Wafer surface defect detection system and method based on line scan camera

Abstract

The application relates to a wafer surface defect detection system and method based on a line-scan camera, which comprises a workbench, a plurality of clamping jaw devices, a line-scan camera assembly, a plurality of avoidance sensing devices, a PLC and a controller. The plurality of clamping jaw devices are annularly distributed on the workbench and surround a clamping area for accommodating a wafer to be detected, each clamping jaw device can be retracted along the radial direction of the clamping area to realize clamping or loosening. The line-scan camera assembly is movably arranged on the workbench and is used for scanning the wafer to be detected. The plurality of avoidance sensing devices are matched with the plurality of clamping jaw devices, and are used for sending a detected position signal when the line-scan camera assembly moves to a preset avoidance point corresponding to a clamping jaw device. The PLC is connected with the plurality of avoidance sensing devices, and receives the position signal sent by the avoidance sensing device through a self-provided IO module.

Description

Technical Field

[0001] This invention relates to the field of electronic technology, and in particular to a wafer surface defect detection system and method based on a line scan camera. Background Technology

[0002] In the field of semiconductor wafer manufacturing, processing and inspection, accurate detection of wafer surface defects is the core link to ensure the yield and performance of semiconductor devices. Among them, 12-inch large-size wafers are the mainstream application specifications in the current semiconductor industry. Their surface defect detection faces multiple technical requirements, such as large-size coverage, high-speed scanning, high-precision identification and non-contact detection. Compared with the detection of 4-inch, 6-inch and 8-inch small-size wafers, the technical difficulty is greatly increased.

[0003] Small-size wafer inspection, due to its small field of view and relatively low precision requirements, often employs area scan cameras for direct imaging to complete full-surface inspection. This eliminates the need for complex image stitching and mechanical structure avoidance designs, and motion control only requires single-axis or simple dual-axis drive, with software timing for imaging triggering meeting the requirements. However, this technical solution cannot adapt to the inspection needs of 12-inch large-size wafers: on the one hand, area scan cameras struggle to achieve high-resolution full coverage of large-size wafers while maintaining inspection speed, easily leading to blind spots or insufficient precision; on the other hand, 12-inch wafers require non-contact clamping and fixation using multiple circumferentially spaced grippers. These grippers can obstruct the inspection field of view and are prone to causing glare and reflections that interfere with imaging. The lack of avoidance or software-delayed avoidance methods in small-size wafer inspection can result in missed defects in the gripper area and decreased image quality, failing to meet the high-precision inspection standards at the 200nm level.

[0004] To address the challenges of inspecting 12-inch large-size wafers, the industry is gradually adopting a method of linear scanning with line scan camera components and multi-image stitching to achieve full surface coverage. By utilizing the high line scan frequency of the line scan camera components, the scanning speed is improved while ensuring inspection accuracy, thus meeting the high-efficiency inspection needs of large-size wafers. However, existing line scanning inspection solutions still have many technical shortcomings in practical applications: First, motion control mostly adopts open-loop or semi-closed-loop drive, resulting in low positioning accuracy and poor speed uniformity of the camera scanning X-axis and translation Y-axis, which can easily lead to image stretching, compression, or splicing misalignment; Second, the trigger signal for line scanning imaging is disconnected from the motion axis drive, mostly triggered by software timing, which cannot be dynamically adjusted according to the actual movement state of the motion axis, resulting in poor imaging synchronization; Third, the avoidance design for gripper occlusion is mostly controlled by software commands, resulting in significant signal transmission delays, untimely avoidance response, and still causing imaging interference in the gripper area, and cannot achieve accurate and independent avoidance of multiple grippers; Fourth, the coordination of front and back synchronous inspection is poor, and the scanning and triggering of the upper and lower line scanning camera components are not synchronized, which can easily lead to deviations in the positioning of defects on the front and back sides, affecting the accuracy of the inspection results.

[0005] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0006] The main objective of this invention is to provide a wafer surface defect detection system and method based on a line scan camera, which aims to solve the aforementioned technical problems in the prior art.

[0007] To achieve the above objectives, the present invention provides a wafer surface defect detection system based on a line scan camera, the wafer surface defect detection system based on a line scan camera comprising: Workbench; Multiple gripper devices are arranged in a ring on the worktable and surround a clamping area for accommodating the wafer to be inspected. Each gripper device can extend and retract radially along the clamping area to clamp or release. A line scan camera assembly is movably mounted on the worktable for scanning the wafer to be inspected; Multiple obstacle avoidance sensing devices, matched with the multiple gripper devices, are used to send a detected position signal when the online scanning camera assembly moves to a preset obstacle avoidance point corresponding to a gripper device; The PLC is connected to the multiple avoidance sensing devices. It receives the position signals sent by the avoidance sensing devices through its own IO module, and autonomously sends a first pulse signal to the corresponding gripper device according to the position signal, so that the corresponding gripper device is driven to extend and retract to avoid the obstacle after receiving the first pulse signal. A controller, electrically connected to the PLC, is configured to acquire multiple images captured by a line scan camera assembly and stitch the multiple images together to form an image of the wafer to be inspected; and to determine whether the wafer to be inspected has defects based on the image of the wafer to be inspected.

[0008] Preferably, in the wafer surface defect detection system based on a line scan camera, each gripper device includes a gripper driver, a gripper motor, and a gripper body; the gripper driver is connected to a PLC and is used to receive a first pulse signal sent by the PLC and drive the gripper motor to rotate according to the first pulse signal; the gripper motor is connected to the gripper body and is used to drive the gripper body to extend and retract when the gripper motor rotates.

[0009] Preferably, the wafer surface defect detection system based on the line scan camera further includes a motion control card, an XY axis driver, and an XY axis motor. The motion control card is used to receive position commands sent by the controller, convert the position commands into a second pulse signal, and send the second pulse signal to the XY axis driver. At the same time, when the motion control card receives the real-time motion position fed back by the XY axis driver, it sends the real-time motion position to the controller. The XY-axis driver and XY-axis motor are used to receive the second pulse signal and drive the XY-axis motor to rotate according to the second pulse signal, so as to drive the line scan camera assembly to realize X-axis scanning or Y-axis translation; the encoder of the XY-axis motor feeds back the real-time motion position to the XY-axis driver in real time, so that the XY-axis driver feeds back the real-time motion position to the motion control card. The motion control card is also used to feed back the real-time motion position to the controller.

[0010] Preferably, the wafer surface defect detection system based on a line scan camera further includes an image acquisition card and a pulse triggering module. The pulse triggering module is connected to the XY axis driver and the image acquisition card respectively, and is used to receive the real-time motion position feedback from the encoder of the XY axis driver in real time, and output a pulse triggering signal to send the pulse triggering signal to the image acquisition card to trigger the image acquisition card to control the line scan camera component to synchronously acquire images. The image acquisition card is also connected to the controller and the line scan camera assembly, and is used to receive image acquisition commands sent by the controller and control the line scan camera assembly to perform image acquisition.

[0011] Preferably, in the wafer surface defect detection system based on the line scan camera, each gripper device corresponds to an avoidance sensing device, and the preset avoidance point corresponding to each gripper device is located in front of the gripper device along the moving path of the line scan camera.

[0012] Preferably, in the wafer surface defect detection system based on a line scan camera, the line scan camera assembly includes an upper line scan camera and a lower line scan camera; In this process, when scanning the wafer to be inspected, the upper scanning camera is positioned above the worktable and corresponds to the front side of the wafer, while the lower scanning camera is positioned below the worktable and corresponds to the back side of the wafer. The upper and lower scanning cameras operate synchronously.

[0013] Preferably, the wafer surface defect detection system based on a line scan camera further includes a wafer robot and a door. The door is located outside the clamping area of ​​the worktable and can be opened and closed. The wafer robot is used to place the wafer to be inspected into the clamping area or to remove the wafer after inspection.

[0014] Preferably, in the wafer surface defect detection system based on a line scan camera, the controller is configured to: The image of the wafer to be inspected is divided into image regions to obtain multiple sub-region images; The multiple sub-region images are used as input and input into a pre-trained convolutional neural network model for the first convolutional feature extraction process to obtain a first feature map set of the multiple sub-region images. Based on the first feature map set, a first feature vector is obtained. The feature weighting module in the convolutional neural network model calculates the feature map weight of each feature map in the first feature map set, so as to select the second feature map set of the multiple sub-region images from the first feature map set, and performs secondary convolutional feature extraction processing on the multiple sub-region images with a different size than the first convolutional feature extraction processing based on the second feature map set to obtain the second feature vector. The first feature vector and the second feature vector are used as inputs to a pre-trained bimodal neural network, which outputs the defect confidence scores of the multiple sub-region images. Based on the defect confidence levels of the multiple sub-region images, the specific regions of surface defects in the wafer image to be inspected are determined.

[0015] Preferably, in the wafer surface defect detection system based on a line scan camera, the controller is configured to: The feature map weight of each feature map in the first feature map set is calculated using the following formula: (1) (2) (3) Where R is the local limit window; This is the defect scaling factor; H is the width of the feature map; W represents the height of the feature map; F k (i, j) represents the pixel value at the i-th row and j-th column position in the k-th feature map; q k The initial weight values ​​for the k-th feature map; wk The weights of the k-th feature map; M(i,j) is a mask matrix of size H×W; f k The stray light threshold of the wafer to be tested; t k This represents the multi-scale aggregated value of the k-th feature map; This is the pseudo-feature suppression coefficient, with a value range of 0.05 to 0.15, used to suppress invalid feature weights caused by stray light.

[0016] To achieve the above objectives, the present invention provides a method for surface defect detection using the aforementioned wafer surface defect detection system based on a line scan camera, the method comprising: The controller sends the first X-axis scan command to the motion control card. The motion control card converts the command into a second pulse signal and sends it to the X-axis driver. The X-axis driver drives the X-axis motor to rotate at a constant speed, causing the line scan camera assembly to move linearly along the X-axis from the first scan start position to the first scan end position. The moving speed matches the scanning speed of the line scan camera. The encoder of the X-axis motor collects the motor rotation position in real time and feeds the position signal back to the X-axis driver. The X-axis driver synchronously transmits the position signal to the pulse trigger module. The pulse trigger module outputs a pulse trigger signal to the image acquisition card according to the preset trigger interval, triggering the line scan camera assembly to synchronously acquire images. The upper and lower line scan cameras respectively acquire one image of the front and back of the wafer. The image is scanned locally and transmitted to the controller for temporary storage in real time. During the movement of the online scanning camera assembly along the X-axis, when it moves to a preset avoidance point corresponding to a certain gripper device, the corresponding avoidance sensing device detects the camera position and immediately transmits the position signal to the PLC's IO module. The PLC does not require software relay; the hardware independently parses the signal and sends a first pulse signal to the gripper driver of the gripper device. The gripper driver drives the gripper motor to rotate, causing the gripper body to rapidly extend and retract radially outward, thereby achieving gripper avoidance and preventing the gripper from entering the camera's scanning field of view. If multiple preset avoidance points are passed along the X-axis scanning path, the corresponding gripper avoidance actions are triggered sequentially. The response time of all avoidance actions is less than 0.01s, with no delay. After the first X-axis scan is completed, the line scan camera assembly moves to the end position of the first scan. The encoder of the X-axis motor feeds back the arrival signal to the controller. The controller sends a Y-axis translation command to the motion control card. The motion control card drives the Y-axis motor to rotate, causing the line scan camera assembly to translate along the Y-axis by a preset distance to the second scan position. After reaching the second scan position, the above X-axis scan and gripper avoidance actions are repeated to complete the second X-axis scan. The upper and lower line scan cameras each acquire one partial scan image. The X-axis linear scan and Y-axis translation are cycled in this way until all wafers to be inspected have been scanned. After scanning is completed, the controller performs image stitching and surface defect detection on the acquired images.

[0017] The present invention has at least the following beneficial effects: The wafer surface defect detection system based on a line scan camera provided by this invention requires no real-time commands from software. From sensor triggering to gripper extension and retraction, the entire process is controlled by a closed-loop hardware link, thereby achieving zero-delay obstacle avoidance and adapting to the high-speed scanning rhythm of the line scan camera components. Pure hardware control is achieved through the cooperation of obstacle avoidance sensing devices and PLC.

[0018] Furthermore, this invention achieves zero latency through pure hardware avoidance control, eliminating intermediate software intervention. These two key designs enable the hardware link to form an independent closed loop.

[0019] Furthermore, the triggering link of the pulse triggering module and the hardware link of the gripper device avoidance work based on the same camera X-axis movement trajectory, realizing dual hardware coordination of "synchronous scanning + synchronous avoidance".

[0020] Furthermore, the wafer surface defect detection system based on a line scan camera provided by this invention can be adapted to high-speed scanning to meet efficiency requirements; the detection of 12-inch large wafers requires high scanning speed, and the pulse triggering module can dynamically adjust the trigger pulse frequency according to the X-axis moving speed, so that it can accurately match the acquisition rhythm regardless of whether the X-axis moves fast or slow, without affecting the detection efficiency; Furthermore, it can ensure high precision and meet the 200nm detection requirements; the PD3000 can precisely control the pulse trigger interval according to the preset detection precision (200nm), so that the camera can accurately acquire every micron / nanometer point when moving along the X-axis, ensuring that the final stitched wafer image has no missed scans and no duplicate scans, meeting the 200nm high-precision detection standard. Furthermore, the wafer surface defect detection system based on a line scan camera provided by this invention operates independently in hardware, improving system stability; the triggering process of the pulse triggering module is completed autonomously by the hardware without the need for software intervention in the detection system. Even if the software is multitasking, it will not affect the accuracy of the acquisition triggering, thus improving the stability of the entire detection system. Attached Figure Description

[0021] Figure 1 A perspective view of an embodiment of the wafer surface defect detection system based on a line scan camera provided by the present invention; Figure 2 This is a schematic diagram of an embodiment of the wafer surface defect detection system based on a line scan camera provided by the present invention; Figure 3 This is a flowchart of an embodiment of the wafer surface defect detection method based on a line scan camera provided by the present invention.

[0022] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0024] In this embodiment of the invention, the term "and / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0026] In this embodiment of the invention, the term "multiple" refers to two or more, and other quantifiers are similar.

[0027] In this invention, unless otherwise stated, directional terms such as "upper," "lower," "top," and "bottom" are generally used in relation to the direction shown in the accompanying drawings, or in relation to the vertical, perpendicular, or gravitational direction of the component itself; similarly, for ease of understanding and description, "inner" and "outer" refer to the inner and outer contours of each component itself, but the above directional terms are not intended to limit this invention.

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details are presented in the embodiments of the present invention to facilitate a better understanding of the invention. However, the technical solutions claimed in the present invention can be implemented even without these technical details and various variations and modifications based on the following embodiments. The division of the following embodiments is for ease of description and should not constitute any limitation on the specific implementation of the present invention. The various embodiments can be combined with and referenced by each other without contradiction.

[0029] This invention provides a wafer surface defect detection system based on a line scan camera, such as... Figure 1 and Figure 2As shown, the wafer surface defect detection system based on a line scan camera includes a worktable 1, multiple gripper devices 2, a line scan camera assembly 3, multiple obstacle avoidance sensors 9, a PLC, a motion control card 7, XY axis drivers and XY axis motors, an image acquisition card 5, a pulse trigger module 6, and a controller 4.

[0030] It should be noted that the wafer surface defect detection system based on a line scan camera provided by this invention is mainly adapted to the front and back defect detection of large-size wafers of at least 12 inches. The detection accuracy can reach 200nm and it can identify a variety of defects such as particles, scratches, dirt, haze, Grind Mark (SG mark), E to E (Edge To Edge), and haze.

[0031] The worktable 1 is used to support various components. The worktable 1 adopts a high-precision marble platform, which has the characteristics of strong vibration resistance and small deformation, providing a stable installation foundation for the entire inspection system. The middle of the worktable 1 has a accommodating through hole for accommodating the wafer to be inspected. At the same time, it can also be adapted to the scanning inspection of the back of the wafer to be inspected by the line scan camera assembly 3 located below, so as to avoid the scanning field of view being blocked by the worktable 1 itself.

[0032] Multiple gripper devices 2 are arranged in a ring on the worktable 1, forming a clamping area for accommodating the wafer to be inspected. Each gripper device 2 can extend or retract radially along the clamping area to clamp or release. More specifically, the clamping area is located within a positioning through-hole, and the multiple gripper devices 2 are evenly distributed in a ring on the inner side of the receiving through-hole of the worktable 1, surrounding the clamping area. Each gripper device 2 includes a gripper driver 81, a gripper motor 82, and a gripper body 83. The gripper driver 81 is connected to a PLC and is used to receive a first pulse signal sent by the PLC and drive the gripper motor 82 to rotate according to the first pulse signal. The gripper motor 82 is connected to the gripper body 83 and is used to drive the gripper body 83 to extend or retract when the gripper motor 82 rotates.

[0033] The gripper driver 81 receives the first pulse signal sent by the PLC and converts it into a motor drive signal to precisely control the rotation angle of the gripper motor 82. The input of the gripper driver 81 is connected to the motion control module of the PLC to receive the first pulse signal; the output of the gripper driver 81 is connected to the gripper motor 82 to send drive signals.

[0034] The gripper motor 82 receives the drive signal from the gripper driver 81 and rotates to drive the gripper body 83 to extend and retract radially along the gripping area. The input of the gripper motor 82 is connected to the gripper driver 81 to receive the drive signal; the output of the gripper motor 82 is connected to the gripper body 83 to drive the mechanical action.

[0035] The gripper body 83 is arranged in a ring to form a wafer clamping area, which is used to extend and retract radially along the wafer clamping area to achieve avoidance (extension) / clamping (retraction). During the avoidance process, it completely leaves the scanning field of view of the line scan camera and is connected to the gripper motor 82 for transmission to perform purely hardware-controlled extension and retraction actions.

[0036] A line scan camera assembly 3 is movably mounted on the worktable 1 for scanning the wafer to be inspected. To facilitate simultaneous scanning of the front and back sides of the wafer, the line scan camera assembly 3 typically includes an upper line scan camera 31 and a lower line scan camera 32. When scanning the wafer 200, the upper line scan camera is positioned above the worktable 1, corresponding to the front side of the wafer 200, while the lower line scan camera 32 is positioned below the worktable 1, corresponding to the back side of the wafer 200. The upper and lower line scan cameras 31 and 32 operate synchronously, thus enabling simultaneous scanning of the front and back sides of the wafer 200.

[0037] The movement of the line scan camera assembly 3 is achieved through a motion control card 7, an XY-axis driver, and an XY-axis motor. The motion control card 7 receives position commands from the controller 4, converts these commands into second pulse signals, and sends these signals to the XY-axis driver. Simultaneously, upon receiving real-time position feedback from the XY-axis driver, the motion control card 7 sends the real-time position to the controller 4. The XY-axis driver receives the second pulse signal and drives the XY-axis motor to rotate, thereby enabling the line scan camera assembly 3 to perform X-axis scanning or Y-axis translation. The encoder of the XY-axis motor provides real-time position feedback to the XY-axis driver, which then feeds this real-time position back to the motion control card 7. The motion control card 7 also feeds the real-time position back to the controller 4. The controller 4 analyzes the real-time position to confirm whether the position is correct, ensuring the accuracy of the scanning position of the line scan camera assembly 3.

[0038] It is worth noting that the gripper body 83 must not contact the front or back of the wafer 200 under inspection during gripping, and must not interfere with the scanning of the line scan camera assembly 3; that is, the gripper body 83 must not appear in the image captured by the line scan camera assembly 3. If it appears in the image captured by the line scan camera assembly 3, the gripper body 83 will cause stray light interference in the image, and the gripper body 83 will be illuminated by the illumination source used for inspection, thus substantially affecting the imaging and defect detection of the surrounding wafer area. Since software control has a delay, it is impossible to guarantee the synchronous avoidance of the gripper body 83 and the scanning action of the line scan camera assembly 3. Therefore, it is necessary to fundamentally prevent the gripper body 83 from entering the camera scanning field of view. More specifically, the gripper avoidance of this invention is achieved through pure hardware control, without secondary software instruction intervention or delayed response, completely avoiding the gripper body 83 from obstructing or interfering with imaging during the scanning of the line scan camera assembly 3.

[0039] Furthermore, it should be noted that the pure hardware control mentioned in this invention does not completely detach itself from the initial software commands. Rather, the software issues only one "allow avoidance" command, and all subsequent avoidance actions, including triggering, execution, and feedback, are autonomously completed through signal interaction between hardware modules. No further real-time commands from the software are required. From sensor triggering to gripper extension and retraction, the entire process is controlled by a closed-loop hardware link, achieving zero-delay avoidance to match the high-speed scanning rhythm of the line scan camera assembly 3. Pure hardware control is achieved through the cooperation of the avoidance sensing device 9 and the PLC.

[0040] Specifically, multiple avoidance sensing devices 9 are matched with the multiple gripper devices 2. When the line scanning camera assembly 3 moves to a preset avoidance point corresponding to a gripper device 2, it sends a detected position signal, thus enabling real-time output of hardware sensing signals. Each avoidance sensing device 9 is matched one-to-one with a gripper device 2, with each device corresponding to a preset avoidance point of a gripper device 2. The preset avoidance point for each gripper device 2 is located in front of the corresponding gripper device 2 along the line scanning camera's moving path. More specifically, the preset avoidance point is located in front of the corresponding gripper device 2 along the X-axis moving path of the line scanning camera assembly 3. The specific distance between the preset avoidance point and the gripper body 83 along the X-axis can be determined based on the moving speed of the line scanning camera assembly 3 and the time required for the gripper body 83 to extend or retract. Typically, it is slightly ahead of the corresponding gripper body 83, allowing the gripper device 2 to avoid the movement in time. The avoidance sensing device 9 is directly electrically connected to the PLC's built-in IO module, directly transmitting position signals to the PLC.

[0041] The PLC is connected to the multiple avoidance sensors 9. It receives position signals from these sensors via its built-in I / O module and autonomously sends a first pulse signal to the corresponding gripper device 2 based on the position signal. Upon receiving the first pulse signal, the corresponding gripper device 2 is driven to extend and retract to avoid an obstacle. The PLC acquires the position signals from the avoidance sensors 9 at high speed via its built-in I / O module; then, the PLC hardware autonomously analyzes the signals and sends the first pulse signal to the corresponding gripper driver 81 via its built-in motion control module (without software instruction triggering). More specifically, the PLC's built-in motion control module directly outputs the pulse control signal without any software instruction relay.

[0042] More specifically, the avoidance mechanism of the gripper device 2 is achieved through closed-loop hardware signal transmission and autonomous hardware execution, without any real-time control commands at the software level. The specific control logic can be divided into three steps: signal acquisition, pulse output, and action execution, all of which are completed in milliseconds with no delay.

[0043] The specific process includes: When the line scan camera assembly 3 performs linear scanning with the X-axis motor, the encoder of the X-axis driver provides real-time feedback on the moving position of the line scan camera assembly 3. When the line scan camera assembly 3 moves to a preset avoidance point corresponding to a certain gripper device 2 (located in front of the gripper along the X-axis scanning path and pre-calibrated), the corresponding avoidance sensing device 9 immediately detects the position of the line scan camera assembly 3, generates a position signal, and directly transmits the position signal to the PLC's built-in IO module. The IO module is a native PLC hardware module, and there is no delay in signal acquisition.

[0044] After the PLC's IO module acquires the position signal of the avoidance sensing device 9, it does not need to detect any instructions sent by the system software. The PLC's internal hardware logic automatically analyzes the signal (determines that "the corresponding gripper body 83 needs to be triggered to avoid") and sends a first pulse signal to the gripper driver 81 corresponding to the gripper body 83 through the PLC's built-in motion control module. The frequency and number of the first pulse signal are calibrated in advance by the PLC program to accurately control the rotation angle of the gripper motor 82 (i.e., the extension and retraction stroke of the gripper body 83).

[0045] After receiving the first pulse signal sent by the PLC, the gripper driver 81 immediately converts the electrical pulse signal into a motor drive signal, driving the gripper motor 82 to rotate at a constant speed. The gripper motor 82 drives the gripper body 83 to extend and retract radially outward along the wafer clamping area through the transmission structure, so that the gripper completely leaves the scanning field of view of the line scan camera and completes the avoidance action. After the camera scans past the gripper position, the avoidance sensing device 9 can no longer detect the camera signal, and the PLC hardware automatically sends a reset pulse signal. The gripper body 83 extends and retracts inward to reset, waiting for the next avoidance trigger.

[0046] Throughout the process, the detection system software only needs to complete system initialization (such as calibrating the preset avoidance point and setting the PLC pulse parameters). No intervention is required during the scanning process. All signals are transmitted directly between hardware. The response time from "the camera reaches the preset avoidance point" to "the gripper completes the avoidance" is determined by the hardware performance, completely eliminating software delay.

[0047] In addition, to ensure precise synchronization between the image acquisition by the line scan camera assembly 3 and the avoidance by the gripper device 2, an image acquisition card 5 and a pulse trigger module 6 are configured. The pulse trigger module 6 is connected to both the XY-axis driver and the image acquisition card 5. It receives real-time motion position feedback from the encoder of the XY-axis driver and outputs a pulse trigger signal, which is then sent to the image acquisition card 5 to trigger the image acquisition card 5 to control the line scan camera assembly 3 to acquire images synchronously. The image acquisition card 5 is also connected to the controller 4 and the line scan camera assembly 3, and receives image acquisition commands sent by the controller 4 to control the line scan camera assembly 3 to perform image acquisition. In operation, controller 4 sends a "start acquisition" command to image acquisition card 5 and a "move X-axis to scan end position" command to motion control card 7. The X-axis motor starts and begins to move. The encoder of the X-axis driver acquires the real-time movement position of the X-axis and continuously feeds it back to pulse trigger module 6. Pulse trigger module 6 analyzes the real-time movement position sent by the encoder in real time and generates a microsecond-level pulse trigger signal at each preset acquisition point along the X-axis, according to the 200nm precision requirement of wafer inspection, and immediately sends it to image acquisition card 5. Image acquisition card 5 receives the pulse trigger signal sent by pulse trigger module 6 and immediately controls line scan camera assembly 3 to complete one image acquisition. During the X-axis movement, the encoder continuously provides feedback, pulse trigger module 6 continuously triggers, and image acquisition card 5 continuously acquires images until the X-axis reaches the scan end position, completing the image acquisition for a single scan. In some embodiments, pulse trigger module 6 is a PD3000.

[0048] By configuring the image acquisition card 5 and the pulse trigger module 6, the core issues of synchronization and accuracy under high-speed scanning were resolved. Compared with software triggering, it can eliminate delay and ensure synchronization. The hardware pulse triggering method has a response time in the microsecond range, completely avoiding the 1-5ms delay of software triggering, and making the X-axis movement and camera acquisition completely synchronized, avoiding image stretching, blurring, and misalignment caused by movement. Furthermore, it can be adapted to high-speed scanning to meet efficiency requirements; the inspection of 12-inch large wafers requires high scanning speed, and the pulse trigger module 6 can dynamically adjust the trigger pulse frequency according to the X-axis movement speed. Whether the X-axis moves fast or slow, it can accurately match the acquisition rhythm without affecting the inspection efficiency. Furthermore, to ensure high precision and meet the 200nm detection requirements, the PD3000 can precisely control the pulse trigger interval according to the preset detection precision (200nm), so that the camera can accurately acquire every micron / nanometer point as it moves along the X-axis, ensuring that the final stitched wafer image has no missed scans or duplicate scans, and meets the 200nm high-precision detection standard. Furthermore, the hardware operates independently, improving system stability; the triggering process of the pulse trigger module 6 is completed autonomously by the hardware without the need for software intervention in the detection system. Even if the software is multitasking, it will not affect the accuracy of the acquisition trigger, thus improving the stability of the entire detection system.

[0049] The controller 4 is electrically connected to the PLC. The controller 4 is configured to acquire multiple images captured by the line scan camera assembly 3 and stitch the multiple images together to form an image of the wafer to be inspected; and to determine whether the wafer to be inspected 200 has defects based on the image of the wafer to be inspected.

[0050] The overall process will be described in detail below: (1) System initialization and material loading preparation After the controller 4 is powered on, it sends a system initialization command. Each module completes self-test and resets to its initial position: the motion control card 7 drives the XY axis motors to rotate, moving the line scan camera assembly 3 to the receiving position (X-axis position 0, Y-axis position 0); the controller 4 sends a gripper receiving command to the PLC, and the motion control module of the PLC sends pulse signals to all gripper drivers 81, causing the gripper motor 82 to drive the gripper body 83 to extend and retract radially outward along the gripping area to the receiving position. At this time, the diameter of the gripping area is slightly larger than that of a 12-inch wafer, which facilitates the wafer robot to unload the wafer; at the same time, the door of the worktable 1 automatically descends, exposing the gripping area, waiting for the wafer robot to load and unload the wafer, and the system enters the loading state.

[0051] (2) Wafer loading and unloading robot After receiving the loading signal from the system, the wafer robot performs a first-pick-then-place action: first, it removes the wafers that have been inspected from workbench 1 and places them at the unloading station, and then it precisely places the 12-inch wafer to be inspected in the center of the clamping area to complete the loading and unloading action; after the wafer robot releases the wafer, it sends a loading completion signal to controller 4.

[0052] (3) Wafer clamping and scanning preparation After receiving the loading completion signal, the controller 4 sends a clamping command to the PLC. The motion control module of the PLC sends a pulse signal to the clamping driver 81, and all the clamping motors 82 rotate synchronously, driving the clamping body 83 to retract radially inward along the clamping area until all the clamping bodies 83 are in contact with the edge of the wafer, realizing concentric clamping of the 12-inch wafer. The clamping force is calibrated by the PLC program to ensure stable clamping without damaging the edge of the wafer. At the same time, the door closes automatically, forming a closed detection environment to avoid interference from external dust and stray light. Subsequently, the motion control card 7 drives the Y-axis motor to rotate, driving the line scan camera assembly 3 to translate along the Y-axis to the first scanning position. The controller 4 sends an image acquisition preparation command to the image acquisition card 5, and the image acquisition card 5 enters the ready-to-trigger state.

[0053] (4) No delay in avoiding the gripper during the first X-axis scan. The controller 4 sends the first X-axis scan command to the motion control card 7. The motion control card 7 converts the command into a second pulse signal and sends it to the X-axis driver. The X-axis driver drives the X-axis motor to rotate at a constant speed, causing the line scan camera assembly 3 to move linearly along the X-axis from the first scan start position to the first scan end position (the moving speed can be set as needed, for example, 50mm / s). The moving speed matches the scanning speed of the line scan camera. The encoder of the X-axis motor collects the motor rotation position in real time and feeds the position signal back to the X-axis driver. The X-axis driver synchronously transmits the position signal to the pulse trigger module 6. The pulse trigger module 6 outputs a pulse trigger signal to the image acquisition card 5 according to the preset trigger interval (e.g., 0.01mm), triggering the line scan camera assembly 3 to synchronously acquire images. The upper line scan camera 31 and the lower line scan camera 32 respectively acquire images. A partial scan image of the front and back sides of a wafer is transmitted in real time to the controller 4 for temporary storage. During the movement of the online scanning camera assembly 3 along the X-axis, when it moves to a preset avoidance point corresponding to a certain gripper device 2, the corresponding avoidance sensing device 9 (e.g., a laser proximity switch) detects the camera position and immediately transmits the position signal to the PLC's IO module. The PLC does not require software relay; the hardware independently parses the signal and sends a first pulse signal to the gripper driver 81 of the gripper device 2. The gripper driver 81 drives the gripper motor 82 to rotate, causing the gripper body 83 to rapidly extend and retract radially outward (extension stroke 5mm), thereby achieving gripper avoidance and preventing the gripper from entering the camera's scanning field of view. If multiple preset avoidance points are passed along the X-axis scanning path, the corresponding gripper avoidance actions are triggered sequentially. The response time of all avoidance actions is less than 0.01s, with no delay.

[0054] (5) Multi-position X-axis scanning and continuous image acquisition After the first X-axis scan is completed, the line scan camera assembly 3 moves to the end position of the first scan. The encoder feeds back the position signal to the controller 4. The controller 4 sends a Y-axis translation command to the motion control card 7. The motion control card 7 drives the Y-axis motor to rotate, causing the line scan camera assembly 3 to translate along the Y-axis by a preset distance (the translation distance is calibrated according to the scanning width of the line scan camera to ensure that there is about 5% overlap between the images of two adjacent X-axis scans, which is convenient for subsequent image stitching), and moves to the second scan position. After reaching the second scan position, the X-axis scan and gripper avoidance movement in step (4) are repeated. The second X-axis scan is completed, and the upper line scan camera 31 and the lower line scan camera 32 each acquire one partial scan image. Following the above method, the X-axis linear scan and Y-axis translation are completed in sequence until the entire wafer has been scanned (in this embodiment, 6 X-axis linear scans and 5 Y-axis translations are required. The upper line scan camera 31 acquires a total of 6 partial images of the front of the wafer, and the lower line scan camera 32 acquires a total of 6 partial images of the back of the wafer. All images are transmitted to the controller 4 in real time for temporary storage. After 6 scans are completed, the entire surface of the front and back of the 12-inch wafer is covered without any dead angles).

[0055] (6) Image stitching and high-precision defect recognition After the scanning is completed (6 times in this embodiment), the controller 4 calls the built-in image stitching algorithm to stitch the acquired images.

[0056] Taking six front and six back images as an example, the system first extracts and matches feature points from the six local images on the same side, and uses overlapping areas to achieve seamless image stitching. Then, an image correction algorithm is used to eliminate minor distortions caused by mechanical movement, ultimately forming a complete front image and a complete back image of the 12-inch wafer. The stitched image has a resolution of 12K×12K, which can clearly present minute features of 200nm and above on the wafer surface. Subsequently, the controller 4 calls the defect recognition algorithm to scan and analyze the stitched front and back images pixel by pixel, matching the features in the image with a preset defect feature library to accurately identify defects such as particles, scratches, dirt, haze, SG marks, E to E, and haze. The system automatically marks and records the type, location, and size of the defects, generates a detection report, and if the detected defects exceed the preset threshold, the system automatically issues an alarm signal.

[0057] (7) Wafer cutting and system reset After defect detection is completed, controller 4 sends a material preparation command, the door of workbench 1 automatically descends, and PLC sends a gripper release command to gripper driver 81. All gripper bodies 83 extend and retract radially outward to the receiving position, releasing the wafer 200 to be inspected. After receiving the material release signal, the wafer robot takes away the inspected wafer and places it in the corresponding station (qualified wafers are placed in qualified stations, and unqualified wafers are placed in rework / scrap stations). After the wafer is taken away, controller 4 sends a system reset command, and all modules are reset to their initial positions, waiting for the next wafer to be loaded, and entering the next inspection cycle.

[0058] Figure 3 A flowchart illustrating the surface defect detection method provided by this invention is shown. This method can be used... Figure 2 This can be performed using the illustrated wafer surface defect inspection system based on a line scan camera or any other suitable computer device. Alternatively, it can be specifically performed using the controller 4 of the wafer surface defect inspection system based on a line scan camera.

[0059] Step S3100 divides the wafer image to be inspected into image regions, resulting in multiple sub-region images. By dividing the wafer image to be inspected into image regions, both fine-grained feature extraction and the computational power consumption of inference per image can be reduced.

[0060] Step S3200 takes the multiple sub-region images as input and inputs them into a pre-trained convolutional neural network model for the first convolutional feature extraction process to obtain a first feature map set of the multiple sub-region images. Based on the first feature map set, a first feature vector is obtained.

[0061] Step S3300 calculates the feature map weight of each feature map in the first feature map set through the feature weighting module in the convolutional neural network model, so as to select the second feature map set of the multiple sub-region images from the first feature map set, and performs secondary convolutional feature extraction processing on the multiple sub-region images with a different size than the first convolutional feature extraction processing based on the second feature map set, to obtain the second feature vector.

[0062] By assigning feature map weights to each feature map in the first feature map set obtained from the first convolution feature extraction, the importance of each feature map in the whole can be determined. Then, based on this importance, the feature maps can be reconstructed, which can extract hidden and less obvious features in the wafer image to be detected, making the extracted features more complete and comprehensive.

[0063] The feature map weight of each feature map in the first feature map set is calculated according to the following formula: (1) (2) (3) Where R is the local limit window; This is the defect scaling factor, with a value ranging from 0.6 to 0.8, used to balance the weight ratio of local extreme features and global average features; H is the width of the feature map; W is the height of the feature map; 1≤i≤H, 1≤j≤W; F k (i, j) represents the pixel value at the i-th row and j-th column position in the k-th feature map; q k The initial weight values ​​for the k-th feature map; w k The weights of the k-th feature map; M(i,j) is a mask matrix of size H×W, where the size and features are given by the mask matrix. Figure 1 The H×W clamping area mask matrix is ​​defined, where M(i,j)=1 indicates that the corresponding position is the stray light interference area around the clamp, and M(i,j)=0 indicates that the corresponding position is the effective detection area of ​​the wafer. f k The stray light threshold value for the wafer to be inspected is 200, ranging from 0.1 to 0.3, and is obtained by pre-calibration using defect-free wafer samples. t k This represents the multi-scale aggregated value of the k-th feature map; This is the pseudo-feature suppression coefficient, with a value range of 0.05 to 0.15, used to suppress invalid feature weights caused by stray light.

[0064] Step S3400 takes the first feature vector and the second feature vector as inputs to a pre-trained bimodal neural network, and outputs the defect confidence scores of the multiple sub-region images. Merging the two feature vectors as input further improves the accuracy of the judgment, allowing for rapid and accurate determination of the defect confidence scores of each sub-region image.

[0065] Step S3500 determines the specific region of the surface defect in the wafer image to be inspected based on the defect confidence scores of the plurality of sub-region images. In some embodiments, a surface defect may be determined to exist in a region when the defect confidence score of a certain region is greater than a certain threshold.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A wafer surface defect detection system based on a line scan camera, characterized in that, include: Workbench; Multiple gripper devices are arranged in a ring on the worktable and surround a clamping area for accommodating the wafer to be inspected. Each gripper device can extend and retract radially along the clamping area to clamp or release. A line scan camera assembly is movably mounted on the worktable for scanning the wafer to be inspected; Multiple obstacle avoidance sensing devices, matched with the multiple gripper devices, are used to send a detected position signal when the online scanning camera assembly moves to a preset obstacle avoidance point corresponding to a gripper device; The PLC is connected to the multiple avoidance sensing devices. It receives the position signals sent by the avoidance sensing devices through its own IO module, and autonomously sends a first pulse signal to the corresponding gripper device according to the position signal, so that the corresponding gripper device is driven to extend and retract to avoid the obstacle after receiving the first pulse signal. A controller, electrically connected to the PLC, is configured to acquire multiple images captured by the line scan camera assembly and stitch the multiple images together to form an image of the wafer to be inspected; The presence of defects in the wafer to be inspected is determined based on the image of the wafer to be inspected.

2. The wafer surface defect detection system based on a line scan camera as described in claim 1, characterized in that, Each gripper device includes a gripper driver, a gripper motor, and a gripper body; the gripper driver is connected to a PLC and is used to receive a first pulse signal sent by the PLC and drive the gripper motor to rotate according to the first pulse signal; the gripper motor is connected to the gripper body and is used to drive the gripper body to extend and retract when the gripper motor rotates.

3. The wafer surface defect detection system based on a line scan camera as described in claim 1, characterized in that, It also includes a motion control card, an XY axis driver, and an XY axis motor. The motion control card is used to receive position commands sent by the controller, convert the position commands into a second pulse signal, and send the second pulse signal to the XY axis driver. At the same time, when the motion control card receives the real-time motion position fed back by the XY axis driver, it sends the real-time motion position to the controller. XY-axis driver and XY-axis motor, wherein the XY-axis driver is used to receive the second pulse signal and drive the XY-axis motor to rotate according to the second pulse signal, so as to drive the line scan camera assembly to realize X-axis scanning or Y-axis translation; The encoder of the XY axis motor feeds back the real-time motion position to the XY axis driver, which in turn feeds back the real-time motion position to the motion control card. The motion control card is also used to feed back the real-time motion position to the controller.

4. The wafer surface defect detection system based on a line scan camera as described in claim 3, characterized in that, It also includes an image acquisition card and a pulse triggering module. The pulse triggering module is connected to the XY axis driver and the image acquisition card respectively. It is used to receive the real-time motion position feedback from the encoder of the XY axis driver in real time, and output a pulse triggering signal. The pulse triggering signal is sent to the image acquisition card to trigger the image acquisition card to control the line scan camera component to synchronously acquire images. The image acquisition card is also connected to the controller and the line scan camera assembly, and is used to receive image acquisition commands sent by the controller and control the line scan camera assembly to perform image acquisition.

5. The wafer surface defect detection system based on a line scan camera as described in claim 1, characterized in that, Each gripper device corresponds to an avoidance sensor, and the preset avoidance point for each gripper device is located in front of the gripper device along the moving path of the line scanning camera.

6. The wafer surface defect detection system based on a line scan camera as described in claim 1, characterized in that, The line scan camera assembly includes an upper line scan camera and a lower line scan camera; In this process, when scanning the wafer to be inspected, the upper scanning camera is positioned above the worktable and corresponds to the front side of the wafer, while the lower scanning camera is positioned below the worktable and corresponds to the back side of the wafer. The upper and lower scanning cameras operate synchronously.

7. The wafer surface defect detection system based on a line scan camera as described in claim 1, characterized in that, It also includes a wafer manipulator and a door, the door being located outside the clamping area of ​​the worktable and capable of opening and closing; the wafer manipulator is used to place wafers to be inspected into the clamping area or to remove wafers that have been inspected.

8. The wafer surface defect detection system based on a line scan camera as described in claim 1, characterized in that, The controller is configured to: The image of the wafer to be inspected is divided into image regions to obtain multiple sub-region images; The multiple sub-region images are used as input and input into a pre-trained convolutional neural network model for the first convolutional feature extraction process to obtain a first feature map set of the multiple sub-region images. Based on the first feature map set, a first feature vector is obtained. The feature weighting module in the convolutional neural network model calculates the feature map weight of each feature map in the first feature map set, so as to select the second feature map set of the multiple sub-region images from the first feature map set, and performs secondary convolutional feature extraction processing on the multiple sub-region images with a different size than the first convolutional feature extraction processing based on the second feature map set to obtain the second feature vector. The first feature vector and the second feature vector are used as inputs to a pre-trained bimodal neural network, which outputs the defect confidence scores of the multiple sub-region images. Based on the defect confidence levels of the multiple sub-region images, the specific regions of surface defects in the wafer image to be inspected are determined.

9. The wafer surface defect detection system based on a line scan camera as described in claim 8, characterized in that, The controller is configured to: The feature map weight of each feature map in the first feature map set is calculated using the following formula: ；（1） ；（2） ；（3） Where R is the local limit window; This is the defect scaling factor; H is the width of the feature map; W represents the height of the feature map; F k (i, j) represents the pixel value at the i-th row and j-th column position in the k-th feature map; q k The initial weight values ​​for the k-th feature map; w k The weights of the k-th feature map; M(i,j) is a mask matrix of size H×W; f k The stray light threshold of the wafer to be tested; t k This represents the multi-scale aggregated value of the k-th feature map; This is the pseudo-feature suppression coefficient, with a value range of 0.05 to 0.15, used to suppress invalid feature weights caused by stray light.

10. A method for surface defect detection using a wafer surface defect detection system based on a line scan camera as described in any one of claims 1 to 9, characterized in that, include: The controller sends the first X-axis scan command to the motion control card. The motion control card converts the command into a second pulse signal and sends it to the X-axis driver. The X-axis driver drives the X-axis motor to rotate at a constant speed, causing the line scan camera assembly to move linearly along the X-axis from the first scan start position to the first scan end position. The moving speed matches the scanning speed of the line scan camera. The encoder of the X-axis motor collects the motor rotation position in real time and feeds the position signal back to the X-axis driver. The X-axis driver synchronously transmits the position signal to the pulse trigger module. The pulse trigger module outputs a pulse trigger signal to the image acquisition card according to the preset trigger interval, triggering the line scan camera assembly to synchronously acquire images. The upper and lower line scan cameras respectively acquire one image of the front and back of the wafer. The image is scanned locally and transmitted to the controller for temporary storage in real time. During the movement of the online scanning camera assembly along the X-axis, when it moves to a preset avoidance point corresponding to a certain gripper device, the corresponding avoidance sensing device detects the camera position and immediately transmits the position signal to the PLC's IO module. The PLC does not require software relay; the hardware independently parses the signal and sends a first pulse signal to the gripper driver of the gripper device. The gripper driver drives the gripper motor to rotate, causing the gripper body to rapidly extend and retract radially outward, thereby achieving gripper avoidance and preventing the gripper from entering the camera's scanning field of view. If multiple preset avoidance points are passed along the X-axis scanning path, the corresponding gripper avoidance actions are triggered sequentially. The response time of all avoidance actions is less than 0.01s, with no delay. After the first X-axis scan is completed, the line scan camera assembly moves to the end position of the first scan. The encoder of the X-axis motor feeds back the arrival signal to the controller. The controller sends a Y-axis translation command to the motion control card. The motion control card drives the Y-axis motor to rotate, causing the line scan camera assembly to translate along the Y-axis by a preset distance to the second scan position. After reaching the second scan position, the above X-axis scan and gripper avoidance actions are repeated to complete the second X-axis scan. The upper and lower line scan cameras each acquire one partial scan image. The X-axis linear scan and Y-axis translation are cycled in this way until all wafers to be inspected have been scanned. After scanning is completed, the controller performs image stitching and surface defect detection on the acquired images.

Images