Positioning mark recognition method, semiconductor inspection system, and storage medium

Abstract

The application provides a positioning mark recognition method, a semiconductor detection system and a storage medium. The method comprises the following steps: acquiring a plurality of phase-shifted interference images representing the surface of a measured object; generating an edge contour enhanced image of the measured object and extracting a reference circular contour as a spatial reference system based on the light intensity change information of each corresponding pixel in the plurality of phase-shifted interference images; combining a dynamic threshold segmentation algorithm and a geometric fitting algorithm to determine the accurate center coordinates of the positioning mark located at the edge of the measured object, and finally calculating the azimuth angle of the positioning mark according to the geometric relationship between the accurate center coordinates and the reference circular contour, so that the center coordinates of the positioning mark are recognized at sub-pixel level precision.

Description

Technical Field

[0001] This application relates to the fields of semiconductor manufacturing and precision testing technology, specifically to a method for identifying positioning markers, a semiconductor testing system, and a storage medium. Background Technology

[0002] In semiconductor manufacturing, wafers must be precisely positioned and aligned before entering critical processes such as photolithography, ion implantation, and dicing. To achieve this, a specific physical mark is typically fabricated on the wafer edge as an orientation reference; this mark is commonly referred to as a positioning marker. Mainstream positioning markers include a "V" notch or an orientation flat. The azimuth angle of this marker determines the relative orientation between the chip pattern and the silicon crystal lattice. The accuracy and reliability of its identification directly affect the overlay accuracy of subsequent processes and the electrical performance of the final device.

[0003] Currently, wafer positioning mark detection primarily relies on machine vision systems, with one mainstream method based on non-interference imaging principles. These methods typically employ highly uniform bright-field or backlighting systems, coupled with industrial cameras (such as CCD or CMOS), to acquire single-frame two-dimensional grayscale images of the wafer. Image processing techniques, such as Canny edge detection, Hough transform, or template-based matching algorithms, are then used to identify the overall wafer outline and the geometric features of the positioning mark from the image, ultimately calculating the mark's angle relative to a reference direction. However, this single-frame static image-based recognition method heavily depends on ideal and stable lighting conditions. In real-world industrial environments, it is prone to failure due to uneven illumination, interference fringes caused by multilayer film structures on the wafer surface, and the inherent low contrast of the marking area, leading to incomplete edge feature extraction or blurred positioning. Furthermore, the algorithm's robustness is challenged when faced with non-ideal conditions such as scratches or particulate contamination on the wafer surface, potentially resulting in misidentification or missed identification.

[0004] Another type of detection method utilizes the principle of interferometry for geometric identification. This method acquires a single-frame interferogram of the wafer using an interferometer, extracting edge features by leveraging the sensitivity of interference fringes to surface contours. However, a single-frame interferogram only reflects the fringe distribution at a specific phase, failing to fully utilize the phase change information contained in multiple frames of interferometric images. In practical applications, this approach has limitations in edge enhancement and noise resistance. When the interference fringes in a single-frame interferogram happen to be at a specific phase, the edge region may exhibit a low modulation amplitude, making contour extraction difficult. Furthermore, single-frame interferograms have limited noise suppression capabilities, making stable identification difficult under low signal-to-noise ratio conditions.

[0005] It is evident that the two existing recognition methods based on single-frame images (including non-interferometric and interferometric images) are limited in accuracy by the signal-to-noise ratio of the image and the sensitivity of traditional edge detection algorithms to noise. They also have inherent biases in locating the geometric center of the marker (especially small notches), making it difficult to meet the increasingly stringent sub-micron alignment tolerance requirements of advanced processes. At the same time, the increased image data volume in pursuit of high resolution also puts pressure on the real-time performance of traditional processing architectures. Summary of the Invention

[0006] This application provides a method for identifying positioning markers, a semiconductor inspection system, and a storage medium, aiming to solve the technical problems of existing wafer positioning marker identification methods based on single-frame static / interference images, such as sensitivity to lighting conditions, weak ability to extract small or low-contrast marker features, limited positioning accuracy, insufficient environmental robustness, and difficulty in meeting the processing speed requirements of high-speed online inspection.

[0007] In a first aspect, embodiments of this application provide a method for identifying a location marker, comprising:

[0008] Acquire multi-frame phase-shifting interferometric images characterizing the surface of the object under test;

[0009] Based on the light intensity change information of each corresponding pixel in the multi-frame phase-shifting interferometric image, an enhanced edge contour image of the object under test is generated. The enhanced image obtained in this way not only highlights the response of the edge region, but also suppresses background noise and interference from uniform regions, which can effectively improve the contrast of the object's contour.

[0010] Extract the reference circular contour of the object under test from the enhanced edge contour image of the object under test to establish a spatial reference system;

[0011] Based on the phase difference information between the reference circular contour and the multi-frame phase-shifting interferometric images, feature analysis is performed on the edge of the object under test to determine the precise center coordinates of the positioning mark located on the edge of the object under test.

[0012] Based on the geometric relationship between the precise center coordinates of the positioning mark and the reference circular outline, the azimuth angle of the positioning mark is determined, achieving sub-pixel level accuracy in identifying the center coordinates of the positioning mark.

[0013] In some embodiments, generating an enhanced edge contour image of the object under test based on the light intensity change information of corresponding pixels in the multi-frame phase-shifting interferometric images includes:

[0014] Calculate the difference between the maximum and minimum grayscale values ​​corresponding to each pixel position in the multi-frame phase-shifting interferometric image to obtain a difference image;

[0015] The difference image is normalized, and the normalized pixel values ​​are mapped to a standard grayscale range to obtain the edge contour enhancement image.

[0016] In some embodiments, extracting the reference circular contour of the object under test from the enhanced edge contour image of the object under test includes:

[0017] The edge contour enhancement image is subjected to Hough circle transform to map the edge information represented by the edge contour enhancement image to the center-radius parameter space;

[0018] Based on the coordinates of each pixel in the edge information and the preset radius parameter range, a voting accumulation is performed on potential circular contours in the parameter space;

[0019] Based on the cumulative voting results, the reference circular profile and its geometric parameters of the object under test are determined.

[0020] In some embodiments, determining the reference circular contour and its geometric parameters of the object under test based on the cumulative voting results includes:

[0021] When the cumulative voting results indicate that there are multiple potential circular contours that meet the preset conditions, the potential circular contour with the highest cumulative intensity is selected as the reference circular contour of the object under test.

[0022] The center coordinates and radius corresponding to the selected potential circular profile are determined as the geometric parameters of the reference circular profile.

[0023] In some embodiments, determining the reference circular contour and its geometric parameters of the object under test based on the cumulative voting results includes:

[0024] When the cumulative voting results indicate that no potential circular contours meeting the preset conditions have been detected, the preset radius parameter range is adjusted.

[0025] Based on the adjusted radius parameter range, the edge contour enhancement image is subjected to Hough circle transformation and voting accumulation again until a reference circular contour and its geometric parameters that meet the preset conditions are determined.

[0026] In some embodiments, the step of performing feature analysis on the edge of the object under test to determine the precise center coordinates of the positioning marker located on the edge of the object under test includes:

[0027] The estimated center coordinates of the positioning mark are calculated based on the center coordinates, radius, and preset reference angle of the reference circular outline.

[0028] A local image region is cropped centered on the estimated center coordinates, and the local image region is filtered and binarized to obtain a binary image that highlights the shape of the positioning mark.

[0029] Extract all closed contours from the binary image and select the contour containing the most pixels as the candidate contour.

[0030] Based on the candidate contour, dynamic thresholding and circle fitting are performed to determine the precise center coordinates.

[0031] In some embodiments, determining the precise center coordinates based on the candidate contour through dynamic thresholding and circle fitting includes:

[0032] Calculate the distance from each pixel in the candidate contour to the center of the reference circular contour;

[0033] Based on the calculation results, a dynamic distance threshold is determined, and points with a distance less than the dynamic distance threshold are selected to form an internal point set.

[0034] Perform a circle fitting on the set of internal points, and use the center coordinates of the fitted circle as the precise positioning center coordinates of the positioning marker.

[0035] In some embodiments, determining the azimuth angle of the positioning marker based on the geometric relationship between the precise center coordinates of the positioning marker and the reference circular outline includes:

[0036] Calculate the direction vector connecting the precise positioning center coordinates of the positioning mark with the center coordinates of the reference circular outline;

[0037] Calculate the angle between the direction vector and a preset reference direction vector;

[0038] Convert the included angle into the azimuth angle of the positioning mark.

[0039] In some embodiments, at least one computational step in the process of generating edge contour enhancement images, vote accumulation, or determining precise center coordinates is accelerated by parallel computing using a graphics processor, thereby improving image preprocessing speed and meeting the real-time requirements of high-speed online inspection in semiconductor production lines.

[0040] Secondly, embodiments of this application provide a semiconductor detection system, comprising:

[0041] A phase-shifting interferometer is used to acquire multi-frame phase-shifting interference images of the surface of an object under test.

[0042] An image processing device, communicatively connected to the phase-shifting interferometer, is configured to perform a method for identifying positioning marks as described in any embodiment of the first aspect, to determine the center coordinates and azimuth of positioning marks formed on the edge of the object under test based on the multi-frame phase-shifting interferometric images.

[0043] In some embodiments, the image processing apparatus includes at least one graphics processor for accelerating the execution of at least one step among generating an edge contour enhancement image, vote accumulation, or circle fitting.

[0044] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer-executable program or instructions, which, when executed by a processor, are used to implement the location identifier identification method as described in any embodiment of the first aspect.

[0045] The location marker recognition method and semiconductor detection system provided in this application utilize the phase and intensity change information contained in multi-frame phase-shifting interferometric images to generate an edge contour enhanced image and extract a reference circular contour as a spatial reference system. Based on this, combined with phase difference information, a dynamic threshold segmentation and geometric fitting algorithm is used to accurately calculate the center coordinates of the location marker at the sub-pixel level, and finally solve its azimuth angle relative to the reference circle, achieving sub-pixel accuracy in recognizing the center coordinates of the location marker. This application overcomes the sensitivity of traditional single-frame non-interferometric images to lighting conditions and the insufficient contrast caused by the single phase information in traditional single-frame interferometric images, improving the robustness of the algorithm to noise and edge blurring. At the same time, by introducing a strategy combining adaptive dynamic thresholding and geometric fitting, it can effectively cope with non-ideal conditions such as scratches and contamination on the wafer surface, avoiding misidentification or missed identification.

[0046] Furthermore, while ensuring high precision, this technical solution improves processing efficiency and real-time performance through parallel acceleration of the graphics processing unit (GPU), meeting the stringent requirements of semiconductor production lines for real-time online inspection. It provides a reliable guarantee for the precise alignment of subsequent processes such as photolithography, and can be adapted to wafers of different sizes and inspection equipment, exhibiting universality and scalability.

[0047] In addition, this application also provides a computer-readable storage medium that has the same beneficial effects as the above-described location identification method. Attached Figure Description

[0048] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0049] Figure 1 This is a schematic diagram of the structure of a semiconductor detection system provided in one embodiment of this application.

[0050] Figure 2 This is a schematic diagram of the optical structure of a phase-shifting interferometer provided in one embodiment of this application.

[0051] Figure 3 This is a flowchart illustrating a method for identifying a location identifier according to one embodiment of this application.

[0052] Figure 4 A flowchart of a location identifier identification method provided in another embodiment of this application.

[0053] Figure 5 A flowchart of a location identifier identification method provided in another embodiment of this application.

[0054] Figure 6 A flowchart of a location identifier identification method provided in another embodiment of this application.

[0055] Figure 7 A flowchart of a location identifier identification method provided in another embodiment of this application.

[0056] Figure 8 A flowchart of a location identifier identification method provided in another embodiment of this application.

[0057] Figure 9 A flowchart of a location identifier identification method provided in another embodiment of this application.

[0058] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0059] The present application will now be described in further detail with reference to the accompanying drawings and specific embodiments. Similar elements in different embodiments are referred to by related similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the present application. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present application are not shown or described in the specification. This is to avoid obscuring the core parts of the present application with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0060] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0061] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, without limiting the number of objects; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship. Unless otherwise specified, the terms "connection" and "linkage" used in this application include both direct and indirect connections (linkages).

[0062] As described in the background section, the electrical performance, mechanical strength, and yield of subsequent processes of semiconductor devices strongly depend on the precise orientation of the crystal lattice. Therefore, before inspecting semiconductor wafers, which have strict requirements for crystal orientation, it is necessary to align the physical positioning marks (such as notches or flat edges) of their edges with the crystal orientation. The wafer manufacturing process gives it a specific crystal orientation (such as...). <110> or <100> The wafer's orientation is determined by the crystal orientation, and the positioning markers on the edge are set to maintain a fixed geometric relationship with this orientation during dicing. Before core processes such as photolithography, etching, and ion implantation, the equipment can adjust the wafer to the preset crystal orientation by identifying and aligning this physical marker. This ensures that the thousands of chips subsequently fabricated on the wafer surface have consistent and correct key characteristics such as carrier mobility, oxidation rate, and piezoelectric effect. It also avoids cracks along the cleavage plane during dicing caused by crystal orientation deviation, thereby guaranteeing product performance and manufacturing yield.

[0063] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0064] Figure 1 This is a schematic diagram of the structure of a semiconductor detection system provided in one embodiment of this application. Figure 1As shown, the semiconductor detection system provided in this embodiment includes at least a phase-shifting interferometer 11 and an image processing device 12.

[0065] In this embodiment, the phase-shifting interferometer 11 is used to acquire multiple frames of phase-shifting interferometric images of the surface of the object under test (such as a semiconductor wafer). Its working principle typically includes: a light beam emitted from a light source is split into a reference beam and a measurement beam by a beam splitter; the measurement beam is projected onto the surface of the object under test and reflected, interfering with the reference beam reflected by a reference mirror to form an interference fringe pattern containing information about the height of the object's surface. The phase-shifting interferometer 11 introduces a known phase change in the optical path through its internal phase shifter, synchronously triggering an image detector to continuously acquire multiple frames of interferometric images with a fixed phase difference, thereby obtaining a sequence of multiple frames of phase-shifting interferometric images characterizing the surface morphology of the object under test.

[0066] The image processing device 12 is a computing device communicatively connected to the phase-shifting interferometer 11, and typically includes one or more processors and a storage unit storing executable computer program instructions. The image processing device 12 is configured to execute a method for identifying positioning marks by processing received multi-frame phase-shifting interferometric images to determine the center coordinates and azimuth angle of positioning marks formed on the edge of the object under test.

[0067] In this embodiment, the processing of the received multi-frame phase-shifting interferometric images includes: First, generating an enhanced edge contour image using the light intensity change information of each corresponding pixel in the image sequence; Second, extracting the reference circular contour of the object under test from the enhanced edge contour image of the object under test, specifically by using Canny operator edge detection, sub-pixel edge fitting, or Hough circle transform to extract the reference circular contour of the object under test from the enhanced image; Then, based on the phase difference information between the reference circular contour and the multi-frame phase-shifting interferometric images, performing feature analysis on the edge of the object under test to determine the precise center coordinates of the positioning mark located on the edge of the object under test, specifically by combining the position information of the reference contour to lock the edge region where the positioning mark is located in the interferometric image, using the phase change characteristics between the mark and the background in the phase difference information, employing dynamic threshold segmentation technology to coarsely extract the mark from the complex edge background, performing sub-pixel level geometric fitting on the segmented mark region, thereby calculating the center coordinates of the positioning mark on the edge of the object under test; Finally, calculating the azimuth angle of the positioning mark relative to the preset reference direction based on the geometric relationship between the center coordinates and the reference circle center.

[0068] In some embodiments, at least one of the processors in the image processing apparatus 12 includes an image processor (i.e., a GPU). By integrating acceleration hardware such as a GPU, computationally intensive steps such as image enhancement, Hough transform voting, and circle fitting are accelerated in parallel to meet the real-time and high-precision requirements of online inspection in semiconductor production lines.

[0069] Graphics processing units (GPUs) were originally designed for graphics rendering. Their core advantage lies in their hundreds or thousands of computing cores, enabling them to execute a large number of parallel threads simultaneously. This allows them to exhibit throughput and computational efficiency far exceeding that of traditional central processing units (CPUs) when handling highly parallel computational tasks. This massively parallel architecture makes GPUs particularly suitable for scenarios that require repeating the same operations on massive amounts of data, such as matrix operations, image processing, physics simulations, and deep learning training. By offloading computationally intensive tasks from the CPU to the GPU, processing time can be significantly reduced, enabling high-performance computing.

[0070] It should be noted that those skilled in the art will understand that the specific hardware structure and implementation of the phase-shifting interferometer 11 and the image processing device 12 can be achieved using existing conventional component configurations. Therefore, their specific circuit configurations or internal module connections will not be elaborated upon here. The core inventive point of this application lies in the positioning marker recognition method executed by the image processing device 12, namely, how to accurately determine the center coordinates and azimuth angle of the edge positioning marker of the measured object based on multiple frames of phase-shifting interferometric images acquired by the phase-shifting interferometer 11 through a combination of enhancement processing, reference contour extraction, dynamic threshold segmentation, and geometric fitting. The core technical solution of this application will be described in detail below with reference to the accompanying drawings.

[0071] Figure 2 This is a schematic diagram of the optical structure of a phase-shifting interferometer provided in one embodiment of this application. Figure 2 As shown, the phase-shifting interferometer 11 provided in this embodiment includes at least a light source assembly 111, an optical path assembly 112, and an imaging assembly 113.

[0072] In this embodiment, the light source assembly 111 is used to provide a coherent illumination beam, the optical path assembly 112 is used to guide the illumination beam to the surface of the object under test and receive the beam reflected back from the surface of the object under test to form interference light, and the imaging assembly 113 is used to acquire the interference light and generate an interference image.

[0073] In some embodiments, the optical path assembly 112 includes at least a beam splitter 1121, a collimating lens 1122, a reference lens 1123, and a relay lens 1124.

[0074] The illumination beam emitted from the light source assembly 111 is reflected / transmitted through the beam splitter 1121 and collimating lens 1122 in sequence before being incident on the reference lens 1123. Part of the beam transmitted through the reference lens 1123 forms the test light and is incident on the surface of the object under test. Part of the beam reflected by the reference lens group forms the reference light. The test light is reflected by the surface of the object under test to form the reflected light. After the reflected light interferes with the reference light, it is transmitted along the original light path through the collimating lens 1122, beam splitter 1121, and relay lens 1124 before being incident on the imaging assembly 113.

[0075] In some embodiments, the optical path assembly 112 further includes a quarter-wave plate 1125 disposed between the beam splitter 1121 and the collimating lens 1122, and an aperture 1126 disposed between the relay lens 1124 and the imaging assembly 113.

[0076] In the phase-shifting interferometer 11, the illumination beam emitted from the light source assembly 111 is split by the beam splitter 1121 and then incident on the quarter-wave plate 1125, which can convert it into circularly polarized light. That is, the test light incident on the surface of the object under test is circularly polarized light. After the test light is reflected by the surface of the object under test, the rotation direction will be reversed. When a stable interference is formed with the reference light, the fringe blurring caused by the abrupt change in reflection phase can be avoided.

[0077] The aperture 1126 set between the relay mirror 1124 and the imaging component 113 is used to precisely limit the beam aperture entering the imaging system, ensuring that only the light energy in the effective interference area (i.e. the part where the test light and the reference light overlap) can pass through, avoiding light interference in the non-interference area at the edge, thereby improving the contrast of the interference fringes.

[0078] It should be noted that the positions of the optical elements in the optical path assembly 112 of the phase-shifting interferometer 11 can be adjusted according to the position of the light source assembly 111. Other optical components can also be set in it to fine-tune the optical path. The main optical functions are all within the protection scope of this embodiment, and no further restrictions are imposed on them.

[0079] Figure 3 This is a flowchart illustrating a location identifier identification method provided in one embodiment of this application. Figure 3 As shown, the location identifier identification method provided in this embodiment specifically includes the following steps:

[0080] Step S31: Obtain multi-frame phase-shifting interferometric images that characterize the surface of the object under test.

[0081] To achieve high-precision identification and crystal orientation alignment of wafer positioning marks, it is first necessary to acquire data that can finely characterize the surface microstructure. Specifically, using equipment such as a phase-shifting interferometer, optical phase-shifting interferometry is employed. By controlling the optical path difference between the reference light and the measurement light, a known phase change is introduced onto the surface of the object under test. Multiple interference images (usually more than three frames) are continuously acquired at specific step sizes (such as equal-interval phase shifts). Phase shifting is mainly achieved through two techniques: mechanical phase shifting and wavelength-tuned phase shifting. Mechanical phase shifting typically uses a piezoelectric ceramic actuator to move the reference mirror or the object under test (such as the wafer) along the optical axis, thereby precisely changing the optical path difference. Wavelength-tuned phase shifting indirectly achieves phase shifting by changing the wavelength of the light source and utilizing the relationship between wavelength change and optical path difference. Regardless of the method used, the core objective is to obtain a sequence of interference images with a fixed phase difference. These images record the fringe patterns formed by the interference of reflected light from the surface of the object under test. Because the phase of each frame is different, the interference fringes are shifted in space, thus forming a sequence of images containing information about the object's surface height and edge contour features. This sequence is the basis for subsequent quantitative calculations of phase information.

[0082] Step S32: Based on the light intensity change information of each corresponding pixel in the multi-frame phase-shifting interferometric image, generate an enhanced edge contour image of the object under test.

[0083] During phase shifting, the grayscale value of the same pixel fluctuates with phase change across different frames, especially at object edges where the reflected light intensity changes more drastically due to surface tilt. Therefore, by analyzing the intensity fluctuation of each pixel over time (i.e., phase shift steps) in multi-frame interferometric images, the phase gradient or modulation information of each pixel can be calculated. Algorithms can then be used to enhance this high-frequency changing signal, resulting in an enhanced image with clear edges and distinct contours corresponding to the measured object. This enhanced image not only highlights the response in edge regions but also suppresses background noise and interference from uniform regions, effectively improving the contrast of the measured object's contour and facilitating subsequent feature extraction for outer circle extraction and notch localization.

[0084] Step S33: Extract the reference circular contour of the object being measured from the enhanced edge contour image of the object being measured.

[0085] After obtaining a clear edge image, since the object being measured is usually designed to be circular or approximately circular (such as a wafer, lens ring, etc.), the outer boundary of the object being measured can be identified and delineated through image segmentation and geometric feature extraction algorithms. The algorithm will select the contour curve that best matches the characteristics of a circle based on constraints such as the continuity, curvature, and closure of the edge, and define it as the reference circular contour. Then, the center coordinates and radius of the reference circle will be calculated from these edge points, which will serve as the geometric reference for subsequent positioning and marking measurements.

[0086] In some embodiments, image segmentation and geometric feature extraction algorithms include Canny operator edge detection, Hough circle transform, or subpixel edge fitting. Canny operator edge detection is a multi-level edge detection algorithm. First, it smooths the image using Gaussian filtering to suppress noise. Then, it calculates the pixel gradient magnitude and direction and uses non-maximum suppression to refine the edges. Finally, it extracts continuous, single-pixel-wide edges through double-threshold detection and connectivity analysis. This method can reliably extract volumetric contours, providing high-quality edge points for geometric feature analysis. Hough circle transform is a circle detection method based on a voting mechanism. It maps edge points in image space to a three-dimensional parameter space consisting of the center coordinates and radius. Each edge point votes on the parameters of the circles it might pass through, and the peak value in the accumulator corresponds to the detected circle. This algorithm can robustly identify circular contours in images, effectively extracting the outer circle reference of a wafer even with partial missing parts or noise interference. Subpixel edge fitting, building upon pixel-level edge detection, utilizes the grayscale distribution of pixels near the edge or the coordinates of edge points for interpolation or curve fitting (such as Gaussian fitting, parabolic fitting, or the moment method) to improve edge localization accuracy to the subpixel level. This method is commonly used in scenarios requiring high detection accuracy and can provide higher-resolution geometric features for measurement.

[0087] Step S34: Based on the phase difference information between the reference circular contour and the multi-frame phase-shifting interferometric images, perform feature analysis on the edge of the object under test to determine the precise center coordinates of the positioning mark located on the edge of the object under test.

[0088] After determining the coordinate system of the reference circular contour of the object under test, the marker area can be further located by combining phase difference information. Specifically, firstly, the edge region where the marker is located is locked in the interferometric image by combining the position information of the reference contour; then, the phase abrupt change characteristics between the marker and the background in the phase difference information (i.e., obtained in step S220) are used to perform feature analysis on the edge of the object under test, including coarse extraction of the marker from the complex edge background and noise removal. In some embodiments, the feature analysis of the edge of the object under test adopts a strategy of dynamic threshold adaptive segmentation and geometric model fitting to calculate the geometric center coordinates of the marker. By fusing the strategy of dynamic threshold adaptive segmentation and geometric model fitting, extremely high positioning accuracy can be obtained even when the marker has slight deformation or occlusion.

[0089] Step S35: Determine the azimuth angle of the positioning mark based on the geometric relationship between the precise center coordinates of the positioning mark and the reference circular outline.

[0090] After obtaining the precise center coordinates of the positioning mark on the edge of the object under test, it needs to be converted into actual angle values ​​to align or adjust the orientation of the object before inspection. Specifically, a polar coordinate system is established with the center of the reference circular contour as the origin, or a reference direction is established through feature points on the reference contour. The angular offset of the positioning mark center coordinates obtained in step S24 relative to this reference system is calculated. By calculating using the arctangent function or vector angle, the azimuth angle of the positioning mark relative to the center of the object is accurately obtained, thus yielding the actual azimuth angle of the notch. This angle directly reflects the current orientation of the object under test, such as the wafer, and provides a precise orientation basis for automated alignment, assembly, or defect detection in the inspection process.

[0091] In summary, the location marker recognition method provided in this embodiment utilizes the phase and intensity change information contained in multi-frame phase-shifting interferometric images to generate an edge contour enhancement image and extract a reference circular contour as a spatial reference system. This process enhances the contour contrast of location markers such as edges and small gaps, overcoming the sensitivity of traditional single-frame non-interferometric images to lighting conditions and the limitations of traditional single-frame interferometric images due to insufficient contrast caused by single phase information. On this basis, dynamic threshold segmentation and geometric fitting algorithms are used in combination with phase difference information to accurately calculate the center coordinates of the location marker at the sub-pixel level, and finally solve its azimuth angle relative to the reference circle. This not only improves the robustness of the algorithm to noise and edge blurring, but also achieves sub-pixel accuracy of the center coordinates of the location marker.

[0092] Furthermore, the location marker recognition method provided in this embodiment can be implemented in parallel using a graphics processing unit (GPU), fully leveraging its massively parallel architecture. During image enhancement, the GPU can simultaneously perform light intensity variation analysis and normalization on millions of pixels, significantly improving image preprocessing speed. In edge detection, the GPU can process parameter mapping and accumulation operations on a large number of edge points in parallel, shortening the time required for circular detection. In dynamic threshold segmentation and geometric fitting, the GPU is also suitable for candidate point distance calculation and parallel iterative optimization. Based on this parallel design, the processing time of the entire location marker recognition process can be significantly compressed, thereby meeting the real-time requirements of high-speed online detection in semiconductor production lines, while simultaneously improving system throughput and response speed without sacrificing accuracy.

[0093] Figure 4 A flowchart illustrating a location identifier identification method provided in another embodiment of this application. For example... Figure 4 As shown, in the above embodiment, step S32, generating an enhanced edge contour image of the object under test based on the light intensity change information of each corresponding pixel in the multi-frame phase-shifting interferometry image, specifically includes the following steps:

[0094] Step S321: Calculate the difference between the maximum and minimum gray values ​​corresponding to each pixel position in the multi-frame phase-shifting interference image to obtain the difference image.

[0095] For any fixed pixel in an interference image sequence, due to the fixed phase difference between each frame, the gray value of the same pixel will periodically change as the interference fringes move. For flat areas of the object's surface, this change is relatively gradual; however, in steep areas such as edges or gaps, the intensity of reflected light is significantly affected by the tilt angle, resulting in a significantly larger fluctuation in gray value across different frames. Therefore, by traversing each pixel position and calculating the difference between the maximum and minimum gray values ​​across multiple frames, this fluctuation amplitude can be quantified. Through difference calculation, the edge responses originally scattered across different phase frames are concentrated into a single difference image. The larger the difference, the higher the probability of a sudden change in height at that location, thus highlighting the object's edge contours from the background while suppressing the influence of background noise and uniform regions.

[0096] Step S322: Normalize the difference image and map the normalized pixel values ​​to the standard grayscale range to obtain an edge contour enhancement image.

[0097] Because pixel values ​​in different regions of the difference image may have a large dynamic range, directly using it may lead to instability or insufficient contrast in subsequent processing. Therefore, the difference image is first normalized. This is typically done by subtracting the global minimum value from the grayscale value of each pixel in the image, then dividing by the difference between the global maximum and minimum values, and finally scaling linearly, i.e., (vv... min ) / (v max -v min This process compresses all pixel values ​​to the [0,1] range, eliminating the influence of extreme values ​​that may exist in the original difference image and making the data distribution more concentrated. Subsequently, the normalized values ​​are linearly mapped to a standard grayscale display range, such as the standard 8-bit grayscale range [0,255], and converted to the corresponding data type, ultimately generating an edge contour enhancement image conforming to conventional image processing formats. In this image, the contrast of the wafer edge and positioning marker area is improved, which is the basis for subsequent baseline circular contour extraction and positioning marker recognition.

[0098] Figure 5 A flowchart illustrating a location identifier identification method provided in another embodiment of this application. For example... Figure 5 As shown in the above embodiment, step S33, extracting the reference circular contour of the object under test from the enhanced edge contour image of the object under test, specifically includes the following steps:

[0099] Step S331: Perform Hough circle transform on the edge contour enhancement image to map the edge information represented by the edge contour enhancement image to the center-radius parameter space.

[0100] Since circular edge detection in image space is often affected by noise and discontinuities, direct fitting is difficult. Therefore, it is necessary to map edge points in image space to a common parameter space. The core of the Hough circle transform is to map edge points in image space to a three-dimensional parameter space consisting of the center coordinates (a, b) and radius r. In edge contour enhancement images, each edge pixel carries its spatial position information. Through the Hough circle transform, each edge pixel (x, y) in the edge contour enhancement image is mapped to a three-dimensional parameter space consisting of the center coordinates (a, b) and radius r according to the mathematical expression of a circle (xa)² + (yb)² = r². Through this mapping process, feature points that originally belonged to the same circle in image space will converge in the same parameter space, transforming the geometric feature detection problem in the image into a peak search problem in the parameter space, which facilitates subsequent circle recognition.

[0101] Step S332: Based on the coordinates of each pixel in the edge information and the preset radius parameter range, accumulate votes for potential circular contours in the parameter space.

[0102] After mapping, a Hough circle transform is performed. First, a reasonable radius parameter range is set based on prior knowledge of the object being measured. This radius parameter range can be the fluctuation range of the standard radius of a wafer, such as the minimum and maximum radii of the circle to be detected. By narrowing the radius range, the search space is reduced, computational efficiency is improved, and the possibility of detection errors is reduced. Subsequently, the Hough circle transform process is implemented through voting accumulation. For each effective edge pixel (x, y) and each possible radius value r in the edge contour enhancement image, its corresponding possible circle center coordinates (a, b) are calculated, and the corresponding position (a, b, r) in the parameter space is accumulated. Whenever a specific (a, b, r) combination is supported by a certain edge point, a "vote" is performed at the corresponding position in the parameter space. This process is equivalent to letting each edge point "vote" for the circle it may belong to. After traversing all edge points, the accumulated votes for each point (a, b, r) in the parameter space reflect the probability that a circle with that point as its center and that value as its radius exists in the image space.

[0103] Step S333: Based on the cumulative voting results, determine the reference circular profile and its geometric parameters of the object being measured.

[0104] After the voting is completed, the distribution of cumulative values ​​in the parameter space reflects the support for different circular hypotheses. Typically, the parameter location corresponding to the true circular contour will form a significant local peak, meaning the cumulative count at that location (a, b, r) is much higher than the surrounding area. By analyzing the voting accumulation results in the parameter space, one or more cumulative peak points with the highest number of votes are identified. These peak points indicate that they have received the most edge pixel support in the parameter space, and therefore are most likely to correspond to a true circular contour in the image space. Peak points that meet preset threshold conditions are selected, and their corresponding coordinates (a, b) and radius r are used as the center coordinates and radius of the reference circle of the measured object, respectively, thus accurately extracting the reference circular contour.

[0105] Figure 6 A flowchart illustrating a location identifier identification method provided in another embodiment of this application. For example... Figure 6 As shown, in the above embodiment, step S333, determining the reference circular contour and its geometric parameters of the object under test based on the cumulative voting results, specifically includes the following steps:

[0106] Step S3331: When the cumulative voting results indicate that there are multiple potential circular contours that meet the preset conditions, select the potential circular contour with the highest cumulative intensity as the reference circular contour of the object being measured.

[0107] In practical applications of the Hough circle transform, image noise, edge discontinuities, or the presence of multiple approximately circular features on the object itself (such as chamfers on wafer edges or arc-shaped structures near positioning markers) can lead to multiple cumulative peaks exceeding a preset threshold in the parameter space, resulting in multiple potential circular contour candidates. To select the one that best reflects the baseline circle of the measured object, this step introduces an optimization mechanism based on cumulative intensity. Cumulative intensity reflects the number of edge pixels in the image that support the circular contour; higher intensity indicates a better fit and higher reliability between the circle and the real edge. Therefore, when multiple candidate circles are detected, the contour with the highest cumulative intensity is selected as the baseline circular contour—that is, the one with the highest number of votes. This minimizes the interference from false circles and ensures that the extracted baseline circle is the optimal fit to the true outer edge of the measured object.

[0108] Step S3332: Determine the center coordinates and radius of the selected potential circular profile as the geometric parameters of the reference circular profile.

[0109] After determining the optimal contour from multiple candidate circles, its positional information in the parameter space needs to be transformed into geometric parameters that can be used for subsequent calculations. Since the Hough parameter space is a three-dimensional space consisting of the center coordinates (a, b) and radius r, the three-dimensional coordinates (a, b, r) corresponding to the optimal contour at the peak point with the highest cumulative intensity are the geometric parameters of the reference circular contour of the measured object. In other words, the center coordinates of the reference circular contour are (a, b), and the radius is r. These coordinates can be used in subsequent positioning marker recognition steps to calculate the positioning marker azimuth, estimate the gap location, perform distance filtering, and establish the object's own coordinate system.

[0110] In some embodiments, the process of determining the reference circular profile and its geometric parameters of the object under test further includes:

[0111] Step S3333: When the cumulative voting result indicates that no potential circular contour that meets the preset conditions has been detected, adjust the preset radius parameter range.

[0112] In actual testing, due to differences in the size of the object being measured, variations in imaging conditions, or mismatches between preset parameters and actual conditions, the Hough circle transform may fail to generate effective peak values ​​exceeding the threshold in the parameter space, leading to detection failure. In actual measurement, if the preset radius range is too narrow, it may not cover actual radius variations caused by processing tolerances, thermal expansion, or different batches of products; if the range is too wide, it may introduce significant noise interference, similarly preventing the formation of effective cumulative peak values. In other words, the currently preset radius parameter range may fail to cover the true dimensions of the actual wafer outline, or the range setting may be too strict. Therefore, when a "no effective circle" detection result occurs, it is necessary to analyze the statistical characteristics of the current voting distribution and dynamically adjust the radius parameter range, for example, by widening the upper and lower limit thresholds of the radius or shifting the radius search range, so that the parameter space can cover the true circular size.

[0113] Step S3334: Based on the adjusted radius parameter range, perform Hough circle transformation and voting accumulation on the edge contour enhancement image again until a reference circular contour and its geometric parameters that meet the preset conditions are determined.

[0114] After adjusting the radius parameter range, step S232 is executed again to perform a Hough circle transform on the same edge contour enhancement image. Based on the adjusted radius range, edge point mapping and parameter space voting accumulation are performed again to search for potential circular peaks in the new parameter space. If a candidate circle that meets the preset intensity threshold appears in the current accumulation result, it is used as the reference circular contour and its geometric parameters are extracted. If no circle is detected, the parameter range is adjusted according to the new voting results, and the above process is repeated until a reliable circular contour is successfully extracted. This adaptive iterative strategy can automatically converge to the correct detection result even when the initial parameter settings are not ideal, thereby improving the adaptability of the method to objects of different sizes and ensuring that the reference circle parameters can be stably obtained under various conditions, thus guaranteeing the robustness of the entire localization and recognition process.

[0115] Figure 7 A flowchart illustrating a location identifier identification method provided in another embodiment of this application. For example... Figure 7 As shown, in the above embodiment, step S34, based on the phase difference information between the reference circular contour and the multi-frame phase-shifting interferometric images, performs feature analysis on the edge of the object under test to determine the precise center coordinates of the positioning marker located on the edge of the object under test. This specifically includes the following steps:

[0116] Step S341: Calculate the estimated center coordinates of the positioning mark based on the center coordinates, radius, and preset reference angle of the reference circular outline.

[0117] Since the approximate orientation of positioning markers (such as wafer notches or flat edges) on the object being measured is fixed, after determining the baseline circular outline of the wafer, the positioning markers can be coarsely located based on known geometric information. Specifically, using the center of the baseline circle as the origin and its radius as the distance, combined with a preset reference angle, the approximate position of the positioning marker in the image is estimated using a polar coordinate to Cartesian coordinate conversion formula. Although the estimated coordinates have limited accuracy, they can significantly reduce the search range for subsequent precise identification, avoiding time-consuming scanning across the entire image and improving the overall efficiency of the algorithm.

[0118] The estimated center coordinates of the positioning marker are ,but:

[0119] ;

[0120] ;

[0121] in, The reference angle is set as a preset angle, (a,b) are the center coordinates of the reference circular profile, and r is the radius of the reference circular profile.

[0122] Step S342: Extract a local image region centered on the estimated center coordinates, and perform filtering and binarization on the local image region to obtain a binary image with the shape of the prominent positioning mark.

[0123] After obtaining the estimated center coordinates, a locally sized image window (e.g., 200×200 pixels) is cropped around this point. This window should be large enough to completely encompass the location marker region and its surrounding edges, limiting the processing scope to a small region of interest. Subsequently, smoothing operations such as filtering algorithms (e.g., median filtering or Gaussian filtering) are used to remove noise interference within this region, smoothing the image. Finally, a thresholding technique (e.g., Otsu's method or adaptive thresholding) is used to convert the grayscale image into a binary image. The segmentation process can mark the shape region of the location marker (usually with lower grayscale) as white / black, and the remaining background region (with higher grayscale) as black / white, thereby separating the location marker region from the background and obtaining the geometric outline of the prominent marker.

[0124] Step S343: Extract all closed contours from the binary image and select the contour containing the most pixels as the candidate contour.

[0125] In a binary image, the location marker region appears as a connected white / black pixel block, with its boundary forming a closed contour. Further contour detection algorithms (such as boundary tracking methods) are used to extract all connected foreground regions in the binary image as independent closed contours. This process may yield multiple contours, including both the true location marker contour and small, cluttered contours caused by noise or incomplete thresholding. Furthermore, since location markers are typically the most prominent structural features within a local region, their corresponding contours often have the largest area (i.e., the most pixels). Therefore, to select the most likely target contour representing the location marker, all contours can be traversed and their pixel counts compared. The contour with the most pixels (i.e., the largest area) can be selected as the candidate contour, thus eliminating interference from small noise spots or non-target structures and locking in the true boundary of the marker.

[0126] Step S344: Based on the candidate contour, perform dynamic threshold segmentation and circle fitting to determine the precise center coordinates.

[0127] Considering that the fixed threshold binarization in step S342 may be affected by uneven local illumination and cause edge shifts, after obtaining the candidate contour in step S343, dynamic threshold segmentation is introduced here to refine the local region of the candidate contour. The segmentation boundary is readjusted according to the gray-level gradient near the contour, making the extracted edge points closer to the real physical edge. Subsequently, these high-precision edge points are subjected to circle fitting (such as least squares circle fitting), and the optimal center coordinates are solved by using the common constraints of all points on the contour. Based on dynamic threshold screening and geometric fitting, the obtained accurate center coordinates can achieve sub-pixel accuracy, providing a reliable data foundation for subsequent azimuth angle calculation.

[0128] Figure 8 A flowchart illustrating a location identifier identification method provided in another embodiment of this application. For example... Figure 8 As shown, in some embodiments, step S344, based on the candidate contour, performs dynamic threshold segmentation and circle fitting to determine the precise center coordinates, specifically including:

[0129] Step S3441: Calculate the distance from each pixel in the candidate contour to the center of the reference circular contour.

[0130] Since the reference circular contour represents the overall position of the object being measured (such as the center of a wafer), and the positioning markers are located near the edge of the object, the points on the candidate contour should theoretically be distributed within a certain radius centered on the reference circle. Typically, the positioning marker area (such as a notch) can be considered as an inwardly recessed area on the wafer edge. Contour points located inside the notch will be significantly offset towards the center relative to the complete circular edge. Therefore, by calculating the Euclidean distance from each pixel in the candidate contour to the center of the reference circular contour, the radial distribution of the quantized contour points relative to the object's center can be determined. This set of calculated distance values ​​directly reflects the radial position of each point on the contour: points with larger distances are closer to the complete circular edge, while points with smaller distances are located deep within the notch's recess. This helps distinguish different parts of the notch and distinguish between true marker points and interference points.

[0131] Step S3442: Based on the calculation results, determine the dynamic distance threshold and select points whose distance is less than the dynamic distance threshold to form an internal point set.

[0132] Since location markers are typically located near the inner or outer edge of an object, the distances from their corresponding contour points to the center of the reference circle should be concentrated within a certain range around the radius of the reference circle. Therefore, based on the distribution of distances from each point on the candidate contour to the center of the reference circle, a reasonable distance threshold can be dynamically determined by calculating the statistical mean, variance, or setting a percentage threshold. Contour points with distances exceeding this threshold are likely to be noise, burrs, or external structures of the object and are therefore discarded; points with distances less than the threshold are retained, forming an internal point set focused on the marker area. This removes interference points far from the edge, improving the purity of subsequent fitting. The advantage of this dynamic threshold lies in its adaptability. Regardless of changes in the actual depth or contour size of the location marker, the threshold can be automatically adjusted based on the current data, ensuring the accuracy and robustness of the screening results.

[0133] Typically, the maximum distance on the candidate contour corresponds to the boundary between the notch and the intact edge, while the minimum distance corresponds to the deepest point of the notch. Therefore, in some embodiments, the dynamic distance threshold can be set as the average of the maximum and minimum values, i.e., T = (max(di) + min(di)) / 2. This intermediate value can effectively divide the contour points into an outer region (close to the intact edge) and an inner region (close to the bottom of the notch). By filtering out points with distances less than this threshold, a set of points focused on the core region of the notch indentation can be obtained.

[0134] Step S3443: Perform circle fitting on the internal point set, and use the center coordinates of the fitted circle as the precise positioning center coordinates of the positioning marker.

[0135] The selected set of internal points represents the true contour region of the positioning marker and exhibits good geometric consistency. Circle fitting is performed on these points (e.g., using least squares to fit a best-fitting circle) to find an optimal circle that minimizes the sum of the squared distances from all points to its edge. Because the fitting process utilizes the overall constraint of multiple points, it effectively offsets the quantization error and residual noise of individual pixels, resulting in sub-pixel level accuracy for the calculated center coordinates. This high-precision center coordinate is the exact location of the positioning marker and will be used for the final azimuth angle calculation.

[0136] Figure 9 A flowchart illustrating a location identifier identification method provided in another embodiment of this application. For example... Figure 9 As shown, in the above embodiment, step S35, determining the azimuth angle of the positioning marker based on the geometric relationship between the precise center coordinates of the positioning marker and the reference circular outline, specifically includes the following steps:

[0137] Step S351: Calculate the direction vector between the precise positioning center coordinates of the connecting positioning mark and the center coordinates of the reference circular outline.

[0138] After obtaining the precise center coordinates of the positioning marker, its spatial location information needs to be converted into angular information that facilitates adjustment of its orientation or attitude. Specifically, starting from the center of the reference circular outline and ending at the precise center coordinates of the positioning marker, a two-dimensional direction vector is constructed through vector subtraction. Its length represents the distance from the marker's center to the center of the circle, and its direction indicates the marker's specific orientation on the wafer circumference. This direction vector transforms the spatial location information of the positioning marker into a vector form that can be used for angle calculation.

[0139] Step S352: Calculate the angle between the direction vector and a preset reference direction vector.

[0140] To determine the actual orientation of a positioning marker, a unified reference standard is needed. The preset reference direction vector is typically defined based on the coordinate system of the measurement system (e.g., the horizontal direction to the right of an image, the direction of the main reference edge of a wafer, etc.). The cosine of the angle between two vectors is calculated using the dot product formula, and then the cross product is used to determine the direction of rotation (clockwise or counterclockwise), thus obtaining the angle between the direction vector and the reference vector. Mathematically, this angle reflects the deviation between the current actual orientation of the positioning marker and the system reference direction.

[0141] In some embodiments, the preset reference direction vector is a unit vector pointing to the six o'clock direction in the image coordinate system.

[0142] In other words, the reference direction in the image coordinate system is defined as the vertically downward six o'clock direction, and it is used as the reference for angle measurement in the form of a unit vector. In practical applications, the standard orientation of a wafer is usually based on its bottom (i.e., the six o'clock direction) as the reference zero point for angle measurement, which also corresponds to the fixed physical direction during image sensor acquisition, such as the downward position that a wafer edge notch or flat edge should point in under ideal alignment. By defining the reference direction vector as a unit vector pointing in this direction, it is ensured that the azimuth angle calculated subsequently has a clear physical meaning, namely the deflection angle of the notch center relative to the standard bottom of the wafer. Using the form of a unit vector further ensures that the reference is only used to represent direction without introducing an additional length scaling factor, making the angle calculation between the direction vector and the reference vector more accurate, thereby providing high-precision angle parameters that meet industry standards for wafer alignment.

[0143] Step S353: Convert the included angle into the azimuth angle of the positioning mark.

[0144] The included angle obtained through vector operations is usually expressed in radians and needs to be converted into angles, a more commonly used unit in engineering applications. During the conversion process, the positive or negative direction of the angle, or its actual pointing direction, must be determined based on the quadrant position of the vector to ensure that the final output azimuth angle accurately reflects the true position of the positioning marker on the wafer circumference. Specifically, according to a preset coordinate system convention, such as taking the reference direction as 0 degrees and counterclockwise as the positive direction, the included angle calculated in step S252 is adjusted for necessary quadrants and corrected for its sign, mapping it to a standard angle range of 0° to 360° (or -180° to 180°). The converted azimuth angle is the final output result. This angle value is the current orientation parameter of the measured object and can be directly used for wafer alignment adjustments in subsequent process equipment, such as controlling a robotic arm to rotate the wafer to a preset standard posture, thereby meeting the stringent requirements for crystal orientation accuracy in processes such as photolithography.

[0145] Assuming the precise positioning center coordinates of the positioning identifier The direction vector relative to the center coordinates (a, b) of the reference circular profile is: The preset reference direction vector is ,but:

[0146] The angle between the direction vector and a preset reference direction vector for:

[0147] ;

[0148] After being converted to a angular format, the final output angle is the azimuth angle of the positioning mark.

[0149] It is worth noting that in the above embodiments, one or more computational steps in the process of generating edge contour enhancement images, accumulating votes, or determining precise center coordinates are accelerated by parallel computation using a graphics processor.

[0150] Due to the large data volume of phase-shifting interferometry images and the need to traverse a large number of pixels and parameter combinations for voting accumulation in the Hough transform, these operations would face significant time overhead if executed serially by a traditional central processing unit (CPU). By introducing a graphics processing unit (GPU) with a large number of computing cores, the computationally intensive steps of generating edge contour enhancement images—including intensity analysis of multiple frames, pixel-level difference calculation, normalization mapping, voting accumulation in the Hough circle transform, and circle fitting calculation and batch calculation of candidate point distances after dynamic threshold screening—are decomposed into thousands of parallel threads for simultaneous processing. This reduces the overall time consumption of image preprocessing, feature extraction, and fitting optimization. By offloading computationally intensive tasks from the CPU to the GPU, this hardware acceleration strategy not only improves the real-time performance of location marker recognition, enabling high-precision measurements to meet the time requirements of online detection, but also provides computational resources to expand the search range or increase the number of iterations while maintaining high resolution, enhancing the practicality and robustness of the method in industrial applications.

[0151] In summary, the positioning marker recognition method provided in this application achieves high-precision and robust positioning of the fine structure of the edge of the measured object by integrating phase-shifting interferometry and image processing technology. First, this method fully utilizes the light intensity variation and phase difference information contained in multi-frame phase-shifting interferometric images to generate an enhanced edge contour image. This overcomes the sensitivity of traditional single-frame non-interferometric images to lighting conditions and the insufficient contrast caused by the single phase information in traditional single-frame interferometric images, thus improving edge contrast and signal-to-noise ratio. Then, it extracts the wafer reference circular contour through Hough circle transform and an adaptive voting mechanism, providing a reliable spatial reference system for subsequent positioning. Based on this, by combining phase difference information and dynamic threshold segmentation, and through candidate contour selection and circle fitting, it achieves sub-pixel-level center coordinate extraction of the positioning marker, enabling stable recognition of low-contrast or small notches. Finally, it accurately calculates the azimuth angle through vector operations, converting the image coordinates into a standardized angle output. Furthermore, while ensuring high precision, this technical solution improves processing efficiency and real-time performance through parallel acceleration of the graphics processing unit (GPU), meeting the stringent requirements of semiconductor production lines for real-time online inspection and providing a reliable guarantee for the precise alignment of subsequent processes such as photolithography.

[0152] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of any embodiment of the above-described location identifier identification method and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0153] The processor can be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0154] Those skilled in the art will understand that all or part of the functions of the various methods in the above embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to achieve the above functions. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be achieved. In addition, when all or part of the functions in the above embodiments are implemented by computer programs, the program can also be stored in a server, another computer, disk, optical disk, flash drive, or external hard drive, etc., and can be downloaded or copied to the memory of a local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be achieved.

[0155] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art, under the guidance of this application, can make several simple deductions, modifications or substitutions based on the spirit of this application and the scope of protection of the claims without departing from the spirit of this application and the claims. All of these are within the protection scope of this application.

Claims

1. A method for identifying location markers, characterized in that, include: Acquire multi-frame phase-shifting interferometric images characterizing the surface of the object under test; Based on the light intensity change information of each corresponding pixel in the multi-frame phase-shifting interferometry image, an enhanced edge contour image of the object under test is generated. Extract the reference circular contour of the object under test from the edge contour enhancement image of the object under test; Based on the phase difference information between the reference circular contour and the multi-frame phase-shifting interferometric images, feature analysis is performed on the edge of the object under test to determine the precise center coordinates of the positioning mark located on the edge of the object under test. The azimuth angle of the positioning mark is determined based on the geometric relationship between the precise center coordinates of the positioning mark and the reference circular outline.

2. The method for identifying location markers according to claim 1, characterized in that, The step of generating an enhanced edge contour image of the object under test based on the light intensity change information of each corresponding pixel in the multi-frame phase-shifting interferometry image includes: Calculate the difference between the maximum and minimum grayscale values ​​corresponding to each pixel position in the multi-frame phase-shifting interferometric image to obtain a difference image; The difference image is normalized, and the normalized pixel values ​​are mapped to a standard grayscale range to obtain the edge contour enhancement image.

3. The method for identifying location markers according to claim 1, characterized in that, Extracting the reference circular contour of the object under test from the enhanced edge contour image of the object under test includes: The edge contour enhancement image is subjected to Hough circle transform to map the edge information represented by the edge contour enhancement image to the center-radius parameter space; Based on the coordinates of each pixel in the edge information and the preset radius parameter range, a voting accumulation is performed on potential circular contours in the parameter space; Based on the cumulative voting results, the reference circular profile and its geometric parameters of the object under test are determined.

4. The method for identifying location markers according to claim 3, characterized in that, The step of determining the reference circular contour and its geometric parameters of the object under test based on the cumulative voting results includes: When the cumulative voting results indicate that there are multiple potential circular contours that meet the preset conditions, the potential circular contour with the highest cumulative intensity is selected as the reference circular contour of the object under test. The center coordinates and radius corresponding to the selected potential circular profile are determined as the geometric parameters of the reference circular profile.

5. The method for identifying location markers according to claim 3, characterized in that, The step of determining the reference circular contour and its geometric parameters of the object under test based on the cumulative voting results includes: When the cumulative voting results indicate that no potential circular contours meeting the preset conditions have been detected, the preset radius parameter range is adjusted. Based on the adjusted radius parameter range, the edge contour enhancement image is subjected to Hough circle transformation and voting accumulation again until a reference circular contour and its geometric parameters that meet the preset conditions are determined.

6. The method for identifying location markers according to claim 1, characterized in that, The step of performing feature analysis on the edge of the object under test to determine the precise center coordinates of the positioning marker located on the edge of the object under test includes: The estimated center coordinates of the positioning mark are calculated based on the center coordinates, radius, and preset reference angle of the reference circular outline. A local image region is cropped centered on the estimated center coordinates, and the local image region is filtered and binarized to obtain a binary image that highlights the shape of the positioning mark. Extract all closed contours from the binary image and select the contour containing the most pixels as the candidate contour. Based on the candidate contour, dynamic thresholding and circle fitting are performed to determine the precise center coordinates.

7. The method for identifying location markers according to claim 6, characterized in that, The step of determining the precise center coordinates based on the candidate contour by performing dynamic threshold segmentation and circle fitting includes: Calculate the distance from each pixel in the candidate contour to the center of the reference circular contour; Based on the calculation results, a dynamic distance threshold is determined, and points with a distance less than the dynamic distance threshold are selected to form an internal point set. Perform a circle fitting on the set of internal points, and use the center coordinates of the fitted circle as the precise positioning center coordinates of the positioning marker.

8. The method for identifying location markers according to claim 1, characterized in that, Determining the azimuth angle of the positioning marker based on the geometric relationship between the precise center coordinates of the positioning marker and the reference circular outline includes: Calculate the direction vector connecting the precise positioning center coordinates of the positioning mark with the center coordinates of the reference circular outline; Calculate the angle between the direction vector and a preset reference direction vector; Convert the included angle into the azimuth angle of the positioning mark.

9. The method for identifying a location marker according to any one of claims 1 to 8, characterized in that, At least one computational step in the process of generating edge contour enhancement images, vote accumulation, or determining precise center coordinates is accelerated by parallel computation using a graphics processor.

10. A semiconductor detection system, characterized in that, include: A phase-shifting interferometer is used to acquire multi-frame phase-shifting interference images of the surface of an object under test. An image processing device, communicatively connected to the phase-shifting interferometer, is configured to perform a method for identifying positioning marks as described in any one of claims 1 to 8, to determine the center coordinates and azimuth angle of positioning marks formed on the edge of the object under test based on the multi-frame phase-shifting interferometric images.

11. The semiconductor detection system according to claim 10, characterized in that, The image processing apparatus includes at least one graphics processor for accelerating the execution of at least one step among generating an edge contour enhanced image, vote accumulation, or circle fitting.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer-executable program or instructions, which, when executed by a processor, are used to implement the identification method of the location marker as described in any one of claims 1 to 8.

Images