A visual-based square silicon wafer pre-alignment method and device

Abstract

The application discloses a kind of square silicon wafer pre-alignment method and device based on vision, which comprises the following steps: calibrating visual acquisition module, establishing the coordinate mapping relationship between image coordinate system and preset rotary table coordinate system;Multiple corner images of square silicon wafer are collected and image segmentation and corner feature extraction are carried out;Determine whether the corner is a normal corner or a positioning cut corner, and obtain the actual corner point or the theoretical corner point before the positioning cut corner is cut off respectively;Convert the corner point coordinates into physical corner point coordinates, calculate the center position and rotation angle of the square silicon wafer, and obtain the center offset and angle offset;Determine the front and back surface state according to the cut length relationship of the positioning cut corner, and generate the surface orientation result accordingly;When the offset exceeds the pre-set allowable range, output the correction instruction and recheck until the pre-alignment requirement is met.The application can improve the stability, adaptability and automation of square silicon wafer pre-alignment.

Description

Technical Field

[0001] This invention relates to the field of semiconductor manufacturing equipment technology, and in particular to a vision-based method and apparatus for pre-aligning square silicon wafers. Background Technology

[0002] Silicon wafer pre-alignment is a crucial step in semiconductor manufacturing equipment. It is typically used to calibrate the position and orientation of silicon wafers before they enter subsequent process chambers, inspection stations, or transport paths. The accuracy of the pre-alignment results affects the stability and consistency of subsequent processing, inspection, bonding, and transport processes. Significant deviations in the center position or angular orientation of the silicon wafer can lead to process area shifts, decreased product consistency, and even transport interference or equipment malfunctions.

[0003] With the development of semiconductor manufacturing and advanced packaging processes, in addition to common circular silicon wafers, square silicon wafers, rectangular silicon wafers, and square or rectangular silicon wafer substrates with positioning chamfers are gradually being used in related process scenarios. The shape characteristics, positioning references, and orientation determination methods of these types of silicon wafers differ from those of circular silicon wafers. Their pre-alignment process not only needs to determine the center position and rotation angle of the silicon wafer, but may also need to identify the facing state of the silicon wafer to ensure that subsequent processes are executed in the predetermined direction.

[0004] Existing silicon wafer pre-alignment equipment typically focuses more on the overall outline or single positioning features of silicon wafers of different shapes when processing them, and does not adequately consider factors such as the spatial relationship between multiple corner features of square silicon wafers, the impact of positioning chamfers on corner references, and the determination of front and back states. When there are placement deviations, corner chamfers, or orientation differences in the silicon wafer, the pre-alignment results are prone to instability, which in turn affects the reliable execution of subsequent processes.

[0005] Furthermore, in automated equipment, a consistent coordinate relationship needs to be maintained between the visual inspection results and the actuators, and the silicon wafer position needs to be corrected based on the inspection results. If there is a lack of effective coordination between coordinate transformation, pose calculation, and correction control, it may lead to inaccurate correction calculations, or the inability to confirm whether the silicon wafer has reached the preset alignment state even after one correction. Therefore, how to achieve stable visual recognition, position and pose determination, orientation judgment, and closed-loop correction for square silicon wafers has become an urgent technical problem to be solved. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present invention aims to provide a vision-based method and apparatus for pre-aligning square silicon wafers, thereby improving the stability of center positioning, angle calibration, orientation determination, and closed-loop correction of square silicon wafers.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a vision-based pre-alignment method for square silicon wafers, comprising: S1. Calibrate the vision acquisition module and establish a coordinate mapping relationship between the image coordinate system of the vision acquisition module and the preset rotary table coordinate system with the center of the rotary table of the alignment actuator as the origin. S2. Place the square silicon wafer on the bearing position of the alignment actuator, so that at least one corner of the square silicon wafer is within the imaging field of view of the vision acquisition module; S3. Acquire multiple corner images of the square silicon wafer through the vision acquisition module, and perform image segmentation and corner feature extraction on the multiple corner images to obtain the image feature coordinates of the corresponding corners; S4. Determine whether the corresponding corner is a normal corner or a positioned chamfered corner based on the image feature coordinates; when it is determined to be a normal corner, extract the actual corner point of the normal corner; when it is determined to be a positioned chamfered corner, extract the chamfered endpoints on both sides of the positioned chamfered corner, and restore the theoretical corner point of the positioned chamfered corner before it was cut off based on the adjacent edges where the chamfered endpoints are located. S5. Based on the coordinate mapping relationship, transform the actual corner points and / or the theoretical corner points from image coordinates to the preset rotary table coordinate system to obtain the coordinates of multiple physical corner points of the square silicon wafer; S6. Calculate the center position and rotation angle of the square silicon wafer based on the coordinates of the multiple physical corner points, and obtain the center offset based on the difference between the center position and the preset target center position, and obtain the angle offset based on the difference between the rotation angle and the preset target angle; S7. Determine the front and back states of the square silicon wafer; wherein, when a positioning chamfer is detected, the square silicon wafer is determined to be in a front or back state based on the cutting length relationship between the positioning chamfer on two adjacent sides; when no positioning chamfer is detected, output an unrecognized state or determine a default facing state according to preset rules; S8. Generate a translation correction amount based on the center offset, generate a rotation correction amount based on the angle offset, and generate an orientation judgment result based on the front and back states; when the center offset and / or the angle offset exceed the corresponding preset allowable range, output a correction command including the translation correction amount and / or the rotation correction amount to the alignment execution mechanism, and re-execute corner image acquisition, corner feature extraction, coordinate transformation and pose calculation after the alignment execution mechanism performs the correction, until the center offset and the angle offset both meet the corresponding preset allowable range.

[0008] Furthermore, the square silicon wafer includes a square silicon wafer, a rectangular silicon wafer, and a square or rectangular silicon wafer substrate with at least one corner cut off to form a positioning chamfer.

[0009] Further, step S1 includes: The calibration board is placed on a detection plane that is consistent with the bearing surface of the square silicon wafer. The calibration board image is acquired, and the image coordinates and their corresponding physical coordinates of multiple calibration corner points in the calibration board are obtained. The mapping matrix of the vision acquisition module is obtained based on the image coordinates and the physical coordinates. When the visual acquisition module includes multiple cameras, the mapping matrix of each camera is calculated separately; The mapping matrix or the mapping matrices of each camera together constitute the coordinate mapping relationship, which is used to convert the pixel coordinates in the corner image into physical coordinates under the preset rotary table coordinate system.

[0010] Further, in step S3, the visual acquisition module includes one camera, two cameras, or four cameras, and the multiple corner images are acquired in any of the following ways: A camera is used in conjunction with the rotating platform to be rotated to multiple preset angle positions in sequence to acquire multiple corner images of the square silicon wafer. Alternatively, two cameras can be used in conjunction with the rotating stage to rotate at one or more preset angles to acquire images of the four corners of the square silicon wafer. Alternatively, four cameras can be used to capture images of the four corners of the bearing position, respectively, and the square silicon wafer can be stationary relative to the four cameras to capture images of the four corners at once.

[0011] Furthermore, when a single camera is used to capture multiple corner images, the rotating platform is positioned at preset angles of 0°, 90°, 180°, and 270° relative to the initial capture position. The corner images acquired at each preset angle position are transformed by the coordinate mapping relationship of the same camera, and then rotation compensation is performed according to the corresponding preset angle position so that the coordinates of the physical corner points corresponding to each corner are unified to the preset rotary table coordinate system.

[0012] Further, step S4 includes: When a valid corner point that meets the preset corner point response threshold and is located within the preset corner interest region is extracted from the corner image, the corresponding corner point is determined to be a normal corner point. When two valid corner points that meet the preset corner point response threshold and are located within the preset corner interest region are extracted from the corner image, and the two valid corner points are located on adjacent silicon wafer edges, the corresponding corner is determined to be a positioning chamfered corner. When no candidate corner points are extracted from the corner image, the number of extracted candidate corner points is greater than 2, or the corner point response value of the candidate corner points does not meet the preset corner point response threshold, the corner features are re-determined by edge line fitting or Hough line detection.

[0013] Further, in step S4, restoring the theoretical corner point before the corner of the positioning cut is removed includes: Edge points located on both sides of the positioning chamfer and belonging to the adjacent edges of the square silicon wafer are extracted. Straight line fitting is performed on the two sets of edge points to obtain the first edge straight line and the second edge straight line. The intersection of the first edge straight line and the second edge straight line is taken as the theoretical corner point before the corner of the positioning chamfer is cut off.

[0014] Further, step S7 includes: When a locating chamfer is detected, the distance from the theoretical corner point to the two chamfer endpoints is calculated and used as the cutting length of the locating chamfer on the two adjacent sides. Compare the size relationship between the two cut lengths to determine the long side and short side of the positioning cut angle; Based on the directional distribution of the long side and the short side in the silicon wafer's own coordinate system determined by the coordinates of the plurality of physical corner points, they are matched with a preset front template or a preset back template to determine the front or back state of the square silicon wafer.

[0015] A vision-based pre-alignment device for square silicon wafers, comprising: The alignment actuator includes a rotary table for carrying a square silicon wafer and for rotating and / or translating the square silicon wafer according to a correction command; The light source module, configured corresponding to the alignment execution mechanism, is used to create a grayscale difference between the square silicon wafer area and the background area. A vision acquisition module, connected to an image processing module, is used to acquire corner images of the square silicon wafer and send the corner images to the image processing module; The calibration module connects the coordinate transformation module and the vision acquisition module, and is used to establish a coordinate mapping relationship between the image coordinate system of the vision acquisition module and the preset rotary table coordinate system with the center of the rotary table as the origin, and send the coordinate mapping relationship to the coordinate transformation module; The image processing module is connected to the theoretical corner point recovery module and the coordinate transformation module, respectively. It is used to extract image feature coordinates based on the corner image and determine the corresponding corner type. When the corresponding corner is a normal corner, the actual corner point is sent to the coordinate transformation module. When the corresponding corner is a locating chamfered corner, the chamfered endpoint and its adjacent edge are sent to the theoretical corner point recovery module. The theoretical corner point recovery module is connected to the coordinate transformation module and is used to recover and locate the theoretical corner point before the chamfered corner is cut off based on the chamfered corner endpoint and its adjacent edge, and send the theoretical corner point to the coordinate transformation module; The coordinate transformation module is connected to the pose calculation module and the front and back recognition module respectively. It is used to convert the actual corner point, the theoretical corner point and the tangent endpoint into physical coordinates according to the coordinate mapping relationship, so as to obtain the physical corner point coordinates, the theoretical corner point physical coordinates and the tangent endpoint physical coordinates. The pose calculation module is connected to the closed-loop control module and is used to calculate the center offset and angle offset based on the physical corner coordinates. The front and back recognition module is connected to the closed-loop control module. It is used to determine the front or back state according to the theoretical corner point physical coordinates and the physical coordinates of the corner endpoint when a positioning chamfered corner is present, and to output an unrecognized state or a default facing state when no positioning chamfered corner is detected. The closed-loop control module is connected to the alignment execution mechanism and the vision acquisition module, respectively. It is used to generate the correction command based on the center offset and the angle offset, generate the orientation judgment result based on the front and back states, and trigger the vision acquisition module to re-acquire the corner image after correction.

[0016] Furthermore, when the visual acquisition module includes a camera, the camera is configured to acquire multiple corner images in conjunction with multiple preset angle rotations of the rotary table; When the visual acquisition module includes two cameras, the two cameras are configured to cooperate with the rotary table to acquire multiple corner images by one or more preset angle rotations; When the vision acquisition module includes four cameras, the four cameras are configured to respectively correspond to the four corner imaging areas of the bearing position, and acquire the four corner images when the square silicon wafer is stationary relative to the four cameras; the vision acquisition module includes a telecentric lens, and the light source module is a backlight light source.

[0017] Compared with the prior art, the present invention has at least the following beneficial effects: 1. This invention acquires multiple corner images of a square silicon wafer and performs coordinate transformation and pose calculation based on the corner features. It can be adapted to square silicon wafers, rectangular silicon wafers, and square or rectangular silicon wafer substrates with positioning chamfers, thus improving the applicability of the pre-alignment method to square silicon wafers.

[0018] 2. When the corner of the positioning cut is detected, the present invention restores the theoretical corner point before it was cut off based on the cut corner endpoint and its adjacent edges, avoiding the deviation in center position and rotation angle calculation caused by directly using the cut corner contour, thus improving the accuracy of square silicon wafer pose calculation.

[0019] 3. This invention determines the front and back state of a square silicon wafer based on the relationship between the cutting lengths of the positioning chamfer on two adjacent sides. It can simultaneously complete orientation identification during the pre-alignment process, reducing the risk of subsequent process abnormalities caused by incorrect silicon wafer orientation.

[0020] 4. The present invention uniformly converts the visual inspection results to a preset rotary table coordinate system, and generates correction instructions based on the center offset and angle offset, so that a clear data correspondence is formed between the visual recognition results and the alignment execution mechanism.

[0021] 5. After correction, the present invention re-executes image acquisition, corner feature extraction, coordinate transformation and pose calculation until the center offset and angle offset meet the preset allowable range, thereby realizing closed-loop pre-alignment control and improving the stability and automation of pre-alignment. Attached Figure Description

[0022] Figure 1 This is a flowchart of the steps of the vision-based square silicon wafer pre-alignment method in this invention; Figure 2 This is a control principle diagram of the vision-based square silicon wafer pre-alignment device in this invention.

[0023] Reference numerals: 1. Visual acquisition module; 2. Image processing module; 3. Calibration module; 4. Coordinate transformation module; 5. Theoretical corner point recovery module; 6. Pose calculation module; 7. Closed-loop control module; 8. Front and back recognition module; 9. Alignment actuator. Detailed Implementation

[0024] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0025] Example 1, as Figure 1 As shown, this embodiment provides a vision-based pre-alignment method for square silicon wafers, used for center positioning, rotation angle calibration, and front / back state identification of square silicon wafers.

[0026] The term "square silicon wafer" includes square silicon wafers, rectangular silicon wafers, and quasi-rectangular silicon wafers, as well as square or rectangular semiconductor substrates with at least one corner cut off to form a positioning chamfer. This definition avoids the misinterpretation of "square silicon wafer" as merely a square silicon wafer and allows for the inclusion of square or rectangular substrates with positioning chamfers.

[0027] In this embodiment, the alignment execution mechanism 9 includes a rotary table and a support position for supporting the square silicon wafer. The vision acquisition module 1 may include one camera, two cameras, or four cameras. The cameras may be mounted perpendicular to the corner edges of the square silicon wafer, preferably using telecentric lenses to reduce the impact of warping, changes in support method, or local height changes of the square silicon wafer on the imaging ratio. The light source is preferably a backlight source, so that the opaque square silicon wafer forms a dark area in the corner image, and the background area forms a bright area, thereby facilitating subsequent image segmentation and corner point extraction.

[0028] The vision-based pre-alignment method for square silicon wafers specifically includes: I. Calibration and Coordinate Mapping Establishment: In step S1, the vision acquisition module 1 is calibrated to establish a coordinate mapping relationship between the image coordinate system of the vision acquisition module 1 and the preset rotary table coordinate system with the center of the rotary table of the alignment actuator 9 as the origin.

[0029] In practice, a calibration board with a checkerboard pattern is placed on a detection plane that aligns with the surface of the square silicon wafer, ensuring that the calibration board plane and the subsequent square silicon wafer detection plane are at the same height. After the camera captures an image of the calibration board, the image coordinates of multiple calibration corner points are extracted from the calibration board image, and the corresponding physical coordinates of each calibration corner point are obtained.

[0030] Image coordinates can be represented as: imgPoints=[[x1,y1],[x2,y2],…,[xn,yn]]; Where imgPoints represents the set of pixel coordinates of the calibrated corner points in the camera image coordinate system; xn represents the horizontal pixel coordinates of the nth calibrated corner point in the image; yn represents the vertical pixel coordinates of the nth calibrated corner point in the image; and n represents the number of corner points involved in the calibration.

[0031] Physical coordinates can be represented as: phyPoints=[[X1,Y1],[X2,Y2],…,[Xn,Yn]]; Wherein, phyPoints represents the set of physical coordinates of the calibration corner points in the calibration plate physical coordinate system or the preset rotary table coordinate system; Xn represents the physical abscissa corresponding to the nth calibration corner point; and Yn represents the physical ordinate corresponding to the nth calibration corner point.

[0032] Based on the correspondence between imgPoints and phyPoints, the mapping matrix Hcamera can be obtained. Hcamera can be a 3×3 homography matrix used to convert image coordinates into physical coordinates. Taking a single camera as an example, Hcamera1 represents the mapping matrix corresponding to the first camera. When the vision acquisition module includes multiple cameras, mapping matrices Hcamera1, Hcamera2, Hcamera3, Hcamera4, etc., are obtained respectively. These mapping matrices together constitute the coordinate mapping relationship.

[0033] Through the above calibration method, the pixels in the subsequent corner images can be transformed into a preset rotary table coordinate system with the center of the rotary table as the origin, so that the visual inspection results and the motion control coordinates of the alignment actuator 9 have a unified reference, which is conducive to improving the accuracy of subsequent pose calculation and correction control.

[0034] II. Placement of square silicon wafers and corner image acquisition: In step S2, the square silicon wafer is placed on the support position of the alignment actuator 9, so that at least one corner of the square silicon wafer is within the imaging field of view of the vision acquisition module 1. During the initial wafer placement, the angular offset of the square silicon wafer is preferably controlled within ±5° to ensure that the corners are still within the camera's field of view and to avoid the corner features being unacceptable.

[0035] In step S3, multiple corner images of the square silicon wafer are acquired by the vision acquisition module 1, and image segmentation and corner feature extraction are performed on the multiple corner images to obtain the image feature coordinates of the corresponding corners.

[0036] In the single-camera implementation, a camera is fixedly mounted at a corresponding corner of a square silicon wafer. The rotating stage is positioned at preset angles of 0°, 90°, 180°, and 270° relative to the initial acquisition position. The first corner image is acquired when the rotating stage is at the initial 0° position; the remaining corner images are acquired when the rotating stage is at 90°, 180°, and 270° positions, respectively. After each acquisition, the corner image coordinates are converted to physical coordinates using a mapping matrix corresponding to the same camera, and rotation compensation is performed according to the corresponding rotation angle to unify the physical corner coordinates of the four corners into the preset rotating stage coordinate system.

[0037] For example, if the initial corner is located in the fourth quadrant, the coordinate symbol can be approximately represented as (+,-); after the rotary table rotates 90° counterclockwise, the corresponding corner can be converted to the first quadrant (+,+); after rotating 180° counterclockwise, the corresponding corner can be converted to the second quadrant (-,+); after rotating 270° counterclockwise, the corresponding corner can be converted to the third quadrant (-,-). This quadrant conversion is used to illustrate that when a single camera rotates to acquire data, the coordinates of multiple corners need to be unified to the same physical coordinate system.

[0038] In the dual-camera implementation, the two cameras, in conjunction with a rotating stage, rotate at one or more preset angles to acquire images of the four corners. The specific rotation angle and number of acquisitions can be determined based on the installation positions of the two cameras. Compared to the single-camera method, this approach reduces the number of rotations and improves acquisition efficiency.

[0039] In the four-camera implementation, the four cameras correspond to the four corner imaging areas of the mounting position. When the square silicon wafer remains stationary relative to the four cameras, the four cameras acquire images of the four corners simultaneously. This method is suitable for scenarios with high equipment cycle time requirements.

[0040] III. Image segmentation, corner type identification, and anomaly recalculation: After corner image acquisition, image segmentation is performed on the corner images. Because the backlight source creates a significant grayscale difference between the square silicon wafer area and the background area, the corner images are usually close to binary images. However, thresholding is still performed to enhance image processing stability. After thresholding, the edge region of the silicon wafer is obtained, and then corner detection is performed within the preset corner region of interest.

[0041] Corner detection can be performed using the `cornerHarris` function from the OpenCV library, or other corner detection algorithms that output corner response values. The corner response value indicates the likelihood that a point in the image is a corner. A preset corner response threshold is used to filter valid corners; a preset region of interest (ROI) is used to limit the corner search range, preventing background noise or non-target edges from being mistakenly identified as corners.

[0042] In step S4, the corresponding corner is determined to be a normal corner or a located chamfered corner based on the image feature coordinates. When a valid corner point that meets the preset corner point response threshold and is located within the preset corner interest region is extracted from the corner image, the corresponding corner is determined to be a normal corner, and the actual corner point of the normal corner is extracted.

[0043] When two valid corner points that meet the preset corner point response threshold and are located within the preset corner region of interest are extracted from the corner image, and the two valid corner points are located on adjacent silicon wafer edges, the corresponding corner is determined to be a positioning chamfered corner. At this time, the two valid corner points serve as the chamfered endpoints on both sides of the positioning chamfer.

[0044] When no candidate corner points are extracted from the corner image, the number of extracted candidate corner points is greater than two, or the corner response values ​​of the candidate corner points do not meet the preset corner response threshold, edge line fitting or Hough line detection is used to re-determine the corner features. This abnormal recalculation mechanism can reduce misjudgments caused by image noise, local reflections, edge damage, or insufficient corner response, and improve the reliability of corner feature extraction.

[0045] IV. Restoring the theoretical corner point of the locating corner: For the positioning of the chamfered corner, since the actual corner has been cut off, directly using the chamfered endpoint in the calculation of the center position and rotation angle will cause the corner reference to shift. Therefore, it is necessary to restore the theoretical corner point before the positioning of the chamfered corner was cut off.

[0046] In step S4, when the corner of the positioning chamfer is detected, edge points located on both sides of the positioning chamfer and belonging to the adjacent edges of the square silicon wafer are extracted. The two sets of edge points come from the two adjacent sides of the positioning chamfer, respectively. Then, straight line fitting is performed on the two sets of edge points to obtain the first edge line and the second edge line. The intersection of the first edge line and the second edge line is taken as the theoretical corner point before the positioning chamfer is cut off.

[0047] For example, when two chamfered endpoints are detected in a corner image, it can be represented as img4[(x4,y4),(x4.1,y4.1)]. Here, img4 represents the image coordinate data extracted from the fourth corner image; (x4,y4) represents the coordinates of the first chamfered endpoint in the image coordinate system; and (x4.1,y4.1) represents the coordinates of the second chamfered endpoint in the image coordinate system. By fitting two edge lines to the adjacent edges containing the two chamfered endpoints, and then finding the intersection of the two edge lines, the theoretical corner point can be obtained.

[0048] Under a specific coordinate system and quadrant, the theoretical corner point can also be approximately determined based on the extreme value relationship of the coordinates of the two tangent endpoints. For example, given the quadrant where the corner is located and the edge direction, the position of the theoretical corner point can be estimated using a combination of extreme values ​​of the horizontal or vertical coordinates. However, this method depends on the coordinate direction and quadrant conditions, and therefore can be considered an optional implementation. The preferred method remains to fit the intersection point to the edge straight line to improve the accuracy of the theoretical corner point recovery.

[0049] By restoring the theoretical corner point, a square silicon wafer with a positioning chamfer can obtain a corner geometric reference consistent with that of a normal corner, thereby reducing the impact of the positioning chamfer on the calculation of center offset and angle offset.

[0050] V. Transformation from image coordinates to physical coordinates: In step S5, based on the coordinate mapping relationship, the actual corner points and / or theoretical corner points are transformed from image coordinates to the preset rotary table coordinate system to obtain the coordinates of multiple physical corner points of the square silicon wafer.

[0051] This embodiment involves three coordinate systems: an image coordinate system, a camera physical coordinate system, and a preset rotary table coordinate system. The image coordinate system is based on the pixel positions in the image captured by the camera; the camera physical coordinate system is established by a calibration plate and is used to express the actual physical position corresponding to the image coordinates; the preset rotary table coordinate system has the center of the rotary table as the origin and is used to express the actual position of the square silicon wafer relative to the alignment actuator 9.

[0052] The Hcamera mapping matrix can be used to convert image coordinates into physical coordinates. The actual corner points of ordinary corners are converted to their corresponding physical corner coordinates; the theoretical corner points of located chamfered corners are also converted to their corresponding physical corner coordinates. If a chamfered corner is detected, the chamfered corner endpoint can be converted to its physical coordinates for front / back view identification.

[0053] Multiple physical corner point coordinates can be represented as P1(x1,y1), P2(x2,y2), P3(x3,y3), and P4(x4,y4). Here, P1, P2, P3, and P4 represent the physical corner points corresponding to the four corners; x1, x2, x3, and x4 represent the x-coordinates of each physical corner point in the preset rotary table coordinate system; and y1, y2, y3, and y4 represent the y-coordinates of each physical corner point in the preset rotary table coordinate system. The four physical corner points can be defined as upper left, upper right, lower right, and lower left corners, or numbered in a clockwise or counterclockwise order.

[0054] VI. Calculation of center position, center offset, and angular offset: In step S6, the center position and rotation angle of the square silicon wafer are calculated based on the coordinates of multiple physical corner points, and the center offset is obtained based on the difference between the center position and the preset target center position, and the angle offset is obtained based on the difference between the rotation angle and the preset target angle.

[0055] When the coordinates of the four physical corner points P1(x1,y1), P2(x2,y2), P3(x3,y3), and P4(x4,y4) are obtained, the center position C of the square silicon wafer can be calculated using the four-corner point averaging method: C=((x1+x2+x3+x4) / 4, (y1+y2+y3+y4) / 4); Where C represents the center position of the square silicon wafer; Cx represents the x-coordinate of the center position C; Cy represents the y-coordinate of the center position C; x1 to x4 represent the x-coordinates of the four physical corner points; and y1 to y4 represent the y-coordinates of the four physical corner points. This formula indicates that the geometric center of the square silicon wafer is determined by averaging the coordinates of the four corner points.

[0056] In some scenarios, the center position can also be calculated using a set of diagonal midpoints, for example: C = ((x1+x3) / 2, (y1+y3) / 2); Here, P1(x1,y1) and P3(x3,y3) are a pair of diagonal points. This formula indicates that the center position of the square silicon wafer is determined by the midpoint of the line connecting the pair of diagonal points.

[0057] The offset of the center of the square silicon wafer relative to the center of the rotating stage, AlignP(px,py), can be expressed as: ΔX = Cx - px; ΔY = Cy - py; Where AlignP represents the center of the rotary stage or the preset target center position; px represents the x-coordinate of the rotary stage center in the preset rotary stage coordinate system; py represents the y-coordinate of the rotary stage center in the preset rotary stage coordinate system; ΔX represents the lateral offset of the square silicon wafer center relative to the rotary stage center; and ΔY represents the longitudinal offset of the square silicon wafer center relative to the rotary stage center. If the rotary stage center has been defined as the origin of the coordinate system during calibration, then px=0, py=0, and ΔX=Cx, ΔY=Cy. The center offset can be represented by ΔX and ΔY together, used to generate the translation correction amount.

[0058] When calculating the rotation angle, the corner points are first sorted according to their relative positions, and then two adjacent physical corner points are selected to form a reference edge. The angle between the reference edge and the X-axis of the preset rotary table coordinate system is taken as the rotation angle of the square silicon wafer. The edge direction angle can be expressed as: θ = atan2(y1-y0, x1-x0); Where θ represents the rotation angle of the square silicon wafer; (x0, y0) represents the physical coordinates of the starting point of the reference edge; (x1, y1) represents the physical coordinates of the ending point of the reference edge; x1-x0 represents the component of the reference edge in the X-axis direction; y1-y0 represents the component of the reference edge in the Y-axis direction; atan2 is the four-quadrant arctangent function, used to determine the direction angle of the reference edge in the complete plane based on the X-axis and Y-axis components. When the result of θ is expressed in radians, the range is (-π, π], where π represents pi; when it needs to be converted to degrees, θ can be multiplied by 180 / π.

[0059] The angular offset is obtained by the difference between the rotation angle of the square silicon wafer and the preset target angle. This angular offset is used to generate a rotation correction amount, enabling the rotary table to compensate for the angle of the square silicon wafer.

[0060] VII. Determining the front and back sides: In step S7, the front and back states of the square silicon wafer are determined. When a corner is detected, the square silicon wafer is determined to be either the front or back state based on the cutting length relationship between the corner and the adjacent two sides; when no corner is detected, an unrecognized state is output or a default facing state is determined according to preset rules.

[0061] Specifically, first calculate the distances from the theoretical corner point to the two chamfer endpoints, which are then used as the cut lengths of the positioning chamfers on the two adjacent sides. These distances can be Euclidean distances. If the physical coordinates of the theoretical corner point are Q(Xq,Yq), and the physical coordinates of the two chamfer endpoints are A(Xa,Ya) and B(Xb,Yb), then the cut lengths on the two adjacent sides can be expressed as: LA= ; LB= ; Where Q represents the theoretical corner point before the corner is cut off; A and B represent the two corner endpoints respectively; LA represents the distance from the theoretical corner point Q to the corner endpoint A; and LB represents the distance from the theoretical corner point Q to the corner endpoint B.

[0062] By comparing the magnitudes of LA and LB, the long and short sides of the positioning chamfer can be determined. Then, based on the directional distribution of the long and short sides in the silicon wafer's own coordinate system determined by multiple physical corner points, they are matched with a preset front or back template to determine whether the square silicon wafer is in a front or back state. This method utilizes the geometric asymmetry of the positioning chamfer for orientation determination, allowing for simultaneous front and back identification during pre-alignment.

[0063] VIII. Calibration Instruction Generation and Closed-Loop Re-verification: In step S8, a translation correction amount is generated based on the center offset, a rotation correction amount is generated based on the angle offset, and an orientation judgment result is generated based on the front and back states. When the center offset and / or angle offset exceed the corresponding preset allowable range, a correction command including the translation correction amount and / or rotation correction amount is output to the alignment execution mechanism 9.

[0064] After the alignment actuator 9 executes the correction command, the system re-executes corner image acquisition, corner feature extraction, coordinate transformation, and pose calculation until the center offset and angle offset both meet the corresponding preset allowable ranges. In the case of detected corners with locating cuts, the system continues to perform corner location judgment and theoretical corner point recovery during the re-inspection process to ensure that the re-inspection calculations are still based on accurate corner geometric references.

[0065] The working principle of this embodiment can be summarized as follows: The process involves establishing a mapping relationship between image coordinates and preset rotary table coordinates through calibration; acquiring multiple corner images of a square silicon wafer using one, two, or four cameras; identifying ordinary corners or locating chamfered corners through image segmentation and corner feature extraction; using actual corner points directly for ordinary corners; and recovering theoretical corner points from the intersection of adjacent edge lines for locating chamfered corners. The actual and theoretical corner points are then converted into physical corner coordinates in the preset rotary table coordinate system. The center position and rotation angle are calculated based on the physical corner coordinates, and the front and back states are determined by combining the geometric relationship of the locating chamfered corners. Finally, correction and judgment results are generated based on the center offset, angle offset, and front and back states, and closed-loop pre-alignment is completed through re-inspection.

[0066] Through the above technical solution, this embodiment can achieve at least the following technical effects: First, it can adapt to square silicon wafers, rectangular silicon wafers, quasi-rectangular silicon wafers, and square or rectangular semiconductor substrates with positioning chamfers, improving the equipment's compatibility with square silicon wafers of different shapes. Second, it can restore the theoretical corner point before the chamfer is cut when the positioning chamfer exists, reducing the calculation error of the center position and rotation angle caused by the chamfer. Third, it can use the cutting length relationship of the positioning chamfer on the adjacent sides to determine the front and back status, improving the automation level of the pre-alignment process. Fourth, it converts the image coordinates into physical coordinates under the preset rotary table coordinate system through coordinate mapping, so that the visual inspection results and the alignment execution mechanism 9 form a clear data correspondence. Fifth, it forms a closed-loop control through correction and re-inspection, improving the accuracy and stability of square silicon wafer pre-alignment.

[0067] Example 2, refer to Figure 2 This embodiment provides a vision-based pre-alignment device for square silicon wafers, used to perform corner image acquisition, corner type determination, coordinate transformation, pose calculation, front and back identification, and closed-loop correction of square silicon wafers. This device can be used for pre-alignment of square silicon wafers, rectangular silicon wafers, and square or rectangular silicon wafer substrates with positioning chamfers.

[0068] The device includes an alignment actuator 9, a light source module, a vision acquisition module 1, a calibration module 3, an image processing module 2, a theoretical corner point recovery module 5, a coordinate transformation module 4, a pose calculation module 6, a front and back recognition module 8, and a closed-loop control module 7.

[0069] The alignment actuator 9 includes a rotary table for supporting a square silicon wafer and for rotating and / or translating the wafer according to calibration instructions. The center of the rotary table serves as the origin of a preset rotary table coordinate system, providing a unified physical reference for subsequent coordinate transformations, center offset calculations, and angular offset calculations. By using the rotary table center as a unified reference, coordinate inconsistencies caused by multi-camera or multi-angle acquisitions can be reduced.

[0070] The light source module is configured to align the actuator 9, creating a grayscale difference between the square silicon wafer area and the background area. Preferably, the light source module is a backlight source, located below the square silicon wafer or below the corner imaging area. Since the square silicon wafer is opaque, the square silicon wafer area appears as a dark area in the corner image, while the background area appears as a bright area. To facilitate image segmentation, the grayscale segmentation threshold can be set to any grayscale value between 80 and 180, or the Otsu thresholding method can be used to automatically obtain the grayscale segmentation threshold. In an 8-bit grayscale image, when a pixel's grayscale value is lower than the grayscale segmentation threshold, it can be identified as a silicon wafer area; when a pixel's grayscale value is higher than the grayscale segmentation threshold, it can be identified as a background area. This setting helps improve the stability of edge extraction.

[0071] The vision acquisition module 1 is connected to the image processing module 2 and is used to acquire corner images of the square silicon wafer and send the corner images to the image processing module 2. The vision acquisition module 1 includes a telecentric lens. The telecentric lens can reduce the impact of slight warping, changes in bearing height, or local posture changes of the square silicon wafer on the corner imaging size, so that the edges and corners in the corner images maintain a more stable geometric proportion, thereby improving the accuracy of physical corner coordinate calculation.

[0072] The calibration module 3 connects to the coordinate transformation module 4 and is used to establish a coordinate mapping relationship between the image coordinate system of the vision acquisition module 1 and a preset rotary table coordinate system with the center of the rotary table as the origin. This coordinate mapping relationship is then sent to the coordinate transformation module 4. The coordinate mapping relationship can be a single-camera mapping matrix or mapping matrices corresponding to multiple cameras. The coordinate mapping relationship output by the calibration module 3 serves as the input to the coordinate transformation module 4, enabling the image coordinates to be converted into physical coordinates in the preset rotary table coordinate system.

[0073] Image processing module 2 is connected to theoretical corner recovery module 5 and coordinate transformation module 4, respectively, and is used to extract image feature coordinates and determine the corresponding corner type based on the corner image. Image processing module 2 can first obtain the silicon wafer edge region based on the grayscale segmentation threshold, and then extract candidate corner points within the preset corner region of interest. The preset corner region of interest can be set according to the camera field of view center and the allowable wafer placement error of the square silicon wafer, for example, covering 40% to 80% of the center region of the corner image. The corner response value can be the Harris corner response value or other corner detection response values, and the preset corner response threshold can be set to 5% to 30% of the normalized corner response maximum value; preferably set to 10%. When the response value of a candidate corner point reaches the preset corner response threshold and is located within the preset corner region of interest, the candidate corner point is considered a valid corner point.

[0074] When the corresponding corner is a normal corner, the image processing module 2 sends the actual corner point to the coordinate transformation module 4; when the corresponding corner is a positioned chamfered corner, the image processing module 2 sends the chamfered endpoint and its adjacent edge to the theoretical corner point recovery module 5. By distinguishing between normal corners and positioned chamfered corners, the chamfered endpoint can be avoided from being directly used as the real corner reference of the square silicon wafer, reducing the error in subsequent pose calculation.

[0075] The theoretical corner point recovery module 5 is connected to the coordinate transformation module 4. It is used to recover the theoretical corner point of the locating corner before it was cut off, based on the chamfer endpoint and its adjacent edges, and then sends the theoretical corner point to the coordinate transformation module 4. The theoretical corner point recovery module 5 can perform straight line fitting on the adjacent edges on both sides of the chamfer, and then use the intersection of the two fitted lines as the theoretical corner point. This process ensures that the square silicon wafer with the locating chamfer can still obtain a corner point reference consistent with that of a normal corner, which is beneficial for subsequent calculations of center offset and angle offset.

[0076] The coordinate transformation module 4 is connected to the pose calculation module 6 and the front / back face recognition module 8, respectively. It converts the actual corner points, theoretical corner points, and tangent endpoints into physical coordinates based on coordinate mapping relationships, obtaining the physical corner point coordinates, the theoretical corner point physical coordinates, and the tangent endpoint physical coordinates. The physical corner point coordinates are sent to the pose calculation module 6; the theoretical corner point physical coordinates and the tangent endpoint physical coordinates are sent to the front / back face recognition module 8. Through this data splitting method, both pose calculation and front / back face recognition have clear data sources.

[0077] The pose calculation module 6 connects to the closed-loop control module 7 and is used to calculate the center offset and angle offset based on the physical corner coordinates. The center offset characterizes the deviation of the center of the square silicon wafer from the preset target center position, and the angle offset characterizes the deviation of the current angle of the square silicon wafer from the preset target angle. The preset allowable range for the center offset can be set to ±0.01mm to ±0.50mm, preferably ±0.05mm, according to the process accuracy requirements; the preset allowable range for the angle offset can be set to ±0.005° to ±0.10°, preferably ±0.02°. By setting the preset allowable range, clear judgment conditions can be provided for the closed-loop control module 7.

[0078] The front / back recognition module 8 is connected to the closed-loop control module 7. When a positioning chamfer is present, it determines the front or back state based on the theoretical corner point's physical coordinates and the chamfer endpoint's physical coordinates. If no positioning chamfer is detected, it outputs an unrecognized state or a default facing state. Specifically, the front / back recognition module 8 calculates the distance from the theoretical corner point to the two chamfer endpoints, obtaining the cutting length of the positioning chamfer on two adjacent sides. Based on the relationship between the two cutting lengths, it determines the long side and short side, and then matches them with a preset front template or a preset back template. Thus, the front / back state can be obtained simultaneously with pose correction, improving the automation of the pre-alignment process.

[0079] The closed-loop control module 7 is connected to the alignment actuator 9 and the vision acquisition module 1, respectively. It generates correction commands based on the center offset and angular offset, generates orientation judgment results based on the front and back states, and triggers the vision acquisition module 1 to re-acquire corner images after correction. The correction commands include translational correction and / or rotational correction. If the center offset exceeds the preset allowable range, the closed-loop control module 7 generates a translational correction; if the angular offset exceeds the preset allowable range, the closed-loop control module 7 generates a rotational correction. After the alignment actuator 9 executes the correction commands, the closed-loop control module 7 triggers the vision acquisition module 1 to re-acquire corner images, causing the device to re-complete corner type judgment, coordinate transformation, pose calculation, and deviation judgment until both the center offset and angular offset fall within their corresponding preset allowable ranges.

[0080] The vision acquisition module 1 can be configured in three ways. When the vision acquisition module 1 includes one camera, the camera is configured to acquire multiple corner images by rotating the rotary table at multiple preset angles. This method reduces the number of cameras and lowers hardware costs. When the vision acquisition module 1 includes two cameras, the two cameras are configured to acquire multiple corner images by rotating the rotary table at one or more preset angles. This method balances cost and acquisition efficiency. When the vision acquisition module 1 includes four cameras, the four cameras are configured to respectively correspond to the four corner imaging areas of the bearing position, and acquire the four corner images while the square silicon wafer is stationary relative to the four cameras. This method reduces rotational acquisition actions and improves equipment cycle time.

[0081] The working principle of this embodiment is as follows: The light source module first establishes a stable grayscale difference between the square silicon wafer area and the background area. The vision acquisition module 1 acquires corner images and sends them to the image processing module 2. The image processing module 2 obtains image feature coordinates based on the corner images and distinguishes between ordinary corners and positioned chamfered corners. The actual corner points corresponding to ordinary corners directly enter the coordinate transformation module 4. The chamfered endpoints and adjacent edges corresponding to positioned chamfered corners first enter the theoretical corner point recovery module 5 to obtain theoretical corner points. The coordinate transformation module 4 converts the actual corner points, theoretical corner points, and chamfered endpoints into physical coordinates based on the coordinate mapping relationship provided by the calibration module 3. The pose calculation module 6 outputs the center offset and angle offset based on the physical corner point coordinates. The front and back recognition module 8 outputs the front and back status based on the physical coordinates of the theoretical corner points and the physical coordinates of the chamfered endpoints. The closed-loop control module 7 generates correction instructions and orientation judgment results based on the center offset, angle offset, and front and back status, and triggers a re-inspection after correction.

[0082] Through the above-described device structure and data flow, this embodiment can achieve the following technical effects: First, the visual inspection data is unified to the preset rotary table coordinate system through the calibration module 3 and the coordinate transformation module 4, which improves the consistency between the visual inspection results and the actions of the actuator. Second, the image processing module 2 and the theoretical corner point recovery module 5 distinguish between ordinary corners and positioning chamfered corners, avoiding corner reference errors caused by positioning chamfers. Third, the front and back face recognition module 8 obtains the front and back face status simultaneously during the pre-alignment process, reducing the process risk caused by incorrect orientation. Fourth, the closed-loop control module 7 triggers re-acquisition and re-inspection after correction, which improves the stability of the center positioning and rotation angle calibration of the square silicon wafer.

[0083] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A vision-based pre-alignment method for square silicon wafers, characterized in that, include: S1. Calibrate the vision acquisition module and establish a coordinate mapping relationship between the image coordinate system of the vision acquisition module and the preset rotary table coordinate system with the center of the rotary table of the alignment actuator as the origin. S2. Place the square silicon wafer on the bearing position of the alignment actuator, so that at least one corner of the square silicon wafer is within the imaging field of view of the vision acquisition module; S3. Acquire multiple corner images of the square silicon wafer through the vision acquisition module, and perform image segmentation and corner feature extraction on the multiple corner images to obtain the image feature coordinates of the corresponding corners; S4. Determine whether the corresponding corner is a normal corner or a positioned chamfered corner based on the image feature coordinates; when it is determined to be a normal corner, extract the actual corner point of the normal corner; when it is determined to be a positioned chamfered corner, extract the chamfered endpoints on both sides of the positioned chamfered corner, and restore the theoretical corner point of the positioned chamfered corner before it was cut off based on the adjacent edges where the chamfered endpoints are located. S5. Based on the coordinate mapping relationship, transform the actual corner points and / or the theoretical corner points from image coordinates to the preset rotary table coordinate system to obtain the coordinates of multiple physical corner points of the square silicon wafer; S6. Calculate the center position and rotation angle of the square silicon wafer based on the coordinates of the multiple physical corner points, and obtain the center offset based on the difference between the center position and the preset target center position, and obtain the angle offset based on the difference between the rotation angle and the preset target angle; S7. Determine the front and back states of the square silicon wafer; wherein, when a positioning chamfer is detected, the square silicon wafer is determined to be in a front or back state based on the cutting length relationship between the positioning chamfer on two adjacent sides; when no positioning chamfer is detected, output an unrecognized state or determine a default facing state according to preset rules; S8. Generate a translation correction amount based on the center offset, generate a rotation correction amount based on the angle offset, and generate an orientation judgment result based on the front and back states; when the center offset and / or the angle offset exceed the corresponding preset allowable range, output a correction command including the translation correction amount and / or the rotation correction amount to the alignment execution mechanism, and re-execute corner image acquisition, corner feature extraction, coordinate transformation and pose calculation after the alignment execution mechanism performs the correction, until the center offset and the angle offset both meet the corresponding preset allowable range.

2. The vision-based pre-alignment method for square silicon wafers according to claim 1, characterized in that, The square silicon wafer includes a square silicon wafer, a rectangular silicon wafer, and a square or rectangular silicon wafer substrate with at least one corner cut off to form a positioning chamfer.

3. The vision-based pre-alignment method for square silicon wafers according to claim 1, characterized in that, Step S1 includes: The calibration board is placed on a detection plane that is consistent with the bearing surface of the square silicon wafer. The calibration board image is acquired, and the image coordinates and their corresponding physical coordinates of multiple calibration corner points in the calibration board are obtained. The mapping matrix of the vision acquisition module is obtained based on the image coordinates and the physical coordinates. When the visual acquisition module includes multiple cameras, the mapping matrix of each camera is calculated separately; The mapping matrix or the mapping matrices of each camera together constitute the coordinate mapping relationship, which is used to convert the pixel coordinates in the corner image into physical coordinates under the preset rotary table coordinate system.

4. The vision-based pre-alignment method for square silicon wafers according to claim 1, characterized in that, In step S3, the visual acquisition module includes one camera, two cameras, or four cameras, and the multiple corner images are acquired in any of the following ways: A camera is used in conjunction with the rotating platform to be rotated to multiple preset angle positions in sequence to acquire multiple corner images of the square silicon wafer. Alternatively, two cameras can be used in conjunction with the rotating stage to rotate at one or more preset angles to acquire images of the four corners of the square silicon wafer. Alternatively, four cameras can be used to capture images of the four corners of the bearing position, respectively, and the square silicon wafer can be stationary relative to the four cameras to capture images of the four corners at once.

5. The vision-based pre-alignment method for square silicon wafers according to claim 1, characterized in that, When a single camera is used to capture multiple corner images, the rotating platform is positioned at preset angles of 0°, 90°, 180°, and 270° relative to the initial capture position. The corner images acquired at each preset angle position are transformed by the coordinate mapping relationship of the same camera, and then rotation compensation is performed according to the corresponding preset angle position so that the coordinates of the physical corner points corresponding to each corner are unified to the preset rotary table coordinate system.

6. The vision-based pre-alignment method for square silicon wafers according to claim 1, characterized in that, Step S4 includes: When a valid corner point that meets the preset corner point response threshold and is located within the preset corner interest region is extracted from the corner image, the corresponding corner point is determined to be a normal corner point. When two valid corner points that meet the preset corner point response threshold and are located within the preset corner interest region are extracted from the corner image, and the two valid corner points are located on adjacent silicon wafer edges, the corresponding corner is determined to be a positioning chamfered corner. When no candidate corner points are extracted from the corner image, the number of extracted candidate corner points is greater than 2, or the corner point response value of the candidate corner points does not meet the preset corner point response threshold, the corner features are re-determined by edge line fitting or Hough line detection.

7. The vision-based pre-alignment method for square silicon wafers according to claim 1, characterized in that, In step S4, restoring the theoretical corner point before the corner of the positioning cut is removed includes: Edge points located on both sides of the positioning chamfer and belonging to the adjacent edges of the square silicon wafer are extracted. Straight line fitting is performed on the two sets of edge points to obtain the first edge straight line and the second edge straight line. The intersection of the first edge straight line and the second edge straight line is taken as the theoretical corner point before the corner of the positioning chamfer is cut off.

8. The vision-based pre-alignment method for square silicon wafers according to claim 1, characterized in that, Step S7 includes: When a locating chamfer is detected, the distance from the theoretical corner point to the two chamfer endpoints is calculated and used as the cutting length of the locating chamfer on the two adjacent sides. Compare the size relationship between the two cut lengths to determine the long side and short side of the positioning cut angle; Based on the directional distribution of the long side and the short side in the silicon wafer's own coordinate system determined by the coordinates of the plurality of physical corner points, they are matched with a preset front template or a preset back template to determine the front or back state of the square silicon wafer.

9. A vision-based pre-alignment device for square silicon wafers, characterized in that, include: The alignment actuator includes a rotary table for carrying a square silicon wafer and for rotating and / or translating the square silicon wafer according to a correction command; The light source module, configured corresponding to the alignment execution mechanism, is used to create a grayscale difference between the square silicon wafer area and the background area. A vision acquisition module, connected to an image processing module, is used to acquire corner images of the square silicon wafer and send the corner images to the image processing module; The calibration module connects the coordinate transformation module and the vision acquisition module, and is used to establish a coordinate mapping relationship between the image coordinate system of the vision acquisition module and the preset rotary table coordinate system with the center of the rotary table as the origin, and send the coordinate mapping relationship to the coordinate transformation module; The image processing module is connected to the theoretical corner point recovery module and the coordinate transformation module, respectively. It is used to extract image feature coordinates based on the corner image and determine the corresponding corner type. When the corresponding corner is a normal corner, the actual corner point is sent to the coordinate transformation module. When the corresponding corner is a locating chamfered corner, the chamfered endpoint and its adjacent edge are sent to the theoretical corner point recovery module. The theoretical corner point recovery module is connected to the coordinate transformation module and is used to recover and locate the theoretical corner point before the chamfered corner is cut off based on the chamfered corner endpoint and its adjacent edge, and send the theoretical corner point to the coordinate transformation module; The coordinate transformation module is connected to the pose calculation module and the front and back recognition module respectively. It is used to convert the actual corner point, the theoretical corner point and the tangent endpoint into physical coordinates according to the coordinate mapping relationship, so as to obtain the physical corner point coordinates, the theoretical corner point physical coordinates and the tangent endpoint physical coordinates. The pose calculation module is connected to the closed-loop control module and is used to calculate the center offset and angle offset based on the physical corner coordinates. The front and back recognition module is connected to the closed-loop control module. It is used to determine the front or back state according to the theoretical corner point physical coordinates and the physical coordinates of the corner endpoint when a positioning chamfered corner is present, and to output an unrecognized state or a default facing state when no positioning chamfered corner is detected. The closed-loop control module is connected to the alignment execution mechanism and the vision acquisition module, respectively. It is used to generate the correction command based on the center offset and the angle offset, generate the orientation judgment result based on the front and back states, and trigger the vision acquisition module to re-acquire the corner image after correction.

10. The vision-based square silicon wafer pre-alignment device according to claim 9, characterized in that, When the visual acquisition module includes a camera, the camera is configured to acquire multiple corner images in conjunction with multiple preset angle rotations of the rotary table; When the visual acquisition module includes two cameras, the two cameras are configured to cooperate with the rotary table to acquire multiple corner images by one or more preset angle rotations; When the vision acquisition module includes four cameras, the four cameras are configured to respectively correspond to the four corner imaging areas of the bearing position, and acquire the four corner images when the square silicon wafer is stationary relative to the four cameras; the vision acquisition module includes a telecentric lens, and the light source module is a backlight light source.

Images