Product single crystal defect locking system and method based on silicon wafer black cross light disappearance principle

Abstract

The present application relates to the technical field of single crystal silicon wafer defect detection, more specifically, to a product single crystal defect locking system and method based on the principle of black cross light disappearance of silicon wafer, which is used to solve the problem that the prior art usually lacks effective verification of the mechanical orthogonal relationship between the fracture characteristics and the local stress direction, and only relies on image intensity or simple threshold to extract the fracture endpoint, which is easy to be interfered by noise, imaging non-uniformity or crystal microstructure, resulting in false fracture misjudgment; the present application fuses polarization response and mechanical priori through a black cross fracture stress verification module, positions the fracture endpoint by modulating the consistency of depth and phase, combines weighted robust fitting to inverse the local stress direction, and strictly verifies the fracture-stress vertical relationship by statistical test, finally takes the verified points as the boundary to drive anisotropic diffusion to generate a physically reasonable, noise-robust and full-field guided feature field, which provides reliable priori for high-precision stress reconstruction.

Description

Technical Field

[0001] This invention relates to the field of single-crystal silicon wafer defect detection technology, and more specifically, to a single-crystal defect locking system and method based on the principle of the disappearance of the black cross light on silicon wafers. Background Technology

[0002] In the process of preparing monocrystalline silicon wafers, the integrity of the crystal directly determines its electrical performance and reliability in integrated circuits or high-efficiency solar cells. Ideal monocrystalline silicon has a highly ordered cubic lattice structure and is optically isotropic under polarized light fields, usually without producing obvious extinction patterns. However, thermal stress and impurity segregation during crystal pulling, as well as damage introduced by mechanical processing such as slicing, can easily lead to crystal defects such as dislocations, slip lines, microtwins, or local polycrystallineization. These defects disrupt the periodicity of the lattice, induce local abnormal birefringence, and manifest as characteristic extinction anomalies in polarized imaging.

[0003] Referring to patent application CN118122643A, a method and system for sorting monocrystalline silicon wafers based on texture width detection is disclosed. The method includes: acquiring an image of the monocrystalline silicon wafer to obtain a texture image; converting the texture image to grayscale to obtain a first texture grayscale image, and performing further processing to obtain a second texture grayscale image; identifying stripes in the second texture grayscale image and marking the stripe pixels to obtain a third texture grayscale image; selecting a texture region to obtain a first texture region image; processing the first texture region image to eliminate noise pixels to obtain a second texture region image, and fitting the stripe pixels on it to obtain several stripe curves; measuring the distance between two adjacent stripe curves and obtaining the texture width of the monocrystalline silicon wafer based on the distance; and selecting qualified monocrystalline silicon wafers. This invention achieves the goal of ensuring the quality of monocrystalline silicon wafer products by detecting the texture width of the lines on the monocrystalline silicon wafers. However, in silicon wafer polarization detection based on the principle of black cross light disappearance, existing technologies usually lack effective verification of the mechanical orthogonality between fracture features and local stress directions. They rely solely on image intensity or simple thresholds to extract fracture endpoints, which is susceptible to noise, imaging inhomogeneity, or crystal microstructure interference, leading to false fracture misjudgments. At the same time, in the full-field stress tensor inversion, global smoothing or fixed isotropic regularization is often used, failing to effectively integrate the directional priors and key point constraints obtained from polarization data. This makes it difficult to accurately preserve stress abrupt changes in the fracture region, while non-fracture regions may be distorted due to excessive smoothing, affecting the consistency between the stress field reconstruction and the actual defect stress distribution.

[0004] To address these issues, this invention proposes a single-crystal defect locking system and method based on the principle of black cross light disappearance in silicon wafers. Summary of the Invention

[0005] The purpose of this invention is to provide a single-crystal defect locking system and method for silicon wafers based on the principle of black cross light disappearance. This invention solves the problems of existing technologies, which often lack effective verification of the mechanical orthogonality between fracture features and local stress directions. They rely solely on image intensity or simple thresholds to extract fracture endpoints, which are easily affected by noise, imaging non-uniformity, or crystal microstructure interference, leading to false fracture misjudgments. At the same time, in the full-field stress tensor inversion, global smoothing or fixed isotropic regularization is often used, which fails to effectively integrate the directional priors and key point constraints obtained from polarization data. This makes it difficult to accurately retain stress abrupt changes in the fracture region, while non-fracture regions may be distorted due to excessive smoothing, affecting the consistency between the stress field reconstruction and the actual defect stress distribution.

[0006] The objective of this invention is achieved through the following technical solution: A single-crystal defect locking system targeting the principle of black cross light disappearance in silicon wafers is applied to a single-crystal silicon wafer production control platform, including: The black cross polarization acquisition and processing module acquires a series of monochromatic light intensity images at different polarization analysis angles during orthogonal circular polarization interferometry imaging, preprocesses the acquired monochromatic light intensity images, and finally outputs a preliminary aligned standardized polarization image stack. The Black Cross Fracture Stress Verification Module extracts the fracture endpoints of the Black Cross Arm from the initially aligned standardized polarization image stack, obtains their local stress direction, verifies whether the fracture tangent is perpendicular to the stress direction, retains the fracture endpoints that satisfy the perpendicular relationship as verified fracture stress points, and generates a guiding feature field. The Black Cross verification stress inversion module, based on the photoelastic effect model, constructs an optimized inversion framework that integrates guided feature field fitting, verified fracture stress point strong constraints, and adaptive spatial smoothing to solve the full-field stress tensor. The stress-driven process root cause localization module matches the full-field stress tensor with the process stress mode knowledge base, and combines it with historical production line testing data to output the ranking of suspected process steps and the results of stress type identification. The executable process diagnostic report module generates a structured diagnostic report based on the full-field stress tensor and the sorting of suspected process steps. The report includes the maximum stress value and location, stress type, suspected process steps, and parameter optimization suggestions.

[0007] In a preferred embodiment of the present invention, the process of extracting the fracture endpoint of the black cross arm and obtaining its local stress direction in the black cross fracture stress verification module includes: Obtain a pre-aligned, standardized polarization image stack. For each position in the image, read the light intensity value at all polarization angles. Calculate the difference between the maximum and minimum light intensity at the position and divide it by the average light intensity to obtain the modulation depth. Multiply the light intensity at each polarization angle by the cosine and sine values, which are twice the corresponding polarization angle, and sum them to obtain two components. Take the square and square root of these components to obtain the polarization phase consistency. Calculate the average and standard deviation of the modulation depth and polarization phase consistency over the entire image. Normalize the two response quantities at each position. Substitute the normalized modulation depth and polarization phase consistency into the breakage confidence formula to calculate the breakage confidence at each position. A fracture confidence map is constructed using fracture confidence values ​​at all locations. Through adaptive threshold segmentation, pixel regions with fracture confidence values ​​exceeding the threshold are extracted to form an initial fracture mask. The mask is then skeletonized to obtain a single-pixel-wide skeleton centerline. The pixels on the skeleton centerline are traversed, and the number of skeleton connections in their eight neighborhoods is counted. If the number is one, it is marked as a fracture endpoint of the black cross arm. A local neighborhood window is defined for each fracture endpoint. Each pixel within the window is assigned a spatial weight and a signal weight. The joint weight is the product of the spatial weight and the signal weight. A weighted robust optimization objective is constructed using the Huber loss function. Through an iterative reweighted least squares algorithm, the average light intensity and the shared local stress direction of all pixels within the window are jointly solved, and the local stress direction corresponding to each fracture endpoint is output.

[0008] In a preferred embodiment of the present invention, the process of verifying whether the fracture tangent is perpendicular to the stress direction in the black cross fracture stress verification module includes: For each fracture endpoint of the black cross arm, obtain the fracture tangent direction and the local stress direction, calculate the cosine of the angle between the two directions, take the absolute value, obtain the standard error of the fracture tangent direction and the standard error of the local stress direction, use the error propagation formula to calculate the standard error of the cosine value from the standard errors of the two directions, divide the absolute value of the cosine by the standard error of the cosine value to obtain the test statistic. The two-sided p-value of the test statistic is calculated based on the standard normal distribution. If the p-value is greater than 5%, the fracture tangent is determined to be perpendicular to the stress direction, and the perpendicularity determination result of the fracture endpoint is output.

[0009] In a preferred embodiment of the present invention, the process of retaining the fracture endpoints that satisfy the vertical relationship as verified fracture stress points and generating a guiding feature field in the black cross fracture stress verification module includes: For each fracture endpoint that satisfies the perpendicular relationship, it is marked as a verified fracture stress point. The corresponding local stress direction is read, and the position and direction are paired to form the Dirichlet boundary condition. The local orientation field of the skeleton centerline and the fracture confidence map are read. The principal axis direction of the diffusion coefficient tensor is determined based on the local orientation field. The anisotropy intensity of the diffusion coefficient is modulated based on the fracture confidence map to generate the diffusion coefficient tensor field. Using all Dirichlet boundary conditions and diffusion coefficient tensor fields as input, the steady-state diffusion equation is established and solved to obtain a numerical solution in the entire image domain. The numerical solution in the entire image domain is then used as the output of the guiding feature field.

[0010] As a preferred embodiment of the present invention, the process of constructing an optimized inversion framework that integrates guided feature field fitting, verified fracture stress point strong constraints, and adaptive spatial smoothing in the black cross verification stress inversion module includes: For each position, obtain the direction of the maximum tensile force and the direction of the guiding feature field at the corresponding position. Determine whether the two directions are the same or orthogonal. If they are neither the same nor orthogonal, and the corresponding position does not belong to the non-endpoint region of the fracture path, calculate the direction inconsistency penalty and add it to the optimization objective. For each verified fracture stress point, obtain the local stress direction and principal stress difference, and add the local stress direction and principal stress difference of the corresponding point as fixed value constraints to the optimization objective. For each pair of adjacent grid points, obtain the fracture confidence value of each point. If the fracture confidence of both points is zero, calculate the difference in stress state between the two points, square it and add it to the optimization objective. If the fracture confidence of any point is greater than zero, skip the smoothing penalty of the corresponding neighborhood pair. After traversing the entire field, form a complete optimization objective function, solve the optimization objective, and output the full field stress tensor inversion result.

[0011] In a preferred embodiment of the present invention, the process of solving the full-field stress tensor in the black cross verification stress inversion module includes: Obtain a regular mesh, obtain the initial value of the stress tensor at each mesh point, obtain the constructed optimization objective function, obtain the preset convergence threshold, enter the iterative loop, construct a symmetric linear equation system using first-order and second-order information, solve the linear equation system to obtain the update amount of the stress tensor in the whole field, and calculate the global norm of the update amount. Determine if the global norm of the update quantity is less than the convergence threshold. If it is less, terminate the iteration and output the three components of the stress tensor of each grid point to form the global stress tensor.

[0012] In a preferred embodiment of the present invention, the process of matching the full-field stress tensor with the process stress mode knowledge base in the stress-driven process root cause localization module includes: Obtain the stress location, stress distribution pattern, symmetry type, maximum stress value, and average stress value corresponding to the full-field stress tensor. Obtain the process stress pattern knowledge base. For each record in the knowledge base, obtain its stress location, stress distribution pattern, symmetry type, and the lower and upper limits of the stress intensity range. Sequentially determine whether the measured stress location is equal to the recorded stress location, whether the measured stress distribution pattern is equal to the recorded stress distribution pattern, and whether the measured symmetry type is equal to the recorded symmetry type. At the same time, determine whether the maximum stress value is greater than or equal to the lower limit of the stress intensity range and less than or equal to the upper limit, and whether the average stress value is greater than or equal to the lower limit and less than or equal to the upper limit. If all judgment results are equal or within the corresponding range, the name of the process step and the name of the stress type currently recorded are added to the matching list. After traversing all records in the knowledge base, the matching list is output.

[0013] In a preferred embodiment of the present invention, the process of outputting the ranking of suspected process steps and the determination of stress type in the stress-driven process root cause localization module, in combination with historical production line detection data, includes: Obtain a list of successfully matched candidate processes. For each candidate process, obtain its process step name and stress type name. Obtain the process execution parameters of the current product and compare them with the process parameters of all stress anomaly events with the current process step name as the root cause in the historical test data. Determine if there are any completely identical records. Count the number of stress anomaly events with the current process step name as the root cause in the most recent month and determine if it is not less than one. Based on the results of the three checks, count the number of successful items as the historical match count. Obtain the historical match count and process step number of all candidate processes. If the historical match counts are the same, sort them by process step number from smallest to largest, and select the top three candidate processes as the final list of suspected processes. For each suspected process, obtain its stress location, stress distribution pattern, symmetry type, maximum stress value, and average stress value. Obtain the cause description of the corresponding stress type from the knowledge base, obtain the recommended engineering points for investigation, and output the process step name, stress type name, stress location, stress distribution pattern, symmetry type, maximum stress value, average stress value, cause description, and recommended engineering points for investigation.

[0014] As a preferred embodiment of the present invention, the product single-crystal defect locking method based on the principle of the disappearance of the black cross light on silicon wafers includes the following steps: Step 1: Acquire a series of monochromatic light intensity images at different polarization analysis angles during orthogonal circular polarization interferometry imaging. Preprocess the acquired monochromatic light intensity images and finally output a preliminarily aligned standardized polarization image stack. Step 2: Extract the fracture endpoints of the black cross arms from the initially aligned standardized polarization image stack, obtain their local stress directions, verify whether the fracture tangent is perpendicular to the stress direction, retain the fracture endpoints that satisfy the perpendicular relationship as verified fracture stress points, and generate a guiding feature field. Step 3: Based on the photoelastic effect model, construct an optimized inversion framework that integrates guided feature field fitting, verified strong constraint of fracture stress points, and adaptive spatial smoothing to solve the full-field stress tensor; Step 4: Match the full-field stress tensor with the process stress mode knowledge base, and combine it with historical production line testing data to output the ranking of suspected process steps and the results of stress type identification. Step 5: Based on the full-field stress tensor and the sorting of suspected process steps, generate a structured diagnostic report containing the maximum stress value and location, stress type, suspected process steps and their parameter optimization suggestions.

[0015] Compared with the prior art, the advantages of this invention are: (1) In this invention, the Black Cross fracture stress verification module integrates polarization response and mechanical prior, locates the fracture endpoint by modulation depth and phase consistency, combines weighted robust fitting to invert the local stress direction, and rigorously verifies the fracture-stress vertical relationship by statistical test. Finally, with the verified point as the boundary, it drives anisotropic diffusion to generate a physically reasonable and noise-resistant full-field guiding feature field, providing a reliable prior for high-precision stress reconstruction. (2) In this invention, the Black Cross verification stress inversion module is based on the photoelastic effect, integrates the guided feature field alignment, the hard constraint of the verified fracture point and the adaptive smoothing driven by the fracture confidence, and constructs a high-fidelity and high-robust full-field stress inversion framework. The dynamic start and stop smoothing term takes into account both the continuity of the non-fracture zone and the preservation of the abrupt change in the fracture zone. Combined with efficient second-order iterative solution, it stably converges to a physically reasonable, detailed and globally consistent stress tensor, which is significantly better than the traditional no-priority or uniform smoothing method. Attached Figure Description

[0016] Figure 1 This is a system block diagram of Embodiment 1 of the present invention; Figure 2 This is a system block diagram of Embodiment 2 of the present invention; Figure 3 This is a flowchart illustrating the steps of the single-crystal defect locking method in this invention. Detailed Implementation

[0017] 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] Example 1: As Figure 1 As shown, the single-crystal defect locking system proposed in this invention, based on the principle of the disappearance of the black cross light on silicon wafers, is applied to a single-crystal silicon wafer production control platform, and includes: The black cross polarization acquisition and processing module acquires a series of monochromatic light intensity images at different polarization detection angles during orthogonal circular polarization interferometry imaging. During the acquisition process, the exposure parameters are dynamically adjusted according to the light intensity distribution in the field of view. The acquired monochromatic light intensity images are preprocessed. The preprocessing operations include flat field correction, noise filtering, and rotation registration based on the orientation of the black cross extinction pattern in the image. Finally, a pre-aligned standardized polarization image stack is output. The Black Cross polarization acquisition and processing module efficiently acquires multi-angle polarization images with high signal-to-noise ratio and high consistency through dynamic exposure control, flat field correction, noise suppression, and automatic rotation registration based on the Black Cross extinction pattern. It also outputs a standardized image stack, which combines accuracy, robustness, and real-time performance, making it suitable for quantitative polarization imaging and dynamic sample observation.

[0019] The Black Cross Fracture Stress Verification Module extracts the fracture endpoints of the Black Cross Arm from the initially aligned standardized polarization image stack, obtains their local stress direction, verifies whether the fracture tangent is perpendicular to the stress direction, retains the fracture endpoints that satisfy the perpendicular relationship as verified fracture stress points, and generates a guiding feature field. The process of extracting the fracture endpoints of the black cross arm from a pre-aligned stack of normalized polarization images and obtaining their local stress directions in the black cross fracture stress verification module includes: Obtain a pre-aligned, standardized polarization image stack. For each location in the image, read the light intensity value at all polarization angles. Calculate the difference between the maximum and minimum light intensity at that location, divide by the average light intensity, to obtain the modulation depth. Multiply the light intensity at each polarization angle by the cosine and sine values, respectively, twice the corresponding polarization angle, and sum them to obtain two components. Take the square and square root of these components to obtain the polarization phase consistency. Calculate the average and standard deviation of the modulation depth and polarization phase consistency over the entire image. Normalize the two response values ​​at each location. Substitute the normalized modulation depth and polarization phase consistency into the breakage confidence formula to calculate the breakage confidence at each location. ,in This represents the break confidence level at image location (x, y). The modulation depth at position (x, y) is defined as the relative amplitude of the change in light intensity with the polarization angle. This represents the average modulation depth across the entire image. This represents the standard deviation of the modulation depth over the entire image. The polarization phase uniformity at position (x,y) is defined as the magnitude of the composite response caused by the abrupt change in the principal stress direction. This represents the average value of polarization phase coherence across the entire image. This represents the standard deviation of polarization phase consistency across the entire image; A fracture confidence map is constructed using fracture confidence values ​​at all locations. Through adaptive threshold segmentation, pixel regions with fracture confidence values ​​exceeding the threshold are extracted to form an initial fracture mask. The mask is then skeletonized to obtain a single-pixel-wide skeleton centerline. Pixels along the skeleton centerline are traversed, and the number of skeleton connections in their eight neighborhoods is counted. If the number is one, it is marked as a fracture endpoint of the black cross arm. A local neighborhood window is defined for each fracture endpoint. The change in light intensity with polarization angle within the window is described by a sinusoidal modulation model, and the model phase is determined by the local stress direction shared by all pixels within the window. Spatial weights are assigned to each pixel within the window. The spatial weights are inversely proportional to the Euclidean distance from the pixel to the fracture endpoint. Signal weights are also assigned. The signal weights are directly proportional to the signal-to-noise ratio of the pixel in each polarization image. The joint weight is the product of the spatial weights and the signal weights. A weighted robust optimization objective is constructed using the Huber loss function. Through an iterative reweighted least squares algorithm, the average light intensity and the shared local stress direction of all pixels within the window are jointly solved, and the local stress direction corresponding to each fracture endpoint is output. The process of verifying whether the fracture tangent is perpendicular to the stress direction in the Black Cross fracture stress verification module includes: For each fracture endpoint of the black cross arm, obtain the fracture tangent direction and the local stress direction, calculate the cosine of the angle between the two directions, take the absolute value, obtain the standard error of the fracture tangent direction and the standard error of the local stress direction, use the error propagation formula to calculate the standard error of the cosine value from the standard errors of the two directions, divide the absolute value of the cosine by the standard error of the cosine value to obtain the test statistic. The two-sided p-value of the test statistic is calculated based on the standard normal distribution. If the p-value is greater than 5%, the fracture tangent is determined to be perpendicular to the stress direction, and the perpendicularity determination result of the fracture endpoint is output. The process of retaining fracture endpoints that satisfy the vertical relationship as verified fracture stress points and generating a guiding feature field in the Black Cross fracture stress verification module includes: For each fracture endpoint that satisfies the perpendicular relationship, it is marked as a verified fracture stress point. The corresponding local stress direction is read, and the position and direction are paired to form the Dirichlet boundary condition. The local orientation field of the skeleton centerline and the fracture confidence map are read. The principal axis direction of the diffusion coefficient tensor is determined based on the local orientation field. The anisotropy intensity of the diffusion coefficient is modulated based on the fracture confidence map to generate the diffusion coefficient tensor field. Using all Dirichlet boundary conditions and diffusion coefficient tensor fields as input, the steady-state diffusion equation is established and solved to obtain the numerical solution in the entire image domain. The numerical solution is strictly equal to the recorded local stress direction at the verified fracture stress point, and is naturally interpolated by the diffusion process in other regions. The numerical solution in the entire image domain is used as the guide feature field output. By deeply integrating polarization optical response and mechanical priors through the Black Cross fracture stress verification module, a high-confidence, physically consistent fracture-stress relationship verification and full-field guided feature generation are achieved. This module constructs a fracture confidence map through modulation depth and polarization phase consistency, accurately locating the fracture endpoints of the Black Cross arm. A robust sine fitting strategy with spatial-signal-noise joint weighting accurately inverts the local stress direction at each endpoint. Statistical hypothesis testing based on error propagation is introduced to rigorously determine the perpendicularity of the fracture tangent to the stress direction, effectively suppressing spurious fracture interference. Finally, using the verified fracture points as Dirichlet boundaries and combining an anisotropic diffusion model, a smooth, continuous, and physically reasonable full-field guided feature field is generated. This provides reliable priors for subsequent high-precision stress reconstruction, crack evolution modeling, or image enhancement, combining physical rigor, noise robustness, and computational scalability.

[0020] The Black Cross verification stress inversion module, based on the photoelastic effect model, constructs an optimized inversion framework that integrates guided feature field fitting, verified fracture stress point strong constraints, and adaptive spatial smoothing to solve the full-field stress tensor. The process of constructing an optimized inversion framework based on the photoelastic effect model in the Black Cross verification stress inversion module includes: (1) Guiding feature field fitting, (2) verified strong constraints on fracture stress points, and (3) adaptive spatial smoothing. For each position, obtain the direction of the maximum tensile force and the direction of the guiding feature field at the corresponding position. Determine whether the two directions are the same or orthogonal. If they are neither the same nor orthogonal, and the corresponding position does not belong to the non-endpoint region of the fracture path, calculate the direction inconsistency penalty and add it to the optimization objective. For each verified fracture stress point, obtain the local stress direction and principal stress difference, and add the local stress direction and principal stress difference of the corresponding point as fixed value constraints to the optimization objective. For each pair of adjacent grid points, obtain the fracture confidence value of each point. If the fracture confidence of both points is zero, calculate the difference in stress state between the two points, square it and add it to the optimization objective. If the fracture confidence of any point is greater than zero, skip the smoothing penalty of the corresponding neighborhood pair. After traversing the entire field, form a complete optimization objective function, solve the optimization objective, and output the full field stress tensor inversion result. The process of solving the full-field stress tensor in the Black Cross verification stress inversion module includes: Obtain a regular grid, obtain the initial value of the stress tensor at each grid point. Each stress tensor contains three independent components. Obtain the constructed optimization objective function and the preset convergence threshold. Enter the iteration loop. Calculate the first and second derivative approximations of the optimization objective based on the current stress tensor. Construct a symmetric linear equation system using the first and second order information. Solve the linear equation system to obtain the update amount of the stress tensor in the entire field. Calculate the global norm of the update amount. Determine if the global norm of the update is less than the convergence threshold. If it is less, terminate the iteration and output the three components of the stress tensor of each grid point to form the global stress tensor. The Black Cross verification stress inversion module is based on the photoelastic effect and organically integrates the guiding feature field, the strong constraints of verified fracture points, and adaptive spatial smoothing to construct a high-fidelity and robust full-field stress tensor inversion framework. During the optimization process, the module ensures that the principal inversion direction is aligned with the guiding feature field through directional consistency penalties, and applies precise directional and principal stress difference hard constraints to verified fracture stress points to guarantee the physical accuracy of key regions. At the same time, it dynamically starts and stops the neighborhood smoothing term based on the fracture confidence, which not only suppresses noise and improves the continuity of the solution in non-fracture regions, but also preserves stress abrupt changes near the fracture to avoid over-smoothing. Combined with an efficient second-order iterative solution strategy, the module can stably converge to a physically reasonable, detail-preserving, and globally consistent full-field stress tensor, which is significantly better than traditional uniform smoothing or inversion methods without prior constraints.

[0021] The stress-driven process root cause localization module matches the full-field stress tensor with the process stress mode knowledge base, and combines it with historical production line testing data to output the ranking of suspected process steps and the results of stress type identification. The process of matching the full-field stress tensor with the process stress mode knowledge base in the stress-driven process root cause localization module includes: Obtain the stress location, stress distribution pattern, symmetry type, maximum stress value, and average stress value corresponding to the full-field stress tensor. Obtain the process stress pattern knowledge base. For each record in the knowledge base, obtain its stress location, stress distribution pattern, symmetry type, and the lower and upper limits of the stress intensity range. Sequentially determine whether the measured stress location is equal to the recorded stress location, whether the measured stress distribution pattern is equal to the recorded stress distribution pattern, and whether the measured symmetry type is equal to the recorded symmetry type. At the same time, determine whether the maximum stress value is greater than or equal to the lower limit of the stress intensity range and less than or equal to the upper limit, and whether the average stress value is greater than or equal to the lower limit and less than or equal to the upper limit. If all judgment results are equal or within the corresponding interval, add the currently recorded process step name and stress type name to the matching list. After completing the traversal of all records in the knowledge base, output the matching list. The process by which the stress-driven process root cause localization module combines historical production line inspection data to output the ranking of suspected process steps and the determination of stress type includes: Obtain a list of successfully matched candidate processes. For each candidate process, obtain its process step name and stress type name. Based on historical test data, check if there are any stress anomaly events with the root cause equal to the current process step name within the last six months. Obtain the process execution parameters of the current product and compare them with the process parameters of all stress anomaly events with the root cause equal to the current process step name in the historical test data to determine if there are any completely identical records. Count the number of stress anomaly events with the root cause equal to the current process step name within the last month and determine if it is not less than one. Based on the results of the three checks, count the number of valid items as the historical match count. Obtain the historical match count and process step number of all candidate processes, sort them from largest to smallest historical match count, and sort them from smallest to largest process step number when the historical match counts are the same. Select the top three candidate processes as the final list of suspected processes. If the total number is less than three, keep all of them. For each suspected process, obtain its stress location, stress distribution pattern, symmetry type, maximum stress value, and average stress value. Obtain the cause description of the corresponding stress type from the knowledge base. Obtain the recommended engineering points for investigation. Output the process step name, stress type name, stress location, stress distribution pattern, symmetry type, maximum stress value, average stress value, cause description, and recommended engineering points for investigation. The stress-driven process root cause localization module performs structured matching of the multidimensional features of the full-field stress tensor with the process stress mode knowledge base to accurately identify candidate processes. It also deeply integrates historical production line inspection data, constructing a quantitative historical matching count based on dimensions such as the frequency of recent anomalies and parameter consistency, enabling intelligent sorting of suspected process steps. Ultimately, it not only outputs high-confidence stress types and corresponding process steps but also simultaneously provides causal explanations and suggested investigation points, forming a closed-loop diagnosis from physical anomalies to engineering root causes. This module combines physical interpretability, data-driven approach, and engineering practicality, significantly improving defect tracing efficiency and process optimization response speed.

[0022] Example 2: The technical solution of this embodiment of the invention differs from that of Example 1 in that: like Figure 2 As shown, the executable process diagnostic report module generates a structured diagnostic report based on the full-field stress tensor and the sorting of suspected process steps. The report includes the maximum stress value and location, stress type, suspected process steps and their parameter optimization suggestions. The process of generating a structured diagnostic report based on the full-field stress tensor and the sorting of suspected process steps in the executable process diagnostic report module includes: Obtain the full-field stress tensor. For each measurement point, obtain two in-plane normal stress components and one in-plane shear stress component. Calculate the sum of squares of the two normal stress components, calculate the product of the two normal stress components, calculate the square of the shear stress component and multiply it by three. Subtract the product of the two normal stress components from the sum of squares, and add three times the square of the shear stress component. Take the square root of the obtained value to obtain the equivalent strength of the corresponding measurement point. Iterate through the equivalent strength of all measurement points, determine the maximum value and its corresponding wafer plane coordinates, obtain the stress type, obtain a list of process steps arranged in descending order of suspicion, and take the process step ranked first in the list as the suspected process step. Obtain the structured knowledge base, search for all records whose process step names are equal to the suspected process step and whose stress type names are equal to the stress type, use the countermeasures in these records as parameter tuning suggestions, and output a structured diagnostic report, including the maximum stress value, corresponding coordinates, stress type, suspected process step, and parameter tuning suggestions. The executable process diagnostic report module automatically calculates the equivalent strength based on the full-field stress tensor, accurately locates the maximum stress value and location, and combines high-confidence stress types and primary suspected process steps to intelligently extract targeted parameter tuning suggestions from the knowledge base. It generates a diagnostic report with a clear structure, physical accuracy, and direct guidance for production line correction, achieving an efficient closed loop from stress detection to process optimization.

[0023] Example 3: The technical solution of this embodiment of the invention differs from that of Example 1 and Example 2 in that: like Figure 3As shown, the product single-crystal defect locking method based on the principle of black cross light disappearance on silicon wafers includes the following steps: Step 1: Acquire a series of monochromatic light intensity images at different polarization analysis angles during orthogonal circular polarization interferometry imaging. Preprocess the acquired monochromatic light intensity images and finally output a preliminarily aligned standardized polarization image stack. Step 2: Extract the fracture endpoints of the black cross arms from the initially aligned standardized polarization image stack, obtain their local stress directions, verify whether the fracture tangent is perpendicular to the stress direction, retain the fracture endpoints that satisfy the perpendicular relationship as verified fracture stress points, and generate a guiding feature field. Step 3: Based on the photoelastic effect model, construct an optimized inversion framework that integrates guided feature field fitting, verified strong constraint of fracture stress points, and adaptive spatial smoothing to solve the full-field stress tensor; Step 4: Match the full-field stress tensor with the process stress mode knowledge base, and combine it with historical production line testing data to output the ranking of suspected process steps and the results of stress type identification. Step 5: Based on the full-field stress tensor and the sorting of suspected process steps, generate a structured diagnostic report containing the maximum stress value and location, stress type, suspected process steps and their parameter optimization suggestions; This method closely follows the physical mechanism of the disappearance of the black cross light, integrating polarization imaging, fracture verification, stress inversion, and process diagnosis to construct a complete chain from defect location to root cause tracing: it extracts credible fracture endpoints through high-precision polarization image preprocessing, combines photoelastic models and guided feature fields to achieve high-fidelity reconstruction of full-field stress, and intelligently identifies stress types and sorts suspected processes by linking process knowledge base and historical data. Finally, it outputs a structured report containing the location and type of maximum stress and optimization suggestions, accurately pinpointing defects in single-crystal silicon wafers and their manufacturing root causes, significantly improving the defect tracing efficiency and process control capabilities of semiconductor production lines.

[0024] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concept, should be covered within the scope of protection of the present invention.

Claims

1. A single-crystal defect locking system based on the principle of black cross light disappearance in silicon wafers, applied to a single-crystal silicon wafer production control platform, characterized in that... include: The black cross polarization acquisition and processing module acquires a series of monochromatic light intensity images at different polarization analysis angles during orthogonal circular polarization interferometry imaging, preprocesses the acquired monochromatic light intensity images, and finally outputs a preliminary aligned standardized polarization image stack. The Black Cross Fracture Stress Verification Module extracts the fracture endpoints of the Black Cross Arm from the initially aligned standardized polarization image stack, obtains their local stress direction, verifies whether the fracture tangent is perpendicular to the stress direction, retains the fracture endpoints that satisfy the perpendicular relationship as verified fracture stress points, and generates a guiding feature field. The Black Cross verification stress inversion module, based on the photoelastic effect model, constructs an optimized inversion framework that integrates guided feature field fitting, verified fracture stress point strong constraints, and adaptive spatial smoothing to solve the full-field stress tensor. The stress-driven process root cause localization module matches the full-field stress tensor with the process stress mode knowledge base, and combines it with historical production line testing data to output the ranking of suspected process steps and the results of stress type identification. The executable process diagnostic report module generates a structured diagnostic report based on the full-field stress tensor and the sorting of suspected process steps. The report includes the maximum stress value and location, stress type, suspected process steps, and parameter optimization suggestions.

2. The single-crystal defect locking system and method for silicon wafers based on the principle of black cross light disappearance as described in claim 1, characterized in that, The process of extracting the fracture endpoint of the black cross arm and obtaining its local stress direction in the black cross fracture stress verification module includes: Obtain a pre-aligned, standardized polarization image stack. For each position in the image, read the light intensity value at all polarization angles. Calculate the difference between the maximum and minimum light intensity at the position and divide it by the average light intensity to obtain the modulation depth. Multiply the light intensity at each polarization angle by the cosine and sine values, which are twice the corresponding polarization angle, and sum them to obtain two components. Take the square and square root of these components to obtain the polarization phase consistency. Calculate the average and standard deviation of the modulation depth and polarization phase consistency over the entire image. Normalize the two response quantities at each position. Substitute the normalized modulation depth and polarization phase consistency into the breakage confidence formula to calculate the breakage confidence at each position. A fracture confidence map is constructed using fracture confidence values ​​at all locations. Through adaptive threshold segmentation, pixel regions with fracture confidence values ​​exceeding the threshold are extracted to form an initial fracture mask. The mask is then skeletonized to obtain a single-pixel-wide skeleton centerline. The pixels on the skeleton centerline are traversed, and the number of skeleton connections in their eight neighborhoods is counted. If the number is one, it is marked as a fracture endpoint of the black cross arm. A local neighborhood window is defined for each fracture endpoint. Each pixel within the window is assigned a spatial weight and a signal weight. The joint weight is the product of the spatial weight and the signal weight. A weighted robust optimization objective is constructed using the Huber loss function. Through an iterative reweighted least squares algorithm, the average light intensity and the shared local stress direction of all pixels within the window are jointly solved, and the local stress direction corresponding to each fracture endpoint is output.

3. The single-crystal defect locking system and method for silicon wafers based on the principle of black cross light disappearance as described in claim 2, characterized in that, The process of verifying whether the fracture tangent is perpendicular to the stress direction in the black cross fracture stress verification module includes: For each fracture endpoint of the black cross arm, obtain the fracture tangent direction and the local stress direction, calculate the cosine of the angle between the two directions, take the absolute value, obtain the standard error of the fracture tangent direction and the standard error of the local stress direction, use the error propagation formula to calculate the standard error of the cosine value from the standard errors of the two directions, divide the absolute value of the cosine by the standard error of the cosine value to obtain the test statistic. The two-sided p-value of the test statistic is calculated based on the standard normal distribution. If the p-value is greater than 5%, the fracture tangent is determined to be perpendicular to the stress direction, and the perpendicularity determination result of the fracture endpoint is output.

4. The single-crystal defect locking system and method for silicon wafers based on the principle of black cross light disappearance as described in claim 3, characterized in that, The process of retaining the fracture endpoints that satisfy the vertical relationship as verified fracture stress points and generating the guiding feature field in the black cross fracture stress verification module includes: For each fracture endpoint that satisfies the perpendicular relationship, it is marked as a verified fracture stress point. The corresponding local stress direction is read, and the position and direction are paired to form the Dirichlet boundary condition. The local orientation field of the skeleton centerline and the fracture confidence map are read. The principal axis direction of the diffusion coefficient tensor is determined based on the local orientation field. The anisotropy intensity of the diffusion coefficient is modulated based on the fracture confidence map to generate the diffusion coefficient tensor field. Using all Dirichlet boundary conditions and diffusion coefficient tensor fields as input, the steady-state diffusion equation is established and solved to obtain a numerical solution in the entire image domain. The numerical solution in the entire image domain is then used as the output of the guiding feature field.

5. The single-crystal defect locking system and method for silicon wafers based on the principle of black cross light disappearance, as described in claim 1, is characterized in that... The process of constructing an optimized inversion framework that integrates guided feature field fitting, verified fracture stress point strong constraints, and adaptive spatial smoothing in the Black Cross verification stress inversion module includes: For each position, obtain the direction of the maximum tensile force and the direction of the guiding feature field at the corresponding position. Determine whether the two directions are the same or orthogonal. If they are neither the same nor orthogonal, and the corresponding position does not belong to the non-endpoint region of the fracture path, calculate the direction inconsistency penalty and add it to the optimization objective. For each verified fracture stress point, obtain the local stress direction and principal stress difference, and add the local stress direction and principal stress difference of the corresponding point as fixed value constraints to the optimization objective. For each pair of adjacent grid points, obtain the fracture confidence value of each point. If the fracture confidence of both points is zero, calculate the difference in stress state between the two points, square it and add it to the optimization objective. If the fracture confidence of any point is greater than zero, skip the smoothing penalty of the corresponding neighborhood pair. After traversing the entire field, form a complete optimization objective function, solve the optimization objective, and output the full field stress tensor inversion result.

6. The single-crystal defect locking system and method for silicon wafers based on the principle of black cross light disappearance, as described in claim 5, is characterized in that... The process of solving the full-field stress tensor in the Black Cross verification stress inversion module includes: Obtain a regular mesh, obtain the initial value of the stress tensor at each mesh point, obtain the constructed optimization objective function, obtain the preset convergence threshold, enter the iterative loop, construct a symmetric linear equation system using first-order and second-order information, solve the linear equation system to obtain the update amount of the stress tensor in the whole field, and calculate the global norm of the update amount. Determine if the global norm of the update quantity is less than the convergence threshold. If it is less, terminate the iteration and output the three components of the stress tensor of each grid point to form the global stress tensor.

7. The single-crystal defect locking system and method for silicon wafers based on the principle of black cross light disappearance as described in claim 1, characterized in that, The process of matching the full-field stress tensor with the process stress pattern knowledge base in the stress-driven process root cause localization module includes: Obtain the stress location, stress distribution pattern, symmetry type, maximum stress value, and average stress value corresponding to the full-field stress tensor. Obtain the process stress pattern knowledge base. For each record in the knowledge base, obtain its stress location, stress distribution pattern, symmetry type, and the lower and upper limits of the stress intensity range. Sequentially determine whether the measured stress location is equal to the recorded stress location, whether the measured stress distribution pattern is equal to the recorded stress distribution pattern, and whether the measured symmetry type is equal to the recorded symmetry type. At the same time, determine whether the maximum stress value is greater than or equal to the lower limit of the stress intensity range and less than or equal to the upper limit, and whether the average stress value is greater than or equal to the lower limit and less than or equal to the upper limit. If all judgment results are equal or within the corresponding range, the name of the process step and the name of the stress type currently recorded are added to the matching list. After traversing all records in the knowledge base, the matching list is output.

8. The single-crystal defect locking system and method for silicon wafers based on the principle of black cross light disappearance, as described in claim 7, is characterized in that... The process by which the stress-driven process root cause localization module combines historical production line testing data to output the ranking of suspected process steps and the determination of stress type includes: Obtain a list of successfully matched candidate processes. For each candidate process, obtain its process step name and stress type name. Obtain the process execution parameters of the current product and compare them with the process parameters of all stress anomaly events with the current process step name as the root cause in the historical test data. Determine if there are any completely identical records. Count the number of stress anomaly events with the current process step name as the root cause in the most recent month and determine if it is not less than one. Based on the results of the three checks, count the number of successful items as the historical match count. Obtain the historical match count and process step number of all candidate processes. If the historical match counts are the same, sort them by process step number from smallest to largest, and select the top three candidate processes as the final list of suspected processes. For each suspected process, obtain its stress location, stress distribution pattern, symmetry type, maximum stress value, and average stress value. Obtain the cause description of the corresponding stress type from the knowledge base, obtain the recommended engineering points for investigation, and output the process step name, stress type name, stress location, stress distribution pattern, symmetry type, maximum stress value, average stress value, cause description, and recommended engineering points for investigation.

9. The single-crystal defect locking system and method for silicon wafers based on the principle of black cross light disappearance as described in claim 1, characterized in that, The process of generating a structured diagnostic report based on the full-field stress tensor and the sorting of suspected process steps in the executable process diagnostic report module includes: Obtain the full-field stress tensor. For each measurement point, obtain two in-plane normal stress components and one in-plane shear stress component. Calculate the sum of squares of the two normal stress components, calculate the product of the two normal stress components, calculate the square of the shear stress component and multiply it by three, subtract the product of the two normal stress components from the sum of squares, and add three times the square of the shear stress component. Take the square root of the obtained value to obtain the equivalent strength of the corresponding measurement point. Iterate through the equivalent strength of all measurement points, determine the maximum value and its corresponding wafer plane coordinates, obtain the stress type, and obtain a list of process steps sorted in descending order of suspicion. Obtain the structured knowledge base, search for all records whose process step names are equal to the suspected process step and whose stress type names are equal to the stress type, use the countermeasures in these records as parameter tuning suggestions, and output a structured diagnostic report.

10. A product single-crystal defect locking method based on the principle of black cross light disappearance in silicon wafers, applied to the product single-crystal defect locking system based on the principle of black cross light disappearance in silicon wafers as described in any one of claims 1-9, characterized in that, Includes the following steps: Step 1: Acquire a series of monochromatic light intensity images at different polarization analysis angles during orthogonal circular polarization interferometry imaging. Preprocess the acquired monochromatic light intensity images and finally output a preliminarily aligned standardized polarization image stack. Step 2: Extract the fracture endpoints of the black cross arms from the initially aligned standardized polarization image stack, obtain their local stress directions, verify whether the fracture tangent is perpendicular to the stress direction, retain the fracture endpoints that satisfy the perpendicular relationship as verified fracture stress points, and generate a guiding feature field. Step 3: Based on the photoelastic effect model, construct an optimized inversion framework that integrates guided feature field fitting, verified strong constraint of fracture stress points, and adaptive spatial smoothing to solve the full-field stress tensor; Step 4: Match the full-field stress tensor with the process stress mode knowledge base, and combine it with historical production line testing data to output the ranking of suspected process steps and the results of stress type identification. Step 5: Based on the full-field stress tensor and the sorting of suspected process steps, generate a structured diagnostic report containing the maximum stress value and location, stress type, suspected process steps and their parameter optimization suggestions.

Images