Regenerated wafer scanning method and scanning system

Abstract

The invention discloses a regenerated wafer scanning method, and relates to the technical field of semiconductor wafer scanning, and the method comprises the steps: obtaining the identification information of a target wafer transmission box when the target wafer transmission box is detected; detecting the in-place state of the wafer through a light curtain sensor, and generating a slot mapping table; controlling a preset micro-rotation mechanism to carry out micro-rotation on the wafer at each slot position within a set angle range, and collecting an annular structure light image; constructing polar coordinates and calculating a first attitude angle of each wafer; driving a micro-rotation mechanism to adjust the wafer to a preset reference angle according to the first attitude angle to obtain a standard attitude wafer; the mechanical arms are driven to be sequentially conveyed to a scanning station; respectively controlling a rotary positioning module to rotate in a first angle range and a second angle range, and acquiring a coded image and a spectral image; the wafer code is analyzed, confidence verification is carried out, and a wafer process type label is identified; and warehousing and recording the wafer codes, the process labels and the transmission box identification information. According to the invention, the storage management efficiency of the regenerated wafer can be effectively improved.

Description

Technical Field

[0001] This application relates to the field of semiconductor wafer scanning technology, and in particular to a method and system for scanning regenerated wafers. Background Technology

[0002] Reclaimed wafers refer to silicon wafers that have been used but have undergone multiple cleaning and polishing processes, including physical and chemical cleaning, to remove the original film layers and defects before being reused. They are widely used in non-functional processes such as equipment testing, process verification, and photolithography masking. As wafer manufacturing costs continue to rise, the utilization rate of reclaimed wafers is gradually increasing, and their inventory management and traceability accuracy have become crucial factors affecting the overall plant operation efficiency. Currently, the mainstream method is manual management. Operators must remove each wafer from the carrier and manually place it on the scanning table. The entire process relies entirely on manual judgment and operation, which is extremely inefficient and prone to alignment deviations and code misreading due to fatigue, viewing angle errors, or light reflection. This leads to incorrect code recognition, resulting in incorrect information entry for reclaimed wafers, and consequently affecting the efficiency of reclaimed wafer inventory management.

[0003] There is currently no good solution to the above problems. Summary of the Invention

[0004] This application provides a method and system for scanning recycled wafers to improve the efficiency of recycled wafer warehousing management.

[0005] To achieve the above objectives, the embodiments of this application adopt the following technical solutions: In a first aspect, a method for scanning reclaimed wafers is provided, applied to a reclaimed wafer scanning system. The reclaimed wafer scanning system includes a machine vision module and a rotation positioning module. The machine vision module includes a first vision unit and a second vision unit. The first vision unit is disposed on a robotic arm, and the second vision unit is disposed directly above the rotation positioning module. The method includes: When the target wafer transport box is detected to have reached the preset position, the identification information of the target wafer transport box is obtained; A preset light curtain sensor is used to detect each slot in the target wafer transfer box to determine the slot mapping table between each slot and the wafer; According to the slot mapping table, the wafers in each slot are micro-rotated within a preset angle range using a micro-rotation mechanism in the target wafer transfer box to obtain ring structured light imaging of each wafer. The polar coordinates of each wafer are determined based on the ring structured light imaging of each wafer, and the first attitude angle corresponding to each wafer is determined based on the polar coordinates of each wafer. Based on the first attitude angle of each wafer, each wafer is adjusted to a preset reference angle using a preset micro-rotation mechanism to obtain a standard attitude wafer; The robotic arm is driven to push each standard-position wafer from the target wafer transfer box to the preset scanning station in a preset order; When a standard posture wafer is detected at the preset scanning station, the preset scanning station is driven to rotate within a first angle range using the rotation positioning module, and the first coded image of each standard posture wafer is acquired through the first vision unit. The preset scanning station is driven to rotate within a second angle range using a rotation positioning module, and the second spectral image of each standard posture wafer is acquired through the second vision unit. The wafer code of each standard posture wafer is determined by the first encoded image, and the coding confidence of each wafer code is verified. The process type label for each standard orientation wafer is determined based on the second spectral image; When the coding confidence level is greater than the preset confidence level, the process type label, wafer code, and target wafer transport box identification information are entered into the database.

[0006] In one possible implementation of the first aspect, the method further includes: Obtain process path and workstation load information; Wafer edge health and wafer warpage values ​​were determined based on ring structured light imaging and second spectral images. The wafer quality grade is determined by combining wafer edge health and wafer warpage value. The wafer quality grade includes defective wafers and qualified wafers. The sorting path is generated by combining process type label, workstation load information and wafer quality grade. The sorting path includes a defective product sorting path and a qualified product sorting path. The target actuator is determined by selecting a preset actuator based on the wafer edge health and wafer warpage value; The execution parameters of the target actuator are set based on the wafer edge health, wafer warpage value, and target actuator. By combining the execution parameters of the target actuator and the sorting path, wafers are sorted to the target workstation to obtain wafer sorting information; The wafer sorting information is stored in the warehouse entry record.

[0007] In one possible implementation of the first aspect, the preset actuator includes an adsorption actuator and a clamping actuator, and the step of selecting the preset actuator and determining the target actuator based on wafer edge health and wafer warpage value includes: When the wafer edge health is greater than the first health threshold and the wafer warpage value is less than the first warpage threshold, the adsorption actuator is selected as the target actuator. When the wafer edge health is less than the second health threshold and the wafer warpage value is greater than the second warpage threshold, the clamping actuator is selected as the target actuator.

[0008] In one possible implementation of the first aspect, the annular structured light imaging is an image of annular light stripes with a specific polarization state, captured by a polarization camera under multiple temperature field conditions and adjusted for polarization state.

[0009] In one possible implementation of the first aspect, determining the polar coordinates of each wafer based on the annular structured light imaging of each wafer, and determining the first attitude angle corresponding to each wafer based on the polar coordinates of each wafer, includes: The reference light intensity value of ring structured light imaging under the reference temperature condition and the corresponding light intensity value of ring structured light imaging under different temperature field conditions, including high temperature environment. The polarization state attenuation coefficient under different temperature field conditions is calculated by combining the light intensity values ​​under different temperature field conditions with the reference light intensity values ​​under the reference temperature condition. The polarization degree temperature correction formula is obtained by fitting the polarization state attenuation coefficient under different temperature field conditions; The theoretical polarization attenuation under high temperature conditions is calculated using the polarization degree temperature correction formula; The thermal radiation interference is determined by comparing the actual light intensity value of the ring structured light imaging with the reference light intensity value corresponding to the theoretical polarization attenuation. By combining thermal radiation interference suppression processing of ring structured light imaging under high temperature environment, a target ring image is obtained. The center coordinate parameters are obtained by performing a circle fitting between the target ring image and the ring structured light imaging under the reference temperature conditions. The center coordinate parameters include the high temperature center coordinates. The theoretical thermal expansion offset is calculated based on the pre-set linear expansion coefficient of the wafer; By combining the theoretical thermal expansion offset, the coordinates of the high-temperature center and the positioning notch of the wafer are corrected to obtain the compensated pole coordinates and polar axis direction; The first attitude angle is determined by combining the compensated pole coordinates and polar axis direction.

[0010] In one possible implementation of the first aspect, determining the first attitude angle by combining the compensated pole coordinates and the polar axis direction includes: Establish a first polar coordinate system based on the compensated pole coordinates and polar axis direction; In the target annular image, filter annular light bar feature points at preset angular intervals, and convert the pixel coordinates of the annular light bar feature points into world coordinates; In the first polar coordinate system, for any annular light stripe feature point in any world coordinate system, the initial polar radius and initial polar angle of the annular light stripe feature point are calculated using the rectangular coordinate transformation relationship based on the polar coordinates corresponding to the annular light stripe feature point. The thermal expansion correction value and the polarization correction value are calculated using the preset thermal expansion correction term and the preset polarization optical path difference correction term, respectively. The final polar diameter is obtained by adding the thermal expansion correction value, the polarization correction value, and the initial polar diameter. The initial polar angle is corrected using a weighted smoothing algorithm based on adjacent three points to obtain the final polar angle; The temperature measurement value corresponding to each annular light stripe feature point and the deviation between the final polar radius and the preset reference radius are obtained, and a thermal distortion basis function term is constructed based on the temperature measurement value. Fourier analysis was performed on the deviation between the final extreme radius phase and the preset reference radius to obtain the first harmonic term of the wafer attitude. Using the final polar angle as the independent variable and the deviation of the final polar radius from the preset reference radius as the observation, a joint model of the observation is constructed by combining the independent variable, the observation, the distortion basis function term, and the first harmonic term. The first attitude angle of the joint model wafer is solved based on the least squares criterion.

[0011] In one possible implementation of the first aspect, determining the process type label for each standard orientation wafer based on the second spectral image includes: A second polar coordinate system is constructed based on the standard orientation wafer, and a corresponding polar coordinate mask is generated according to the second polar coordinate system. The polar coordinate mask is used to avoid the wafer notch area and the wafer clamping area. The effective process region in the polar coordinate mask is extracted by combining the polar coordinate mask and the second spectral image; The correspondence between process type labels and wafer spectral data is determined based on preset historical annotation information, which includes wafer samples of process types and corresponding spectral images. By combining the correspondence relationship, the mutual information maximization algorithm is used to select the sensitive region of the process type in the effective process region; Based on preset historical annotation information, determine the mutual information measurement between sensitive areas and process type labels; Each sensitive region is assigned a corresponding weight parameter based on the classification contribution between the sensitive region and the process type label. The classification contribution is obtained by measuring the mutual information between the sensitive region features and the process type label. The sensitive region undergoes spectral data processing steps to obtain a multidimensional spectral feature set of the sensitive region; The multidimensional spectral feature set is matched with the preset process spectral dictionary and input into the preset supervised discrimination model to obtain the process posterior probability corresponding to each sensitive region. The joint process posterior probability distribution is calculated by combining the weight parameters corresponding to each sensitive region and the process posterior probability. Obtain prior information for standard-position wafers, including wafer process production data from a pre-defined manufacturing execution system; The confidence level of the process type is obtained by Bayesian fusion of prior information and joint posterior probability. When the confidence level of the process type is greater than the preset confidence level of the process type, output the process type label of the standard orientation wafer.

[0012] In one possible implementation of the first aspect, the spectral data processing step includes: Obtain the reflectance spectrum of each sensitive region; The reflectance spectrum of each sensitive region is decomposed into the basic endmember spectrum using a constrained nonnegative endmember decomposition algorithm. The actual physical parameters of the wafer film layer corresponding to the sensitive region are obtained by combining the basic end-member spectrum and the preset multilayer thin film interference model through physical inversion. Empirical spectral features corresponding to each sensitive region are extracted based on the actual physical parameters of the wafer film layer; The differential characteristics of empirical spectral features are calculated based on the characteristics of a reference wafer; For any sensitive region, spatial distribution features are extracted, and the spatial distribution features, differential features, and empirical spectral features are concatenated to obtain multidimensional spectral features.

[0013] Secondly, this application provides a machine-readable storage medium storing instructions that cause a machine to perform the above-described regenerated wafer scanning method.

[0014] Thirdly, this application provides an electronic device, comprising: The memory is configured to store instructions; and The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the regenerated wafer scanning method described above.

[0015] The above technical solution utilizes a light curtain sensor to determine the slot mapping table, accurately locating the correspondence between each slot and the wafer. A micro-rotation mechanism allows the wafer to rotate within a preset angle range, generating ring-shaped structured light imaging. This provides a more comprehensive and clearer view of the wafer's features, effectively reducing blind spots caused by varying initial wafer orientations and improving the accuracy of wafer feature detection. Based on the ring-shaped structured light imaging, the wafer's polar coordinates are determined, yielding a first orientation angle. The wafer is then adjusted to a preset reference angle to obtain a standard orientation wafer. This ensures that the wafer is in a uniform and detection-friendly orientation during subsequent scanning, avoiding interference from orientation differences. A robotic arm pushes the standard orientation wafers to a preset scanning station in a predetermined sequence, improving wafer transport efficiency. Simultaneously, a rotation positioning module drives the scanning station to rotate within different angle ranges, enabling the first and second vision units to acquire different types of images, significantly shortening the scanning time for each wafer and improving overall scanning efficiency. The process of quickly determining the wafer code and verifying its confidence level using the first coded image, and then determining the process type label based on the second spectral image, helps improve the efficiency of production management and quality traceability. When the code confidence level is greater than a preset confidence level, the process type label, wafer code, and target wafer transport box identification information are recorded in the warehouse, providing data support for full-process wafer traceability. When a wafer has a quality problem or the production process needs to be traced, this warehouse information can be used to quickly locate the specific wafer, its transport box, and the corresponding process type, improving the efficiency of warehouse recording.

[0016] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0017] Figure 1 A schematic flowchart of a regenerated wafer scanning method provided in an embodiment of this application; Figure 2 This is a schematic diagram of a regenerated wafer scanning method provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0019] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0020] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0021] Figure 1 The illustration schematically shows a flow chart of a regenerated wafer scanning method according to an embodiment of this application. Figure 1 As shown, this application provides a method for scanning regenerated wafers, applied to a regenerated wafer scanning system. The regenerated wafer scanning system includes a machine vision module and a rotation positioning module. The machine vision module includes a first vision unit and a second vision unit. The first vision unit is disposed on a robotic arm, and the second vision unit is disposed directly above the rotation positioning module. The method may include the following steps.

[0022] S101. When the target wafer transport box is detected to be at a preset position, the identification information of the target wafer transport box is obtained. S102. Use a preset light curtain sensor to detect each slot in the target wafer transfer box and determine the slot mapping table between each slot and the wafer. S103. According to the slot mapping table, the wafers in each slot are micro-rotated within a preset angle range using the micro-rotation mechanism in the target wafer transfer box to obtain the ring structured light imaging of each wafer. S104. Determine the polar coordinates of each wafer based on the ring structured light imaging of each wafer, and determine the first attitude angle corresponding to each wafer according to the polar coordinates of each wafer. S105. Based on the first attitude angle of each wafer, each wafer is adjusted to a preset reference angle using a preset micro-rotation mechanism to obtain a standard attitude wafer. S106. Drive the robotic arm to push each standard posture wafer from the target wafer transfer box to the preset scanning station in a preset order; S107. When a standard posture wafer is detected at the preset scanning station, the preset scanning station is driven to rotate within a first angle range using the rotation positioning module, and the first coded image of each standard posture wafer is acquired through the first vision unit. S108. The preset scanning station is driven to rotate within the second angle range by the rotation positioning module, and the second spectral image of each standard posture wafer is acquired by the second vision unit. S109. Determine the wafer code of each standard posture wafer through the first encoded image, and verify the encoding confidence of each wafer code; S110. Determine the process type label for each standard orientation wafer based on the second spectral image; S111. When the coding confidence level is greater than the preset confidence level, the process type label, wafer code, and target wafer transfer box identification information are entered into the database.

[0023] When the target wafer transport box is detected to have reached the preset position, its identification information is acquired. The target wafer transport box is a specific container used for storing and transporting wafers; the preset position is a specific location or area pre-defined by the system. Subsequent scanning processes will only begin when the target wafer transport box is moved to this preset position. The identification information is a unique identifier for the target wafer transport box, which can be a barcode or QR code. By acquiring this identification information, the system can accurately identify which specific transport box is currently being processed.

[0024] A pre-set light curtain sensor is used to detect each slot within the target wafer transfer box, establishing a slot mapping table between each slot and the wafer. The pre-set light curtain sensor is a device that uses photoelectric principles for detection. The target wafer transfer box typically has multiple slots for placing wafers, arranged in a specific pattern and spatial position. The light curtain sensor detects each slot to determine if a wafer is present, obtaining approximate wafer position information within the slot. The slot mapping table is a data table or structure used to record the correspondence between each slot within the target wafer transfer box and the wafers within it.

[0025] Based on the slot mapping table, a pre-set micro-rotation mechanism within the target wafer transfer box is used to micro-rotate the wafers in each slot within a preset angle range to obtain ring structured light imaging for each wafer. The pre-set micro-rotation mechanism is a device installed within the target wafer transfer box that enables the wafer to rotate at minute angles. This mechanism can be a rotary table installed within the wafer transfer box, comprising a rotary motor, a reduction gear, and an angle feedback device. The micro-rotation mechanism is connected to the wafer clamping device in each slot to achieve precise micro-rotation of the wafer within the preset angle range. By controlling the rotation speed and rotation angle of the rotary motor, the micro-rotation mechanism maintains a stable position for the wafer during micro-rotation and monitors the rotation angle in real time through the angle feedback device, ensuring that the accuracy of each wafer rotating to the target angle meets the requirements of ring structured light imaging. The micro-rotation mechanism enables ring structured light acquisition and wafer reference attitude adjustment. It can also employ a servo motor-driven rotary platform or a stepper motor-driven precision rotary table, and can be combined with position sensors or encoders to achieve closed-loop control of the rotation angle to meet the needs of different wafer scanning systems. The preset angle range refers to the different angles from which images may be acquired for different wafer inspection needs. By micro-rotating the wafer within the preset angle range, the machine vision module can acquire images of the wafer from different angles, thus providing a more comprehensive view of the wafer's features. The micro-rotation mechanism drives the wafer to rotate according to the preset angle range. During rotation, the wafer's orientation continuously changes, allowing the machine vision module to image the wafer at different rotation angles. For example, the micro-rotation mechanism first rotates the wafer 10 degrees from an initial angle, pauses for a period of time to allow the vision module to image, then rotates it another 10 degrees, and so on, until the upper limit of the preset angle range is reached. Through multiple micro-rotations, wafer images are acquired from different angles, capturing the annular light spot features of the wafer to obtain annular structured light imaging of each wafer.

[0026] Next, based on the ring structured light imaging of each wafer, an edge detection algorithm is used to extract the boundary lines of the bright spots in the ring structured light imaging, obtaining the coordinates of the edge points in the ring structured light imaging. These edge point coordinates are the positions of the wafer edges on the image. The set of edge points of the wafer edges is used to calculate the center coordinates and average radius of the edge using the least squares circle fitting method. After obtaining the center parameters, a polar coordinate system is constructed with the center as the pole. For each edge point (x1, y1) and the center (a, b), the distance between the edge point and the center is calculated, which can be calculated using Euclidean distance, to obtain the polar radius r. The angle between the line connecting this point and the center and the reference direction is obtained as the polar angle θ. The reference direction can be set according to the actual situation (e.g., horizontal to the right of the image). Finally, the polar coordinates of the wafer are determined, and each point on the wafer edge is represented by polar coordinates (r, θ). With the polar angle θ as the horizontal axis and the polar radius r as the vertical axis, all (r, θ) data points are connected to form a continuous curve, obtaining the wafer edge structure feature distribution curve. The system analyzes the distribution curves of wafer edge structural features to identify structural abrupt changes, such as the angular position of wafer notches. Notch features typically manifest as abrupt changes in edge radius, sharp changes in brightness, texture interruptions, or a phase anomaly regions. Based on the polar angle corresponding to this feature and the angle difference between this angle and the system-defined standard reference direction θ, the system calculates the first orientation angle of the current wafer. .

[0027] In this embodiment, based on the first attitude angle identified for each wafer, the system further invokes the micro-rotation control module to perform high-precision rotation adjustment of the wafer, aligning the wafer attitude with the reference angle set by the system. This reference angle alignment can be set according to the position of the robotic arm, facilitating the robotic arm to push the wafer and thus obtaining a wafer with standardized attitude. After the system completes the calculation of the first attitude angle for each wafer, it inputs this attitude angle as a control parameter into the micro-rotation control module. The control module generates a corresponding rotation command based on the difference Δθ = θ1 - α between the first attitude angle α and the preset reference angle θ1, driving the micro-rotation mechanism to perform angle compensation. The wafer is rotated to the reference angle set by the system using the micro-rotation mechanism, resulting in a wafer with a standard attitude.

[0028] After completing the attitude calibration of each wafer, the driven robotic arm sequentially pushes standard-position wafers from the target wafer transport box to the preset scanning station in a preset order to complete subsequent operations such as image acquisition, code recognition, or film analysis. Specifically, the system reads the slot layout of the target wafer transport box and identifies the slot number where the current wafer is located. Next, the driven robotic arm moves to the front of the corresponding slot and controls the end effector to align with the edge of the wafer, pushing the standard-position wafer along the guide rail, causing the wafer to leave the slot and slide into the preset guide channel. The wafer moves in the guide channel and finally enters and stops in the positioning area of ​​the scanning station, where the sensing module (such as a light curtain or ToF sensor) in the scanning station confirms the wafer's positioning status.

[0029] When the scanning control system detects that a standard-position wafer is already present at the preset scanning station, it activates the rotation positioning module to rotate the preset scanning station where the standard-position wafer is placed within a first angle range. Simultaneously, the first vision unit performs the scanning preparation process. The first angle range is the angle interval most likely to exist within the pre-set encoding, such as ±10°. Because the wafer is in a standard orientation, the encoding position deviation will not be too large, and a small rotation range is sufficient for coverage. During the wafer rotation, the first vision unit continuously or intermittently acquires image data of specific areas on the wafer surface, especially the encoding area contained in the back of the wafer. During rotation, to ensure clear and readable encoded images, the vision system can combine mechanisms such as autofocus and dynamic exposure adjustment to obtain the first encoded image of each standard-position wafer. In this embodiment, the first encoded image is the image of the wafer's encoding area.

[0030] Next, the preset scanning station is driven to rotate within a second angle range by the rotation positioning module, and multi-angle spectral image acquisition is achieved in conjunction with the second vision unit for functions such as wafer surface film detection, micro-defect identification, and material composition analysis. Unlike the first angle range used in the coding and recognition stage, the current rotation operation uses the second angle range, which is preferably set to cover at least three different viewing angles, such as 0°~120° or -45°~+45°, to obtain spectral response information with viewing angle differences. The second angle range can be set according to process objectives. The second vision unit is equipped with a multi-channel spectral imaging device, preferably a line-scan or area-array spectral camera. The second vision unit can acquire spectral images of the wafer surface in visible light, near-infrared, ultraviolet, or multi-channel fused bands at different rotation angles. During scanning, by controlling the wafer to stop at each set angle, the second vision unit is activated to complete the acquisition of one frame or multiple consecutive frames of spectral images. After acquisition, the acquired images are time-stamped and angle-labeled to obtain the second spectral image of each standard-position wafer.

[0031] After acquiring the first coded image, the system further extracts the coded information of the wafer surface from the first coded image using an image recognition algorithm, and verifies the confidence level of the extraction results to ensure the accuracy and reliability of the coded recognition. The image recognition algorithm can be an OCR algorithm or a DataMatrix code. Specifically, the first coded image typically contains an identification code area located at the center or edge of the back side of the wafer, which can be a QR code, a DataMatrix code, etc. The system calls a preset image decoding module to preprocess the coded image, including grayscale conversion, noise reduction, contrast enhancement, geometric correction, and boundary extraction. Then, the decoding algorithm is executed to extract the coded content and generate the corresponding coded string. While extracting the wafer code, the OCR algorithm or QR code decoding algorithm automatically outputs a confidence score for each character or the entire code during recognition. The algorithm scores each character. For example, when recognizing "W", if it is considered to be 90% likely to be "W" and 10% likely to be "V" after comparing with the character library, then the confidence score for that character is 90%. The confidence score for the entire code is the average of the confidence scores of all characters. For example, for a string of 10 characters, the average confidence score is 92%. For QR codes, the algorithm scores based on the completeness of the dot matrix recognition. For example, if a QR code has 3 positioning points, and all 3 are clearly recognized and the dot matrix is ​​complete, the confidence score may be 95%. If one positioning point is blurry, the confidence score will drop below 70%.

[0032] Based on the second spectral image, spectral analysis is performed on the wafer surface to determine the process type label corresponding to each standard-position wafer. The process type label indicates the dominant process category of the wafer's current or previous surface, such as deposition, photolithography, etching (dry / wet), polishing, cleaning / baking, etc. The second spectral image refers to a wafer image acquired through a spectral camera or multi-band imaging device. Each pixel, in addition to its spatial location, contains information such as reflected light intensity / transmitted light intensity at the corresponding wavelength. Pixel-level spectral reflectance features are extracted from the second spectral image of the standard-position wafer. A machine learning classification model is used to compare these spectral features with a pre-defined process type sample library to identify the current process type label of the wafer surface. The process type label characterizes the manufacturing stages the wafer has undergone, such as film deposition, etching, polishing, or thermal treatment. It can also be compared with process route information in the manufacturing execution system to achieve process traceability and anomaly detection. When the confidence level of the identified wafer code is higher than a preset threshold, the system automatically binds the wafer's process type label, coding information, and the identification data of the corresponding target wafer transport box, and performs an entry record.

[0033] Figure 2 This paper illustrates a structural schematic diagram of a regenerated wafer scanning method provided in an embodiment of this application, as shown below. Figure 2 As shown, it includes a first vision unit 10, a second vision unit 20, and a rotation positioning module 30.

[0034] By using a light curtain sensor to determine the slot mapping table, the correspondence between each slot and the wafer can be accurately located. A micro-rotation mechanism allows the wafer to be rotated within a preset angle range to obtain ring-shaped structured light imaging, which can more comprehensively and clearly display the wafer's features. This effectively reduces blind spots caused by different initial wafer postures, thereby improving the accuracy of wafer feature detection. Based on the ring-shaped structured light imaging, the polar coordinates of the wafer are determined, and the first posture angle is obtained. The wafer is then adjusted to a preset reference angle to obtain a standard posture wafer. This ensures that the wafer is in a uniform and detection-friendly posture during subsequent scanning, avoiding interference from posture differences. A driven robotic arm pushes the standard posture wafers to a preset scanning station in a preset sequence, improving the efficiency of wafer transfer. Simultaneously, a rotation positioning module drives the scanning station to rotate within different angle ranges, allowing the first and second vision units to acquire different types of images, significantly shortening the scanning time for each wafer and improving overall scanning efficiency. The wafer code is quickly determined using the first coded image, and the code confidence is verified. The process type label is determined based on the second spectral image, which helps improve the efficiency of production management and quality traceability. When the coding confidence level exceeds the preset confidence level, the process type label, wafer code, and target wafer transport box identification information are entered into the inventory, providing data support for end-to-end wafer traceability. When a wafer experiences quality issues or requires tracing the production process, this inventory information can be used to quickly locate the specific wafer, its transport box, and the corresponding process type, improving the efficiency of inventory recording.

[0035] In one embodiment of this invention, the method further includes: S210. Obtain process path and workstation load information; S220. Determine wafer edge health and wafer warpage value based on ring structured light imaging and second spectral images; S230. The wafer quality grade is determined by combining the wafer edge health and wafer warpage value. The wafer quality grade includes defective wafers and qualified wafers. S240. Combine process type label, workstation load information and wafer quality grade to generate sorting paths. The sorting paths include defective product sorting paths and qualified product sorting paths. S250. Select the preset actuator and determine the target actuator based on the wafer edge health and wafer warpage value; S260. Set the execution parameters of the target actuator based on the wafer edge health, wafer warpage value, and target actuator. S270. Combining the execution parameters of the target actuator and the sorting path, the wafers are sorted to the target workstation to obtain wafer sorting information; S280. Store the wafer sorting information in the warehouse entry record.

[0036] In this embodiment, the system can interface with a Manufacturing Execution System (MES) to obtain process path information related to the wafer to be inspected, as well as the workstation load information for each corresponding process node, to assist in intelligent decision-making for subsequent wafer scheduling, diversion, sorting, and warehousing path planning. Upon receiving the identification information of the target wafer transport box, based on the wafer's identification code within the transport box, a query request is initiated to the interfaced MES system or process flow database to obtain the complete process flow path information corresponding to that wafer. The process path information includes, but is not limited to, process flow number and version number, process step sequence, process type corresponding to each step, and key process parameter requirements. After obtaining the candidate workstations corresponding to the current process step of the wafer, real-time operating load information for each candidate workstation is obtained from the scheduling management system or workstation control module.

[0037] Next, wafer edge health is determined using ring structured light imaging. Wafer edge health is a comprehensive score or grade index used to quantify the geometric integrity and surface condition of the wafer edge region. Surface condition refers to the presence of contamination, residue, abnormal films, etc. A normal, healthy wafer edge will exhibit regular and symmetrical characteristics in the ring structured light image. When problems such as microcracks or wear exist at the edge, the reflection and refraction characteristics of the ring light change, resulting in irregular edge contours and abnormal light intensity distribution in the image. Image analysis algorithms can quantify these abnormal features to assess edge health. These algorithms can include techniques such as histogram equalization and median filtering. Image analysis algorithms are used to filter and remove noise from the image, and the Sobel operator is used to extract the wafer edge contour and abnormal light intensity regions. For example, certain health indicators can be set, which can be determined by process requirements. Edges exceeding these indicators are considered unhealthy. Subsequently, wafer warpage is determined using a second spectral image. The second spectral image is an optical image of the wafer surface, containing certain spectral parameters, such as the absorption coefficient in a specific wavelength band and the spectral phase caused by changes in refractive index. When the second spectrum is projected onto the wafer surface, reflection occurs. If the wafer surface has a perfectly flat warp, the phase distribution of the spectrum in the image will be uniform. If there is a bulge or depression at a point on the wafer, the optical path (distance the light travels) of the reflected light will change. For example, for every 1 μm increase in height, the phase difference increases by a fixed value, determined by the spectral wavelength. By analyzing the phase difference of different pixels in the second spectral image using algorithms (such as phase unwrapping algorithms), the three-dimensional morphology of the wafer can be obtained, and the wafer warp value can be calculated. The wafer warp value is a quantitative indicator of the degree of non-planar deformation caused by internal stress (such as processing, temperature changes, material differences, etc.), and is primarily used to measure the flatness of the wafer surface.

[0038] After obtaining the wafer edge health and wafer warpage values, the system determines the wafer quality grade based on these values. The system sets edge health and warpage thresholds according to the company's process requirements. If both the wafer edge health and warpage values ​​are greater than or equal to their respective thresholds, the wafer is considered a qualified wafer. If either the wafer edge health or warpage value is less than its corresponding threshold, the wafer is considered a defective wafer.

[0039] In a semiconductor manufacturing plant, there are multiple production stations, such as photolithography, etching, doping, and packaging stations. Station load information refers to the current workload of each station, including equipment status and task queues. Different chip manufacturing processes have different characteristics, such as logic chip manufacturing processes and memory chip manufacturing processes. Even within the same broad category of processes, there are subdivisions, such as FinFET and FD-SOI processes in logic chips. Process type labels are used to identify the specific process type that the wafer is applying to or undergoing. Specifically, based on wafer quality level as the primary criterion, wafers are first divided into qualified and defective groups to exclude the possibility of unqualified wafers entering a certain path. Qualified wafers are only allowed to enter the qualified product sorting path and are prohibited from flowing to the rework or scrap area. Defective wafers are only allowed to enter the defective product sorting path and are prohibited from flowing to normal process stations. Next, for the classified qualified and defective groups, the target process is determined according to the process type labels, clarifying the general direction of the path. For example, the process type label of a qualified wafer directly corresponds to the next normal process, locking the path direction. The process type label for a Grade A wafer is "Photolithography Process" → the path direction is locked to "High-end Photolithography Line Area". The process type label for a defective wafer corresponds to "Processing Process", locking the path direction. For example, the process type label for a Grade C wafer is "Edge Grinding Rework" → the path direction is locked to "Rework Line". Based on the determined target process, combined with workstation load information, specific workstations with remaining capacity are selected, and the final destination of the path is determined to avoid wasted capacity. For example, workstations with a load rate ≤85% are prioritized to reserve buffer capacity and avoid congestion caused by sudden tasks. If all workstations of the same type are at full capacity, the workstation that releases capacity earliest is assigned, specifically determined by the system predicting the time it will take for the workstation to complete its current task.

[0040] The target actuator is determined by selecting a preset actuator based on the wafer edge health and wafer warpage value. In semiconductor manufacturing or wafer processing scenarios, an actuator refers to a device or apparatus capable of performing specific operations on the wafer. In this embodiment, the preset actuators are suction actuators and clamping actuators, which are a series of actuators predefined and configured by the factory according to different process requirements, wafer characteristics, and equipment functions. Suction actuators use the principle of vacuum adsorption, utilizing suction cups to contact the wafer surface and create a vacuum environment to adsorb and fix the wafer; clamping actuators use mechanical grippers to clamp and fix the wafer from the edge. If the wafer warpage value is small, suction actuators are preferred because they cause less damage to the wafer surface and better protect the wafer's quality during handling and other operations. If the wafer warpage value is large but the edge health is good, clamping actuators are selected. Although clamping actuators have a potential risk of edge pressure, their warpage tolerance advantage is more important in this case, and the risk of edge damage can be reduced by precisely controlling the gripper force.

[0041] After determining the edge health and warpage value of each wafer, the execution parameters of the target actuator are adaptively set based on the corresponding quality characteristics of the wafer to achieve precise handling control of wafers of different quality levels, thereby improving the stability and safety of the overall system. When the system detects that the edge health of a wafer is below the edge health threshold (e.g., there is obvious edge chipping or edge defects), the system will automatically reduce the action intensity of the target actuator, for example, by reducing the clamping force or vacuum suction, limiting the moving speed and acceleration, to prevent wafer breakage or handling failure due to edge damage. For wafers with large warpage values, the clamping path and angle or adsorption area of ​​the elastic clamping mechanism can be adjusted to improve the adaptability to wafers with non-ideal shapes and ensure stability during the gripping process. The above action parameters can be set through the programmable interface in the actuator controller, or dynamically adjusted in real time by the edge computing module by sending control commands based on the wafer evaluation results. For high-health, low-warpage wafers, a rigid adsorption end effector is used; for low-health, high-warpage wafers, a flexible adsorption + elastic clamping composite end effector is preferred. By setting the execution parameters of the target actuator, the action mode and control parameters of the target actuator can be dynamically set according to the actual wafer state, thereby improving the stability, safety and adaptability of the handling process and avoiding operational errors or physical damage caused by fixed execution parameters.

[0042] After setting the execution parameters of the target actuator, the system further combines the acquired target sorting path information to complete the precise sorting operation of wafers from the current workstation to the target workstation, and generates wafer sorting information including wafer physical state, transport path, actuator parameters, etc., for recording and traceability. The system obtains the target process path and sorting strategy for each wafer from the manufacturing execution system, including the target workstation number, to form a sorting path list. Based on the previously set target actuator execution parameters, the system dynamically adjusts the control commands of the target actuators, driving the end effector according to the path commands to precisely transport each wafer along the preset path to its corresponding target workstation, such as a temporary storage position, an inspection position, or a re-inspection position. After sorting, the system records the following information in a structured manner to form wafer sorting information, such as the wafer's unique code, wafer quality grade, and the actual sorting path and target workstation number. Finally, after completing the identification, evaluation, and sorting operations for each wafer, all processing information for that wafer is saved as data records in the wafer warehousing database or record table for subsequent traceability, query, quality analysis, or inventory management.

[0043] By combining wafer edge health and wafer warpage value to determine wafer quality level, identify target actuators, and set parameters for the target actuators, a more comprehensive and detailed assessment of wafer quality can be achieved. This optimizes sorting paths, improves actuator efficiency, and helps analyze whether the problem lies in the sorting process or the quality of the wafer itself, thus enabling targeted improvements.

[0044] In one embodiment of this invention, the preset actuator includes an adsorption actuator and a clamping actuator. Selecting the preset actuator and determining the target actuator based on wafer edge health and wafer warpage value includes: S310. When the wafer edge health is greater than the first health threshold and the wafer warpage value is less than the first warpage threshold, select the adsorption actuator as the target actuator. S320. When the wafer edge health is less than the second health threshold and the wafer warpage value is greater than the second warpage threshold, select the clamping actuator as the target actuator.

[0045] After image acquisition and parameter extraction of the wafer, the edge health and warpage value of the wafer are first determined based on the ring structured light image and spectral image, respectively. Then, the detected parameters are compared with preset reference thresholds. If the wafer edge health is greater than the first health threshold and the wafer warpage value is less than the first warpage threshold, the wafer is considered to have good mechanical stability and edge structural integrity, and is not prone to breakage or edge chipping. In this case, an adsorption-type end effector is preferentially selected as the target actuator. The first health threshold is a standard value for measuring the edge health of the wafer, determined through extensive experiments, production experience, and comprehensive analysis of chip manufacturing process requirements. The first warpage threshold is also a standard value for measuring the degree of wafer warpage, determined based on chip manufacturing processes and the operating characteristics of the adsorption-type actuator. It should be noted that not all actuators with adsorption functions can be used. The selected adsorption actuator must meet the matching requirements of wafer size, warpage, and edge health, including but not limited to the following parameters: the suction cup vacuum pressure should be sufficient to ensure stable wafer adsorption during handling without damage; the hardness of the suction cup material should be suitable for the wafer surface and edge structure to prevent indentation or edge breakage; the suction cup diameter should match the wafer size and warpage state to ensure uniform and stable force. Therefore, only when the various operating parameters of the adsorption actuator meet the above conditions can it be used as the target actuator for wafer handling. By determining the target actuator, this embodiment can safely and reliably select a suitable adsorption actuator to achieve precise wafer handling and subsequent processing operations while ensuring wafer mechanical stability and edge integrity.

[0046] When a wafer has edge defects and significant warpage, it is no longer suitable for suction handling. A more stable, gripping actuator with stronger coverage is selected for transport to reduce the risk of wafer drop and breakage. Based on the assessment of wafer edge health and warpage value, the suitability of suction handling is determined. If the wafer edge health is detected to be below a preset second health threshold and the wafer warpage value is above a preset second warpage threshold, the wafer is deemed to have minor edge chipping, cracks, or other issues, and its overall deformation is high. Therefore, it is unsuitable for vacuum or electrostatic suction actuators to avoid uneven suction, suction failure, or drops during transport. Thus, the system automatically selects a more stable, stronger, and wider-coverage gripping end effector as the target actuator. The second health threshold is another standard value corresponding to the first health threshold, but it represents a lower edge health requirement, indicating a worse wafer edge condition. Similarly, the second warpage threshold represents a higher standard for wafer warpage, indicating more severe wafer warpage. The second warpage threshold and the second health threshold can be determined based on extensive experimental and production experience, as well as a comprehensive analysis of chip manufacturing process requirements. It is important to note that not all actuators with clamping functions are suitable. The selected clamping actuator must meet the matching requirements of wafer size, edge defects, and warpage degree, including but not limited to the following parameters: the clamping force threshold should be sufficient to clamp the wafer while avoiding indentations or edge damage; the clamping coverage angle should ensure wafer stability and uniform force distribution; the clamping jaw shape and material hardness should be adapted to the wafer surface and edge structure to avoid slippage or damage. Therefore, only when the operating parameters of the clamping actuator meet the above conditions can it be used as the target actuator for wafer handling. By determining the target actuator, this embodiment can safely and reliably select a suitable clamping actuator to achieve precise wafer handling and subsequent processing operations even when wafer edges are defective or severely warped.

[0047] Selecting the appropriate actuator based on wafer edge health and wafer warpage value can improve the safety of wafer handling, reduce handling failures or repetitive operations caused by actuator mismatch, and thus improve overall production efficiency.

[0048] In one embodiment of this invention, the polar coordinates of each wafer are determined based on the ring structured light imaging of each wafer, and a first attitude angle corresponding to each wafer is determined based on the polar coordinates of each wafer, including: S401. Collect the reference light intensity value of ring structured light imaging under the reference temperature condition and the corresponding light intensity value of ring structured light imaging under different temperature field conditions, including high temperature environment. S402. Calculate the polarization state attenuation coefficient under different temperature field conditions by combining the light intensity values ​​under different temperature field conditions with the reference light intensity values ​​under the reference temperature conditions. S403. The polarization degree temperature correction formula is obtained by fitting the polarization state attenuation coefficient under different temperature field conditions. S404. Calculate the theoretical polarization attenuation under high temperature conditions using the polarization degree temperature correction formula; S405. Compare the actual light intensity value of the ring structured light imaging with the reference light intensity value corresponding to the theoretical polarization attenuation to determine the thermal radiation interference. S406. Combine the thermal radiation interference amount to perform thermal radiation interference suppression processing on the ring structured light imaging under high temperature environment to obtain the target ring image. S407. Perform circle fitting between the target ring image and the ring structured light imaging under the reference temperature conditions to obtain the circle center coordinate parameters, including the high temperature circle center coordinates. S408. Calculate the theoretical thermal expansion offset based on the preset linear expansion coefficient of the wafer; S409. The coordinates of the high-temperature center and the positioning gap of the wafer are corrected by combining the theoretical thermal expansion offset to obtain the compensated pole coordinates and polar axis direction. S410. Determine the first attitude angle by combining the compensated pole coordinates and polar axis direction.

[0049] In one embodiment of this invention, to improve the imaging stability and edge recognition accuracy of the ring structured light image under different ambient temperatures, the ring structured light imaging is performed under multiple temperature field conditions (including room temperature, high temperature, or local thermal interference environment). A preset polarization adjustment module adjusts the polarization state of the structured light source to give it a specific polarization direction. Then, a polarization camera acquires an image of the wafer, obtaining a ring light stripe image with a specific polarization state. This image not only reflects the geometric contour information of the wafer edge but also contains polarization response information related to material properties, film state, and micro-deformation. This information can be used for high-precision analysis and judgment in subsequent stages such as attitude angle detection, warpage correction, and edge health assessment.

[0050] Subsequently, ring-shaped structured light images were acquired under various temperature conditions (including but not limited to 40℃, 60℃, 80℃, and high-temperature environments), and their light intensity values ​​were extracted and compared with reference light intensity values. Next, the polarization state attenuation coefficient under different temperature conditions was calculated by combining the light intensity values ​​under different temperature conditions with the reference light intensity value under the reference temperature condition. By comparing the light intensity difference between the reference temperature and the target temperature field, the intensity attenuation ratio of polarized light at a specific temperature was calculated. This coefficient can be used to eliminate the interference of temperature on polarized light imaging, ensuring the accuracy of wafer defect (such as microcracks) and warpage detection based on polarized light across the entire temperature range. The core calculation of the polarization state attenuation coefficient is through the relative attenuation coefficient formula, as shown below:

[0051] Where Kt represents the polarization state attenuation coefficient under the target temperature field; It represents the light intensity value at the target temperature t; and I0 represents the reference light intensity value at the reference temperature. For example, when the target temperature t = 300℃, It = 153, I0 = 180, Kt = 153, I0 = 180, and Kt = 153. 300° =1-153 / 180=0.15, which means that at 300℃, the intensity of a ring light in a specific polarization state is reduced by 15% after reflection from the wafer compared to the reference temperature. This reduction is caused by temperature.

[0052] A polarization degree temperature correction formula is derived by fitting the polarization state attenuation coefficient under different temperature field conditions. This involves acquiring polarization imaging data of ring structured light under various temperature environments, calculating the polarization state attenuation coefficient relative to a reference temperature at each temperature, and then mathematically fitting these temperature-attenuation data points to obtain a correction function for polarization degree as a function of temperature. The polarization degree temperature correction formula is used in actual detection to dynamically compensate for image intensity and polarization degree under high / low temperature conditions, thereby improving the stability and accuracy of tasks such as attitude angle, edge health, and notch recognition. The polarization state attenuation coefficient under each temperature field is calculated using the aforementioned relative attenuation coefficient formula, and the final data is compiled into a discrete dataset of temperature-polarization state attenuation coefficients. This discrete dataset is then fitted using a quadratic function model, which has the following form: Kt=a×T 2 +b×T+c Where a (quadratic coefficient), b (linear coefficient), and c (constant term) are the fitting parameters to be determined, T is the temperature (°C), and Kt is the polarization state attenuation coefficient.

[0053] Next, using the least squares method and Matlab software, the sum of squared residuals between the fitted curve and the discrete data points is minimized. The values ​​of a, b, and c in the quadratic function model are then solved. The discrete dataset of the temperature-polarization state attenuation coefficient is input into the quadratic function model, resulting in multiple equations. These equations are solved using an algorithm to obtain the fitting parameters. Substituting these parameters into the quadratic function model yields the temperature fitting formula for the polarization state attenuation coefficient. For example, in the discrete dataset of the temperature-polarization state attenuation coefficient: When T = -10℃ and Kt = -0.011, -0.011 = a × (-10) 2 +b×(-10)+c; When T=100℃ and Kt=0.078, 0.078=a×100 2 +b×100+c; When T=300℃ and Kt=0.150, 0.150=a×300 2 +b×300+c; When T=500℃ and Kt=0.222, 0.222=a×500 2 +b×500+c.

[0054] The discrete dataset of temperature-polarization state attenuation coefficients was obtained through a large amount of historical experimental data, with fitting residuals within ±2×10⁻⁶. -4 Within, the coefficient of determination R 2 =0.997, verifying the effectiveness and stability of the model within this temperature range. The fitting residuals were obtained by calculating the difference between the observed values ​​and the fitted values, and their absolute values ​​were all less than ±2×10. -4 The coefficient of determination, obtained by comparing the sum of squares of the residuals with the sum of squares of the observed values, is 0.997, indicating that the temperature correction formula can effectively reflect the relationship between polarization degree and temperature within the set temperature range. Through calculation, the final fitting parameter is obtained as a≈5.1×10⁻⁶. -7 b≈1.9×10 -4 Since c≈-0.005, substituting this into the quadratic function model yields the temperature fitting formula for the polarization state attenuation coefficient: Kt=5.1×10 -4 ×T 2 +1.9×10 -4 ×T-0.005.

[0055] The theoretical polarization attenuation under high-temperature conditions is calculated using the polarization degree temperature correction formula. This formula is derived by measuring the polarization state attenuation coefficient under multiple temperature field conditions and fitting the data, describing the quantitative relationship between the polarization state attenuation coefficient and temperature. It is known that the polarization degree temperature correction formula describes the quantitative relationship between the polarization state attenuation coefficient Ht and temperature (T). If the polarization degree (P) TThe relationship between P and the polarization state attenuation coefficient is P T =P0×Ht(P T P0 is the degree of polarization at temperature T, and P0 is the degree of polarization at the reference temperature. The theoretical polarization attenuation can be expressed as the degree of polarization P0 at the reference temperature and the degree of polarization at high temperature T. high Theoretical polarization degree P Thigh The difference, i.e., the polarization attenuation, is P0 - P Thigh For example, calculating the high temperature T. high =Polarization state attenuation coefficient HT at 150℃ high By substituting HT into the polarization degree temperature correction formula high =5.1×10 -4 ×T 2 +1.9×10 -4 The calculation is performed in ×T-0.005, HT high =0.0349; Next, calculate the theoretical degree of polarization P at high temperature. Thigh According to P T =P0×Kt, given P0=0.9, HT high =0.0349, then P Thigh =0.9 × 0.0349 = 0.0314775. Calculating the theoretical polarization attenuation, the polarization attenuation is = P0 - P Thigh =0.9-0.0314775=0.8685225. Here, polarization attenuation represents the degree to which the polarization characteristics of light are weakened or changed after the interaction between the wafer and the temperature field. It can be measured by the polarization attenuation coefficient; the smaller the polarization attenuation coefficient, the greater the polarization attenuation.

[0056] The thermal radiation interference is determined by comparing the actual light intensity value of the ring structured light imaging with the reference light intensity value corresponding to the theoretical polarization attenuation. Under high-temperature conditions, the actual light intensity value I of the ring structured light imaging is... actual It is affected by polarization state attenuation and thermal radiation. Polarization state attenuation occurs when high temperatures alter the material properties of an object, leading to a decrease in the polarization state of light, which in turn theoretically reduces the light intensity (corresponding to the theoretical light intensity value I). theoretical Thermal radiation occurs when a high-temperature object (such as a wafer at high temperature) radiates heat outward, producing thermal radiation light. This thermal radiation light is superimposed on the imaging light intensity of the ring structured light, causing the actual light intensity value I to change. actual The deviation from the theoretical light intensity value I determined solely by polarization state attenuation theoretical Specifically, firstly, at a reference temperature T0 (e.g., 25°C), multiple ring-shaped structured light imaging measurements are performed on the wafer, and the average value is taken as the reference light intensity value I0. For example, after measurement and statistics, I0 is determined to be 1000 (the unit is set according to the actual measurement, such as any light intensity unit). Next, the theoretical light intensity value I is calculated. theoreticalDetermine the current high-temperature ambient temperature T high (e.g., 150℃), calculate the theoretical polarization attenuation using the polarization degree temperature correction formula, based on I theoretical =I0×(1-theoretical polarization attenuation), calculate I theoretical =1000×(1-0.8)=200. Measure the actual light intensity value I. actual In T high Ring structured light imaging was performed on the actual tested wafer at an environment of 150℃, and the light intensity value I was read and recorded by the camera. actual Assume that the measurement yields I. actual =250. Substitute into formula I interference =I actua -I theoretical =250-200=50. This indicates that at a high temperature of 150℃, thermal radiation increases the light intensity by an additional 50 units (according to actual units), meaning the thermal radiation interference is 50.

[0057] When performing ring structured light imaging in a high-temperature environment, the thermal radiation radiated from the equipment or wafer surface will superimpose a large amount of invalid light signals into the image, interfering with the clarity and contrast of the structured light stripes, and thus affecting the accuracy of notch recognition and attitude angle extraction. Therefore, it is necessary to first calculate the amount of thermal radiation interference based on the difference between the theoretical polarized brightness after temperature correction and the measured image brightness, and then perform targeted suppression in the image domain or frequency domain to remove the interference components and restore a target ring image that can be used for image recognition and analysis. The calculated amount of thermal radiation interference is manifested in the image as a uniform or non-uniform brightness shift, and the brightness value of each pixel can be adjusted. If it is a uniform increase in brightness, the average interference amount can be subtracted from the measured image brightness. In more complex cases, if the interference amount is different in different areas of the image (e.g., strong thermal radiation interference at the image edges and weak in the middle), the interference amount of each pixel can be calculated as ΔL(x,y) (x,y are pixel coordinates), and the measured brightness L can be adjusted accordingly. 实测 (x,y) is corrected pixel by pixel: L 修正 (x,y)=L 实测 (x,y)-ΔL(x,y) is used to obtain the target ring image by correcting the pixels.

[0058] Subsequently, the target annular image is fitted with the annular structured light image under the reference temperature condition to obtain the center coordinate parameters. These center coordinate parameters include the high-temperature center coordinates. In this embodiment, the target annular image is obtained under high-temperature conditions after thermal radiation interference suppression processing. For the target annular image (at high temperature), a circle fit is performed on its annular structured light edge pixels. A least-squares circle fit algorithm can be used to extract the center coordinates and radius parameters from the image, denoted as the high-temperature center coordinates and high-temperature radius. For example, in the wafer inspection embodiment, the annular contour of the wafer may undergo slight changes due to thermal expansion and other factors at high temperatures. By extracting the edge pixels of the annular light in the high-temperature target annular image and performing a circle fit, the high-temperature center coordinates (x, y, z) are obtained. 高温 y 高温 ) and high temperature radius r 高温 Similarly, a circle fit was performed on the ring structured light image (wafer ring structured light image at room temperature) under the reference temperature condition to obtain the reference circle center coordinates (x... 基准 y 基准 (x) 基准 y 基准 ) and reference radius r 基准 Through the above circle fitting operation, two sets of circle center coordinates were obtained. Among them, the high-temperature circle center coordinates are the result of circle fitting of the target ring image under high temperature environment, recording the geometric center position of the ring structure after the influence of high temperature.

[0059] The theoretical thermal expansion offset is calculated based on the preset linear expansion coefficient of the wafer. The linear expansion coefficient is a physical quantity characterizing the thermal expansion properties of an object, representing the relative change in its length when the temperature changes by 1°C. The preset linear expansion coefficient is set according to the wafer material; for example, the linear expansion coefficient of a silicon wafer is approximately 2.6 × 10⁻⁶. -6 / °C. The temperature rise is calculated based on the difference between the real-time temperature and the reference temperature under high-temperature conditions. The radius R is extracted from the circle fitting in the image. ref (Radius of a circle at the reference temperature), calculate the theoretical thermal expansion offset ΔR = preset linear expansion coefficient × R ref × Temperature rise. The theoretical offset (ΔCx) that maps ΔR to the center coordinates of the circle is used as a thermal expansion compensation factor when comparing the center of the target image with the center of the reference image. This is used to remove the components caused by thermal expansion from the observed offset.

[0060] When performing ring structured light imaging on a wafer at high temperatures, the wafer's center position and the location notch position in the image may shift due to thermal expansion, leading to errors in subsequent attitude recognition or notch alignment. To eliminate this imaging error caused by thermal expansion, this implementation scheme first calculates the theoretical thermal expansion offset based on the material's linear expansion coefficient and the actual temperature rise. Based on this, the center coordinates fitted in the high-temperature image and the wafer notch position identified in the image are corrected at the coordinate level, thus obtaining the pole coordinates (i.e., the thermally compensated center) and polar axis direction (i.e., the direction of the notch connection line after thermal compensation) in the thermally compensated polar coordinate system. Specifically, this involves calculating the theoretical thermal expansion offset based on the wafer's linear expansion coefficient, its dimensions at the reference temperature (e.g., diameter), and the current high temperature, resulting in the theoretical offset of the wafer's spatial position (e.g., center coordinates) due to thermal expansion, such as the offset Δx in the x and y directions. 理论 Δy 理论 The theoretical thermal expansion offset is the positional shift of the wafer under ideal conditions due to thermal expansion. However, the high-temperature center coordinates (x...) 高温 y 高温 This includes actual thermal expansion and other errors such as imaging. By subtracting the theoretical thermal expansion offset, the coordinates of the high-temperature circle's center can be corrected, making it closer to the position caused only by other non-thermal expansion factors. The corrected center coordinates (polar coordinates) (x...) 补偿 ,y 补偿 The calculation is as follows: x 补偿 =x 高温 -Δx 理论 y 补偿 =y 高温 -Δy 理论

[0061] Then, the coordinates (x, y) of the positioning notch under high temperature were first determined. 缺口高温 y 缺口高温 ) converted relative to the high temperature center (x) 高温 ,y 高温 polar coordinates (r) 高温 ,θ 高温 ):

[0062]

[0063] Considering theoretical thermal expansion, the wafer expands overall (center shift Δx). 理论 Δy 理论 Assuming the position of the notch relative to the wafer (distance and angle relative to the center) changes under thermal expansion alone: ​​the radius r becomes r = r 高温+Δr, where Δr is calculated based on the linear expansion coefficient and the effect of temperature change on the distance from the notch to the center. Generally, for uniform expansion, Δr = αr. 基准 ΔT, r 基准 It is the distance from the notch to the center of the circle at the reference temperature, with the polar angle θ remaining unchanged. However, a simpler method is to use the compensated center (x) 补偿 ,y 补偿 Let be the pole, and calculate the coordinates of the center of the circle relative to the notch after compensation: x 缺口补偿 =x 缺口高温 -Δx 理论 y 缺口补偿 =y 缺口高温 -Δy 理论 Then (x) 缺口补偿 ,y 缺口补偿 Convert to polar coordinates (r) relative to the compensated center (pole). 补偿 ,θ 补偿 ):

[0064]

[0065] Here θ 补偿 This refers to the corrected polar axis direction (if the polar axis direction is defined as the positive x-axis direction, θ). 补偿 θ represents the angle of the notch relative to the polar axis; if the polar axis direction has other definitions, such as according to a fixed direction in the process, θ 补偿 Then calculate according to the corresponding definition.

[0066] Subsequently, based on the vector relationship between the corrected wafer notch position and the compensated center coordinates, the compensated polar axis direction is calculated. A polar coordinate system is then established using the compensated pole coordinates as the pole and the compensated polar axis direction as the polar axis. The angle between this polar axis direction and the system reference direction is calculated and defined as the wafer's first attitude angle. This first attitude angle reflects the wafer's rotational orientation in the current coordinate system and is used for subsequent wafer attitude standardization and micro-rotation calibration operations. The angle between this polar axis direction and the system reference direction can be calculated based on the difference between the angle at the current wafer notch position and the angle of the system reference direction.

[0067] By processing the ring structured light imaging under high temperature conditions to accurately determine the polar coordinates and first attitude angle of each wafer, the measurement accuracy, positioning accuracy and production stability of wafers under high temperature conditions can be significantly improved, thereby ensuring the positioning accuracy of wafers.

[0068] In one embodiment of this invention, determining the first attitude angle by combining the compensated pole coordinates and the polar axis direction includes: S501. Establish a first polar coordinate system based on the compensated pole coordinates and polar axis direction; S502. Select ring light bar feature points in the target ring image at preset angle intervals, and convert the pixel coordinates of the ring light bar feature points into world coordinates; S503. In the first polar coordinate system, for any annular light stripe feature point in any world coordinate system, calculate the initial polar radius and initial polar angle of the annular light stripe feature point using the rectangular coordinate transformation relationship based on the polar coordinates corresponding to the annular light stripe feature point. S504. Calculate the thermal expansion correction value and polarization correction value using the preset thermal expansion correction term and the preset polarization optical path difference correction term, respectively. S505. The final polar diameter is obtained by adding the thermal expansion correction value, polarization correction value and initial polar diameter. S506. Correct the initial polar angle using the adjacent three-point weighted smoothing algorithm to obtain the final polar angle; S507. Obtain the temperature measurement value corresponding to each annular light stripe feature point and the deviation between the final polar radius and the preset reference radius, and construct the thermal distortion basis function term based on the temperature measurement value; S508. Perform Fourier analysis on the deviation between the final extreme radius phase and the preset reference radius to obtain the first harmonic term of the wafer attitude. S509. Using the final polar angle as the independent variable and the deviation of the final polar radius from the preset reference radius as the observation, a joint model of the observation is constructed by combining the independent variable, the observation, the distortion basis function term, and the first harmonic term. S510, Solve the first attitude angle of the joint model wafer based on the least squares criterion.

[0069] First, a first polar coordinate system is established based on the compensated pole coordinates and the polar axis direction. Specifically, the compensated pole coordinates (x, y, y) are used as the coordinates. 补偿 ,y 补偿 Let O be the origin (pole O). Establish a polar coordinate system with a defined polar axis direction. The polar axis direction can be the positive x-axis or determined based on the corrected positioning notch direction. For any point Q on the plane, which can be another feature point on the wafer, a reference point in the process equipment, etc., its position in this polar coordinate system can be represented as Q(r). Q ,θ Q ), where r Q It is point Q to the extreme point (x) 补偿 ,y 补偿 The distance θ Q θ is the angle θ traversed by rotating counterclockwise from the positive polar axis to line segment OQ.Q It can be obtained through the arctangent function θ Q =arctan2(y Q -y 补偿 ,x Q -x 补偿 Therefore, in this embodiment, the first polar coordinate system refers to a local polar coordinate system established with the compensated center of the circle as the pole, the compensated polar axis direction as the zero-degree direction (0°), and counterclockwise as the positive direction.

[0070] Next, for each target annular image, a circumferential scan is performed with the wafer center as the pole, at preset angular intervals (e.g., every 10°). The Canny edge detection algorithm is used to extract the light stripe feature points at the corresponding angles from the annular light stripes. The annular light stripe feature points can be stripe edge points, intensity center points, or local extrema points. The target annular image refers to a standardized annular structured light image that can be used for analysis, obtained under high-temperature conditions after processing such as thermal radiation suppression and pole and polar axis correction. The image contains clearly identifiable annular structured light stripes. Selecting annular light stripe feature points at preset angular intervals involves scanning the circular structured light image around the center (pole) at equal angular intervals, such as selecting a feature point every 5°, 10°, or 15°. The preset angular interval can be pre-set according to detection requirements and accuracy requirements, specifying the angular interval at which feature points are selected on the annular circumference. Subsequently, the pixel coordinates of the annular light stripe feature points are converted to world coordinates. Each feature point is initially located in the pixel coordinate system of the image and is affected by camera angle, distortion, etc. To perform real-world calculations, the coordinates need to be converted to actual world coordinates. This can be done by applying homography or depth estimation (such as Time-of-Flight or stereo vision) to calculate the spatial location. A homography matrix is ​​a transformation matrix that maps a two-dimensional image coordinate system to another two-dimensional plane (usually the world plane), suitable for situations where the camera and the target surface are coplanar or approximately parallel. The pixel coordinates of the annular light stripe feature points can be converted to world coordinates using homography and other methods.

[0071] In the first polar coordinate system, for any annular fringe feature point in world coordinates, the initial polar radius and initial polar angle of the feature point are calculated using the Cartesian coordinate transformation relationship based on the polar coordinates corresponding to the feature point. In other words, the position of the feature point in world coordinates is transformed into the initial polar radius and initial polar angle in the first polar coordinate system using mathematical conversion formulas between polar and Cartesian coordinates. For example, the pole of the first polar coordinate system is the compensated wafer center world coordinate (x...). 补偿 ,y 补偿 ), for example (x 补偿 ,y 补偿The actual value is calculated based on a correction, with the polar axis being the positive x-axis. Through coordinate transformation, the world coordinates of a ring-shaped light stripe feature point are obtained as (X,Y)=(120,110). This represents the position of the feature point in the actual detection space, such as a point on the wafer edge. Using the above polar diameter calculation formula, r is calculated to be approximately 22.36 mm. r is the initial polar diameter of the ring-shaped light stripe feature point in the first polar coordinate system, representing the distance from the feature point to the pole (wafer center). Next, using the above polar angle calculation formula, θ is calculated to be approximately 26.6°. θ is the initial polar angle of the ring-shaped light stripe feature point in the first polar coordinate system, representing the angle from the polar axis (positive x-axis direction) counterclockwise to the line connecting the feature point and the pole.

[0072] By combining the linear expansion coefficient of the wafer material and the temperature difference between the actual operating temperature and the reference temperature, a corresponding thermal expansion correction value is calculated using a preset thermal expansion correction term. This thermal expansion correction value is used to correct the offset of the wafer's edge contour or center coordinates under high-temperature conditions. Secondly, considering the relative change in the polarization path length in the optical system under high-temperature conditions, a preset polarization path difference correction term is used to quantify and compensate for the intensity changes in the structured light image, obtaining a polarization correction value. These two correction values ​​are used in subsequent polar coordinate calculations and attitude angle fitting steps, respectively, to achieve dual suppression of image thermal distortion and polarization interference, thereby improving the stability and robustness of the entire ring structured light recognition module. The preset thermal expansion correction term is: Thermal expansion correction value = αD0(T-T0), for example, given the linear expansion coefficient of the wafer material α = 2.6 × 10⁻⁶. -6 / °C, reference temperature T0=25°C (room temperature), current high temperature T=400°C, wafer diameter D0=300mm at reference temperature, can be measured using high-precision measuring equipment at reference temperature. Calculate thermal expansion correction value = 2.6×10 -6 / °C×300mm×(400-25)°C=0.2925mm, this 0.2925mm is the thermal expansion correction value, representing the theoretical increase in wafer diameter due to thermal expansion. Next, the polarization correction value is calculated using a preset polarization path difference correction term, MT, where M is an empirical coefficient, calibrated through extensive previous experiments to be M=1×10. -10 m / °C, where T is the temperature. The preset polarization path difference correction term is derived through experimental and theoretical analysis for the polarized ring structured light used in this detection system. For example, if the current temperature T = 400°C, then the polarization correction value is: Polarization correction value = MT = 1 × 10⁻⁶ m / °C. -10 m / °C × 400°C = 4 × 10 -8 m=40nm.

[0073] The final polar diameter is calculated by adding the thermal expansion correction value, the polarization correction value, and the initial polar diameter. The thermal expansion correction value is calculated based on the linear expansion coefficient of the wafer material and the temperature difference, and is used to compensate for image shift caused by the expansion of the wafer's physical dimensions. The polarization path difference correction value is calculated by considering the current image's polarization contrast change, and is used to compensate for the interference of high-temperature polarization distortion on image edge features. The final polar diameter is obtained by adding the initial polar diameter, the thermal expansion correction value, and the polarization correction value.

[0074] The initial polar angle is corrected using a weighted smoothing algorithm with adjacent three points to obtain the final polar angle. However, factors such as high-temperature thermal disturbances and image edge blurring can cause local jumps or discontinuities in the polar angle value, affecting the accuracy of subsequent attitude angle and notch orientation calculations. Therefore, a weighted smoothing algorithm with adjacent three points is introduced for polar angle correction. Specifically, for each polar angle point θ... i Introduce two adjacent points θ before and after it. i-1 θ i+1 The three points are weighted and averaged according to preset weight coefficients w1, w2, and w3 to obtain the final polar angle. The weight values ​​satisfy w1 + w2 + w3 = 1. The preset weight coefficients can be adjusted according to the actual situation, such as assigning a higher weight to the midpoint if it is considered more reliable.

[0075] The system acquires the temperature measurement value and the deviation between the final polar radius and the preset reference radius for each annular light stripe feature point, and constructs a thermal distortion basis function term based on the temperature measurement value. Specifically, it extracts the compensated final polar radius Ri for each annular light stripe feature point in the target annular image and compares it with the ideal radius R0 of the feature point in the cold state in the reference wafer to obtain the corresponding polar radius deviation value ΔRi = Ri - R0. Simultaneously, it acquires the surface temperature Ti at the location of each feature point using an infrared thermal imager or an embedded temperature sensing device, forming multiple temperature-deformation deviation data pairs {Ti, ΔRi}. Based on this, the system uses a polynomial fitting method to construct the thermal distortion basis function term f. 热(T) For example, a quadratic function model can be used: f 热(T) =a1T 2 +b1T+c1 Where a1, b1, and c1 are fitting parameters, calculated from the sample dataset using the least squares method. 热(T) The thermal distortion basis function term is used to characterize the radial thermal expansion trend of the wafer under different temperature fields. It can be subsequently applied to deformation prediction in areas without images or to perform global warpage compensation; T is the temperature value.

[0076] a1, b1, and c1 are calculated from the sample dataset using the least squares method, and then the fitted parameters are substituted into the quadratic function model to obtain the thermal distortion basis function term.

[0077] Subsequently, Fourier analysis was performed on the deviation between the final polar radius and the preset reference radius to obtain the first harmonic term of the wafer attitude. Specifically, for multiple annular light stripe feature points, the deviation at each point was calculated to form a deviation sequence, and the corresponding polar angle at each point was recorded to obtain a polar angle sequence. Discrete Fourier transform was performed on the deviation sequence and the polar angle sequence. The deviation was regarded as a function Δr(θ) of the polar angle (discrete sampling), and a Fourier series expansion was performed, the Fourier series expansion of which is:

[0078] Where a0 is the DC component (average deviation), a k and b k denoted as the coefficient of the kth harmonic; Δr(θ) is a function of the polar angle; θ is the polar angle.

[0079] Next, the coefficients a are calculated using the Discrete Fourier Transform. k and b k When k=1, the coefficients a1 and b1 of the first harmonic term are obtained. The first harmonic term is expressed as:

[0080] It can be further expressed in amplitude and phase form:

[0081] Where A1 is the amplitude of the first harmonic. The first harmonic phase. The first harmonic amplitude and first harmonic phase are the first harmonic terms of the wafer orientation.

[0082] Using the final polar angle as the independent variable and the deviation of the final polar radius from the preset reference radius as the observation, a joint model of the observation is constructed by combining the independent variable, the observation, the distortion basis function term, and the first harmonic term. In other words, to achieve decoupling of wafer attitude information and thermal distortion effect, the system uses the final polar angle value θi of each light stripe feature point in the image as the input variable and the deviation ΔRi of its final polar radius from the preset reference radius as the observation to construct the joint model, as shown below:

[0083] in, These are thermal distortion basis function terms; It is a first harmonic term; For fitting residuals; By fitting the parameters of the joint model using the least squares method, the simultaneous extraction of thermal distortion variables and attitude angle parameters is achieved, thereby improving the accuracy of wafer image decoupling modeling under complex thermal environments.

[0084] The constructed observational joint model uses the final polar angle as the input variable and the deviation of the polar radius of the annular light stripe feature point relative to the reference radius as the observed value, establishing a mathematical model including a thermal distortion function term and a first-order harmonic term. Subsequently, the least squares criterion is used to solve the parameters of this joint model, obtaining the cosine coefficient A1 and sine coefficient B1 in the first-order harmonic term. By substituting A1 and B1 into the formula:

[0085] The first attitude angle of the wafer is determined to characterize the overall tilt direction and tilt magnitude of the wafer in the imaging plane. This attitude angle serves as a key parameter for subsequent operations such as clamping, alignment, and warehouse attitude correction.

[0086] By calculating the first orientation angle of the wafer, the wafer's orientation can be accurately determined, providing a precise positioning basis for subsequent processing and inspection.

[0087] In one embodiment of this example, determining the process type label for each standard orientation wafer based on the second spectral image includes: S601. Construct a second polar coordinate system based on the standard orientation wafer, and generate a corresponding polar coordinate mask according to the second polar coordinate system. The polar coordinate mask is used to avoid the wafer notch area and the wafer clamping area. S602. Combine the polar coordinate mask and the second spectral image to extract the effective process region in the polar coordinate mask; S603. Determine the correspondence between process type labels and wafer spectral data based on preset historical labeling information. The preset historical labeling information includes wafer samples of process types and corresponding spectral images. S604. Combining the correspondence relationship, the mutual information maximization algorithm is used to select the sensitive area of ​​the process type in the effective process area; S605. Determine the mutual information measurement between sensitive areas and process type labels based on preset historical annotation information; S606. Assign corresponding weight parameters to each sensitive region based on the classification contribution between the sensitive region and the process type label. The classification contribution is obtained by measuring the mutual information between the sensitive region features and the process type label. S607. Perform spectral data processing on the sensitive area to obtain a multidimensional spectral feature set of the sensitive area; S608. Match the multidimensional spectral feature set with the preset process spectral dictionary and input it into the preset supervised discrimination model to obtain the process posterior probability corresponding to each sensitive region. S609. Combine the weight parameters and process posterior probabilities corresponding to each sensitive region to calculate the joint process posterior probability distribution. S610. Obtain prior information for the standard orientation wafer, including pre-set wafer process production data from the manufacturing execution system. S611. Perform Bayesian fusion of prior information and joint posterior probability to obtain the confidence level of process type. S612. When the confidence level of the process type is greater than the preset confidence level of the process type, output the process type label of the standard orientation wafer.

[0088] First, a second polar coordinate system is constructed based on the standard-position wafer. This second polar coordinate system uses the center of the standard-position wafer as its pole and a pre-defined polar axis direction (consistent with the positioning notch direction). Unlike the first polar coordinate system, the second polar coordinate system is an ideal coordinate system used for reference and standardization. Specifically, the pole of the second polar coordinate system is the center of the standard-position wafer, and the polar axis direction is set to be consistent with the positioning notch direction of the standard-position wafer, assuming the positioning notch polar angle is 0°. Based on the second polar coordinate system, a polar coordinate mask region is defined. The polar coordinate mask is a mask designed in the image space after polar coordinate transformation to avoid abnormal areas such as notches and gripper occlusion. The mask region is used to shield the notch and gripping areas of the wafer. Specifically, the angle range where the notch is located (e.g., 160°~200°) is shielded in the polar angle direction, and the gripper occlusion area (e.g., where r>98%R) is shielded in the polar radius direction. The generated mask is applied to the unfolded image in the form of a Boolean matrix. Pixels within the masked area are not processed in subsequent image analysis, thereby improving the stability and accuracy of the analysis.

[0089] A polar coordinate mask is generated based on a second polar coordinate system to mask notch and clamping regions in the wafer image. The mask marks valid and invalid regions in the image in Boolean matrix form. Combined with the polar coordinate mask, the valid image region within the mask's coverage area is extracted from the second spectral image. The second spectral image is a wafer edge image acquired through a multispectral camera or polarization imaging device, carrying spectral features related to the film structure and surface reflectance properties. After polar coordinate unfolding of the second spectral image, the polar coordinate mask and the unfolded image are matched point-by-point to extract all pixel information within the mask's effective range, forming the effective process region image. The effective process region is used for subsequent analysis and extraction of key parameters such as wafer edge health, film thickness uniformity, and process defect characteristics.

[0090] The system determines the correspondence between process type labels and wafer spectral data based on preset historical annotation information, which includes wafer samples of process types and their corresponding spectral images. In this embodiment, the system has a preset historical annotation information database, which includes multiple wafer samples with known process type labels and their corresponding spectral image data. Based on this database, the spectral distribution characteristics of wafer samples under different process types are analyzed, and key spectral parameters for process identification are extracted, such as frequency-selective reflectance, band energy ratio, and derivative spectrum peak value. Combining the key spectral parameters with the corresponding process type labels, a mapping model between process type and wafer spectral data is constructed. The model can employ a dimensionality reduction clustering algorithm based on principal component analysis or a neural network classifier for automatic identification and sorting of wafers with unknown processes in subsequent steps.

[0091] Based on the previously established correspondence between process type labels and wafer spectral data, regional information analysis is performed on the identified effective process regions. A mutual information maximization algorithm is then used to calculate the mutual information value between the spectral features of each sub-region and the process type label. The mutual information maximization algorithm determines which regions have stronger process discrimination capabilities by measuring the information gain between regional spectral features and process labels. Specifically, the mutual information maximization algorithm is implemented based on the logic for determining the sample size of historical labeled information and the threshold selection method. The historical labeled information sample size is the number of process type labels corresponding to the spectral features of each sub-region in the dataset used for training and calculating mutual information. This sample size can be historical production data or experimental data, where each sample contains the spectral measurement results of a certain sub-region and the accurate process type to which the sample belongs. The mutual information maximization algorithm receives these prepared historical labeled information samples as input. For the spectral features of each sub-region, the algorithm performs correlation analysis with the process type label to obtain the trained mutual information maximization algorithm. Subsequently, after calculating the mutual information value of the spectral features of each sub-region, a threshold is needed to distinguish regions with stronger process discrimination capabilities (i.e., sensitive regions) from regions with weaker discrimination capabilities. Regions with mutual information values ​​higher than this threshold are selected as sensitive regions. Threshold selection can be based on cross-validation, dividing the dataset into training and validation sets. Mutual information is calculated on the training set, and different thresholds are tried. Then, the effectiveness of sensitive region selection under different thresholds is evaluated on the validation set, and the best-performing threshold is selected. Regions are sorted based on mutual information values, and those with mutual information values ​​higher than a preset threshold are selected as sensitive regions for the target process type. The sensitive region for a process type refers to the sub-region in the effective process area where the mutual information between the detected data and the current process quality indicator is the largest. Changes in the detected data in this region are most sensitive to process quality fluctuations and can most accurately reflect whether the process is normal. Sensitive regions can serve as input regions for subsequent process classification models, spectral feature extraction modules, or sorting strategy rules, thereby improving the accuracy and computational efficiency of wafer process identification.

[0092] The mutual information metric between sensitive regions and process type labels is determined based on pre-set historical annotation information. This pre-set historical annotation information is information manually or automatically labeled on a large number of data samples in a specific domain. Mutual information, a concept in information theory, measures the correlation between two random variables, representing the degree to which the uncertainty of one random variable is reduced after knowing information about the other. Specifically, a set of wafer samples with known process labels and their corresponding spectral image data are acquired to construct a historical annotation information set. This historical annotation information includes the process type labels of multiple wafer samples and the pixel value distribution in their spectral images. Next, the spectral images are divided into multiple candidate regions, and the corresponding spectral feature vector is extracted for each candidate region. Then, combining the pre-set historical annotation information, the mutual information value between the spectral features and the process label of each candidate region is calculated using a mutual information metric algorithm. The mutual information value is used to measure whether the region is a "sensitive region," that is, whether its spectral changes are highly discriminative in identifying the process type.

[0093] Each sensitive region is assigned a corresponding weight parameter based on its classification contribution to the process type label. The classification contribution is obtained by measuring the mutual information between the sensitive region's features and the process type label. Specifically, based on previously obtained historical labeling information, the spectral characteristics of each sensitive region are acquired, and combined with the process type label corresponding to each wafer sample. The mutual information value between the feature distribution of each sensitive region and the process label is calculated using a mutual information measurement method. A higher mutual information value indicates a higher contribution of the sensitive region to distinguishing different process types. This mutual information value is used as the classification contribution index for the sensitive region, and corresponding weight parameters are assigned accordingly. If a sensitive region has a high classification contribution, its corresponding weight parameter is large; conversely, if the classification contribution is low, the weight parameter is small. For example, in a product quality assessment model, there are multiple sensitive regions. Their classification contributions are obtained by calculating their mutual information with the process type label. Then, a higher weight parameter is assigned to the chip interface region, and a lower weight parameter is assigned to the battery mounting region, indicating that the chip interface region has a greater impact when assessing product quality.

[0094] Subsequently, for the pre-selected sensitive region, the raw reflectance spectrum under multi-band spectral images is processed, specifically through steps such as spectral normalization, noise reduction and smoothing, feature point extraction, and frequency domain transformation. By processing the spectral data of this sensitive region, a multi-dimensional feature set reflecting its spectral behavior is extracted. This multi-dimensional spectral feature set includes, but is not limited to, frequency-selective reflectance values, inter-spectral ratio features, derivative spectral peak and valley features, spectral slope change points, principal component coefficient values, and spectral curve shape factors.

[0095] The multidimensional spectral feature set is matched with a preset process spectral dictionary and input into a preset supervised discriminant model to obtain the process posterior probability corresponding to each sensitive region. Specifically, the multidimensional spectral feature set extracted from the sensitive region is matched with multiple standard process spectral features in the preset process spectral dictionary, and the similarity score between spectral features is calculated using methods such as cosine similarity. The preset process spectral dictionary is a spectral template library built from a large number of wafer samples of known process types, recording typical spectral features or statistical models corresponding to different processes. Next, the feature vector corresponding to each sensitive region is used as input and fed into a pre-trained supervised discriminant model. The supervised discriminant model is a machine learning classification model trained using labeled data, such as SVM, KNN, LDA, neural networks, etc. In this embodiment, the pre-trained supervised discriminant model can be an SVM model, whose input is the spectral feature vector of the sub-region, and its input dimension is consistent with the dimension of the spectral features. The model's hyperparameters, such as kernel function type, penalty coefficient C, and kernel parameter γ, can be preset or optimized using the training dataset through cross-validation. The loss function used is the standard SVM classification loss function, which minimizes the classification error of the training samples and improves the model's discriminative ability. The posterior probability distribution of each sensitive region corresponding to various process labels is obtained through model inference, serving as the basis for process identification. The process posterior probability is the model's predicted probability for each process label, reflecting the confidence level or classification confidence that the sensitive region belongs to a certain process.

[0096] Next, prior information for the standard-position wafer is acquired. This prior information includes pre-defined wafer process production data from the manufacturing execution system. This process production data for the standard wafer specifically includes, but is not limited to, the wafer's production batch number, the various process steps and their parameters, and process data such as temperature, pressure, spin coating speed, and equipment model during wafer processing. This prior information is used to construct a reference polar coordinate model, attitude angle range, and edge health threshold for the standard-position wafer, and serves as a comparison benchmark for subsequent detection of wafer attitude angles and notch offsets.

[0097] The process type confidence score is obtained by performing Bayesian fusion of prior information and joint posterior probabilities. This involves combining historical experience (prior probabilities from existing knowledge) with the joint posterior probabilities of current detection data, using Bayesian formulas to correct and fuse probabilities, ultimately determining the degree of confidence that the wafer's process type is true. Bayesian fusion is a mathematical method used to synthesize prior probabilities and observed information (posterior probabilities) to arrive at a more reliable judgment. The process type confidence score indicates the likelihood that the current wafer belongs to a particular process type; a higher value indicates a greater probability of belonging to that process. The process type confidence score can quantify the current process stage of the wafer, such as photolithography or annealing, overcoming the limitations of relying solely on prior experience or current data, and improving the accuracy of process type determination.

[0098] Finally, to ensure the accuracy and stability of wafer process identification in this embodiment, a process type confidence level determination mechanism is introduced. When the confidence level of the process type obtained by fusing the spectral matching model with prior information is higher than a preset confidence threshold (e.g., 85%), the current identification result is considered to have sufficient confidence, and the system will automatically output a process type label for the wafer and mark it as a standard orientation wafer. The preset confidence threshold can be set according to the requirements for process identification accuracy in actual production. This label can be used for subsequent process path confirmation, workstation sorting, and quality analysis. If the identification confidence level does not reach the threshold, no label is output, and the process proceeds to a fault-tolerant process such as manual review, image re-acquisition, or adjustment of identification parameters.

[0099] By constructing a second polar coordinate system and using a mask to avoid non-process areas, a mutual information maximization algorithm is used to select sensitive regions and assign weights. This is combined with Bayesian fusion of prior information and posterior probability, ultimately outputting the process type label when the confidence level meets the standard. The overall effect is to achieve high-precision, adaptive, and robust classification of wafer process types, while improving algorithm efficiency and providing reliable process type identification for wafer manufacturing.

[0100] In one embodiment of this invention, the spectral data processing step includes: S701. Obtain the reflectance spectrum of each sensitive region; S702. The reflectance spectrum in each sensitive region is decomposed into the basic endmember spectrum using the constrained nonnegative endmember decomposition algorithm. S703. Combine the basic end-member spectrum and the preset multilayer thin film interference model to perform physical inversion to obtain the actual physical parameters of the wafer film layer corresponding to the sensitive region. S704. Extract the empirical spectral features corresponding to each sensitive region based on the actual physical parameters of the wafer film layer; S705. Calculate the differential characteristics of empirical spectral features based on the characteristics of the reference wafer; S706. Extract spatial distribution features for any sensitive region, and combine the spatial distribution features, differential features, and empirical spectral features to obtain multidimensional spectral features.

[0101] First, the reflectance spectrum of each sensitive region is obtained. This can be achieved by scanning the wafer surface using a multi-channel spectral camera or imaging spectrometer to acquire the reflectance values ​​of each pixel at multiple wavelengths. Based on a pre-marked process-sensitive region mask, the multi-band reflection intensity of each pixel within that region is extracted, forming the reflectance spectrum data for that sensitive region. The sensitive region refers to an area on the wafer surface or in an image that significantly responds to process parameters or defect classification results; these are typically key detection areas identified through feature selection or mutual information analysis. The reflectance spectrum refers to the intensity of light reflected from the surface of this region under illumination at different wavelengths. The reflectance spectrum can reveal the optical properties of the material.

[0102] Secondly, the reflectance spectrum of each sensitive region is decomposed into basic endmember spectra using a constrained nonnegative endmember decomposition algorithm. The basic endmember spectra can be understood as a set of typical, indivisible spectral components. The reflectance spectrum of each sensitive region can be considered as a linear mixture of multiple endmember spectra in different proportions. Constrained nonnegative endmember decomposition is a spectral unmixing algorithm used to decompose a mixed spectrum into several nonnegative endmember spectra and their corresponding nonnegative weights. For the multi-band reflectance spectrum data acquired in each sensitive region, the constrained nonnegative endmember decomposition algorithm is used for decomposition. This algorithm unmixes the observed reflectance spectrum using the following model: L=E×A Where L is the observed reflectance spectrum matrix, E is the basic endmember spectrum matrix, and A is the spatial distribution coefficient matrix of the corresponding endmember.

[0103] L (observed reflectance spectrum matrix) corresponds to the multi-band reflectance data of the sensitive region, with the dimension being the number of samples in the sensitive region × the number of bands, such as 100 sensitive regions × 50 wavelength points, which is the input data for the algorithm; E (basic endmember spectrum matrix) corresponds to the endmember spectrum of the pure material, with the dimension being the number of endmembers × the number of bands, such as 3 endmembers, silicon, oxide layer, photoresist × 50 wavelength points, which is the basic spectral unit required by the algorithm; A (spatial distribution coefficient matrix) corresponds to the weight ratio of each endmember in each sensitive region, with the dimension being the number of endmembers × the number of samples in the sensitive region or the number of samples in the sensitive region × the number of endmembers, which is the weight parameter required by the algorithm. The following constraints are introduced during the solution process: all spectral values ​​and weight coefficients must be non-negative (physical reflectance cannot be negative), and a sparsity regularization term is added to limit the number of activated endmembers. A priori film spectral dictionary is introduced to guide the regularization of the decomposition results. Ultimately, this decomposition process can output the spectra of multiple basic endmembers contained in each sensitive region and their distribution weights, providing accurate spectral component inputs for subsequent physical inversion and process classification.

[0104] Next, the actual physical parameters of the wafer layers corresponding to the sensitive regions are determined by combining the basic endmember spectra and the preset multilayer thin-film interference model. Specifically, for each sensitive region, the basic endmember spectra extracted by the non-negative endmember decomposition algorithm are combined with the preset multilayer thin-film interference theoretical model for spectral physical inversion. The basic endmember spectra are the pure spectral components extracted from the reflectance spectrum in the previous step using the non-negative endmember decomposition algorithm, representing the ideal spectral response of a material or film structure. The preset multilayer thin-film interference model is a theoretical model of the interference phenomenon generated by different film structures when light is incident. Given the thickness, refractive index, and extinction coefficient of the film, the reflectance spectrum at that location at different wavelengths can be calculated. The preset multilayer thin-film interference model is a theoretical model based on the interference phenomenon generated by different film structures when light is incident. The model input includes parameters such as the thickness, refractive index, extinction coefficient, and total number of film layers for each layer. Based on these input parameters, by establishing an interlayer optical transfer matrix or a reflection-transmission matrix method, the reflectance spectrum at any location at different wavelengths can be calculated, thereby obtaining the optical response characteristics of each film layer combination. To calculate the reflectance spectrum from film parameters, this embodiment employs a numerical inversion algorithm to optimize the film parameters. The inversion algorithm includes least-squares fitting and iterative optimization. It aims to match the simulated reflectance spectrum as closely as possible to the actual measured spectrum, enabling the calculation of the reflectance spectrum at different wavelengths at a given location based on the preset film thickness, refractive index, and extinction coefficient. Next, using the basic endmember spectrum as the target spectrum, the system iteratively optimizes the film thickness, refractive index, and other parameters in the model using a nonlinear least-squares fitting algorithm until the sum of squared residuals between the theoretical reflectance spectrum and the target spectrum is minimized. Finally, the system outputs a set of physical parameters for the wafer film structure corresponding to the sensitive region, including the thickness, refractive index, and other optical constants of each film layer, providing fundamental physical input for subsequent empirical feature extraction and process classification.

[0105] Empirical spectral features corresponding to each sensitive region are extracted based on the actual physical parameters of the wafer film layers. These features are derived from the physical parameters of the wafer film layers obtained during the physical inversion stage, specifically including the thickness, refractive index, and extinction coefficient of each film layer. For example, the reflectance value at a characteristic wavelength is selected based on process experience, choosing a specific wavelength that is most sensitive to the film layer parameters, and extracting the reflectance at that wavelength as a feature. Empirical spectral features include, but are not limited to, reflectance values ​​at several typical wavelength bands, reflectance ratios between different wavelength bands, and peaks, valleys, and slope change points extracted based on the first and second derivatives of the reflectance spectrum curve. These empirical spectral features are used to describe the typical morphology of the spectral response in the sensitive region, providing input for subsequent differential analysis, feature stitching, and process classification.

[0106] First, standard wafer data samples from the Manufacturing Execution System (MES) are retrieved, and empirical spectral features are extracted for each sensitive region to obtain a baseline feature vector. These empirical spectral features include, but are not limited to, parameters such as frequency-selective reflectance, ratio characteristics, derivative spectral peaks, and spectral slope change points. Second, for the sensitive region in the current wafer under test, the actual empirical spectral features of that region are obtained using the same feature extraction method. Subsequent differences are then performed along the corresponding dimensions to obtain differential feature vectors. These differential features are used to quantify the deviation trend of the sensitive region along the spectral dimensions, aiding in subsequent anomaly detection, feature weighting, and classification processes.

[0107] Secondly, for any identified sensitive region, the system extracts spatial distribution features based on its spatial distribution characteristics and reflectance intensity changes in the second spectral image. These features include the maximum spectral reflectance value and its pixel coordinates, the region's gray-level centroid location, horizontal and vertical gray-level gradient statistics, and higher-order statistical features such as skewness and kurtosis of the gray-level distribution. These spatial distribution features enhance the system's robustness in sensing spectral distribution non-uniformity caused by differences in wafer fabrication processes. After obtaining the spatial distribution features, the spatial distribution features, differential features, and empirical spectral features are concatenated to obtain multidimensional spectral features. These three types of feature vectors are then sequentially concatenated to form a complete multidimensional spectral feature vector.

[0108] The physical parameters of wafer films can be extracted through spectral data processing. By decomposing the reflectance spectrum, inverting film parameters, extracting spectral features, and stitching them together, a multidimensional spectral feature set is finally obtained. This helps to improve the accuracy of wafer process type identification, enhances the sensitivity to process differences, and provides strong support for subsequent classification.

[0109] This application provides a machine-readable storage medium storing instructions that cause a machine to perform the above-described regenerated wafer scanning method.

[0110] This application also provides an electronic device, including: The memory is configured to store instructions; and The processor is configured to retrieve instructions from memory and, when executing instructions, to implement the aforementioned regenerated wafer scanning method.

[0111] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0112] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0113] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0114] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0115] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0116] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0117] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0118] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0119] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for scanning regenerated wafers, characterized in that, An application is made in a reclaimed wafer scanning system, which includes a machine vision module and a rotation positioning module. The machine vision module includes a first vision unit and a second vision unit. The first vision unit is mounted on a robotic arm, and the second vision unit is positioned directly above the rotation positioning module. The method includes: When the target wafer transport box is detected to have reached the preset position, the identification information of the target wafer transport box is obtained; A preset light curtain sensor is used to detect each slot in the target wafer transfer box to determine the slot mapping table between each slot and the wafer; According to the slot mapping table, the wafers in each slot are micro-rotated within a preset angle range using a micro-rotation mechanism in the target wafer transfer box to obtain ring structured light imaging of each wafer. The polar coordinates of each wafer are determined based on the ring structured light imaging of each wafer, and the first attitude angle corresponding to each wafer is determined based on the polar coordinates of each wafer. Based on the first attitude angle of each wafer, each wafer is adjusted to a preset reference angle using a preset micro-rotation mechanism to obtain a standard attitude wafer; The robotic arm is driven to push each standard-position wafer from the target wafer transfer box to the preset scanning station in a preset order; When a standard posture wafer is detected at the preset scanning station, the preset scanning station is driven to rotate within a first angle range using the rotation positioning module, and the first coded image of each standard posture wafer is acquired through the first vision unit. The preset scanning station is driven to rotate within a second angle range using a rotation positioning module, and the second spectral image of each standard posture wafer is acquired through the second vision unit. The wafer code of each standard posture wafer is determined by the first encoded image, and the coding confidence of each wafer code is verified. The process type label for each standard orientation wafer is determined based on the second spectral image; When the coding confidence level is greater than the preset confidence level, the process type label, wafer code, and target wafer transport box identification information are entered into the database.

2. The method according to claim 1, characterized in that, The method further includes: Obtain process path and workstation load information; Wafer edge health and wafer warpage values ​​were determined based on ring structured light imaging and second spectral images. The wafer quality grade is determined by combining wafer edge health and wafer warpage value. The wafer quality grade includes defective wafers and qualified wafers. The sorting path is generated by combining process type label, workstation load information and wafer quality grade. The sorting path includes a defective product sorting path and a qualified product sorting path. The target actuator is determined by selecting a preset actuator based on the wafer edge health and wafer warpage value; The execution parameters of the target actuator are set based on the wafer edge health, wafer warpage value, and target actuator. By combining the execution parameters of the target actuator and the sorting path, wafers are sorted to the target workstation to obtain wafer sorting information; The wafer sorting information is stored in the warehouse entry record.

3. The method according to claim 2, characterized in that, The preset actuators include adsorption actuators and clamping actuators. The step of selecting the preset actuator and determining the target actuator based on wafer edge health and wafer warpage value includes: When the wafer edge health is greater than the first health threshold and the wafer warpage value is less than the first warpage threshold, the adsorption actuator is selected as the target actuator. When the wafer edge health is less than the second health threshold and the wafer warpage value is greater than the second warpage threshold, the clamping actuator is selected as the target actuator.

4. The method according to claim 1, characterized in that, The ring structured light imaging is an image of a ring of light stripes with a specific polarization state, captured by a polarization camera under multiple temperature field conditions and after polarization state adjustment.

5. The method according to claim 4, characterized in that, The process of determining the polar coordinates of each wafer based on ring structured light imaging, and determining the first attitude angle corresponding to each wafer based on its polar coordinates, includes: The reference light intensity value of ring structured light imaging under the reference temperature condition and the corresponding light intensity value of ring structured light imaging under different temperature field conditions, including high temperature environment. The polarization state attenuation coefficient under different temperature field conditions is calculated by combining the light intensity values ​​under different temperature field conditions with the reference light intensity values ​​under the reference temperature condition. The polarization degree temperature correction formula is obtained by fitting the polarization state attenuation coefficient under different temperature field conditions; The theoretical polarization attenuation under high temperature conditions is calculated using the polarization degree temperature correction formula; The thermal radiation interference is determined by comparing the actual light intensity value of the ring structured light imaging with the reference light intensity value corresponding to the theoretical polarization attenuation. By combining thermal radiation interference suppression processing of ring structured light imaging under high temperature environment, a target ring image is obtained. The center coordinate parameters are obtained by performing a circle fitting between the target ring image and the ring structured light imaging under the reference temperature conditions. The center coordinate parameters include the high temperature center coordinates. The theoretical thermal expansion offset is calculated based on the pre-set linear expansion coefficient of the wafer; By combining the theoretical thermal expansion offset, the coordinates of the high-temperature center and the positioning notch of the wafer are corrected to obtain the compensated pole coordinates and polar axis direction; The first attitude angle is determined by combining the compensated pole coordinates and polar axis direction.

6. The method according to claim 5, characterized in that, The determination of the first attitude angle by combining the compensated pole coordinates and polar axis direction includes: Establish a first polar coordinate system based on the compensated pole coordinates and polar axis direction; In the target annular image, filter annular light bar feature points at preset angular intervals, and convert the pixel coordinates of the annular light bar feature points into world coordinates; In the first polar coordinate system, for any annular light stripe feature point in any world coordinate system, the initial polar radius and initial polar angle of the annular light stripe feature point are calculated using the rectangular coordinate transformation relationship based on the polar coordinates corresponding to the annular light stripe feature point. The thermal expansion correction value and the polarization correction value are calculated using the preset thermal expansion correction term and the preset polarization optical path difference correction term, respectively. The final polar diameter is obtained by adding the thermal expansion correction value, the polarization correction value, and the initial polar diameter. The initial polar angle is corrected using a weighted smoothing algorithm based on adjacent three points to obtain the final polar angle; The temperature measurement value corresponding to each annular light stripe feature point and the deviation between the final polar radius and the preset reference radius are obtained, and a thermal distortion basis function term is constructed based on the temperature measurement value. Fourier analysis was performed on the deviation between the final extreme radius phase and the preset reference radius to obtain the first harmonic term of the wafer attitude. Using the final polar angle as the independent variable and the deviation of the final polar radius from the preset reference radius as the observation, a joint model of the observation is constructed by combining the independent variable, the observation, the distortion basis function term, and the first harmonic term. The first attitude angle of the joint model wafer is solved based on the least squares criterion.

7. The method according to claim 1, characterized in that, The process type label for each standard orientation wafer determined based on the second spectral image includes: A second polar coordinate system is constructed based on the standard orientation wafer, and a corresponding polar coordinate mask is generated according to the second polar coordinate system. The polar coordinate mask is used to avoid the wafer notch area and the wafer clamping area. The effective process region in the polar coordinate mask is extracted by combining the polar coordinate mask and the second spectral image; The correspondence between process type labels and wafer spectral data is determined based on preset historical annotation information, which includes wafer samples of process types and corresponding spectral images. By combining the correspondence relationship, the mutual information maximization algorithm is used to select the sensitive region of the process type in the effective process region; Based on preset historical annotation information, determine the mutual information measurement between sensitive areas and process type labels; Each sensitive region is assigned a corresponding weight parameter based on the classification contribution between the sensitive region and the process type label. The classification contribution is obtained by measuring the mutual information between the sensitive region features and the process type label. The sensitive region undergoes spectral data processing steps to obtain a multidimensional spectral feature set of the sensitive region; The multidimensional spectral feature set is matched with the preset process spectral dictionary and input into the preset supervised discrimination model to obtain the process posterior probability corresponding to each sensitive region. The joint process posterior probability distribution is calculated by combining the weight parameters corresponding to each sensitive region and the process posterior probability. Obtain prior information for standard-position wafers, including wafer process production data from a pre-defined manufacturing execution system; The confidence level of the process type is obtained by Bayesian fusion of prior information and joint posterior probability. When the confidence level of the process type is greater than the preset confidence level of the process type, output the process type label of the standard orientation wafer.

8. The method according to claim 7, characterized in that, The spectral data processing steps include: Obtain the reflectance spectrum of each sensitive region; The reflectance spectrum of each sensitive region is decomposed into the basic endmember spectrum using a constrained nonnegative endmember decomposition algorithm. The actual physical parameters of the wafer film layer corresponding to the sensitive region are obtained by combining the basic end-member spectrum and the preset multilayer thin film interference model through physical inversion. Empirical spectral features corresponding to each sensitive region are extracted based on the actual physical parameters of the wafer film layer; The differential characteristics of empirical spectral features are calculated based on the characteristics of a reference wafer; For any sensitive region, spatial distribution features are extracted, and the spatial distribution features, differential features, and empirical spectral features are concatenated to obtain multidimensional spectral features.

9. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores instructions for causing the machine to perform the regenerated wafer scanning method according to any one of claims 1 to 8.

10. An electronic device, characterized in that, include: The memory is configured to store instructions; as well as The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the regenerated wafer scanning method according to any one of claims 1 to 8.

Images