Methods, systems, devices, and storage media for polishing equipment to detect debris

Abstract

The application provides a method, system, device and storage medium for detecting debris of a polishing device, applied to a silicon wafer polishing process, and comprises the following steps: monitoring a polishing area of the polishing device, and acquiring images of a surface of the polishing area in real time; processing the acquired images to obtain parameter characteristics of the debris; judging a defect risk level based on the parameter characteristics of the debris, and if a preset alarm condition is reached, the polishing device performs a preset response action. The application has the beneficial effect of realizing real-time detection and intelligent judgment of debris of the polishing device, can realize continuous monitoring without affecting the operation of the polishing device, significantly improves the detection automation level, reduces the dependence on manual operation, discovers abnormal debris in a timely manner to trigger machine alarm shutdown, and prevents process loss caused by the spread of debris.

Description

Technical Field

[0001] This invention belongs to the field of semiconductor manufacturing equipment technology, and in particular relates to a method, system, device and storage medium for detecting debris in polishing equipment. Background Technology

[0002] Chemical Mechanical Polishing (CMP) is a crucial process in semiconductor manufacturing for achieving global planarization of the wafer surface. Its polishing quality directly impacts the accuracy and yield of subsequent processes such as photolithography and deposition. CMP equipment typically operates continuously for extended periods, involving the coordinated action of various components including polishing pads, polishing fluids, carrier disks, and ring-shaped fixtures. In actual production, factors such as incoming material surface defects, wafer slippage during processing, process disturbances, and mechanical wear can easily generate debris, flaking particles, or foreign matter within the polishing area. If these debris remain on the polishing pad surface or adhere to the wafers, they can cause severe scratches, pits, or uneven material removal, leading to wafer scrap or reduced yield.

[0003] More seriously, in the continuous operation environment of CMP equipment, if the fragmentation problem is not detected and handled in a timely manner, it will spread as the process continues, causing a chain reaction of contamination to subsequent batches of wafers, and even leading to abnormal equipment wear and unplanned downtime. The fragmentation problem not only affects process stability, but also causes significant economic losses such as equipment downtime for maintenance and batch scrapping of products.

[0004] Currently, fragment detection in the industry largely relies on manual inspections or periodic shutdowns for checks. This results in long inspection cycles, low efficiency, and the risk of subjective misjudgment, making it difficult to meet the real-time monitoring requirements of continuous production. Although some equipment has introduced particle detection or liquid monitoring modules, they mostly focus on the statistical analysis of particle concentration in the polishing fluid, failing to accurately identify the specific location, shape, and distribution of fragments. Summary of the Invention

[0005] In view of the above problems, the present invention provides a method, system, device and storage medium for detecting debris in polishing equipment, so as to solve the above or other problems existing in the prior art.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: On the one hand, the present invention provides a method for detecting debris in polishing equipment, applied in the process of silicon wafer polishing, including the following steps:

[0007] Monitor the polishing area of ​​the polishing equipment and acquire images of the polishing area surface in real time;

[0008] The acquired image is processed to obtain the parametric features of the fragments;

[0009] The defect risk level is determined based on the parameter characteristics of the fragments. If the preset alarm conditions are met, the polishing equipment will execute the preset response action.

[0010] Further processing of the acquired image to obtain the parametric features of the fragments includes:

[0011] The acquired image is preprocessed to obtain the detection region;

[0012] A convolutional neural network is used to process the preprocessed image, locate the defect area, and identify the defect type.

[0013] Furthermore, the convolutional neural network is a deep convolutional neural network with an added feature pyramid network structure, which processes the preprocessed image including:

[0014] The convolution operation of the convolutional layer generates corresponding feature maps of several scales;

[0015] By fusing feature maps at several scales through a feature pyramid network, a multi-scale feature map is constructed to identify defect areas.

[0016] The system processes candidate defects within the defect area and outputs the detection results.

[0017] Furthermore, the defective areas include:

[0018] Crack initiation point, propagation direction, and termination point;

[0019] The spatial extent and edge contours of the fragments;

[0020] Abnormal brightness and / or abnormal darkness distribution in the image.

[0021] Furthermore, the test results included:

[0022] Crack type and confidence level;

[0023] Crack length, direction, and morphology;

[0024] Number of fragments, size, and location coordinates.

[0025] Furthermore, the defect risk level is generated based on crack length, fragment area, density, and confidence level, including:

[0026] Level 1 risk: minor cracks or tiny fragments;

[0027] Level 2 risk: Moderate defect.

[0028] Furthermore, based on the parameter characteristics of the fragments, the defect risk level is determined. If the preset alarm conditions are met, the polishing equipment executes preset response actions, including:

[0029] If the defect level reaches Level 1 risk, the polishing equipment will trigger an alarm signal to prompt the operator to perform a manual re-inspection.

[0030] If the defect level reaches level two risk, the polishing equipment will execute a shutdown response and stop.

[0031] Furthermore, the convolutional neural network collects samples including small fragments, polishing slurry-covered fragments, and reflective fragments when constructing the dataset, and labels the fragments, polishing slurry, and silicon wafer body.

[0032] Furthermore, the preprocessing of the acquired image to obtain the detection region includes:

[0033] The acquired image is processed by Gaussian filtering to suppress background noise and image sensor noise;

[0034] Adaptive histogram equalization, gamma correction, or illumination normalization are applied to the Gaussian filtered image to reduce brightness unevenness.

[0035] ROI extraction is performed based on the boundary line between the silicon wafer edge and the background to obtain the detection area;

[0036] The image of the detected region is processed and converted into a standard format that can be processed by convolutional neural networks.

[0037] On the other hand, the present invention also provides a system for detecting debris in polishing equipment, comprising,

[0038] The image acquisition module is used to monitor the polishing area of ​​the polishing equipment and acquire images of the surface of the polishing area in real time.

[0039] The image processing module, connected to the image acquisition module, is used to process the acquired images and obtain the parameter characteristics of the fragments;

[0040] The control module, connected to the image processing module, is used to determine the defect risk level based on the parametric characteristics of the fragments, and to control the polishing equipment to execute preset response actions according to the determination results.

[0041] Furthermore, the image processing module includes:

[0042] The image preprocessing unit is used to preprocess the acquired image;

[0043] The feature extraction unit is used to generate corresponding feature maps of several scales through the convolution operation of the convolutional layer;

[0044] The classification and recognition unit is used to fuse feature maps at several scales through a feature pyramid network to construct a multi-scale feature map and identify defect areas.

[0045] The result output unit is used to process candidate defects within the defect area, output the detection results, and interact with the control module.

[0046] Furthermore, the image acquisition module includes an industrial camera, a light source assembly, and a protective component. The industrial camera is housed within the protective component, which has a window to allow the industrial camera to acquire images. The light source assembly is located around the industrial camera.

[0047] Furthermore, the system also includes a human-machine interface module, which is connected to the image acquisition module, image processing module, and control module to display the acquired images, image processing output results, and the operating status of the polishing equipment.

[0048] The present invention also provides an electronic device, including a memory and a processor, wherein the processor is coupled to the memory, reads and executes instructions in the memory, thereby enabling the electronic device to implement the above-described method for detecting debris in a polishing device.

[0049] The present invention also provides a readable storage medium storing a computer program, which, when executed, implements the above-described method for detecting debris in a polishing device.

[0050] By employing the aforementioned technical solution, this method for detecting debris in polishing equipment combines industrial cameras and image recognition technology to achieve real-time detection and intelligent judgment of debris. By deploying industrial cameras in key areas of the polishing equipment and combining image recognition and intelligent analysis algorithms, it enables real-time capture and early warning of debris. This allows for continuous monitoring without affecting the operation of the polishing equipment, significantly improving the level of automation, reducing reliance on manual labor, and promptly detecting abnormal debris to trigger machine alarms and shutdowns. This prevents process losses caused by debris spread, thereby ensuring the stability of the polishing process and product quality. Upon detecting abnormal debris, it can link with the polishing equipment's control system to automatically issue shutdown or deceleration commands, preventing equipment damage or product scrapping caused by debris spread. It can replace manual inspection, achieving closed-loop automated operation of debris detection, alarm, and machine control, significantly improving industrial automation, reducing manual intervention and production losses, increasing equipment uptime, and ensuring stable and reliable image acquisition and recognition, adapting to complex humidity and vibration environments. Attached Figure Description

[0051] Figure 1 This is a logic flowchart of a method for detecting debris in a polishing device according to an embodiment of the present invention. Detailed Implementation

[0052] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0053] Figure 1 The diagram illustrates a logic flowchart of an embodiment of the present invention. This embodiment relates to a method and system for detecting debris in a polishing equipment, an electronic device, and a readable storage medium. It is used to detect debris within the polishing equipment during the polishing process of a silicon wafer. By processing the image of the silicon wafer surface in the detection area, real-time detection and classification of debris during the polishing process can be achieved, preventing wafer damage and equipment contamination caused by debris diffusion, and improving the safety and automation level of the polishing equipment operation.

[0054] A method for detecting debris in polishing equipment, such as Figure 1 As shown, the process applied to silicon wafer polishing includes the following steps:

[0055] The polishing area of ​​the polishing equipment is monitored, and images of the polishing area surface are acquired in real time. These images can display the surface condition of the polishing area in real time, such as whether there are fragments, scratches, pits and other defects on the silicon wafer surface. The acquired images are then processed to detect whether there are fragments in the polishing equipment, thus achieving non-contact fragment detection during the silicon wafer polishing process.

[0056] The acquired image is processed to obtain the parametric features of the fragments. In this step, a trained convolutional neural network is used to process the acquired image and extract the parametric features of the fragments, including the edge contour of the fragments, the distribution of the fragments, the area size of the fragments, the number of fragments, etc.

[0057] Based on the parameter characteristics of the fragments, the defect risk level is determined. If the preset alarm conditions are met, the polishing equipment executes a preset response action to clear or remove the fragments. If the preset alarm conditions are not met, the polishing equipment continues polishing, repeating the above steps of image acquisition, image processing, and defect risk level determination for the polishing area, thereby achieving automated fragment detection. The detection system continuously monitors the polishing area screen to determine whether the fragments have been cleared or removed from the risk area. When the polishing area screen returns to normal and stabilizes for a period of time, the detection system can automatically or manually restore the operation of the polishing equipment, thus achieving a complete closed-loop control of detection-response-feedback. This enables closed-loop automated operation of fragment detection, alarm, and polishing equipment control during silicon wafer polishing, improving the level of automation.

[0058] Specifically, in the step of monitoring the polishing area of ​​the polishing equipment and acquiring images of the polishing area surface in real time, the polishing area surface is continuously image acquired in real time, mainly acquiring images of the silicon wafer surface within the polishing area.

[0059] In this step, when acquiring images of the polished area surface in real time, the interval between acquiring two adjacent images can be selected and set according to actual needs, such as acquiring an image of the polished area surface every 1 second.

[0060] The steps of processing the acquired image to obtain the parametric features of the fragments include:

[0061] The acquired image is preprocessed to obtain the detection region. Subsequent processing mainly focuses on the detection region to reduce the amount of unnecessary computation in subsequent image processing and improve the speed of image processing.

[0062] This step includes:

[0063] The acquired image is processed by Gaussian filtering to suppress background noise and image sensor noise in order to maintain the continuity of the fragment crack structure;

[0064] Adaptive histogram equalization, gamma correction, or illumination normalization are applied to the Gaussian filtered image to reduce brightness unevenness and enhance the local contrast of the crack.

[0065] The ROI is extracted based on the boundary line between the silicon wafer edge and the background to obtain the detection area. The silicon wafer area is located by edge detection method, and the detection area is cropped to reduce the amount of unnecessary computation. The edge detection method can be HED (Holistically-Nested Edge Detection), BDCN (Bi-Directional Cascade Network), DiffusionEdge, Canny edge detection algorithm, etc., which are existing technologies and will not be described in detail here.

[0066] The image of the detected region is processed by converting it into a standard format that can be processed by the convolutional neural network, so that the convolutional neural network can then process the image and extract fragment parameter features.

[0067] A convolutional neural network is used to process the preprocessed image, locate the defect region, and identify the defect type. This convolutional neural network is a deep convolutional neural network with an added feature pyramid network structure. It is a deep learning model, which is a pre-trained model that can directly process the preprocessed image.

[0068] During training, the deep convolutional neural network with a feature pyramid network structure is adapted to the specific characteristics of silicon wafer polishing scenarios. This involves constructing a high-quality, highly adaptable dataset to lay the foundation for model iteration, improve the accuracy of the model's output, and enhance detection precision.

[0069] In adapting to different scenarios, the physical characteristics of silicon wafer fragments, core detection requirements, and imaging environment constraints are clearly defined to ensure targeted detection. Silicon wafer fragments are characterized by high reflectivity, irregular flake shapes, and susceptibility to polishing slurry coverage. The core detection requirement is to avoid damage to polishing equipment and product defects caused by silicon wafer fragments. Imaging environment constraints include the image acquisition module being located inside the polishing equipment, and the image acquisition module being affected by polishing slurry interference and light source reflection, impacting the quality of the acquired images. Therefore, when constructing the dataset, sample collection covers the actual operating conditions of the polishing equipment, focusing on collecting samples including small fragments, polishing slurry-covered fragments, and reflective fragments to ensure sample diversity. Labeling includes not only fragments but also interference areas such as polishing slurry and the silicon wafer itself, assisting the model in learning interference features. By training the sample data and adding new data, the model's ability to specifically address pain points in silicon wafer polishing scenarios is enhanced, such as simulating polishing slurry coverage, reflective fluctuations, and expanding small fragment samples, thereby improving the model's generalization ability.

[0070] When using a convolutional neural network to process a preprocessed image, the following steps are included:

[0071] The convolutional operations of the convolutional layers generate corresponding feature maps of several scales, including extracting local features through shallow convolutional operations and extracting structural and semantic features through deep convolutional operations, thereby extracting feature maps of several scales at different depths of the convolutional neural network.

[0072] In the process of extracting local features through shallow convolution operations, the preprocessed image is input into the first convolutional layer of the convolutional neural network. This layer includes multiple convolutional kernels (such as 3×3 or 5×5) to perform local region scanning on the preprocessed image, extracting the grayscale variation patterns of each small region. The ReLU activation function is used to enhance the nonlinear representation, enabling the model to capture complex crack edges or fragment textures. The convolutional kernel is essentially a filter that enhances the response to local patterns in the preprocessed image. High-frequency features (such as crack and fragment edges) produce strong responses after convolution, while low-frequency flat regions (such as the background) have weak responses, which is beneficial for distinguishing defects from the background.

[0073] Shallow convolutions capture local features, such as local texture or edge information. When extracting structural and semantic features through deep convolutions, these local features can be combined into high-level semantic features, enabling the model to understand crack propagation, fragment distribution, and defect levels. Specifically, shallow convolutions capture the most basic and local details in an image, such as the "vertical lines" and "diagonal edges" of crack edges, the "micro-textures" of the background metal surface, and the "isolated bright spot edges" of sensor noise. These features themselves do not represent "cracks" but are merely the most basic components of the image. Mid-level convolutions combine the scattered basic features extracted by shallow convolutions according to "spatial correlation," for example, "vertical lines + diagonal edges" form an "L-shaped structure," and "continuous horizontal edges" form a "long strip texture." Deep convolutions further combine and correlate the "component-level features" (such as crack fragments) of mid-level convolutions to form "global abstract features" that represent the essence of the target. For example, multiple "local crack fragments" are combined in a "continuous extension" relationship to form a "complete crack structure" (e.g., "a continuous strip from left to right, with clear edges and a certain width in the middle"). This feature has been abstracted to the point that "regardless of the specific location or width of the crack in the image, as long as it conforms to this structure, it is a crack." Shallow, mid-level, and deep convolutions can all extract corresponding scale feature maps, resulting in several scale feature maps. ,in, The resolution gradually decreases, for example, to 1 / 4, 1 / 8, 1 / 16, 1 / 32, etc.

[0074] By fusing feature maps at several scales using a Feature Pyramid Network (FPN), a multi-scale feature map is constructed to identify defect regions. The FPN is used for bottom-up and top-down fusion to construct the multi-scale feature map. Each multi-scale feature map For each A 3x3 convolution is then performed to eliminate the aliasing effect introduced by upsampling. A feature pyramid network is used to fuse feature maps at several scales, allowing the high-resolution details of the shallow layers and the high semantic information of the deeper layers to compensate for each other, enhancing the detection capability of defects of different sizes (microcracks and large cracks, scattered particles and large fragments). The model automatically locates and identifies defect regions using pre-trained feature channels and an attention mechanism. Here, the automatically identifiable defect regions include:

[0075] Crack initiation point, propagation direction, and termination point;

[0076] The spatial extent and edge contours of the fragments;

[0077] Abnormal brightness and / or abnormal darkness distribution in the image.

[0078] Candidate defects within the defect region are processed, and the detection results are output, including candidate region boxes for cracks and fragments. Non-maximum suppression (NMS) is used to eliminate duplicate boxes, yielding the final detection result, which includes:

[0079] Crack type and confidence level;

[0080] Crack length, direction, and morphology;

[0081] Number of fragments, size, and location coordinates.

[0082] In determining the defect risk level based on the parametric characteristics of fragments, if a preset alarm condition is met, the polishing equipment executes a preset response action. The defect risk level is generated based on crack length, fragment area, density, and confidence level, and can be classified and set according to actual production requirements. Defect risk levels include:

[0083] Level 1 risk: minor cracks or tiny fragments;

[0084] Level 2 risk: Moderate defect.

[0085] The risk level is determined based on the parameter characteristics of the fragments. If the preset alarm conditions are met, it indicates the presence of abnormal fragments. The polishing equipment then receives a control signal and executes preset response actions, including:

[0086] If the defect level reaches Level 1 risk, the polishing equipment will respond to the alarm signal and prompt the operator to perform a manual re-inspection: the polishing equipment can immediately slow down or stop the polishing operation to prevent the spread of debris, trigger the alarm device, and prompt the operator to perform a manual re-inspection through sound and light signals or the host computer interface;

[0087] If the defect level reaches level two risk, the polishing equipment will execute a shutdown response action to stop the machine, which may trigger a cleaning or recycling procedure to protect the polishing pad and the workpiece surface.

[0088] Once the debris is removed from the risk area, and the detection screen returns to normal and stabilizes for a period of time, the detection system can automatically or manually resume the operation of the polishing equipment, thereby achieving a complete closed-loop control of detection-response-feedback.

[0089] A system for detecting debris in polishing equipment, comprising,

[0090] The image acquisition module is used to monitor the polishing area of ​​the polishing equipment and acquire images of the surface of the polishing area in real time.

[0091] The image processing module is connected to the image acquisition module, receives images transmitted by the image acquisition module, processes the acquired images, and obtains the parameter characteristics of the fragments.

[0092] The control module is connected to the image processing module, receives the output results of the image processing module, determines the defect risk level based on the parameter characteristics of the fragments, and controls the polishing equipment to execute preset response actions according to the determination results.

[0093] The aforementioned image acquisition module includes an industrial camera, a light source assembly, and a protective component. The light source assembly is located around the industrial camera to ensure the ambient lighting brightness required for the camera to take pictures. The industrial camera is housed within the protective component, which has a window allowing the camera to acquire images. This image acquisition module is installed inside the polishing equipment, in a critical detection area, to continuously acquire images of the polished surface.

[0094] The industrial cameras mentioned above are commercially available products, and can be selected and configured according to actual needs. No specific requirements are given here.

[0095] The aforementioned protective component is a shell structure with internal space to accommodate the industrial camera, protecting it. The shape of the protective component can be selected and set according to actual needs; specific requirements are not specified here. This protective component is a sealed, one-piece molded structure made of corrosion-resistant aluminum alloy. Its surface is fluorinated, giving it waterproof, dustproof, and anti-condensation properties, making it suitable for the working environment inside polishing equipment.

[0096] A window is provided at the front end of the protective component (the end facing the inspection area). This window uses a waterproof and anti-fog protective film to ensure that the window can adapt to the working environment inside the polishing equipment and avoid interference with the industrial camera's ability to take pictures of the inspection area through the window. The connection between the window and the protective component housing is sealed to prevent moisture from entering the protective component and causing corrosion or other damage to the industrial camera. This waterproof and anti-fog window film is a commercially available product; its selection depends on actual needs, and specific requirements are not specified here.

[0097] Given the limited space, restricted installation location, and humid environment inside polishing equipment, an industrial camera is connected to the inner wall of the equipment via an adjustable structure. This allows the camera to rotate relative to the inner wall, enabling pitch and rotation movements and adjusting its position. This allows for fine-tuning of the viewing angle within the confined space, allowing the camera to capture images of different locations within the polishing area. This adjustable structure is a universal ball joint, such as a gimbal, a commercially available product. The specific configuration can be selected based on actual needs; no specific requirements are specified here.

[0098] The aforementioned light source components employ either a ring light source or a side-emitting light source to enhance surface reflection differences and improve the contrast of fragment features. These light source components are commercially available products and can be selected and configured according to actual needs.

[0099] The image processing module described above includes:

[0100] The image preprocessing unit is used to preprocess the acquired image;

[0101] The feature extraction unit is used to generate corresponding feature maps of several scales through the convolution operation of the convolutional layer;

[0102] The classification and recognition unit is used to fuse feature maps at several scales through a feature pyramid network to construct a multi-scale feature map and identify defect areas.

[0103] The result output unit is used to process candidate defects within the defect area, output the detection results, and interact with the control module.

[0104] To further optimize the design, the device for detecting debris in polishing equipment also includes a human-machine interface module. This module is connected to the image acquisition module, image processing module, and control module, and is used to display the acquired images, image processing output results, and the operating status of the polishing equipment. In some feasible embodiments, the human-machine interface module can be a touchscreen, a commercially available product, which can be selected and configured according to actual needs; specific requirements are not specified here.

[0105] This embodiment also provides an electronic device, including a memory and a processor, wherein the processor is coupled to the memory, reads and executes instructions in the memory, so that the electronic device implements the above-described method for detecting debris in a polishing device.

[0106] This embodiment also provides a readable storage medium storing a computer program, which, when executed, implements the above-described method for detecting debris in a polishing device.

[0107] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the methods, processes, or functions described in the embodiments of the present invention are generated.

[0108] The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0109] The computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)). Where there is no conflict, the solutions of the above embodiments can be combined.

[0110] By employing the aforementioned technical solution, this method for detecting debris in polishing equipment combines industrial cameras and image recognition technology to achieve real-time detection and intelligent judgment of debris. By deploying industrial cameras in key areas of the polishing equipment and combining image recognition and intelligent analysis algorithms, it enables real-time capture and early warning of debris. This allows for continuous monitoring without affecting the operation of the polishing equipment, significantly improving the level of automation, reducing reliance on manual labor, and promptly detecting abnormal debris to trigger machine alarms and shutdowns. This prevents process losses caused by debris spread, thereby ensuring the stability of the polishing process and product quality. Upon detecting abnormal debris, it can link with the polishing equipment's control system to automatically issue shutdown or deceleration commands, preventing equipment damage or product scrapping caused by debris spread. It can replace manual inspection, achieving closed-loop automated operation of debris detection, alarm, and machine control, significantly improving industrial automation, reducing manual intervention and production losses, increasing equipment uptime, and ensuring stable and reliable image acquisition and recognition, adapting to complex humidity and vibration environments.

[0111] The embodiments of the present invention have been described in detail above, but the content described is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. A method for detecting debris in polishing equipment, characterized in that: Applied to the silicon wafer polishing process, it includes the following steps: Monitor the polishing area of ​​the polishing equipment and acquire images of the polishing area surface in real time; The acquired image is processed to obtain the parametric features of the fragments; The defect risk level is determined based on the parameter characteristics of the fragments. If the preset alarm conditions are met, the polishing equipment will execute the preset response action.

2. The method for detecting debris in polishing equipment according to claim 1, characterized in that: The step of processing the acquired image to obtain the parameter features of the fragments includes: The acquired image is preprocessed to obtain the detection region; A convolutional neural network is used to process the preprocessed image, locate the defect area, and identify the defect type.

3. The method for detecting debris in polishing equipment according to claim 2, characterized in that: The convolutional neural network is a deep convolutional neural network with an added feature pyramid network structure. The processing of the preprocessed image includes: The convolution operation of the convolutional layer generates corresponding feature maps of several scales; By fusing feature maps at several scales through a feature pyramid network, a multi-scale feature map is constructed to identify defect areas. The system processes candidate defects within the defect area and outputs the detection results.

4. The method for detecting debris in a polishing device according to claim 3, characterized in that: The defective region includes: Crack initiation point, propagation direction, and termination point; The spatial extent and edge contours of the fragments; Abnormal brightness and / or abnormal darkness distribution in the image.

5. The method for detecting debris in a polishing device according to claim 3 or 4, characterized in that: The detection results include: Crack type and confidence level; Crack length, direction, and morphology; Number of fragments, size, and location coordinates.

6. The method for detecting debris in a polishing device according to any one of claims 1-4, characterized in that: The defect risk level is generated based on crack length, fragment area, density, and confidence level, including: Level 1 risk: minor cracks or tiny fragments; Level 2 risk: Moderate defect.

7. The method for detecting debris in a polishing device according to claim 6, characterized in that: Based on the parameter characteristics of the fragments, the defect risk level is determined. If the preset alarm conditions are met, the polishing equipment executes preset response actions, including: If the defect level reaches the first-level risk level, the polishing equipment will trigger an alarm signal to prompt the operator to perform a manual re-inspection. If the defect level reaches the level two risk, the polishing equipment will execute a shutdown response action and stop.

8. The method for detecting debris in a polishing device according to claim 2, 3, or 4, characterized in that: The convolutional neural network collects samples including small fragments, polishing fluid-covered fragments, and reflective fragments when constructing the dataset, and labels the fragments, polishing fluid, and silicon wafer body.

9. The method for detecting debris in a polishing device according to claim 8, characterized in that: The preprocessing of the acquired image to obtain the detection region includes: The acquired image is processed by Gaussian filtering to suppress background noise and image sensor noise; Adaptive histogram equalization, gamma correction, or illumination normalization are applied to the Gaussian filtered image to reduce brightness unevenness. The ROI is extracted based on the boundary line between the silicon wafer edge and the background to obtain the detection area; The image of the detected region is processed to convert it into a standard format that can be processed by a convolutional neural network.

10. A system for detecting debris in polishing equipment, characterized in that: include, The image acquisition module is used to monitor the polishing area of ​​the polishing equipment and acquire images of the surface of the polishing area in real time. An image processing module, connected to the image acquisition module, is used to process the acquired image and obtain the parameter characteristics of the fragments; The control module, connected to the image processing module, is used to determine the defect risk level based on the parameter characteristics of the fragments, and control the polishing equipment to perform preset response actions according to the determination result.

11. The system for detecting debris in a polishing device according to claim 10, characterized in that: The image processing module includes: The image preprocessing unit is used to preprocess the acquired image; The feature extraction unit is used to generate corresponding feature maps of several scales through the convolution operation of the convolutional layer; The classification and recognition unit is used to fuse feature maps at several scales through a feature pyramid network to construct a multi-scale feature map and identify defect areas. The result output unit is used to process candidate defects within the defect area, output the detection results, and interact with the control module.

12. The system for detecting debris in a polishing device according to claim 10 or 11, characterized in that: The image acquisition module includes an industrial camera, a light source assembly, and a protective component. The industrial camera is housed within the protective component, which has a window to allow the industrial camera to acquire images. The light source assembly is located around the industrial camera.

13. The system for detecting debris in a polishing device according to claim 10 or 11, characterized in that: The system also includes a human-machine interface module, which is connected to the image acquisition module, the image processing module, and the control module, and is used to display the acquired images, image processing output results, and the operating status of the polishing equipment.

14. An electronic device, characterized in that: The device includes a memory and a processor, the processor being coupled to the memory to read and execute instructions in the memory, causing the electronic device to implement the method for detecting debris in a polishing device as described in any one of claims 1-9.

15. A readable storage medium, characterized in that: The readable storage medium stores a computer program that, when executed, implements the method for detecting debris in a polishing device according to any one of claims 1-9.

Images