A semiconductor dust removal method, system, device, and storage medium

By linking two-level two-dimensional detection and condition-triggered three-dimensional detection, the problem of distinguishing between dust and defects in semiconductor wafer inspection has been solved, realizing an efficient and accurate inspection and cleaning process, and improving production efficiency and automation level.

CN122872003BActive Publication Date: 2026-11-17JIANWEI SEMICONDUCTOR (SHENZHEN) CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611363747.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-09-04
Publication Date
2026-11-17
Estimated Expiration
2046-09-04

AI Technical Summary

Technical Problem

Existing semiconductor wafer inspection technologies cannot effectively distinguish between dust and defects, resulting in a high false positive rate. Furthermore, the inspection and dust removal processes are disconnected, leading to low efficiency and failing to meet mass production requirements.

Method used

The system employs a two-level two-dimensional detection and a condition-triggered three-dimensional detection linkage architecture. First, two-dimensional detection is used to locate abnormal areas and perform preliminary cleaning. Then, two-dimensional re-inspection is used to verify the cleaning effect. Three-dimensional detection is triggered only for areas that do not meet the standards, and the abnormality type is accurately determined by combining three-dimensional morphological features.

Benefits of technology

It achieves accurate differentiation between micro-dust and defects, reduces the amount of 3D inspection data, improves inspection efficiency and automation level, reduces the risk of human intervention, and meets mass production requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122872003B_ABST
    Figure CN122872003B_ABST
Patent Text Reader

Abstract

This application provides a semiconductor dust removal method, control system, device, and storage medium. The method includes acquiring first two-dimensional morphological data of the semiconductor wafer surface; determining first coordinate information of at least one abnormal region of the semiconductor wafer based on the first two-dimensional morphological data; removing dust from the at least one abnormal region based on the first coordinate information; acquiring second two-dimensional morphological data of the semiconductor wafer surface after dust removal; determining second coordinate information of the abnormal region based on the second two-dimensional morphological data; acquiring three-dimensional topographic data of the abnormal region based on the second coordinate information; determining the abnormal type of the abnormal region based on the three-dimensional topographic data; and generating a corresponding treatment strategy for the semiconductor wafer. This application achieves the screening, identification, and automated classification processing of micro-dust and defects on the wafer surface through a linkage architecture of two-dimensional detection and condition-triggered three-dimensional detection.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of semiconductor manufacturing technology, specifically to a semiconductor dust removal control method, system, device, and storage medium. Background Technology

[0002] In semiconductor wafer and chip manufacturing, the cleanliness and microstructure quality of the wafer surface directly determine the device yield and reliability. As semiconductor process nodes continue to evolve towards the nanometer scale, even minute contamination and damage on the wafer surface can lead to short circuits, open circuits, or photolithography alignment failures. Therefore, between each step of wafer manufacturing, the wafer surface must be inspected and cleaned to ensure that the wafer surface meets the process cleanliness requirements before entering the next process.

[0003] Currently, foreign object detection on semiconductor surfaces mainly relies on traditional two-dimensional optical microscopy or bright / dark field scattering detection techniques. However, these existing technologies have the following significant drawbacks: First, traditional 2D inspection cannot effectively distinguish between surface dust and defect types. 2D image inspection is mainly based on grayscale changes, contrast, or two-dimensional contour features. For dust particles adhering to the wafer surface and defects such as scratches, pits, and edge chipping caused by the wafer itself, they often present similar contrast anomalies in 2D images, making accurate differentiation difficult. If dust is misjudged as irreversible defects, it will lead to the incorrect scrapping of wafers, resulting in material waste and increased production costs. If defects are misjudged as dust and physical cleaning is performed blindly, not only will the defects not be eliminated, but surface scratches may also be aggravated, further reducing wafer yield.

[0004] Secondly, existing wafer dust removal methods are mostly "blind sweeping" cleaning, which means that the entire wafer surface is cleaned uniformly without distinguishing the specific location and type of abnormal areas. On the one hand, this method unnecessarily cleans areas without abnormalities, increasing process time and the risk of secondary contamination or damage to the wafer; on the other hand, blind sweeping cleaning cannot identify and skip areas with existing physical defects on the wafer surface, and may even cause defects to expand due to physical contact of cleaning components or airflow impact, resulting in irreversible wafer scrapping.

[0005] Third, existing technologies lack an automatic feedback and linkage mechanism between detection results and dust removal actuators. 2D detection equipment and dust removal equipment operate independently; after detection, manual judgment or transmission of results through an intermediate system is required, making it impossible to automatically convert detection results into control commands for the dust removal actuators. This not only reduces the automation level and production efficiency of the production line but also introduces the risk of human error, preventing detection data from directly serving the optimization of dust removal strategies.

[0006] Fourth, although 3D inspection technology can provide height information of the wafer surface, helping to distinguish between dust and defects, the point-by-point 3D scanning of the entire wafer is extremely slow, and the inspection time is far from meeting the production line's capacity requirements. Due to the large area of ​​semiconductor wafers and the dense density of 3D inspection points, if a full 3D scan of the entire wafer is performed, the inspection time for a single wafer will reach tens of minutes or even hours, making it impractical for mass production lines. Therefore, how to significantly improve inspection efficiency while ensuring inspection accuracy, and achieve an effective balance between speed and accuracy, is a technical challenge that urgently needs to be solved in this field.

[0007] To address the aforementioned issues, some existing technologies attempt to employ a strategy of first performing 2D detection and positioning, followed by 3D re-inspection of the positioned area, in order to reduce the 3D scanning range. However, in such solutions, dust removal is often performed after 3D detection; that is, all detection and judgment are completed before dust removal, resulting in a disconnect between the detection and dust removal processes and an inability to form a closed-loop feedback. Specifically, if a detailed 3D inspection has already been completed before dust removal, the cleaning effect cannot be known after dust removal, making it impossible to perform adaptive secondary cleaning or adjust the cleaning strategy based on the cleaning effect. Furthermore, due to the high cost and time-consuming nature of 3D detection, placing 3D detection before dust removal unnecessarily forces many areas that could have been confirmed as clean with simple 2D re-inspection to enter the 3D detection process, still resulting in an efficiency bottleneck.

[0008] Therefore, how to provide a detection and dust removal linkage scheme that can quickly locate and initially clean abnormal areas through two-dimensional detection, screen out areas that still need to be treated through secondary two-dimensional detection after dust removal, and trigger three-dimensional detection only when necessary for accurate type determination, so as to effectively reduce the amount of data processing for three-dimensional detection, improve the overall detection efficiency while ensuring classification accuracy, and automatically generate corresponding differentiated treatment strategies based on the determination results, is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0009] This application provides a dust removal control method, system, device, and storage medium, aiming to solve the problems of the prior art. To achieve the above objectives, this application provides the following technical solution: In a first aspect, this application provides a semiconductor dust removal method, comprising the following steps: S1. Obtain the first two-dimensional morphological data of the semiconductor wafer surface; S2. Based on the first two-dimensional morphological data, determine the first coordinate information of at least one abnormal region of the semiconductor wafer; S3. Based on the first coordinate information, perform dust removal on at least one of the abnormal areas; S4. Obtain the second two-dimensional morphological data of the semiconductor wafer surface after dust removal; S5. Determine the second coordinates of the abnormal region based on the second two-dimensional morphological data. information; S6. Obtain the three-dimensional topographic data of the abnormal region based on the second coordinate information; S7. Determine the anomaly type of the abnormal region based on the three-dimensional topography data, and generate a corresponding handling strategy for the semiconductor wafer. Specifically, the first two-dimensional topography data determines the first coordinate information of at least one abnormal region of the semiconductor wafer, including: S21. Control the two-dimensional detection module to perform a panoramic scan of the surface of the semiconductor wafer to obtain a first two-dimensional image of the semiconductor wafer; S22. The first two-dimensional image is divided into blocks, and the image features of each block region are extracted. S23. Compare the image features of each segmented region with a preset abnormal image feature threshold. The comparison is performed, and regions whose image features exceed the preset threshold are marked as suspected abnormal regions; S24. Extract the pixel coordinates of the suspected abnormal region in the first two-dimensional image; S25. Based on the calibration parameters of the two-dimensional detection module, the pixel coordinates are mapped to the physical coordinates of the suspected abnormal area on the semiconductor wafer surface, and the physical coordinates are used as the first coordinate information. S26. The suspected abnormal area is identified as the abnormal area.

[0010] Specifically, the calibration parameters of the two-dimensional detection module include at least one of the following: the intrinsic parameter matrix, extrinsic parameter matrix, distortion coefficient, and pixel equivalent of the two-dimensional detection module; wherein, the intrinsic parameter matrix is ​​used to describe the internal optical characteristics of the camera, such as focal length, principal point coordinates, and pixel size; the extrinsic parameter matrix is ​​used to describe the spatial pose of the camera relative to the support platform or wafer surface; the distortion coefficient is used to correct image distortion caused by the camera lens; and the pixel equivalent is used to convert pixel distances in the image into physical distances; through the above calibration parameters, a mapping relationship from any pixel coordinate in the first two-dimensional image to the physical coordinates of the semiconductor wafer surface can be established, thereby achieving accurate positioning of abnormal areas.

[0011] Specifically, determining the first coordinate information of at least one abnormal region of the semiconductor wafer based on the first two-dimensional morphological data further includes: The abnormal region on the surface of the semiconductor wafer is identified by comparing the first two-dimensional morphological data with the reference morphological data. The identified abnormal regions are subjected to morphological processing, which includes at least one of dilation, erosion, opening, and closing operations, to eliminate isolated noise points and connect adjacent abnormal pixel regions to obtain a complete abnormal region contour. The first coordinate information is determined based on the position of the abnormal region contour on the semiconductor wafer surface.

[0012] Specifically, the second two-dimensional morphological data of the semiconductor wafer surface after dust removal is obtained. include: The two-dimensional detection module is controlled to perform a re-inspection scan on the surface of the semiconductor wafer after dust removal to obtain a second two-dimensional image; or, based on the first coordinate information, the two-dimensional detection module is controlled to move to the abnormal area to perform a local scan to obtain a local two-dimensional image of the abnormal area.

[0013] Specifically, based on the second two-dimensional morphological data, determining the second coordinate information of the abnormal region includes: S51. Compare the second two-dimensional morphological data with the baseline morphological data to determine whether the abnormal area still has an anomaly; S52. If the abnormal area still has an abnormality after dust removal, the outline of the abnormal area that still has an abnormality is extracted, and the second coordinate information is determined according to the position of the outline on the semiconductor wafer surface. S53. If the abnormal area is not abnormal after dust removal, the abnormal area is marked as a clean area, and the subsequent processing of the abnormal area is terminated.

[0014] Specifically, based on the second coordinate information, the three-dimensional topographic data of the abnormal region is obtained, including: S61. Based on the second coordinate information, control the three-dimensional detection module to move to the abnormal area where the abnormality still exists; S62. Perform a three-dimensional scan on the abnormal area to obtain a three-dimensional topographic image of the abnormal area; S63. Extract the three-dimensional geometric features of the abnormal region from the three-dimensional topography image, and use the three-dimensional geometric features as the three-dimensional topography data.

[0015] Specifically, extracting the three-dimensional geometric features of the abnormal region from the three-dimensional topography image, and using the three-dimensional geometric features as the three-dimensional topography data, includes: S631, Define the reference plane of the abnormal region as... The highest or lowest point of the abnormal region is defined as... ; S632, the highest point of the abnormal region or lowest point Relative to the reference plane The height extreme value H is taken as the first parameter; S633. The distribution dimension D of the abnormal region in the direction parallel to the reference plane is taken as the second parameter; S634. Based on the ratio of the height extreme value H to the spread size D, determine the steepness index R of the abnormal region, R=|H| / D; S635. The height extreme value H, the spread size D, and the steepness index R are used as the three-dimensional geometric features.

[0016] Specifically, the display dimension D is determined by establishing a height section at a preset height offset △H above or below the reference plane. The height offset ΔH = k × H, where k is a preset proportionality coefficient, 0 <k<1; The abnormal region is intercepted at the height section. The outline of the section is calculated; the maximum span between any two points on the cross-sectional outline is calculated, and the maximum span is used as the display dimension D.

[0017] This application provides a semiconductor dust removal control system, comprising: a first two-dimensional detection module for acquiring first two-dimensional morphological data of the semiconductor wafer surface; a first positioning module for determining first coordinate information of at least one abnormal region of the semiconductor wafer based on the first two-dimensional morphological data; a dust removal module for removing dust from at least one of the abnormal regions based on the first coordinate information; a second two-dimensional detection module for acquiring second two-dimensional morphological data of the semiconductor wafer surface after dust removal; a second positioning module for determining second coordinate information of the abnormal region based on the second two-dimensional morphological data; a three-dimensional detection module for acquiring three-dimensional topographic data of the abnormal region based on the second coordinate information; and a classification module for determining the abnormal type of the abnormal region based on the three-dimensional topographic data and generating a corresponding treatment strategy for the semiconductor wafer.

[0018] This application provides a semiconductor dust removal device in three aspects, including: a device frame assembly; a motion assembly disposed on the device frame assembly; a support platform disposed on the motion assembly for fixing and supporting a semiconductor wafer; a detection assembly disposed on the motion assembly for scanning the surface of the semiconductor wafer to collect first two-dimensional morphological data, second two-dimensional morphological data, and three-dimensional topographic data; a dust removal assembly disposed on the motion assembly for cleaning dust from the surface of the semiconductor wafer; and a dust removal control system connected to the detection assembly, the motion assembly, and the dust removal assembly.

[0019] This application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the semiconductor dust removal method. The semiconductor dust removal method includes the following steps: S1. Obtain the first two-dimensional morphological data of the semiconductor wafer surface; S2. Based on the first two-dimensional morphological data, determine the first coordinate information of at least one abnormal region of the semiconductor wafer; S3. Based on the first coordinate information, perform dust removal on at least one of the abnormal areas; S4. Obtain the second two-dimensional morphological data of the semiconductor wafer surface after dust removal; S5. Determine the second coordinate information of the abnormal region based on the second two-dimensional morphological data; S6. Obtain the three-dimensional topographic data of the abnormal region based on the second coordinate information; S7. Determine the anomaly type of the abnormal region based on the three-dimensional topography data, and generate a corresponding processing strategy for the semiconductor wafer.

[0020] Compared with existing technologies, the semiconductor dust removal method, control system, equipment, and storage medium provided by this invention, through a linkage architecture of two-dimensional detection and condition-triggered three-dimensional detection, have the following significant and beneficial technical effects: First, this invention employs an architecture combining two-level two-dimensional detection with condition-triggered three-dimensional detection. First, a rapid panoramic scan of the entire wafer is performed using a first two-dimensional detection to coarsely locate abnormal areas. After dust removal, a second two-dimensional detection is used for re-inspection, triggering three-dimensional detection only on areas that still exhibit abnormalities after dust removal. Since most micro-dust particles are effectively removed after the first dust removal in actual production, only a very small number of abnormal areas require the three-dimensional detection stage. Compared to a comprehensive three-dimensional scan of the entire wafer, this invention significantly reduces the data processing volume for three-dimensional detection and substantially shortens the detection time per wafer, meeting the capacity requirements of mass production lines. Furthermore, compared to a scheme that performs three-dimensional detection before dust removal, this invention places three-dimensional detection after dust removal, eliminating the need for many already cleaned areas to undergo the three-dimensional detection process, further improving overall efficiency.

[0021] Secondly, this invention uses a three-dimensional detection module to acquire the three-dimensional geometric features of the abnormal region, such as the height extreme value H, the distribution size D, and the steepness index R. Based on these features, a classification model is used to accurately determine the anomaly type. Microparticles are foreign deposits with a clear physical interface to the wafer surface, appearing as a sudden bulge from the reference plane, characterized by high steepness and large edge gradients. In contrast, physical defects such as scratches, pits, and edge chips are structural deficiencies or deformations of the wafer's own material, exhibiting a gradual transition with the surrounding normal wafer surface, characterized by low steepness and small edge gradients. Based on the essential differences in these three-dimensional morphological features, this invention can accurately distinguish between microparticles and defects, avoiding the problem of misjudging microparticles as defects due to similar grayscale in traditional 2D inspection, leading to incorrect wafer rejection. It also avoids the problem of misjudging defects as microparticles and blindly cleaning them, resulting in defect expansion or secondary damage to the wafer surface.

[0022] Third, this invention establishes a closed-loop feedback mechanism by first using two-dimensional detection to locate abnormal areas and drive the dust removal components to perform targeted cleaning, and then using a second two-dimensional detection to verify the cleaning effect. For areas that fail to meet cleaning standards, the system triggers three-dimensional detection for precise judgment and generates corresponding disposal strategies; for areas that meet cleaning standards, the system terminates subsequent processing steps to avoid invalid detection. This closed-loop mechanism allows detection results to be automatically fed back to the dust removal actuators, completing the entire process from anomaly detection to cleaning verification and classification disposal without manual intervention. This significantly improves the automation level and production efficiency of the production line and reduces labor costs.

[0023] Fourth, this invention quantifies the overall quality of wafers through a weighted combination of contamination and defect evaluation indicators. Based on the evaluation results, wafers are classified into four levels: Grade A (Qualified), Grade B (Forced Dust Level), Grade C (Re-inspection), and Grade D (Scrap). Each level corresponds to a different disposal strategy. Wafers with fatal defects are directly classified as scrap and subsequent processing is terminated, avoiding wasted processing time on scrapped wafers. Wafers with only micro-dust contamination are classified as requiring dust removal, and the dust removal components are activated for re-dust removal. Wafers that still fail to meet standards after secondary cleaning or are at the edge of the problem are classified as re-inspection, and are manually reviewed by operators. Through the above multi-level classification and refined disposal strategy, automated and refined management of wafer quality control is achieved, improving the overall efficiency and yield management level of the production line. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 This is a flowchart of the semiconductor dust removal method in the embodiments of this application; Figure 2 for Figure 1 The flowchart of S2 shown below; Figure 3 for Figure 1 The flowchart of S5 shown is shown below; Figure 4 for Figure 1 The flowchart of S6 shown is shown; Figure 5 for Figure 4 The flowchart of S63 shown is shown; Figure 6 A first-angle schematic diagram of the three-dimensional geometric morphology of dust on the surface of a semiconductor wafer; Figure 7 A second-angle schematic diagram of the three-dimensional geometric morphology of dust on the surface of a semiconductor wafer; Figure 8 This is a structural block diagram of the dust removal control system in the embodiments of this application; Figure 9 This is a three-dimensional structural diagram of the dust removal equipment in the embodiments of this application. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0026] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicator will also change accordingly.

[0027] Furthermore, the use of terms such as "first," "second," etc., in this application is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least three, such as three, four, etc., unless otherwise explicitly specified.

[0028] In this application, unless otherwise expressly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise expressly limited. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0029] Furthermore, the technical solutions of the various embodiments of this application can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this application.

[0030] This invention provides a semiconductor dust removal method applicable to surface quality inspection and cleaning processes in semiconductor wafer manufacturing. The method employs a linked architecture of two-dimensional detection and condition-triggered three-dimensional detection. First, a first-dimensional detection performs a rapid panoramic scan of the wafer surface and locates abnormal areas, driving the dust removal component to perform targeted cleaning. Then, a second-dimensional detection verifies the cleaning effect. Only areas that fail to meet cleaning standards are triggered for precise type determination in three-dimensional detection, generating differentiated treatment strategies. This achieves rapid screening, accurate identification, and automated grading of micro-dust and defects on the wafer surface.

[0031] Please refer to Figure 1 The present invention provides a flowchart of a semiconductor dust removal method. Step S1: Obtain the first two-dimensional morphological data of the semiconductor wafer surface.

[0032] In some embodiments of the present invention, acquiring the first two-dimensional morphological data of the semiconductor wafer surface includes: controlling a support platform to move the semiconductor wafer to a position below a two-dimensional detection module; the two-dimensional detection module performing a panoramic scan of the semiconductor wafer with the aid of an illumination source to acquire the first two-dimensional morphological data of the semiconductor wafer. This embodiment uses a two-dimensional detection module for initial inspection, which has the advantages of fast scanning speed and wide coverage, making it suitable for quickly locating suspicious abnormal areas on large-area wafer surfaces and improving overall detection efficiency.

[0033] In some embodiments of the present invention, the two-dimensional detection module is a 2D camera. At least one set of 2D cameras is arranged above and / or below the transmission path of the semiconductor wafer. The 2D cameras synchronously expose and acquire data on the surface of the semiconductor wafer to obtain first two-dimensional morphological data. The first two-dimensional morphological data includes grayscale images and / or color images. In this embodiment, a high-resolution area array industrial camera is used in conjunction with a telecentric optical lens. In another embodiment, the 2D camera can also be a line array camera used in conjunction with the stepping motion of the support platform to achieve panoramic scanning. Line array cameras have higher scanning speed and resolution, making them suitable for rapid inspection of large-area wafers.

[0034] In some embodiments of the present invention, after acquiring the first morphology data of the semiconductor wafer, the method further includes: preprocessing the first morphology data, wherein the preprocessing includes at least one of dark field correction, bright field correction, noise filtering, contrast enhancement, and image sharpening. This processing can eliminate systematic errors and random noise introduced during image acquisition, thereby improving the accuracy of subsequent abnormal region identification.

[0035] Please refer to Figure 2 Step S2: Based on the first two-dimensional morphological data, determine the first coordinate information of at least one abnormal region of the semiconductor wafer.

[0036] In some embodiments of the present invention, determining the first coordinate information of at least one abnormal region of the semiconductor wafer based on the first two-dimensional morphological data includes: S21. Control the two-dimensional detection module to perform a panoramic scan of the surface of the semiconductor wafer to obtain a first two-dimensional image of the semiconductor wafer; S22. The first two-dimensional image is divided into blocks, and the image features of each block region are extracted. S23. Compare the image features of each block region with a preset abnormal image feature threshold, and mark the block regions whose image features exceed the preset threshold as suspected abnormal regions; S24. Extract the pixel coordinates of the suspected abnormal region in the first two-dimensional image; S25. Based on the calibration parameters of the two-dimensional detection module, the pixel coordinates are mapped to the physical coordinates of the suspected abnormal area on the semiconductor wafer surface, and the physical coordinates are used as the first coordinate information. S26. The suspected abnormal area is identified as the abnormal area.

[0037] In some embodiments of the present invention, the calibration parameters of the two-dimensional detection module include at least one of the following: an intrinsic parameter matrix, an extrinsic parameter matrix, a distortion coefficient, and a pixel equivalent of the two-dimensional detection module. The intrinsic parameter matrix describes the internal optical characteristics of the camera, such as its focal length, principal point coordinates, and pixel size; the extrinsic parameter matrix describes the spatial pose of the camera relative to the support platform or wafer surface; the distortion coefficient corrects image distortion caused by the camera lens; and the pixel equivalent converts pixel distances in the image into physical distances. Through these calibration parameters, a mapping relationship can be established from any pixel coordinate in the first two-dimensional image to the physical coordinates of the semiconductor wafer surface, thereby achieving precise localization of abnormal regions.

[0038] In some embodiments of the present invention, during the mapping process, the pixel coordinates are first distorted to eliminate positional deviations introduced by lens distortion; then, the distorted pixel coordinates are converted into coordinates in the camera coordinate system; finally, the coordinates in the camera coordinate system are converted into physical coordinates on the wafer surface using the extrinsic parameter matrix. Through the above multi-level coordinate transformation, the accuracy of the first coordinate information can be ensured to meet the positioning requirements of subsequent dust removal steps. In some embodiments of the present invention, determining the abnormal region of the semiconductor wafer based on the first two-dimensional morphological data further includes: comparing the first two-dimensional morphological data with reference morphological data to identify the abnormal region on the surface of the semiconductor wafer; performing morphological processing on the identified abnormal region, the morphological processing including at least one of dilation, erosion, opening, and closing operations to eliminate isolated noise points and connect adjacent abnormal pixel regions to obtain a complete abnormal region outline; and determining the first coordinate information based on the position of the abnormal region outline on the surface of the semiconductor wafer.

[0039] The reference morphological data refers to the surface morphological data of a defect-free standard semiconductor wafer, or the design morphological data of the semiconductor wafer. This embodiment identifies anomalous regions by comparing them with the reference morphological data, effectively eliminating interference from non-abnormal features such as inherent patterns, textures, and markings on the wafer surface, thus reducing the false alarm rate. The image difference combined with binarization threshold segmentation method has low computational cost and fast processing speed, making it suitable for online real-time detection scenarios. In another embodiment, a machine learning-based anomaly detection method can also be used, employing a pre-trained classifier to detect anomalous regions in the image. This method can adapt to more complex background changes and achieve higher detection accuracy. Morphological processing effectively filters out isolated false detection points caused by image noise, while connecting adjacent anomalous pixels into complete anomalous regions, providing accurate region boundaries for subsequent localization and dust removal.

[0040] Step S3: Based on the first coordinate information, perform dust removal on at least one of the abnormal areas.

[0041] In some embodiments of the present invention, the step of removing dust from at least one of the abnormal areas based on the first coordinate information includes: controlling a motion component to drive a dust removal component to move to the abnormal area based on the first coordinate information; and performing a dust removal operation on the abnormal area using physical dust removal methods and / or chemical dust removal methods. The abnormal area undergoes a first dust removal treatment using physical dust removal methods and / or chemical dust removal methods. The physical dust removal methods include, but are not limited to, any one or more combinations of airflow purging, ultrasonic cleaning, brushing, and vacuum adsorption. In this embodiment, the first dust removal uses airflow purging, where high-speed clean air blows loose dust particles from the surface of the abnormal area away from the wafer surface. This method is non-contact cleaning and will not cause mechanical damage to the wafer surface, making it suitable as a preliminary dust removal method. In another embodiment, the first dust removal can also use an adhesive rod contact cleaning method, where the adhesive material at the tip of the adhesive rod contacts the surface of the abnormal area, using adhesion to remove the dust particles. This method has a better removal effect on dust particles with strong adhesion.

[0042] Step S4: Obtain the second two-dimensional morphological data of the semiconductor wafer surface after dust removal.

[0043] In some embodiments of the present invention, obtaining the second two-dimensional morphological data of the semiconductor wafer surface after dust removal includes: controlling the two-dimensional detection module to perform a re-inspection scan on the semiconductor wafer surface after dust removal to obtain a second two-dimensional image; or controlling the two-dimensional detection module to move to the abnormal area for local scanning according to the first coordinate information to obtain a local two-dimensional image of the abnormal area.

[0044] In some embodiments of the present invention, a re-inspection scan is performed on the surface of the semiconductor wafer after dust removal, which can accurately evaluate the dust removal effect, determine whether the micro dust has been completely removed, and provide a basis for determining whether to trigger three-dimensional detection.

[0045] Please refer to Figure 3 Step S5: Determine the second coordinate information of the abnormal region based on the second two-dimensional morphological data, including: S51. Compare the second two-dimensional morphological data with the baseline morphological data to determine whether the abnormal area still has an anomaly; S52. If the abnormal area still has an abnormality after dust removal, the outline of the abnormal area that still has an abnormality is extracted, and the second coordinate information is determined according to the position of the outline on the semiconductor wafer surface. S53. If the abnormal area is found to be normal after dust removal, then the abnormal area will be... The area is marked as clean, and the subsequent processing of the abnormal area is terminated.

[0046] This step verifies the dust removal effect through a second 2D inspection, forming a closed-loop feedback loop of inspection, dust removal, and re-inspection. For areas that meet the cleaning standards, the system automatically terminates the subsequent processing flow to avoid unnecessary 3D inspection; for areas that do not meet the cleaning standards, the system extracts their second coordinate information to provide a positioning basis for subsequent condition-triggered 3D inspection. Through this mechanism, a large number of already cleaned areas do not need to enter the 3D inspection process, effectively reducing the amount of data processing required for 3D inspection and significantly improving the overall inspection efficiency.

[0047] Please refer to Figure 4 Step S6: Obtain the three-dimensional topographic data of the abnormal region based on the second coordinate information, including: S61. Based on the second coordinate information, control the three-dimensional detection module to move to the abnormal area where the abnormality still exists; S62. Perform a three-dimensional scan on the abnormal area to obtain a three-dimensional topographic image of the abnormal area; S63. Extract the three-dimensional geometric features of the abnormal region from the three-dimensional topography image, and use the three-dimensional geometric features as the three-dimensional topography data.

[0048] In some embodiments of the present invention, the three-dimensional scanning of the anomalous region includes: using a three-dimensional camera to capture images of the anomalous region from multiple perspectives, acquiring at least two three-dimensional images of the anomalous region from different perspectives, and generating a three-dimensional topographic image of the anomalous region using a multi-view three-dimensional reconstruction algorithm. The multi-view three-dimensional reconstruction algorithm includes, but is not limited to, any one of: stereo matching algorithm, structured light three-dimensional reconstruction algorithm, and time-of-flight three-dimensional reconstruction algorithm. This embodiment uses a multi-view three-dimensional reconstruction method to acquire three-dimensional topographic images, enabling the acquisition of surface information of the anomalous region from multiple angles, resulting in high reconstruction accuracy and suitability for fine detection of micron- and submicron-level anomalous regions.

[0049] In another embodiment, the three-dimensional scanning of the anomalous region includes: using a laser confocal microscope to scan the anomalous region layer by layer, acquiring sequential images of the anomalous region on different focal planes, and generating the three-dimensional topography image through focal plane superposition and height extraction algorithms. The advantage of this method is its extremely high vertical resolution, making it suitable for detecting anomalous regions with nanometer-level height undulations. In another embodiment, the three-dimensional scanning of the anomalous region includes: using a white light interferometer to perform interferometry on the anomalous region, acquiring interference fringe images of the anomalous region, and generating the three-dimensional topography image by performing phase calculation on the interference fringe images. The vertical resolution of the white light interferometer can reach 0.1 nm, making it suitable for detecting wafer surfaces with atomic-level flatness.

[0050] In another embodiment, the three-dimensional scanning of the abnormal region includes: using an atomic force microscope to perform probe scanning on the abnormal region, obtaining the probe deflection amount at each point on the surface of the abnormal region, and generating the three-dimensional topography image based on the probe deflection amount.

[0051] Atomic force microscopy has atomic-level resolution in all three dimensions, making it one of the highest-resolution methods for surface morphology detection.

[0052] Those skilled in the art can select a suitable three-dimensional imaging method based on actual detection accuracy requirements, detection speed requirements, equipment costs, and sample characteristics.

[0053] Those skilled in the art can select a suitable 3D imaging method based on actual detection accuracy requirements, detection speed requirements, equipment costs, and sample characteristics.

[0054] Please refer to Figure 5 In some embodiments of the present invention, extracting the three-dimensional geometric features of the abnormal region from the three-dimensional topography image and using the three-dimensional geometric features as the three-dimensional topography data includes: S631, Define the reference plane of the abnormal region as... The highest or lowest point of the abnormal region is defined as... ; S632, the highest or lowest point of the abnormal region. Relative to the reference plane The height extreme value H is taken as the first parameter; S633. The distribution dimension D of the abnormal region in the direction parallel to the reference plane is taken as the second parameter; S634. Based on the ratio of the height extreme value H to the spread size D, determine the steepness index R of the abnormal region, R=|H| / D; S635, the height to extreme value H, the spread size D, and the steepness index The symbol R is used as the three-dimensional geometric shape feature.

[0055] like Figure 6 As shown, the reference plane of the abnormal region is defined as... This includes: performing planar fitting on the normal regions surrounding the abnormal regions in the three-dimensional topography image, and determining the fitted plane as the reference plane. The normal region is the region in the three-dimensional topography image other than the abnormal region, or it is the neighborhood region within a preset range around the abnormal region.

[0056] The reason why this application uses the reference plane The technique of fitting a plane around the anomalous region, rather than directly using the theoretical horizontal plane of the wafer, offers several advantages. Semiconductor wafers are not perfectly flat; during manufacturing, transportation, and processing, their surfaces may exhibit macroscopic morphological changes such as overall warping, local bending, or uneven thickness. If the theoretical horizontal plane of the wafer is directly used as a reference, the wafer's own warping and bending will be mistakenly included in the height extremum H of the anomalous region, leading to a systematic bias in the measurement of H. By fitting a local plane around the normal region surrounding the anomalous region, the macroscopic undulations of the wafer itself can be effectively decoupled from the microscopic morphology of the anomalous region. This allows the height extremum H to accurately reflect the degree of protrusion or depression of the anomalous region relative to its local background, thereby significantly improving the accuracy of subsequent anomaly type identification.

[0057] In some embodiments of the present invention, the selection method of the normal region includes: forming an annular region by extending a preset pixel distance outward from the centroid of the abnormal region, and using the annular region as the normal region for planar fitting. This method can ensure the reference plane. The fitting is not affected by the morphology of the abnormal region itself. In another embodiment, the normal region may also be selected as an annular neighborhood with a preset width (e.g., 5 μm to 20 μm) on the periphery of the abnormal region, which is more applicable to scenarios where abnormal regions are relatively densely distributed.

[0058] In some alternative embodiments of the present invention, the reference surface is not limited to a flat surface, and may also be a curved surface obtained by fitting the local topography around the abnormal region. For example, when the surface of the semiconductor wafer itself has a certain bend or warp, the reference surface obtained by fitting a quadric surface or a higher-order curved surface can more accurately reflect the local reference morphology, and is particularly suitable for anomaly detection scenarios at the edge region of the wafer.

[0059] In some embodiments of the present invention, the defining the highest point or the lowest point of the abnormal region as comprises: traversing the height values pixel by pixel within the abnormal region of the three-dimensional topographic image, and determining the pixel point with the maximum height value as the highest point of the abnormal region ; or determining the pixel point with the minimum height value as the lowest point of the abnormal region .

[0060] In some embodiments of the present invention, if the abnormal region is in a raised morphology, the highest point thereof is extracted with respect to the reference surface as a height-direction extreme value H (H is a positive value); if the abnormal region is in a recessed morphology, the lowest point thereof is extracted with respect to the reference surface as a depth-direction extreme value H (H is a negative value). That is, the first parameter H may be a positive value (for a protrusion) or a negative value (for a depression), the sign thereof is used to characterize the convex-concave property of the abnormal region, and the absolute value thereof is used to characterize the magnitude of the height or depth of the abnormal region. By uniformly expressing both protrusions and depressions with the height-direction extreme value H, the subsequent calculation of the steepness index R has a unified mathematical form for both the protrusion and depression cases, which simplifies the input feature processing of the classification model.

[0061] In some embodiments of the present invention, as Figure 6 and Figure 7 shown, the spread dimension D is determined in the following manner: a height cross-section is established at a preset height offset ΔH above or below the reference surface , where the height offset ΔH = k×H, k being a preset proportionality coefficient, 0 < k < 1; the contour of the abnormal region at the height cross-section is intercepted; the maximum span between any two points on the cross-sectional contour is calculated, and the maximum span is taken as the spread dimension D.

[0062] The reason why the embodiments of the present invention adopt the height section is Instead of directly using the abnormal region on the reference plane, the layout dimension D is determined by extracting the outline of the abnormal region and calculating its maximum span. The technical principle behind projecting the profile dimensions on the reference plane lies in the fact that the wafer surface itself is not an ideal plane, exhibiting macroscopic warping and microscopic roughness. If the abnormal region is directly projected onto the reference plane... The projection profile on the reference plane determines the display size D, then the reference plane The inherent undulations of the wafer itself cause significant fluctuations in the edge position of the projected profile, resulting in a substantial deviation of the measured value of D from the true distribution scale of the anomaly region. This is especially true for shallow anomalies with low heights, where the surface morphology is on a similar order of magnitude to the roughness of the wafer itself, and the reference plane... Even minute fluctuations can lead to drastic changes in the projected profile, causing the display dimension D to lose repeatability and comparability.

[0063] For anomalous regions with complex three-dimensional morphology, such as scratches with trailing edges or irregularly shaped aggregates, their position on the reference surface... The projected contour on the map typically contains numerous fragmented branches and irregular edges, making it difficult to accurately define its extent. If the span of the projected contour is directly measured, different operators or different algorithms may obtain significantly different D values, which seriously affects the consistency and stability of classification features.

[0064] More importantly, for anomalous regions of particulate matter, their typical morphology is a three-dimensional body that is approximately hemispherical or conical, with a height cross-section... The outline at the point presents a clear, continuous, and regularly defined closed shape; while for physical defect types, such as shallow scratches and lattice damage, the outline at the height section... The contours at these locations often fail to form closed figures, or their shapes are extremely elongated and dispersed. Therefore, a height section is used. Using the maximum span of the outline at a certain point as the display dimension D can effectively filter out the reference plane. The interference caused by the undulations of the upper wafer itself can be accurately located to pinpoint the effective distribution range of the abnormal region. Furthermore, the integrity and closure of the cross-sectional profile can provide crucial geometric feature inputs for the classification model to determine the anomaly type in subsequent steps. Due to the height of the cross-section... Set away from the reference plane At this location, the influence of the wafer's own roughness has been greatly attenuated, while the morphological information of the abnormal region is clearly preserved. This effectively decouples the wafer's own micro-undulations from the true scale of the abnormal region, significantly improving the measurement accuracy and repeatability of the second parameter D.

[0065] In some embodiments of the present invention, the specific value of the height offset ΔH can be optimized and adjusted according to the actual classification accuracy requirements. When a larger value is taken (e.g., 60%H or 70%H), the height cross-section... By getting closer to the top of the anomalous region, the outermost contour of the particulate matter can be obtained, resulting in a more accurate measurement of the effective distribution range of the particulate matter; when taking smaller values ​​(such as 30%H or 40%H), the height section... Being closer to the reference plane ensures that more abnormal areas obtain effective cross-sectional profiles. In practical applications, the optimal k value can be determined through experimental calibration based on the semiconductor wafer's process node and surface roughness requirements.

[0066] In another embodiment, the height offset ΔH is a fixed value that does not change with variations in the height extreme value H. For example, the height offset ΔH can be set to a fixed height value, such as 1 micrometer, 3 micrometers, or 5 micrometers, depending on the process precision requirements of the semiconductor wafer. This approach is suitable for scenarios where the measurement of the height extreme value H itself has significant uncertainties, and using a fixed height section ensures the comparability of the spread dimension D across different abnormal regions.

[0067] In some embodiments of the present invention, the abnormal region is intercepted at the height section. The profile at the location is used to calculate the maximum span between any two points on the cross-sectional profile, including: The height section The anomaly region is obtained by intersecting its three-dimensional topography with the cross section at the height of the cross section. The cross-sectional profile is defined; any two points are taken on the cross-sectional profile, and the distance between the two points is calculated; the maximum value among all calculated distances between two points is determined as the spreading dimension D. In some embodiments of the present invention, the maximum span between any two points on the cross-sectional profile is the Freette diameter of the cross-sectional profile.

[0068] Using the Freette diameter as a measure of the spread dimension D ensures that the value of D is not affected by the shape of the cross-sectional profile, and has a unified mathematical definition and calculation standard.

[0069] In some embodiments of the present invention, the ratio of the height extreme value H to the spread size D constitutes the steepness index R of the anomalous region, i.e., R = |H| / D. The steepness index R is used to characterize the morphological steepness of the anomalous region in the direction perpendicular to the wafer surface. The larger the R value, the steeper the anomalous region; the smaller the R value, the gentler the anomalous region.

[0070] The reason why the steepness index R is constructed as the core parameter of the three-dimensional geometric morphology of the abnormal region in this embodiment of the invention is that the morphological distribution of different types of abnormal regions in three-dimensional space is fundamentally different. For micro-dust particles, their attachment morphology on the wafer surface is usually manifested as a local protrusion structure with obvious height protrusions. The height extreme value H is relatively large compared with its distribution size D, so the steepness index R is large.

[0071] For physical defects (such as scratches), the typical morphology is a shallow and long groove. The height extreme value H (i.e., scratch depth) is usually small, while the spread size D (i.e., scratch length) is usually large, resulting in a small steepness index R. For edge chipping defects, although they may have a large depth, their spread size D is also correspondingly large, and the R value is usually between that of particles and scratches. Therefore, the steepness index R alone can be used to effectively classify most abnormal areas initially, providing the most discriminative geometric feature for the input of subsequent classification models.

[0072] In this embodiment of the invention, the steepness index R is not only used to characterize the geometry of abnormal regions, but also directly used for accurate determination of the anomaly type. According to the present invention, the essential difference between dust and physical defects in three-dimensional morphology lies in the following: dust is an additional object grown on the wafer reference plane, with a clear interface between it and the wafer surface, thus possessing a steep edge gradient E and a large steepness R; while physical defects are material regions "missing" or "deformed" above or below the wafer reference plane, exhibiting a gradual transition between them and the surrounding normal wafer surface, thus possessing a gentle edge gradient E and a smaller steepness R. Based on this essential physical difference, by setting reasonable steepness and edge gradient thresholds, accurate differentiation between particulate matter and physical defects can be achieved, avoiding misjudgments caused by similar grayscale in traditional 2D detection.

[0073] Furthermore, in some embodiments of the present invention, in addition to extracting the height extreme value H, the distribution size D, and the steepness index R, the volume V, projected area A, edge gradient E, and curvature features of the anomalous region are also extracted simultaneously. The volume V is obtained by integrating the anomalous region along the height direction and is used to characterize the overall scale of the anomalous region; the projected area A is the area of ​​the anomalous region on the reference plane. The projected area on the wafer surface is used to characterize the coverage of the anomalous region on the wafer surface; the edge gradient E is the height change rate at the edge of the anomalous region, used to characterize the steepness of the anomalous region boundary; the curvature feature is the principal curvature value at each point on the surface of the anomalous region, used to characterize the surface bending characteristics of the anomalous region. The fusion of the above multidimensional geometric features provides richer discriminative information for the classification model, further improving the accuracy of anomalous type discrimination.

[0074] Step S7: Determine the anomaly type of the abnormal region based on the three-dimensional topography data, and generate a corresponding processing strategy for the semiconductor wafer.

[0075] In some embodiments of the present invention, determining the anomaly type of the abnormal region based on the three-dimensional topography data includes: inputting the three-dimensional topography data into a preset classification model; if the abnormal region is a convex shape relative to a reference surface, and its steepness index is greater than a preset threshold, and the edge gradient conforms to the steepness feature, then it is determined to be an attachment type; if the abnormal region is a concave shape, or although it is a convex shape, its steepness index is less than a preset threshold, then it is determined to be a defect type.

[0076] In some embodiments of the present invention, the classification model is a classification model built based on machine learning algorithms, including but not limited to: support vector machines, random forests, gradient boosting trees, and convolutional neural networks. The classification model is pre-trained through the following steps: obtaining a training dataset, which includes morphological feature samples of anomalous regions on the wafer surface labeled with anomaly types; inputting the training dataset into an initial classification model for iterative training, adjusting model parameters until the model reaches a preset classification accuracy. In this embodiment, the random forest model can handle high-dimensional features and is not prone to overfitting, making it suitable as a classification model.

[0077] The specific operating logic is as follows: when the control component inputs the above-mentioned three-dimensional geometric features into the preset decision classification model, if the height extreme value H of the abnormal area is greater than 0 (protruding shape), and the steepness index R is greater than the preset threshold (indicating that the object's facade is steep and protrudes upwards), and the edge gradient E is greater than the preset edge threshold and the contour compactness conforms to the granular geometric features, then the classification model determines the abnormal area as a type of surface attachment that can be removed, such as dust.

[0078] If the height extreme value H of the abnormal region is less than 0 (concave morphology), or although it is a convex morphology (H>0) but its steepness R is less than the preset threshold, it indicates that it presents a gently rising or large-area flattening feature, or manifests as a long strip of deep groove, then the classification model determines that the abnormal region is abnormal. These are irreversible defect types, such as scratches, chipping, pits, and lattice defects.

[0079] This application uses the following criteria to determine whether a particle is a protrusion, has a steepness index exceeding a threshold, and exhibits a characteristic edge gradient: these are all criteria for classifying a particle as an adhering object. The adhering of microparticles to the wafer surface is considered an accumulation of foreign matter. There is a clear physical interface between the particle's base and the wafer surface; therefore, the particle's three-dimensional morphology appears as a sudden bulge from the wafer's reference plane, with a large edge gradient E (i.e., a drastic height change at the edge) and a large steepness index R (i.e., a large height extreme relative to the spread size). In contrast, physical defects, such as scratches, are structural deficiencies in the wafer's own material. The bottom of the scratch exhibits a gradual arc-shaped transition with the surrounding normal wafer surface, resulting in a smaller edge gradient E and a smaller steepness index R. Furthermore, for some special defect types, such as protrusion-type defects like lattice protrusions and epitaxial growth anomalies, although they exhibit a protruding morphology, they are continuous with the wafer body material and lack a clear interface. Therefore, their edge gradient E and steepness index R are small, and they will not be misclassified as adhering objects. Therefore, by comprehensively judging the characteristics of three dimensions—convexity / concavity, steepness R, and edge gradient E—it is possible to achieve high-precision differentiation between attachment types and defect types.

[0080] In some embodiments of the present invention, the input features of the classification model include at least one or more combinations of the following three-dimensional geometric feature parameters: height extrema H, spread size D, steepness index R, volume V, projected area A, edge gradient E, and curvature feature. In this embodiment, after inputting the above-mentioned multi-dimensional feature parameters into the classification model, the classification model outputs anomaly type labels for the abnormal regions based on pre-trained classification decision boundaries. By fusing the above-mentioned multi-dimensional features, the accuracy and robustness of anomaly type discrimination can be further improved.

[0081] It should be noted that those skilled in the art will understand that the specific values ​​of the aforementioned preset threshold and preset edge threshold can be adjusted according to the process requirements of different substrate materials or different process nodes.

[0082] To verify the technical effectiveness of the wafer dust identification and removal scheme based on preset thresholds and preset edge thresholds described in this application, comparative experiments were conducted using 12-inch unpatterned silicon wafers and patterned wafers. Dust particle size range: standard polystyrene microspheres with a test particle size of 0.1μm to 20μm and metallic foreign objects from the actual production environment.

[0083] Edge region definition: The wafer edge chamfer is defined as a ring-shaped region from 0mm to 3.0mm from the outermost edge of the wafer. Comparative Example 1 is a traditional two-dimensional inspection and fixed dust removal; Comparative Example 2 is a three-dimensional topography inspection but only uses the height extreme value H to determine the type, without using the preset steepness threshold and edge gradient threshold. The embodiments of this application adopt three-dimensional topography feature extraction, joint determination of preset steepness threshold and edge gradient threshold, and a multi-level weighted hierarchical strategy. The experimental data comparison and analysis table is as follows: Test metrics Comparative Example 1 Comparative Example 2 Examples of this application Dust Recognition Accuracy (%) 82.4% 91.5% 98.7% False alarm rate (%) of chamfered edge area 34.2% 28.6% 0.3% Dust removal rate (%) 76.5% 79.1% 88.2% wafer edge damage 1.8% 1.5% 0.1% Average time for single-chip detection and processing (s) 1.8s 2.1s 1.6s

[0084] Based on the experimental data above, the following conclusions can be drawn: In this embodiment, by simultaneously employing preset steepness thresholds and edge gradient thresholds for anomaly type determination, the micro-dust recognition accuracy reaches 98.7%, significantly higher than Comparative Example 1 (82.4%) and Comparative Example 2 (91.5%); the false alarm rate in the chamfered edge area is reduced to 0.3%, far lower than Comparative Example 1 (34.2%) and Comparative Example 2 (28.6%). This indicates that the preset edge threshold has a significant effect on reducing the false alarm rate. Simultaneously, due to the reduction in the false alarm rate, the average detection and processing time per piece is also correspondingly shortened, decreasing from 2.1s in Comparative Example 2 to 1.6s.

[0085] The particulate matter adhesion type refers to the abnormal area being formed by micro-dust particles from the external environment adhering to the wafer surface. The defect type includes physical damage type and structural damage type. The physical damage type refers to the abnormal area being formed by structural damage to the wafer surface material itself. The structural damage includes at least one of scratches, pits, cracks, and edge chipping.

[0086] The step of determining the anomaly type of the abnormal region based on the three-dimensional topographic data further includes: when the confidence level output by the discrimination model is lower than a preset confidence threshold, a manual review process is triggered, whereby an operator manually determines the type of the abnormal region. This mechanism, as a supplement to automatic determination, can effectively reduce the risk of misjudgment caused by model uncertainty and improve the overall reliability of the determination.

[0087] In some embodiments of the present invention, the processing strategy for generating the semiconductor wafer includes: if the abnormality type of the abnormal area is determined to be an attachment type, a dust removal execution strategy is generated; the dust removal execution strategy includes at least one of the following: determining the dust removal method according to the morphological characteristics of the attachment type, determining the dust removal parameters according to the size of the attachment type, and driving the dust removal component to move to the corresponding position to perform a secondary dust removal action according to the second coordinate information of the attachment type.

[0088] In some embodiments of the present invention, the processing strategy for generating the semiconductor wafer further includes: selecting a corresponding evaluation index calculation method based on the anomaly type. If the anomaly area is determined to be of the attachment type, a contamination evaluation index is calculated to evaluate the degree of impact of the anomaly type on the quality of the semiconductor wafer; if the anomaly area is determined to be of the defect type, a defect evaluation index is calculated to evaluate the degree of impact of the anomaly type on the quality of the semiconductor wafer.

[0089] In one embodiment, calculating the pollution assessment index and the defect assessment index includes: the pollution assessment index is calculated by weighted combination of the geometric volume of a single attachment type, the total number of attachment types on a single wafer, the effective area of ​​the semiconductor wafer, the total coverage area of ​​attachment types, the local maximum attachment type density, and the directional distribution index of attachment types. The specific calculation formula is as follows:

[0090] in, and The first The height value of each dust particle and the diameter of the bottom median line, where N is the total number of attachment types on a single wafer. This represents the total number of dust particles. For the effective area of ​​a semiconductor wafer, The total coverage area of ​​the attachment types. This represents the local maximum density of attachment types. This is a directional distribution index for the type of attachment. , , , , These are the preset weighting coefficients.

[0091] In one embodiment, the defect evaluation index is calculated by a weighted combination of the defect depth or height, defect opening diameter, total defect volume, total number of defects, effective wafer area, and total defect area. The specific calculation formula is as follows:

[0092] in, For the first The depth or height value of a defect. For the first The opening diameter of a defect, M is the total volume of the defect. This represents the total number of defects. For the effective area of ​​the wafer, β1, β2, and β3 represent the total defect area, and β1, β2, and β3 are preset weighting coefficients.

[0093] In some embodiments of the present invention, determining the degree of impact of the abnormal region on the quality of the semiconductor wafer based on the abnormality type further includes: when the abnormal region simultaneously exhibits particulate matter adhesion type abnormalities and defect type abnormalities, weighting and combining the contamination evaluation index and the defect evaluation index to obtain a comprehensive quality impact evaluation index. The weighting coefficients of the weighted combination can be set according to the specific application scenario and process requirements of the semiconductor wafer.

[0094] In some embodiments of the present invention, generating the quality grade of the semiconductor wafer and the corresponding handling strategy based on the degree of quality impact includes: If the defect evaluation index Reach the preset first value The semiconductor wafer is directly classified as scrap, defined as grade D, and subsequent dust removal operations are terminated; if the defect evaluation index... Reaching the second value But lower than the first value , It was determined to be a re-inspection level, and defined as a Class C re-inspection. When the defect evaluation index is lower than the second value Verify the pollution assessment indicators If the pollution assessment indicators Exceeding the set dust removal threshold If the dust level is determined to be below the set dust removal threshold, it is defined as Class B forced dust removal, and the dust removal components are driven to perform dust removal operations; If the result is deemed acceptable and allowed to proceed to the next process, it is defined as Grade A Qualified.

[0095] This solution prioritizes intercepting Class D wafers with fatal damage and terminates meaningless dust removal operations, reducing wasted labor time. At the same time, it combines the size, total number, distribution density, and directional aggregation characteristics of the attached materials to perform weighted scoring, achieving a refined flow of Class A qualified, Class B forced dust removal, Class C re-inspection, and Class D scrap, which greatly improves the production line collaboration efficiency and yield management level.

[0096] In some embodiments of the present invention, the disposal strategy is associated with the quality level.

[0097] For example: When the quality grade of the semiconductor wafer is Grade A (qualified), the handling strategy is as follows: The semiconductor wafers were marked as qualified and released. When the quality grade of the semiconductor wafer is B-level forced dust removal or C-level re-inspection. The handling strategy is to perform dust removal or manual inspection and repair on the semiconductor wafer; When the semiconductor wafer is classified as Grade D and deemed scrap, the disposal strategy is as follows: The semiconductor wafers are marked as scrap and recycled.

[0098] In some embodiments of the present invention, the handling strategy further includes: generating an abnormal region distribution map based on the position coordinates of the abnormal region on the semiconductor wafer, and outputting the abnormal region distribution map together with the quality grade. The abnormal region distribution map can be used to guide subsequent process adjustments and quality improvements.

[0099] Please refer to Figure 8 This application provides a semiconductor dust removal control system, comprising: a first two-dimensional detection module for acquiring first two-dimensional morphological data of the surface of a semiconductor wafer; a first positioning module for determining first coordinate information of at least one abnormal region of the semiconductor wafer based on the first two-dimensional morphological data; a dust removal module for removing dust from at least one of the abnormal regions based on the first coordinate information; a second two-dimensional detection module for acquiring second two-dimensional morphological data of the surface of the semiconductor wafer after dust removal; a second positioning module for determining second coordinate information of the abnormal region based on the second two-dimensional morphological data; a three-dimensional detection module for acquiring three-dimensional topographic data of the abnormal region based on the second coordinate information; and a classification module for determining the abnormal type of the abnormal region based on the three-dimensional topographic data and generating a corresponding treatment strategy for the semiconductor wafer.

[0100] The dust removal control system integrates two-dimensional detection, positioning, dust removal, three-dimensional detection, and classification decision-making functions into a single control system. This achieves fully automated closed-loop control of the entire process, from two-dimensional coarse screening and positioning, to fixed-point dust removal, two-dimensional re-inspection and verification, condition-triggered three-dimensional fine inspection, and differentiated treatment. The first two-dimensional detection module and the first positioning module are responsible for quickly locating abnormal areas. The dust removal module performs initial cleaning. The second two-dimensional detection module and the second positioning module are responsible for verifying the cleaning effect and screening out areas that still need to be treated. The three-dimensional detection module only performs fixed-point fine inspection on areas that have not met the cleaning standards. The classification module accurately distinguishes dust from defects based on three-dimensional geometric features and generates differentiated treatment strategies. The modules are closely connected through data flow and control commands, completing the entire process from wafer detection to cleaning verification and classification treatment without manual intervention.

[0101] Please refer to Figure 9This application provides a semiconductor dust removal device in three aspects, including: a device frame assembly 1; a motion assembly 2 disposed on the device frame assembly; a support platform 3 disposed on the motion assembly 2 for fixing and supporting a semiconductor wafer 10; a detection assembly 5 disposed on the motion assembly 2 for scanning the surface of the semiconductor wafer to collect first two-dimensional morphological data, second two-dimensional morphological data, and three-dimensional topographic data; a dust removal assembly 4 disposed on the motion assembly 2 for cleaning dust from the surface of the semiconductor wafer 10; and a dust removal control system 6 connected to the detection assembly 5, the motion assembly 2, and the dust removal assembly 4.

[0102] The dust removal equipment integrates detection, dust removal, and hierarchical decision-making functions into a single platform, achieving a fully integrated operation from two-dimensional initial inspection and positioning to fixed-point dust removal execution, two-dimensional re-inspection verification, condition-triggered three-dimensional fine inspection, and classification and disposal. Driven by the motion component 2, the carrier platform 3 moves the semiconductor wafer 10 sequentially to the inspection station and the dust removal station, eliminating the need to transfer wafers between different devices and significantly reducing wafer transfer time and the risk of secondary contamination between processes. The detection component 5 can integrate a two-dimensional camera and a three-dimensional sensor into the same component, performing two-dimensional and three-dimensional scanning tasks in a time-sharing manner under the scheduling of the control system. The dust removal control system 6, as the core decision-making unit of the equipment, coordinates the scanning actions of the detection component 5, the movement and cleaning actions of the dust removal component 4, and the output of anomaly type judgment and disposal strategies, ensuring seamless connection between processes. The equipment provided in this application has a compact structure and high functional integration, effectively reducing the equipment footprint and manufacturing cost. It is suitable for scenarios involving online inspection and cleaning of wafer surface quality in semiconductor manufacturing production lines. Multiple units can be configured to operate in parallel according to production line capacity requirements, exhibiting good scalability and adaptability.

[0103] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A semiconductor dust removal method, characterized in that, include: Acquire first two-dimensional morphological data of the semiconductor wafer surface; Based on the first two-dimensional morphological data, determine the first coordinate information of at least one abnormal region of the semiconductor wafer; Based on the first coordinate information, at least one of the abnormal areas is dusted, and the dust removal is carried out using a sticky rod contact cleaning method. Obtain second two-dimensional morphological data of the semiconductor wafer surface after dust removal; Determining the second coordinate information of the abnormal region based on the second two-dimensional morphological data includes: comparing the second two-dimensional morphological data with the reference morphological data to determine whether the abnormal region still has an anomaly; if the abnormal region still has an anomaly after dust removal, extracting the outline of the abnormal region that still has an anomaly, and determining the second coordinate information based on the position of the outline on the semiconductor wafer surface; if the abnormal region does not have an anomaly after dust removal, marking the abnormal region as a clean region and terminating the subsequent processing of the abnormal region. Based on the second coordinate information, obtain the three-dimensional topographic data of the abnormal region; Based on the three-dimensional topography data, the anomaly type of the abnormal region is determined, and a corresponding treatment strategy is generated for the semiconductor wafer.

2. The semiconductor dust removal method according to claim 1, characterized in that, Based on the first two-dimensional morphological data, determining the first coordinate information of at least one abnormal region of the semiconductor wafer includes: The control module performs a panoramic scan of the surface of the semiconductor wafer to obtain a first two-dimensional image of the semiconductor wafer. The first two-dimensional image is divided into blocks, and the image features of each block region are extracted. The image features of each segmented region are compared with a preset abnormal image feature threshold, and the segmented regions whose image features exceed the preset abnormal image feature threshold are marked as suspected abnormal regions. Extract the pixel coordinates of the suspected abnormal region in the first two-dimensional image; According to the calibration parameters of the two-dimensional detection module, the pixel coordinates are mapped to the physical coordinates of the suspected abnormal area on the semiconductor wafer surface, and the physical coordinates are used as the first coordinate information; The suspected abnormal area is identified as the abnormal area.

3. The semiconductor dust removal method according to claim 2, characterized in that, The calibration parameters of the two-dimensional detection module include at least one of the following: the intrinsic parameter matrix, extrinsic parameter matrix, distortion coefficient, and pixel equivalent of the two-dimensional detection module; The intrinsic parameter matrix describes the camera's internal optical characteristics, such as focal length, principal point coordinates, and pixel size; the extrinsic parameter matrix describes the camera's spatial pose relative to the support platform or wafer surface; the distortion coefficient corrects image distortion caused by the camera lens; and the pixel equivalent converts pixel distances in the image into physical distances. By using the above calibration parameters, a mapping relationship can be established from the coordinates of any pixel in the first two-dimensional image to the physical coordinates on the surface of the semiconductor wafer, thereby achieving precise location of abnormal areas.

4. The semiconductor dust removal method according to claim 3, characterized in that, Based on the first two-dimensional morphological data, determining the first coordinate information of at least one abnormal region of the semiconductor wafer further includes: The abnormal region on the surface of the semiconductor wafer is identified by comparing the first two-dimensional morphological data with the reference morphological data. The identified abnormal regions are subjected to morphological processing, which includes at least one of dilation, erosion, opening, and closing operations, to eliminate isolated noise points and connect adjacent abnormal pixel regions to obtain a complete abnormal region contour. The first coordinate information is determined based on the position of the abnormal region contour on the semiconductor wafer surface.

5. The semiconductor dust removal method according to claim 1, characterized in that, Obtaining second two-dimensional morphological data of the semiconductor wafer surface after dust removal includes: The control module performs a second two-dimensional scan on the surface of the semiconductor wafer after dust removal to obtain a second two-dimensional image; or Based on the first coordinate information, the two-dimensional detection module is controlled to move to the abnormal area to perform local scanning and obtain a local two-dimensional image of the abnormal area.

6. The semiconductor dust removal method according to claim 1, characterized in that, Based on the second coordinate information, the three-dimensional topographic data of the abnormal region is obtained, including: Based on the second coordinate information, the three-dimensional detection module is controlled to move to the abnormal area where the abnormality still exists; The abnormal region is scanned in three dimensions to obtain a three-dimensional topographic image of the abnormal region. The three-dimensional geometric features of the abnormal region are extracted from the three-dimensional topography image, and the three-dimensional geometric features are used as the three-dimensional topography data.

7. The semiconductor dust removal method according to claim 6, characterized in that, Extracting the three-dimensional geometric features of the abnormal region from the three-dimensional topography image, and using the three-dimensional geometric features as the three-dimensional topography data, includes: The reference plane of the abnormal region is defined as O1, and the highest or lowest point of the abnormal region is defined as O2. The height extreme value H of the highest or lowest point O2 of the abnormal region relative to the reference plane O1 is used as the first parameter. The distribution dimension D of the abnormal region in the direction parallel to the reference plane is used as the second parameter; The steepness index R of the abnormal region is determined based on the ratio of the height extreme value H to the spread size D, where R = |H| / D; The height extreme value H, the spread size D, and the steepness index R are used as the three-dimensional geometric features.

8. The semiconductor dust removal method according to claim 7, characterized in that, The fabric size D is determined in the following way: A height section Z1 is established at a preset height offset △H above or below the reference plane, wherein the height offset △H = k × H, and k is a preset proportionality coefficient, 0. <k<1; The outline of the abnormal region at the height section Z1 is extracted; Calculate the maximum span between any two points on the contour at the height section, and use the maximum span as the display dimension D.

9. A semiconductor dust removal control system, characterized in that, include: The first two-dimensional detection module is used to acquire the first two-dimensional morphological data of the semiconductor wafer surface; The first positioning module is used to determine the first coordinate information of at least one abnormal region of the semiconductor wafer based on the first two-dimensional morphological data. A dust removal module is used to remove dust from at least one of the abnormal areas based on the first coordinate information; The second two-dimensional detection module is used to acquire the second two-dimensional morphological data of the semiconductor wafer surface after dust removal; The second positioning module is used to determine the second coordinate information of the abnormal area based on the second two-dimensional morphological data, including: comparing the second two-dimensional morphological data with the reference morphological data to determine whether the abnormal area still has an anomaly; if the abnormal area still has an anomaly after dust removal, extracting the outline of the abnormal area that still has an anomaly, and determining the second coordinate information based on the position of the outline on the semiconductor wafer surface; if the abnormal area does not have an anomaly after dust removal, marking the abnormal area as a clean area and terminating the subsequent processing of the abnormal area. A three-dimensional detection module is used to obtain the three-dimensional shape data of the abnormal area based on the second coordinate information; The classification module is used to determine the anomaly type of the abnormal region based on the three-dimensional topography data, and to generate a corresponding treatment strategy for the semiconductor wafer.

10. A semiconductor dust removal device, characterized in that, include: Equipment frame components; A motion component, which is disposed on the device frame assembly. A support platform, which is mounted on the motion component, is used to fix and support the semiconductor wafer; A detection component, which is disposed on the motion component, is used to scan the surface of the semiconductor wafer to acquire first two-dimensional morphological data, second two-dimensional morphological data and three-dimensional topography data; A dust removal component, which is disposed on the moving component, is used to clean dust from the surface of the semiconductor wafer; and The dust removal control system as described in claim 9 is connected to the detection component, the motion component, and the dust removal component, respectively.

11. A computer-readable storage medium, characterized in that, The device contains a computer program that, when executed by a processor, implements the semiconductor dust removal method as described in any one of claims 1 to 8.

Citation Information

Patent Citations

  • Grinding equipment and grinding method

    CN106680288A

  • Surface particle defect detection and removal method and computer readable storage medium

    CN119324158A