / β when using the light source to detect defects in this batch of SiC wafers. If α / β > γ, the light source is considered effective for detecting defects in this batch of SiC wafers; otherwise, the light source is considered ineffective. In historical data, if a light source is effective for detecting defects in any batch of SiC wafers, the light source is considered a template light source; otherwise, the light source is considered a non-template light source. By deeply mining historical detection data and using the actual defect situation as the core judgment criterion, template light sources and non-template light sources are accurately classified. This design eliminates consistently ineffective light sources from the source, avoiding interference with subsequent detection processes and ensuring the targetedness and effectiveness of light source selection. The judgment process abandons subjective experience dependence and uses actual detection results as the measurement standard, ensuring that the selected template light sources have practical defect detection capabilities, providing high-quality data support for subsequent feature-matching-based selection of suitable light sources. It strictly controls the quality of the light source from the initial stage, which significantly improves the rigor and scientific nature of the overall defect detection process and lays a solid foundation for optimizing the accuracy of subsequent detection.

[0006] Furthermore, in step S2, light source data is collected for M template light sources, where M is the total number of template light sources. The light source data is preprocessed and normalized to obtain light source features. The light source features of the m-th template light source are {A}. 1_m A 2_m ,…,A n_m ,…,A N_m}, where N is the number of light source features for any template light source, A n_mGiven the nth light source feature of the mth template light source, the system then retrieves the valid batch X of SiC wafers from the mth template light source in historical data. These batches of SiC wafers are considered usable batches for the mth template light source. The SiC wafer feature of the xth batch of SiC wafers is {B}. 1_x B 2_x ,…,B y_x ,…,B Y_x}, where Y represents the total number of SiC wafer features in any batch of SiC wafers, B y_x This represents the y-th SiC wafer feature of the x-th batch of SiC wafers. A template light source database is established based on template light sources and the SiC wafer features of available batches of SiC wafers. By systematically analyzing the feature correlation between template light sources and corresponding valid batches of SiC wafers, core data support is built for the testing process. Preprocessing and normalization of the light source data eliminates redundant information interference, allowing the light source features to more accurately reflect their testing attributes. Integrating the template light sources with the corresponding SiC wafer features clearly establishes the adaptation correlation between the two, avoiding mismatches between light sources and material properties. The constructed template light source database breaks down the information silos between light sources and testing objects, providing a structured data foundation for rapid matching of suitable light sources, reducing the cost of repetitive data collection and analysis, and shifting light source selection from fragmented experience to data-driven approaches, thus improving the standardization and efficiency of the testing process.

[0007] Furthermore, in step S3, based on the template light source database analysis, the constraints of the m-th template light source on the available light sources are analyzed. For the m-th template light source, the SiC wafer characteristics of batch X SiC wafers are called, and the y-th SiC wafer characteristic of batch X SiC wafers is {B y_1 B y_2 ,…,B y_x ,…,B y_X Then, the influence C of the y-th SiC wafer feature on the m-th template light source is calculated. y_m : ; Where b is the average value of the y-th SiC wafer characteristic in batch X, when C y_m >C y_0 If the condition is met, determine if the y-th SiC wafer feature affects the m-th template light source; otherwise, determine if the y-th SiC wafer feature has no effect on the m-th template light source, where C y_0 To determine the threshold value for the influence of the y-th SiC wafer feature on the m-th template light source, we obtain the limiting feature range of the m-th template light source on the y-th SiC wafer feature. If it is determined that the y-th SiC wafer feature has an influence on the m-th template light source, we use the IQR method to evaluate the y-th SiC wafer feature {B} of batch X SiC wafers. y_1 B y_2,…,B y_x ,…,B y_X Data cleaning is performed to obtain a cleaned feature set. The minimum value of the cleaned feature set is used as the lower limit of the feature range, and the maximum value of the cleaned feature set is used as the upper limit of the feature range. If it is determined that the y-th SiC wafer feature has no effect on the m-th template light source, the feature range is set to a set of real numbers. Substituting y = 1, 2, ..., Y, the constraint feature range {D} of the m-th template light source on the Y features of the SiC wafer. 1_m D 2_m ,…,D y_m ,…,D Y_m}, where D y_m To define the limiting feature range of the m-th template light source for the y-th SiC wafer feature, we substitute m=1,2,…,M one by one to obtain the limiting feature ranges of the M template light sources for the Y SiC wafer features, thus establishing a limiting feature range database. By deeply exploring the intrinsic relationship between template light sources and SiC wafer features, we accurately defined the key features affecting the light source detection effect, eliminated the interference of irrelevant features, and made the light source adaptation standard more targeted. The data cleaning process effectively removed abnormal information, ensuring that the limiting feature ranges conform to actual detection needs, and improving the scientificity and reliability of the standard. The constructed limiting feature range database provides a clear basis for the rapid matching of the current batch of SiC wafers with template light sources, transforming ambiguous adaptation judgments into clear range comparisons, reducing errors caused by subjective experience, further consolidating the data-driven foundation of the detection process, and improving the accuracy and efficiency of light source selection.

[0008] Furthermore, in step S4, when a defect detection task for the current batch of SiC wafers is received, feature acquisition is performed on the current batch of SiC wafers to obtain the SiC wafer features {B1, B2, ..., B} of the current batch of SiC wafers. y ,…,B Y}, where B y Let B represent the y-th SiC wafer feature of the current batch of SiC wafers. The method for determining the matching relationship between the m-th template light source and the current batch of SiC wafers is as follows: If the y-th SiC wafer feature B of the current batch of SiC wafers... y The limiting feature range D of the m-th template light source on the y-th SiC wafer y_mWithin the current batch of SiC wafers, the system determines whether the y-th SiC wafer feature matches the m-th template light source; otherwise, it determines that the y-th SiC wafer feature does not match the m-th template light source. If any SiC wafer feature in the current batch does not match the m-th template light source, then the m-th template light source is determined to be incompatible with the current batch of SiC wafers; otherwise, the m-th template light source is determined to match the current batch of SiC wafers, and the m-th template light source becomes a candidate light source for the current batch of SiC wafers. This standardized feature acquisition and range comparison logic achieves precise matching and screening between the current batch of SiC wafers and template light sources. By acquiring complete features of the SiC wafers to be tested and combining them with an established database of limited feature ranges, clear comparison rules are used to determine light source suitability, avoiding bias caused by subjective judgment. The rigorous design of determining a match only if all features match effectively eliminates incompatible light sources with mismatched features, ensuring that the candidate light sources have reliable testing potential. This step transforms the complex light source adaptation judgment into a clear feature range verification, simplifying the screening process, quickly identifying candidate light sources that meet the requirements, and providing a high-quality candidate pool for further selection of the optimal detection light source, thereby improving the overall targeting and efficiency of the detection process.

[0009] Furthermore, in step S5, each of the m=1,2,…,M template light sources is substituted to determine the matching relationship between the M template light sources and the current batch of SiC wafers, resulting in Q candidate light sources for the current batch of SiC wafers. If Q is not equal to 0, it is determined that there are candidate light sources for the current batch of SiC wafers. The historical defect misjudgment rate of the Q candidate light sources is then called. The historical defect misjudgment rate is the ratio of the number of defects misjudged by the candidate light source in historical data to the actual number of defects obtained from the SiC wafer through microscopy. The candidate light source with the lowest historical defect misjudgment rate is then selected as the detection light source for the current batch of SiC wafers. In scenarios where candidate light sources exist, the optimal detection light source is locked through scientific optimization logic, further improving the reliability of defect detection. This step uses historical defect misjudgment performance as the core screening criterion, calling the past detection data of the candidate light sources to intuitively reflect the actual detection quality of each light source, avoiding the detection risks caused by blind selection. Selecting the light source with the lowest historical misjudgment rate as the final detection light source fundamentally reduces the probability of misjudgment in the current batch detection, making the detection results more consistent with the real situation. Its quantitative screening method based on historical data eliminates the uncertainty of subjective experience judgment, making the selection of light sources more objective and convincing. At the same time, it quickly focuses on the optimal option, shortens the decision-making process, and ensures both accuracy and efficiency, providing key support for completing defect detection tasks with high quality.

[0010] Furthermore, in step S6, if Q equals 0, it is determined that there is no candidate light source for the current batch of SiC wafers, and the detection light source for the current batch of SiC wafers is analyzed. The analysis method for the detection light source is as follows: for the m-th template light source, if the y-th SiC wafer feature B of the current batch of SiC wafers... y The limiting feature range D of the m-th template light source on the y-th SiC wafer y_m Within this process, the number of matching features for the m-th template light source is incremented by 1; otherwise, the number of matching features for the m-th template light source is incremented by 0. Substituting each of these values ​​into y=1,2,…,Y, we obtain the total number of matching features for the m-th template light source, and thus the degree of matching between the m-th template light source and the current batch's m-th template light source. Substituting each of these values ​​into m=1,2,…,M, we obtain the degree of matching between the M template light sources and the current batch's m-th template light source. The template light source with the highest degree of matching is selected as the detection light source for the current batch of SiC wafers. This provides a scientific light source selection scheme for scenarios where there are no perfectly matching candidate light sources, ensuring the continuity and adaptability of the detection process. This step evaluates the fit between the template light source and the current batch of SiC wafers by quantifying the number of matching features, taking into account light sources with partially matching features, and avoiding detection interruptions due to the lack of perfectly compatible light sources. Using the degree of matching as the screening criterion, the template light source with the highest degree of matching is prioritized, minimizing detection errors caused by mismatched light sources and making the selected light source closer to the characteristics of the current batch of SiC wafers. It abandons the blind selection when there is no suitable light source, and provides reliable support for detection in special scenarios through objective feature matching and quantitative analysis. It not only ensures the smooth progress of the detection work, but also takes into account the accuracy of the detection results, further improving the practicality and fault tolerance of the overall detection method. At the same time, it reduces the debugging difficulty for managers.

[0011] The semi-transparent SiC wafer defect detection system based on multi-light source collaboration includes: a historical data light source classification module, a template light source database construction module, a limited feature range database module, a light source matching judgment module, a candidate light source matching optimization module, and a no-candidate-light source matching optimization module. The historical data light source classification module is used to classify light sources into template light sources and non-template light sources based on historical data; The template light source database construction module is used to preprocess and normalize the light source data to obtain light source features, call the SiC wafer features detected by the light source, and establish a template light source database. The restricted feature range database module is used to analyze the restrictions of template light sources on available light sources based on the template light source database, and to establish a restricted feature range database based on the restrictions. The light source matching determination module is used to determine the matching relationship between the template light source and the current batch of SiC wafers when receiving a defect detection task for the current batch of SiC wafers. The candidate light source matching and optimization module is used to obtain the candidate light source for the current batch of SiC wafers according to the matching relationship when there are candidate light sources for the current batch of SiC wafers, and select the detection light source considering the historical defect misjudgment rate. The no-candidate-light-source matching and optimization module is used to select a detection light source when there is no candidate light source for the current batch of SiC wafers, taking into account the matching degree between the characteristics of the current batch of SiC wafers and the template light source.

[0012] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: Firstly, by deeply mining and analyzing historical testing data, effective template light sources are selected, thus avoiding interference from invalid light sources on the testing results from the source. Secondly, during the testing process, the actual defect conditions are used as the calibration basis, and multi-dimensional feature matching ensures the compatibility between the selected light source and the SiC wafer under test, effectively reducing false positives and false negatives. Thirdly, it does not rely on a fixed testing mode with a single light source, but dynamically selects the optimal light source based on the actual testing results, making the defect detection results more consistent with the actual quality condition of the SiC wafer, providing a reliable basis for subsequent quality control and processing, and significantly improving the credibility and rigor of defect detection work.

[0013] On the one hand, by fully considering the differences in characteristics between different batches of SiC wafers, a comprehensive feature matching mechanism has been established to achieve precise matching between the light source and the SiC wafer to be tested. Whether it's a SiC wafer with standard characteristics or a batch with special features, a suitable testing light source can be found through feature comparison, eliminating the need for frequent adjustments to testing parameters or replacement of testing equipment. This flexible matching method breaks through the limitations of traditional testing methods on SiC wafer characteristics, reduces the operational complexity of the testing process, and avoids the problem of low testing efficiency caused by light source mismatch. It can easily cope with diverse production testing scenarios, ensuring the smooth progress of the production process.

[0014] On the other hand, relying on historical data to build a comprehensive database system has formed standardized testing logic and operating procedures, significantly reducing reliance on human experience and lowering the operational threshold. The database can be continuously optimized as testing data accumulates, constantly improving the adaptability and accuracy of the testing methods, and possessing good scalability and growth potential. Simultaneously, through a scientific light source selection mechanism, invalid testing steps are reduced, and the testing cycle is shortened, improving testing quality while maintaining efficiency. This testing solution, which balances practicality and growth potential, can well adapt to the quality testing needs of large-scale production, helping enterprises achieve more efficient quality control and cost control. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural diagram of the semi-transparent SiC wafer defect detection system based on multi-light source collaboration of the present invention; Figure 2 This is a flowchart of the defect detection method for semi-transparent SiC wafers based on multi-light source collaboration according to the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 and Figure 2 This invention provides a technical solution: a defect detection method for semi-transparent SiC wafers based on multi-light source collaboration, comprising the following steps: S1. Based on historical data, light sources are divided into template light sources and non-template light sources; S2. After preprocessing and normalizing the light source data, the light source characteristics are obtained. The SiC wafer characteristics detected by the light source are called to establish a template light source database. S3. Analyze the constraints of template light sources on available light sources based on the template light source database, and establish a database of constraint feature ranges based on the constraints. S4. When receiving the defect detection task of the current batch of SiC wafers, determine the matching relationship between the template light source and the current batch of SiC wafers; S5. When there are candidate light sources for the current batch of SiC wafers, obtain the candidate light sources for the current batch of SiC wafers according to the matching relationship, and select the detection light source considering the historical defect misjudgment rate. S6. If there is no candidate light source for the current batch of SiC wafers, select the detection light source by considering the matching degree between the characteristics of the current batch of SiC wafers and the template light source.

[0018] In step S1, based on historical data, when using a light source to detect defects in a batch of SiC wafers, the number of defects detected by the light source, α, is recorded. The actual number of defects in the SiC wafers, β, is obtained using a microscope, thus obtaining the defect rate α / β when using the light source to detect defects in this batch of SiC wafers. If α / β > γ, the light source is considered effective for detecting defects in this batch of SiC wafers; otherwise, the light source is considered ineffective. In historical data, if a light source is effective for detecting defects in any batch of SiC wafers, the light source is considered a template light source; otherwise, the light source is considered a non-template light source. By deeply mining historical detection data and using the actual defect situation as the core judgment criterion, template light sources and non-template light sources are accurately classified. This design eliminates consistently ineffective light sources from the source, avoiding interference with subsequent detection processes and ensuring the targeted and effective selection of light sources. The judgment process abandons subjective experience dependence and uses actual detection results as the measurement standard, ensuring that the selected template light sources have practical defect detection capabilities, providing high-quality data support for subsequent feature-matching-based selection of suitable light sources. It strictly controls the quality of the light source from the initial stage, which significantly improves the rigor and scientific nature of the overall defect detection process and lays a solid foundation for optimizing the accuracy of subsequent detection.

[0019] In step S2, light source data is collected for M template light sources, where M is the total number of template light sources. The light source data is preprocessed and normalized to obtain the light source features. The light source feature of the m-th template light source is {A}. 1_m A 2_m ,…,A n_m ,…,A N_m}, where N is the number of light source features for any template light source, A n_m Given the nth light source feature of the mth template light source, the system then retrieves the valid batch X of SiC wafers from the mth template light source in historical data. These batches of SiC wafers are considered usable batches for the mth template light source. The SiC wafer feature of the xth batch of SiC wafers is {B}. 1_x B 2_x ,…,B y_x ,…,B Y_x}, where Y represents the total number of SiC wafer features in any batch of SiC wafers, B y_xThis represents the y-th SiC wafer feature of the x-th batch of SiC wafers. A template light source database is established based on template light sources and the SiC wafer features of available batches of SiC wafers. By systematically analyzing the feature correlation between template light sources and corresponding valid batches of SiC wafers, core data support is built for the testing process. Preprocessing and normalization of the light source data eliminates redundant information interference, allowing the light source features to more accurately reflect their testing attributes. Integrating the template light sources with the corresponding SiC wafer features clearly establishes the adaptation correlation between the two, avoiding mismatches between light sources and material properties. The constructed template light source database breaks down the information silos between light sources and testing objects, providing a structured data foundation for rapid matching of suitable light sources, reducing the cost of repetitive data collection and analysis, and shifting light source selection from fragmented experience to data-driven approaches, thus improving the standardization and efficiency of the testing process.

[0020] In step S3, based on the template light source database analysis, the constraints of the m-th template light source on the available light sources are analyzed. For the m-th template light source, the SiC wafer characteristics of batch X SiC wafers are called. The y-th SiC wafer characteristic of batch X SiC wafers is {B y_1 B y_2 ,…,B y_x ,…,B y_X Then, the influence C of the y-th SiC wafer feature on the m-th template light source is calculated. y_m : ; Where b is the average value of the y-th SiC wafer characteristic in batch X, when C y_m >C y_0 If the condition is met, determine if the y-th SiC wafer feature affects the m-th template light source; otherwise, determine if the y-th SiC wafer feature has no effect on the m-th template light source, where C y_0 To determine the threshold value for the influence of the y-th SiC wafer feature on the m-th template light source, we obtain the limiting feature range of the m-th template light source on the y-th SiC wafer feature. If it is determined that the y-th SiC wafer feature has an influence on the m-th template light source, we use the IQR method to evaluate the y-th SiC wafer feature {B} of batch X SiC wafers. y_1 B y_2 ,…,B y_x ,…,B y_X Data cleaning is performed to obtain a cleaned feature set. The minimum value of the cleaned feature set is used as the lower limit of the feature range, and the maximum value of the cleaned feature set is used as the upper limit of the feature range. If it is determined that the y-th SiC wafer feature has no effect on the m-th template light source, the feature range is set to a set of real numbers. Substituting y = 1, 2, ..., Y, the constraint feature range {D} of the m-th template light source on the Y features of the SiC wafer. 1_mD 2_m ,…,D y_m ,…,D Y_m}, where D y_m To define the limiting feature range of the m-th template light source for the y-th SiC wafer feature, we substitute m=1,2,…,M one by one to obtain the limiting feature ranges of the M template light sources for the Y SiC wafer features, thus establishing a limiting feature range database. By deeply exploring the intrinsic relationship between template light sources and SiC wafer features, we accurately defined the key features affecting the light source detection effect, eliminated the interference of irrelevant features, and made the light source adaptation standard more targeted. The data cleaning process effectively removed abnormal information, ensuring that the limiting feature ranges conform to actual detection needs, and improving the scientificity and reliability of the standard. The constructed limiting feature range database provides a clear basis for the rapid matching of the current batch of SiC wafers with template light sources, transforming ambiguous adaptation judgments into clear range comparisons, reducing errors caused by subjective experience, further consolidating the data-driven foundation of the detection process, and improving the accuracy and efficiency of light source selection.

[0021] In step S4, when a defect detection task for the current batch of SiC wafers is received, feature acquisition is performed on the current batch of SiC wafers to obtain the SiC wafer features {B1, B2, ..., B} of the current batch of SiC wafers. y ,…,B Y}, where B y Let B represent the y-th SiC wafer feature of the current batch of SiC wafers. The method for determining the matching relationship between the m-th template light source and the current batch of SiC wafers is as follows: If the y-th SiC wafer feature B of the current batch of SiC wafers... y The limiting feature range D of the m-th template light source on the y-th SiC wafer y_mWithin the current batch of SiC wafers, the system determines whether the y-th SiC wafer feature matches the m-th template light source; otherwise, it determines that the y-th SiC wafer feature does not match the m-th template light source. If any SiC wafer feature in the current batch does not match the m-th template light source, then the m-th template light source is determined to be incompatible with the current batch of SiC wafers; otherwise, the m-th template light source is determined to match the current batch of SiC wafers, and the m-th template light source becomes a candidate light source for the current batch of SiC wafers. This standardized feature acquisition and range comparison logic achieves precise matching and screening between the current batch of SiC wafers and template light sources. By acquiring complete features of the SiC wafers to be tested and combining them with an established database of limited feature ranges, clear comparison rules are used to determine light source suitability, avoiding bias caused by subjective judgment. The rigorous design of determining a match only if all features match effectively eliminates incompatible light sources with mismatched features, ensuring that the candidate light sources have reliable testing potential. This step transforms the complex light source adaptation judgment into a clear feature range verification, simplifying the screening process, quickly identifying candidate light sources that meet the requirements, and providing a high-quality candidate pool for further selection of the optimal detection light source, thereby improving the overall targeting and efficiency of the detection process.

[0022] In step S5, each of the m=1,2,…,M template light sources is substituted to determine the matching relationship between the M template light sources and the current batch of SiC wafers, resulting in Q candidate light sources for the current batch of SiC wafers. If Q is not equal to 0, it is determined that there are candidate light sources for the current batch of SiC wafers. The historical defect misjudgment rate of the Q candidate light sources is then called. The historical defect misjudgment rate is the ratio of the number of defects misjudged by the candidate light source in historical data to the actual number of defects obtained from the SiC wafer through microscopy. The candidate light source with the lowest historical defect misjudgment rate is then selected as the detection light source for the current batch of SiC wafers. In scenarios where candidate light sources exist, the optimal detection light source is locked through scientific optimization logic, further improving the reliability of defect detection. This step uses historical defect misjudgment performance as the core screening criterion, calling the past detection data of candidate light sources to intuitively reflect the actual detection quality of each light source, avoiding the detection risks caused by blind selection. Selecting the light source with the lowest historical misjudgment rate as the final detection light source fundamentally reduces the probability of misjudgment in the current batch of detection, making the detection results more consistent with the real situation. Its quantitative screening method based on historical data eliminates the uncertainty of subjective experience judgment, making the selection of light sources more objective and convincing. At the same time, it quickly focuses on the optimal option, shortens the decision-making process, and ensures both accuracy and efficiency, providing key support for completing defect detection tasks with high quality.

[0023] In step S6, if Q equals 0, it is determined that there is no candidate light source for the current batch of SiC wafers. The detection light source for the current batch of SiC wafers is analyzed. The analysis method for the detection light source is as follows: For the m-th template light source, if the y-th SiC wafer feature B of the current batch of SiC wafers... y The limiting feature range D of the m-th template light source on the y-th SiC wafer y_m Within this process, the number of matching features for the m-th template light source is incremented by 1; otherwise, the number of matching features for the m-th template light source is incremented by 0. Substituting each of these values ​​into y=1,2,…,Y, we obtain the total number of matching features for the m-th template light source, and thus the degree of matching between the m-th template light source and the current batch's m-th template light source. Substituting each of these values ​​into m=1,2,…,M, we obtain the degree of matching between the M template light sources and the current batch's m-th template light source. The template light source with the highest degree of matching is selected as the detection light source for the current batch of SiC wafers. This provides a scientific light source selection scheme for scenarios where there are no perfectly matching candidate light sources, ensuring the continuity and adaptability of the detection process. This step evaluates the fit between the template light source and the current batch of SiC wafers by quantifying the number of matching features, taking into account light sources with partially matching features, and avoiding detection interruptions due to the lack of perfectly compatible light sources. Using the degree of matching as the screening criterion, the template light source with the highest degree of matching is prioritized, minimizing detection errors caused by mismatched light sources and making the selected light source closer to the characteristics of the current batch of SiC wafers. It abandons the blind selection when there is no suitable light source, and provides reliable support for detection in special scenarios through objective feature matching and quantitative analysis. It not only ensures the smooth progress of the detection work, but also takes into account the accuracy of the detection results, further improving the practicality and fault tolerance of the overall detection method. At the same time, it reduces the debugging difficulty for managers.

[0024] A defect detection system for semi-transparent SiC wafers based on multi-light source collaboration includes: a historical data light source classification module, a template light source database construction module, a limited feature range database module, a light source matching judgment module, a candidate light source matching optimization module, and a no-candidate-light source matching optimization module. The historical data light source classification module is used to classify light sources into template light sources and non-template light sources based on historical data; The template light source database construction module is used to preprocess and normalize the light source data to obtain the light source characteristics, call the SiC wafer characteristics detected by the light source, and establish the template light source database. The Limitation Feature Range Database module is used to analyze the limitations of template light sources on available light sources based on the template light source database, and to establish a limitation feature range database based on the limitations. The light source matching determination module is used to determine the matching relationship between the template light source and the current batch of SiC wafers when receiving a defect detection task for the current batch of SiC wafers; The candidate light source matching and optimization module is used to obtain the candidate light source for the current batch of SiC wafers based on the matching relationship when there are candidate light sources for the current batch of SiC wafers, and select the detection light source considering the historical defect false judgment rate. The no-candidate-light-source matching and optimization module is used to select the detection light source when there is no candidate light source for the current batch of SiC wafers, taking into account the matching degree between the characteristics of the current batch of SiC wafers and the template light source.

[0025] Example 1: In the industrial defect detection scenario of semi-transparent SiC wafers, the historical data review process is initiated first. Records of past detection using various light sources are extracted, and the detection results for different batches of materials using each type of light source are extracted. Simultaneously, actual defect information confirmed by professional microscopic equipment is retrieved. Based on this, the actual effectiveness of each type of light source in the corresponding batch detection is calculated. According to preset validity standards, it is determined whether each type of light source meets the standards for detecting each batch of materials. If a certain type of light source performs effectively in any batch detection, it is included in the template light source category; the rest are classified as non-template light sources. This process filters out light source types with practical detection value from the source.

[0026] Subsequently, comprehensive data collection was conducted on the selected template light sources, covering their key attribute information. Through professional data processing methods, the collected light source data was purified and standardized, eliminating redundant and interfering information and extracting core features that accurately reflect the detection characteristics of the light sources. Simultaneously, the characteristics of relevant batches of materials that had been effectively tested in the past for each type of template light source were retrieved. The core features of the template light sources were then correlated and integrated with the characteristics of the corresponding compatible materials, constructing a complete and detailed template light source database, providing fundamental data support for subsequent adaptation and matching.

[0027] Based on the established template light source database, this study delves into the intrinsic relationships between various template light sources and the characteristics of compatible materials. The impact of each material characteristic on the detection performance of the corresponding template light source is evaluated, key influencing features are identified, and interference from irrelevant features is eliminated. Outlier removal is performed on the data related to key influencing features, clarifying the compatibility range of key material features for each type of template light source; features without influence are left unrestricted. Finally, the feature compatibility range information of all template light sources is integrated to establish a unified database of restricted feature ranges, providing clear standards for subsequent matching judgments.

[0028] When a defect detection task is received for a new batch of SiC wafers, a comprehensive feature acquisition process is first performed on the batch of materials to obtain their core attribute information. The acquired features of the current batch of materials are then compared one by one with the compatibility standards of various template light sources in the feature range database. Only when all features of the current batch of materials meet the compatibility range of a certain type of template light source is that template light source listed as a candidate light source for the current batch, ensuring that the candidate light source has basic compatibility.

[0029] If candidate light sources exist, historical test records for each candidate light source are retrieved, and their misjudgment rates in previous tests are calculated. This is used as the core indicator for evaluating the quality of the light source test. The light source with the best misjudgment performance is then selected and determined as the final test light source for the current batch of materials, maximizing the accuracy of the test results.

[0030] If no perfectly suitable candidate light source is found, a quantitative evaluation is performed on the feature matching between various template light sources and the current batch of materials. The number of features that match each type of template light source with the current batch of materials is counted to determine the matching degree of each type of light source. The template light source with the highest matching degree is selected as the detection light source to ensure that the detection work can proceed smoothly, while minimizing detection errors caused by insufficient matching degree.

[0031] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary sensing device embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A defect detection method for semi-transparent SiC wafers based on multi-source collaborative methods, characterized in that: The method includes the following steps: S1. Based on historical data, light sources are divided into template light sources and non-template light sources; S2. After preprocessing and normalizing the light source data, the light source characteristics are obtained. The SiC wafer characteristics detected by the light source are called to establish a template light source database. S3. Analyze the constraints of template light sources on available light sources based on the template light source database, and establish a database of constraint feature ranges based on the constraints. S4. When receiving the defect detection task of the current batch of SiC wafers, determine the matching relationship between the template light source and the current batch of SiC wafers; S5. When there are candidate light sources for the current batch of SiC wafers, obtain the candidate light sources for the current batch of SiC wafers according to the matching relationship, and select the detection light source considering the historical defect misjudgment rate. S6. If there is no candidate light source for the current batch of SiC wafers, select the detection light source by considering the matching degree between the characteristics of the current batch of SiC wafers and the template light source.

2. The defect detection method for semi-transparent SiC wafers based on multi-source collaborative processing according to claim 1, characterized in that: In step S1, in the historical data, when using a light source to detect defects in a batch of SiC wafers, the number of defects α found by the light source is recorded. The actual number of defects β of the SiC wafers is obtained by microscopy, thereby obtaining the defect rate α / β when using the light source to detect defects in this batch of SiC wafers. If α / β > γ, it is determined that the light source is effective in detecting this batch of SiC wafers; otherwise, it is determined that the light source is ineffective in detecting this batch of SiC wafers. In the historical data, if there is a light source that is effective in detecting any batch of SiC wafers, it is determined that the light source is a template light source; otherwise, it is determined that the light source is a non-template light source.

3. The defect detection method for semi-transparent SiC wafers based on multi-source collaborative processing according to claim 2, characterized in that: In step S2, light source data is collected for M template light sources, where M is the total number of template light sources. The light source data is preprocessed and normalized to obtain the light source features. The light source feature of the m-th template light source is {A}. 1_m A 2_m ,…,A n_m ,…,A N_m }, where N is the number of light source features for any template light source, A n_m Given the nth light source feature of the mth template light source, the system then retrieves the valid batch X of SiC wafers from the mth template light source in historical data. These batches of SiC wafers are considered usable batches for the mth template light source. The SiC wafer feature of the xth batch of SiC wafers is {B}. 1_x B 2_x ,…,B y_x ,…,B Y_x }, where Y represents the total number of SiC wafer features in any batch of SiC wafers, B y_x This represents the y-th SiC wafer feature of the x-th batch of SiC wafers. A template light source database is established based on the template light source and the SiC wafer features of available batches of SiC wafers.

4. The defect detection method for semi-transparent SiC wafers based on multi-source collaborative processing according to claim 3, characterized in that: In step S3, based on the template light source database analysis, the constraints of the m-th template light source on the available light sources are analyzed. For the m-th template light source, the SiC wafer characteristics of batch X SiC wafers are called. The y-th SiC wafer characteristic of batch X SiC wafers is {B y_1 B y_2 ,…,B y_x ,…,B y_X Then, the influence C of the y-th SiC wafer feature on the m-th template light source is calculated. y_m : ； Where b is the average value of the y-th SiC wafer characteristic in batch X, when C y_m >C y_0 If the condition is met, determine if the y-th SiC wafer feature affects the m-th template light source; otherwise, determine if the y-th SiC wafer feature has no effect on the m-th template light source, where C y_0 To determine the threshold value for the influence of the y-th SiC wafer feature on the m-th template light source, we obtain the limiting feature range of the m-th template light source on the y-th SiC wafer feature. If it is determined that the y-th SiC wafer feature has an influence on the m-th template light source, we use the IQR method to evaluate the y-th SiC wafer feature {B} of batch X SiC wafers. y_1 B y_2 ,…,B y_x ,…,B y_X Data cleaning is performed to obtain a cleaned feature set. The minimum value of the cleaned feature set is used as the lower limit of the feature range, and the maximum value of the cleaned feature set is used as the upper limit of the feature range. If it is determined that the y-th SiC wafer feature has no effect on the m-th template light source, the feature range is restricted to the set of real numbers.

5. The defect detection method for semi-transparent SiC wafers based on multi-source collaborative processing according to claim 4, characterized in that: Substituting y = 1, 2, ..., Y, the constraint feature range {D} of the m-th template light source on the Y features of the SiC wafer. 1_m D 2_m ,…,D y_m ,…,D Y_m }, where D y_m Let m be the limiting feature range of the m-th template light source on the y-th SiC wafer feature. Substitute m=1,2,…,M one by one to obtain the limiting feature range of the M template light sources on the Y SiC wafer features, and then establish a limiting feature range database.

6. The defect detection method for semi-transparent SiC wafers based on multi-source collaborative processing according to claim 5, characterized in that: In step S4, when a defect detection task for the current batch of SiC wafers is received, feature acquisition is performed on the current batch of SiC wafers to obtain the SiC wafer features {B1, B2, ..., B} of the current batch of SiC wafers. y ,…,B Y }, where B y Let B represent the y-th SiC wafer feature of the current batch of SiC wafers. The method for determining the matching relationship between the m-th template light source and the current batch of SiC wafers is as follows: If the y-th SiC wafer feature B of the current batch of SiC wafers... y The limiting feature range D of the m-th template light source on the y-th SiC wafer y_m Within the current batch of SiC wafers, determine if the y-th SiC wafer feature matches the m-th template light source; otherwise, determine if the y-th SiC wafer feature does not match the m-th template light source. If any SiC wafer feature in the current batch does not match the m-th template light source, then determine if the m-th template light source does not match the current batch of SiC wafers; otherwise, determine if the m-th template light source matches the current batch of SiC wafers, and the m-th template light source is the candidate light source for the current batch of SiC wafers.

7. The defect detection method for semi-transparent SiC wafers based on multi-source collaborative processing according to claim 6, characterized in that: In step S5, each of the M template light sources (m=1,2,…,M) is substituted to determine the matching relationship between them and the current batch of SiC wafers, thus obtaining Q candidate light sources for the current batch of SiC wafers. If Q is not equal to 0, it is determined that there are candidate light sources for the current batch of SiC wafers. The historical defect misjudgment rate of the Q candidate light sources is then called. The historical defect misjudgment rate is the ratio of the number of defects misjudged by the candidate light source in the historical data to the actual number of defects of the SiC wafers obtained by microscope. The candidate light source with the lowest historical defect misjudgment rate is then selected as the detection light source for the current batch of SiC wafers.

8. The defect detection method for semi-transparent SiC wafers based on multi-source collaborative processing according to claim 6, characterized in that: In step S6, if Q equals 0, it is determined that there is no candidate light source for the current batch of SiC wafers. The detection light source for the current batch of SiC wafers is analyzed. The analysis method for the detection light source is as follows: For the m-th template light source, if the y-th SiC wafer feature B of the current batch of SiC wafers... y The limiting feature range D of the m-th template light source on the y-th SiC wafer y_m For each m-th template light source, increment the number of matching features by 1; otherwise, increment the number of matching features by 0. Substitute each of these into y=1,2,…,Y to obtain the total number of matching features for the m-th template light source, and then obtain the matching degree of the m-th template light source with the current batch of m-th template light sources. Substitute each of these into m=1,2,…,M to obtain the matching degree of the M template light sources with the current batch of m-th template light sources, and select the template light source with the highest matching degree as the detection light source for the current batch of SiC wafers.

9. A defect detection system for semi-transparent SiC wafers based on multi-source collaborative operation, wherein the system is applied to the defect detection method for semi-transparent SiC wafers based on multi-source collaborative operation as described in any one of claims 1-8, characterized in that: The system includes: a historical data light source classification module, a template light source database construction module, a restricted feature range database module, a light source matching determination module, a candidate light source matching optimization module, and a no-candidate-light source matching optimization module; The historical data light source classification module is used to classify light sources into template light sources and non-template light sources based on historical data; The template light source database construction module is used to preprocess and normalize the light source data to obtain light source features, call the SiC wafer features detected by the light source, and establish a template light source database. The restricted feature range database module is used to analyze the restrictions of template light sources on available light sources based on the template light source database, and to establish a restricted feature range database based on the restrictions. The light source matching determination module is used to determine the matching relationship between the template light source and the current batch of SiC wafers when receiving a defect detection task for the current batch of SiC wafers. The candidate light source matching and optimization module is used to obtain the candidate light source for the current batch of SiC wafers according to the matching relationship when there are candidate light sources for the current batch of SiC wafers, and select the detection light source considering the historical defect misjudgment rate. The no-candidate-light-source matching and optimization module is used to select a detection light source when there is no candidate light source for the current batch of SiC wafers, taking into account the matching degree between the characteristics of the current batch of SiC wafers and the template light source.

Images