Wafer surface defect detection method and system based on artificial intelligence

Abstract

The invention provides a wafer surface defect detection method and system based on artificial intelligence, and the method comprises the steps: obtaining a first light source scheme, configuring a controllable light source unit, carrying out the image collection of a target wafer through combining with an image collection unit, and obtaining a first image set; according to the first image set, identifying a plurality of first defect points of the target wafer, and generating first defect distribution; generating a second light source scheme based on the first defect distribution, configuring the controllable light source unit again, and performing image acquisition on the target wafer again in combination with the image acquisition unit to obtain a second image set; and according to the second image set, identifying a plurality of second defect points of the target wafer, marking the defect type of each second defect point, and generating the surface defect distribution of the target wafer in combination with the first defect distribution. The technical problem that in the prior art, the detection rate and accuracy of wafer surface defects are poor is solved.

Description

Technical Field

[0001] The present invention relates to the field of image detection, and particularly to a method and system for detecting wafer surface defects based on artificial intelligence. Background Art

[0002] During the current wafer surface defect detection process, fixed lighting parameters are usually used to collect images of the wafer surface, and then defect recognition is performed based on the collected images. However, different types of defects are generated on different wafer surfaces, and it is difficult for fixed lighting parameters to capture all types of defects. For example, some defects are not obvious under a certain fixed lighting parameter, resulting in the collected images not being able to display the defects well, thereby leading to poor detection rate and accuracy of wafer surface defects. Summary of the Invention

[0003] In view of the technical problem of poor detection rate and accuracy of wafer surface defects in the prior art, the present invention provides a method and system for detecting wafer surface defects based on artificial intelligence.

[0004] The technical solutions of the present invention for solving the above technical problems are as follows: In a first aspect, the present invention provides a method for detecting wafer surface defects based on artificial intelligence, including: Obtaining a first light source scheme, configuring the controllable light source unit, and collecting images of the target wafer in combination with the image collection unit to obtain a first image set; Identifying a plurality of first defect points of the target wafer according to the first image set, and generating a first defect distribution; Generating a second light source scheme based on the first defect distribution, reconfiguring the controllable light source unit, and collecting images of the target wafer again in combination with the image collection unit to obtain a second image set; Identifying a plurality of second defect points of the target wafer according to the second image set, labeling the defect types of each of the second defect points, and generating the surface defect distribution of the target wafer in combination with the first defect distribution.

[0005] In a second aspect, the present invention provides a system for detecting wafer surface defects based on artificial intelligence, including: An initial light source module, configured to obtain a first light source scheme, configure the controllable light source unit, and collect images of the target wafer in combination with the image collection unit to obtain a first image set; A defect analysis module, configured to identify a plurality of first defect points of the target wafer according to the first image set, and generate a first defect distribution; An optimized light source module is used to generate a second light source scheme based on the first defect distribution, reconfigure the controllable light source unit, and combine the image acquisition unit to perform image acquisition on the target wafer again to obtain a second image set. An integrated output module is used to identify multiple second defect points of the target wafer according to the second image set, label the defect types of each second defect point, and combine the first defect distribution to generate the surface defect distribution of the target wafer.

[0006] The beneficial effects of the present invention are as follows: Compared with the prior art, the present application first obtains a first light source scheme, configures a controllable light source unit, and combines an image acquisition unit to perform image acquisition on a target wafer to obtain a first image set, realizing the intelligent reuse of historical data, improving the efficiency and accuracy of image acquisition, and providing necessary support for subsequent defect analysis. Secondly, according to the first image set, multiple first defect points of the target wafer are identified, a first defect distribution is generated, the distribution status of most defects on the wafer surface is obtained, and a necessary basis for obtaining the subsequent second defect distribution is provided. Thirdly, based on the first defect distribution, a second light source scheme is generated, the controllable light source unit is reconfigured, and the image acquisition unit is combined to perform image acquisition on the target wafer again to obtain a second image set, effectively identifying potential defects, which is a reliable supplement to the first defect distribution, and improving the accuracy and precision of defect detection. Finally, according to the second image set, multiple second defect points of the target wafer are identified, the defect types of each second defect point are labeled, and combined with the first defect distribution, the surface defect distribution of the target wafer is generated, obtaining the accurate surface defect distribution of the target wafer.

[0007] Through the above technical solutions, the present application dynamically matches the light configuration items from historical data according to the raw material characteristics and production process characteristics of the target wafer, and fully considers potential defects, significantly improving the detection rate and accuracy of wafer surface defect detection. Description of the Drawings

[0008] Figure 1 It is a schematic flow chart of the method for detecting wafer surface defects based on artificial intelligence provided by the present invention; Figure 2 It is a block diagram of the wafer defect recognition network in the method for detecting wafer surface defects based on artificial intelligence provided by the present invention; Figure 3 It is a schematic structural diagram of the system for detecting wafer surface defects based on artificial intelligence provided by the present invention.

[0009] In the drawings, the components represented by each reference numeral are as follows: Initial light source module 11, defect analysis module 12, optimized light source module 13, integrated output module 14. Detailed implementation manners

[0010] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0011] In the description of the present invention, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0012] In the description of the present invention, the term "for example" is used to mean "serving as an example, illustration, or explanation". Any embodiment described as "for example" in the present invention is not necessarily construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the present invention. Details are set forth for the purpose of explanation in the following description. It should be understood that those of ordinary skill in the art can recognize that the present invention can be implemented without these specific details. In other instances, well-known structures and processes are not described in detail to avoid unnecessary details from obscuring the description of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.

[0013] Embodiment 1, as Figure 1 shown, the embodiment of the present invention provides a method for detecting defects on the surface of a wafer based on artificial intelligence, including: S10: Obtain a first light source scheme, configure the controllable light source unit, and combine the image acquisition unit to perform image acquisition on the target wafer to obtain a first image set; In the traditional process of detecting defects on the surface of a wafer, fixed lighting parameters are usually used to perform image acquisition on the surface of the wafer, and then defect recognition is performed based on the acquired images. However, different types of defects are generated on the surfaces of different wafers, and it is difficult for fixed lighting parameters to capture all types of defects, resulting in low accuracy of defect recognition.

[0014] In view of the above problems, according to the raw material characteristics and production process characteristics of the target wafer, this application retrieves multiple sample defect distributions, then obtains multiple historical light source schemes corresponding to the multiple sample defect distributions from the historical wafer production database, and then selects the illumination configuration items with a historical configuration frequency greater than the preset configuration frequency from the multiple historical light source schemes, aggregates them to form the first light source scheme, configures the controllable light source unit, and combines with the image acquisition unit to perform image acquisition on the target wafer to obtain the first image set.

[0015] Specifically, step S10 in the method includes: Obtain the raw material characteristics and production process characteristics of the target wafer; Based on the raw material characteristics and the production process characteristics, retrieve the historical wafer production database, screen out multiple sample wafers, and extract the sample defect distributions of the multiple sample wafers to obtain multiple sample defect distributions; Generate the first light source scheme for the target wafer according to the multiple sample defect distributions.

[0016] In the embodiment of this application, first, the raw material characteristics and production process characteristics of the target wafer are obtained. Among them, the obtained raw material characteristics include multi-dimensional feature vectors such as the wafer substrate material (such as single crystal silicon / polycrystalline silicon / compound semiconductor), doping type (such as N-type / P-type), crystal structure (such as crystal orientation, defect density), etc., which can reflect the inherent physical properties and optical characteristics of the wafer; the production process characteristics include multi-dimensional feature vectors such as lithography accuracy (such as 7nm process node), etching process (such as anisotropic / isotropic etching), thin film deposition (such as CVD / PVD process parameters), mechanical polishing (such as pressure, time), etc., which can reflect the potential defect types formed during the wafer manufacturing process. Exemplarily, the material feature vector of the obtained GaN wafer is [compound semiconductor, N-type doping, crystal orientation <0001>], and the production process feature vector of the 7nm FinFET wafer is [EUV lithography, copper interconnect, CMP pressure 5psi, low-k dielectric layer].

[0017] Secondly, based on the raw material characteristics and production process characteristics, retrieve the historical wafer production database, screen out multiple sample wafers, and extract the sample defect distributions of the multiple sample wafers to obtain multiple sample defect distributions. Among them, the historical wafer production database is established based on historical multi-dimensional data. Further, the retrieval process can adopt a hybrid metric method that combines cosine similarity and Euclidean distance for matching in the historical wafer production database: for the raw material characteristics (nominal data), calculate the semantic correlation using cosine similarity (e.g., the similarity between "silicon-based" and "GaN-based" is 0.3, and the similarity between "silicon-based" and "polycrystalline silicon" is 0.9); for the production process characteristics (numerical data), use the normalized Euclidean distance to measure the difference (e.g., when the CMP pressure difference > 2 psi, the similarity decays by 20%). In this way, 30 - 50 sample wafers with a similarity ≥ 0.85 (this threshold is determined based on a large number of sample data) are screened out. Exemplarily, the feature vector of the raw material is [Si-based material, N-type doping, 7nm EUV lithography, copper interconnect, CMP pressure 5 psi], and 42 samples with a similarity ≥ 0.85 are screened out from the historical wafer production database using the hybrid metric method that combines cosine similarity and Euclidean distance.

[0018] Furthermore, extract the sample defect distributions of the multiple sample wafers. Among them, the defect distribution includes the spatial distribution, type distribution, and depth distribution of the defects. Specifically, for each sample wafer, extract its three-dimensional defect distribution: 1. Spatial distribution: Establish a polar coordinate system with the wafer center as the origin (e.g., with an accuracy of ±1 μm), divide the wafer surface into a 500×500 grid (e.g., with a resolution of 100 μm), and calculate the defect density of each grid. 2. Type distribution: Based on morphological features (such as aspect ratio) and spectral response, classify the defect types into particle contamination (environmental dust, equipment wear), scratches (mechanical contact, abnormal polishing process), voids (chemical vapor deposition (CVD) process defects), etc., and then count the proportion of the number of each type of defect. For example, the particle contamination of a certain wafer accounts for 45%, the scratches account for 30%, and the voids account for 25%. 3. Depth distribution: Utilize the penetration characteristics of multi-wavelength light sources (such as 940nm infrared light penetrating 50μm of the silicon layer), combined with optical sectioning technology, to construct a defect depth probability distribution function (e.g., P(d≤10μm)=0.7 indicates that 70% of the defects are within the surface layer of 10μm). Finally, the DBSCAN clustering algorithm can be used to identify the defect dense areas and generate a multi-dimensional defect distribution including spatial density, type proportion, and depth characteristics. Exemplarily, extract the defect distribution data of multiple sample wafers respectively, and integrate them through the DBSCAN clustering algorithm as follows: spatial density (the proportion of defects in the edge area is 40%), defect type (particle contamination accounts for 45%, scratches account for 30%, voids account for 25%), depth distribution (the proportion of subsurface defects is 25%), forming a three-dimensional defect distribution, which reflects the possible defect distribution on the wafer surface under the current raw material characteristics and production process characteristics.

[0019] Thus, based on the raw material characteristics and production process characteristics of the target wafer, the historical wafer production database is retrieved to screen out multiple similar sample wafers. Then, by extracting the defect distributions of the multiple similar sample wafers, a bridge is built between the historical data and the real-time detection requirements, providing a necessary data basis for determining the light source scheme subsequently.

[0020] Finally, according to the defect distributions of multiple samples, a first light source scheme for the target wafer is generated. Specifically, multiple historical light source schemes corresponding to the defect distributions of multiple samples are obtained from the historical wafer production database. Each historical scheme sequence includes multiple illumination configuration items. Finally, a first light source scheme for the target wafer is generated according to the multiple historical light source schemes.

[0021] Specifically, the step of "generating the first light source scheme for the target wafer according to the defect distributions of the multiple samples" includes: Obtaining multiple historical light source schemes corresponding to the defect distributions of the multiple samples from the historical wafer production database. Each of the historical light source schemes includes multiple illumination configuration items. Each illumination configuration item includes the wavelength parameter, angle parameter, polarization parameter, and illumination mode parameter used by the controllable light source unit in a single image acquisition. Forming the first light source scheme according to the multiple historical light source schemes.

[0022] In the embodiment of the present application, first, multiple historical light source schemes corresponding to the defect distributions of multiple samples are obtained from the historical wafer production database. Among them, the light source scheme is a sequence. Each historical light source scheme includes multiple illumination configuration items. Each illumination configuration item includes the wavelength parameter (such as 250 - 1100 nm), angle parameter (such as incident angle 0 - 90°, azimuth angle 0 - 360°), polarization parameter (such as linear polarization, circular polarization), and illumination mode parameter (such as dark field / bright field / combination mode) used by the controllable light source unit in a single image acquisition. Exemplarily, the defect distributions of multiple samples are spatial density (defect ratio in the edge area is 40%), defect type (particle contamination accounts for 45%, scratch accounts for 30%, void accounts for 25%), and depth distribution (subsurface defect ratio is 25%). Accordingly, multiple corresponding historical light source schemes are obtained from the historical wafer production database. Each historical light source scheme includes multiple illumination configuration items. Among them, a certain illumination configuration item is: 532 nm wavelength (wavelength parameter), 45° incident angle (angle parameter), horizontal polarization (polarization parameter), and annular dark field illumination (illumination mode parameter). Thus, a mapping relationship between the defect distribution and the light source scheme is established.

[0023] Secondly, based on the multiple historical light source schemes, the first light source scheme is formed. Specifically, the union of the lighting configuration items in the multiple historical light source schemes is taken to obtain a lighting configuration set. Then, the lighting configuration set is traversed, and the historical configuration frequency of each lighting configuration item in the lighting configuration set is counted. Finally, the lighting configuration items with a historical configuration frequency greater than the preset configuration frequency are selected and summarized to form the first light source scheme. In this way, by screening the lighting configuration items with a historical configuration frequency greater than the preset configuration frequency in the multiple historical light source schemes, the first light source scheme is obtained, providing necessary support for the subsequent optimization of the light source scheme.

[0024] Specifically, the step of "forming the first light source scheme based on the multiple historical light source schemes" includes: Taking the union of the lighting configuration items in the multiple historical light source schemes to obtain a lighting configuration set; Traversing the lighting configuration set, sequentially extracting each lighting configuration item, and counting the historical configuration frequency of each lighting configuration item in the multiple historical light source schemes; Selecting the lighting configuration items with a historical configuration frequency greater than the preset configuration frequency and summarizing them to form the first light source scheme.

[0025] In the embodiments of the present application, first, the union of the lighting configuration items in the multiple historical light source schemes is taken to obtain a lighting configuration set. Exemplarily, the union of all the lighting configuration items in the multiple historical light source schemes is taken to obtain 100 lighting configuration sets, and the lighting configuration set reflects all the historical lighting configurations corresponding to the sample defect distribution.

[0026] Secondly, the lighting configuration set is traversed, and each lighting configuration item is sequentially extracted, and the historical configuration frequency of each lighting configuration item in the multiple historical light source schemes is counted. Among them, the historical configuration frequency of each lighting configuration item = the number of occurrences of this lighting configuration item / the total number of lighting configuration sets. Exemplarily, the lighting configuration set is 100. Among them, the lighting configuration item 1 (532nm, incident angle 45°, horizontal polarization, annular dark field) appears 55 times in the lighting configuration set, and the lighting configuration item 2 (940nm, incident angle 15°, circular polarization, backscattering illumination) appears 42 times in the lighting configuration set. Then, the historical configuration frequency of the lighting configuration item 1 = 55 / 100 = 0.55, and the historical configuration frequency of the lighting configuration item 2 = 42 / 100 = 0.42. The historical configuration frequency reflects the proportion of this lighting configuration item in the historical lighting scheme. The higher the historical configuration frequency, the better this lighting configuration item.

[0027] Finally, select the illumination configuration items with historical configuration frequencies greater than the preset configuration frequency, and summarize them to form the first light source scheme. Herein, the recommended preset configuration frequency in this application is 0.4, and those skilled in the art can dynamically adjust it according to the actual sample size. Further, select the illumination configuration items with historical configuration frequencies greater than the preset configuration frequency (such as 0.4), and summarize them to form the first light source scheme. Among them, the summarization process can be sorted in descending order according to the angle parameters of the illumination configuration items. In this way, the first light source scheme is generated. Exemplarily, the preset configuration frequency is 0.4. Based on this, it is screened out that the historical configuration frequency of illumination configuration item 1 (532nm, incident angle 45°, horizontal polarization, annular dark field) is 0.55, and the historical configuration frequency of illumination configuration item 2 (940nm, incident angle 15°, circular polarization, backscattering illumination) is 0.42. Then, the first light source scheme is obtained by sorting in descending order according to the angle parameters: illumination configuration item 1, illumination configuration item 2. In this way, inefficient configurations are eliminated, and a better illumination configuration is obtained.

[0028] Further, based on the obtained first light source scheme, configure the controllable light source unit, and combine the image acquisition unit to perform image acquisition on the target wafer to obtain the first image set. Specifically, according to the first light source scheme, sequentially convert the multiple illumination configurations (multiple wavelength parameters, angle parameters, polarization parameters, illumination mode parameters) in the first light source scheme into specific control instructions for the controllable light source unit, and combine the image acquisition unit (such as a line array camera with a resolution of 12K), and sequentially acquire the wafer images under different illumination configurations. Finally, summarize the multiple wafer images to obtain the first image set. In this way, reliable feature inputs are provided for the subsequent defect recognition network.

[0029] In summary, compared with the prior art, according to the raw material characteristics and production process characteristics of the target wafer, this application retrieves multiple sample defect distributions, then obtains multiple historical light source schemes corresponding to the multiple sample defect distributions from the historical wafer production database, and then selects the illumination configuration items with historical configuration frequencies greater than the preset configuration frequency from the multiple historical light source schemes, and summarizes them to form the first light source scheme. Finally, based on the obtained first light source scheme, configure the controllable light source unit, and combine the image acquisition unit to perform image acquisition on the target wafer to obtain the first image set. In this way, by screening multiple illumination configurations in historical similar data and acquiring images accordingly, the intelligent reuse of historical data is realized, the efficiency and accuracy of image acquisition are improved, and necessary support is provided for subsequent defect analysis.

[0030] S20: Identify multiple first defect points of the target wafer according to the first image set, and generate a first defect distribution; The foregoing steps obtained the first image set. To more intuitively reflect the distribution of defects on the surface of the target wafer, multiple defect points of the target wafer, their corresponding spatial coordinate positions, and defect types can be identified based on the first image set.

[0031] To address the above problems, based on the first image set, this application uses a wafer defect recognition network to identify multiple first defect points, their corresponding coordinates, and defect types in multiple target wafer images, and finally marks them on the three-dimensional structure of the target wafer to generate the first defect distribution.

[0032] Specifically, step S20 in the method includes: Invoke the wafer defect recognition network, and sequentially input multiple first images in the first image set into the wafer defect recognition network to obtain multiple first defect points, the coordinate positions of each first defect point in the target wafer, and the defect type corresponding to each first defect point; Obtain the three-dimensional structure of the target wafer; Based on the coordinate positions of each first defect point in the target wafer, mark the multiple first defect points and the defect type corresponding to each first defect point on the three-dimensional structure to generate the first defect distribution.

[0033] In the embodiment of this application, first, the pre-trained wafer defect recognition network is invoked, and multiple first images in the first image set (one image is collected for each illumination configuration in the first light source scheme) are sequentially input into the wafer defect recognition network to obtain multiple first defect points, the coordinate positions (three-dimensional coordinates) of each first defect point in the target wafer, and the defect type corresponding to each first defect point (such as particle contamination, scratch, void, etc.). Exemplarily, a certain first image in the first image set is input into the wafer defect recognition network to obtain multiple first defect points, their corresponding coordinate positions, and defect types. Among them, the coordinate of a certain defect point is (5nm, -12nm, 3nm), and the defect type is a void, thus achieving the precise positioning and type recognition of the target wafer defects.

[0034] Secondly, obtain the three-dimensional structure of the target wafer. Specifically, by constructing a three-dimensional model of the target wafer, its three-dimensional structure is obtained. Exemplarily, the three-dimensional structure of the wafer can be constructed based on design files (such as GDS files) or process simulation data (such as TCAD), with an accuracy of ±5nm, and then a three-dimensional coordinate system (x, y, z) is established. In this way, the three-dimensional structure of the target wafer is obtained.

[0035] Finally, based on the coordinate positions of each first defect point in the target wafer, multiple first defect points and the defect types corresponding to each first defect point are marked on the three-dimensional structure to generate a first defect distribution. Specifically, based on the coordinate positions of each first defect point in the target wafer, all first defect points (x, y, z, defect type) are marked in the three-dimensional structure of the target wafer to form an intuitive three-dimensional first defect distribution.

[0036] Further, the steps for constructing the "wafer defect recognition network" include: Collect historical wafer images in the historical wafer production database to form a set of sample wafer images; Perform image defect annotation on the set of sample wafer images to generate a set of sample wafer defects and a set of sample defect types; Use machine learning to generate a defect presence recognition branch based on the set of sample wafer images and the set of sample wafer defects, and generate a defect type judgment branch based on the set of sample wafer images and the set of sample defect types; Connect and configure the defect presence recognition branch and the defect type judgment branch to obtain the wafer defect recognition network.

[0037] In the embodiment of the present application, first, historical wafer images are collected in the historical wafer production database to form a set of sample wafer images. Among them, the image collection process is random sampling, that is, the collected images include both defective images and defect-free images to improve the accuracy and robustness of model detection. Exemplarily, first, 5000 historical wafer images (such as covering the process from 28nm to 3nm) are randomly collected from the historical database to form a set of sample wafer images.

[0038] Secondly, perform image defect annotation on the set of sample wafer images to generate a set of sample wafer defects and a set of sample defect types. Exemplarily, each wafer image is marked with the corresponding defect type by manual annotation to obtain a set of sample wafer defects and a set of sample defect types. Further, an automatic annotation module can be used for automatic annotation to reduce labor costs. Exemplarily, common defect types can be initially marked based on threshold segmentation, and then complex defects (such as micro scratches) are corrected by those skilled in the art. At the same time, weak supervision annotation (only marking the image-level label "defect exists") is introduced for training the detection branch.

[0039] Thirdly, as Figure 2As shown, machine learning is used to generate a defect presence recognition branch based on a set of sample wafer images and a set of sample wafer defects, and a defect type judgment branch is generated based on the set of sample wafer images and the set of sample defect types. Specifically, two branches are constructed based on machine learning. One is the defect presence recognition branch, which is used to identify whether there are defects in the image. The other is the defect type judgment branch, which is used to judge the specific defect type of the image with defects. Among them, only when the defect presence recognition branch judges that there are defects in the image will the image enter the defect type judgment branch for type judgment. If the image has no defects, it will be directly output as "none".

[0040] Exemplarily, the defect presence recognition branch can be based on the improved YOLOv7 framework, using ResNet50 as the backbone network. The first 10 convolutional layers extract multi-scale features. Among them, the shallow layers (Conv1-Conv2_x) capture the edge details of defects at the 0.3μm level (such as the pixel-level contour of scratches), and the deep layers (Conv3_x) expand the receptive field through dilated convolutions to identify cluster-like defects above 50μm (such as large-area contamination). The embedded CBAM attention module suppresses the interference of normal textures on the wafer surface through the dual attention mechanisms of channels and space, and focuses on the defect area (such as enhancing the feature response of particle contamination by 2 times). The feature fusion part adopts the FPN+PANet structure to generate 3-level feature maps (P3-P5), which respectively correspond to small, medium, and large defect detections. The detection head outputs the defect bounding box coordinates (accuracy ±0.5μm) and confidence through Focal Loss (solving the imbalance between positive and negative samples) and GIoU Loss (improving the regression accuracy of extremely sized boxes), and locates the defect position after non-maximum suppression.

[0041] Exemplarily, the defect type judgment branch is based on the underlying features extracted by the first 5 convolutional layers of ResNet50. The spatial dimension is compressed through global average pooling, and semantic information such as the texture roughness and edge directionality of the defects is retained (such as the difference between the linear texture of metal bridging and the circular contour of particle contamination). The fully connected layer undergoes two non-linear transformations (such as 2048 dimensions → 16 dimensions), and combines Softmax to output the probability distribution of 16 types of defects. The cross-entropy loss is used to optimize the classification accuracy, and differential feature weights are designed especially for visually similar defects (such as subsurface voids and surface pits) (such as enhancing the penetration depth feature of infrared images).

[0042] Furthermore, the first 5 layers of convolutional parameters are shared by the two branches. For the defect presence recognition branch, subsequent spatial localization is enhanced through FPN + PANet, and for the defect type judgment branch, global pooling is used to focus on semantic abstraction, forming position-semantic complementarity. The total loss function balances the detection and classification tasks with a weight of 0.6:0.4. In this way, by sharing the first 5 layers of convolutional parameters, the model size is reduced, and by first judging whether there are defects and then identifying the defect type, the efficiency of defect recognition is improved.

[0043] Finally, the defect presence recognition branch and the defect type judgment branch are connected and configured to obtain a wafer defect recognition network. Specifically, the wafer image is input into the wafer defect recognition network. First, the defect presence recognition branch judges whether there are defects in the image. If there are no defects, "none" is directly output. If there are defects, it enters the defect type judgment branch for specific defect type judgment. In this way, defect recognition is performed on the wafer image.

[0044] In summary, compared with the prior art, based on the first image set, through the wafer defect recognition network, this application identifies multiple first defect points, their corresponding coordinates, and defect types of multiple target wafer images, and finally marks them on the three-dimensional structure of the target wafer to generate the first defect distribution. In this way, the distribution of most defects on the wafer surface is obtained, and a necessary basis is provided for the subsequent acquisition of the second defect distribution.

[0045] S30: Based on the first defect distribution, generate a second light source scheme, reconfigure the controllable light source unit, and combine the image acquisition unit to perform image acquisition on the target wafer again to obtain a second image set; In the foregoing steps, the first image set is collected. The first image set can reflect most of the defects of the target wafer, but there may still be other potential defect distributions on the target wafer.

[0046] To address the above problems, based on the first defect distribution, this application retrieves the historical wafer production database, filters out multiple historical wafer samples, eliminates the first defect distribution, obtains the potential defect distribution, extracts the first potential defect points, their corresponding coordinates, and defect types from the potential defect distribution, then retrieves multiple groups of candidate lighting configuration items with detection rate identifiers according to the first potential defect type, selects the candidate lighting configuration item with the highest detection rate as the first preferred lighting configuration item and adds it to the second light source scheme, and finally configures the controllable light source unit and combines the image acquisition unit to perform image acquisition on the target wafer again to obtain a second image set.

[0047] Specifically, step S30 in the method includes: Retrieve the historical wafer production database according to the first defect distribution and filter out multiple historical wafer samples; Extract the sample complete defect distribution of the multiple historical wafer samples, and take the union of the sample complete defect distributions of the multiple historical wafer samples to form an overall predicted defect distribution; Exclude the first defect distribution from the overall predicted defect distribution to obtain a potential defect distribution; Generate the second light source scheme according to the potential defect distribution.

[0048] In the embodiments of the present application, first, according to the first defect distribution, the historical wafer production database is retrieved to screen out multiple historical wafer samples. Specifically, based on the first defect distribution of the target wafer (the coordinate positions of the first defect points in the target wafer and the corresponding defect types), multiple historical wafer samples are screened out from the historical database, wherein the multiple historical wafer samples screened out include most of the defect distributions in the first defect distribution.

[0049] Secondly, extract the sample complete defect distribution of the multiple historical wafer samples, and take the union of the sample complete defect distributions of the multiple historical wafer samples to form an overall predicted defect distribution. Specifically, take the union of the complete defect distributions (including potential defects not found in the first detection) of the multiple historical wafer samples screened out to form an overall predicted defect distribution including the first defect distribution and other potential defect distributions.

[0050] Thirdly, exclude the first defect distribution from the overall predicted defect distribution to obtain a potential defect distribution. Specifically, exclude the first defect distribution of the detected defects from the overall predicted defect distribution, and retain the potential defect types that exist in the historical samples but are not recognized in the first detection of the target wafer.

[0051] Finally, generate the second light source scheme according to the potential defect distribution. Specifically, extract the first potential defect points and their corresponding coordinates and defect types in the potential defect distribution, then retrieve and obtain multiple groups of candidate illumination configuration items with detection rate identifiers corresponding to the first potential defect type according to the first potential defect type, then select the candidate illumination configuration item with the highest detection rate as the first preferred illumination configuration item, and finally add the first preferred illumination configuration item to the second light source scheme.

[0052] Specifically, the "generating the second light source scheme according to the potential defect distribution" includes: Extract the first potential defect points in the potential defect distribution, and extract the first potential coordinates and the first potential defect types of the first potential defect points; According to the first potential defect type, retrieve the defect-illumination mapping database to obtain multiple groups of candidate illumination configuration items corresponding to the first potential defect type, and each candidate illumination configuration item has a detection rate identifier; Arrange the multiple groups of candidate illumination configuration items in descending order according to the detection rate identifier, and select the candidate illumination configuration item with the highest detection rate as the first preferred illumination configuration item; Add the first preferred illumination configuration item to the second light source scheme.

[0053] In the embodiment of the present application, first, extract the first potential defect points from the potential defect distribution, and extract the first potential coordinates and the first potential defect types of the first potential defect points. Exemplarily, traverse the potential defect distribution to extract all the first potential defect points and their corresponding first potential coordinates (such as (4nm, -10nm, 5nm)) and the first potential defect types (such as interlayer bridging).

[0054] Secondly, according to the first potential defect type, retrieve the defect-illumination mapping database to obtain multiple groups of candidate illumination configuration items corresponding to the first potential defect type, and each candidate illumination configuration item has a detection rate identifier. Among them, the defect-illumination mapping database contains the mapping relationship between different defect types and illumination detection parameters. The detection rate = the number of true defects detected / the total number of defects actually present, which can reflect the accuracy of defect detection under this illumination configuration. The larger the detection rate, the better the defect detection effect, and the more excellent this illumination configuration item is. Exemplarily, according to the first potential defect type, retrieve the defect-illumination mapping database to obtain multiple groups of candidate illumination configuration items corresponding to the first potential defect type, and each candidate illumination configuration item has a detection rate identifier. For example, candidate illumination configuration item 1 (500nm, incident angle 35°, horizontal polarization, annular dark field, detection rate 80%), candidate illumination configuration item 2 (900nm, incident angle 25°, circular polarization, backscattering illumination, detection rate 50%).

[0055] Thirdly, arrange the multiple groups of candidate illumination configuration items in descending order according to the detection rate identifier, and select the candidate illumination configuration item with the highest detection rate as the first preferred illumination configuration item. Exemplarily, the multiple groups of candidate illumination configuration items retrieved include: candidate illumination configuration item 1 (500nm, incident angle 35°, horizontal polarization, annular dark field, detection rate 80%), candidate illumination configuration item 2 (900nm, incident angle 25°, circular polarization, backscattering illumination, detection rate 50%). Arranged in descending order according to the detection rate identifier: candidate illumination configuration item 1, candidate illumination configuration item 2. Among them, the larger the detection rate, the better the defect detection effect, and the more excellent this illumination configuration item is. Thus, select the candidate illumination configuration item with the highest detection rate (candidate illumination configuration item 1) as the first preferred illumination configuration item.

[0056] Finally, add the first preferred illumination configuration item to the second light source scheme. Exemplarily, add the first preferred illumination configuration item - candidate illumination configuration item 1 (500 nm, incident angle 35°, horizontal polarization, annular dark field) to the second light source scheme. In this way, the second light source scheme is the illumination configuration item with the highest detection rate for the potential defect distribution. Based on this, the image can be collected, and the potential defects can be effectively identified, improving the accuracy and precision of defect detection.

[0057] Further, according to the second light source scheme, configure the controllable light source unit again, and combine with the image acquisition unit to collect images of the target wafer again to obtain a second image set. Specifically, according to the second light source scheme, convert the illumination configuration (including wavelength parameters, angle parameters, polarization parameters, illumination mode parameters) into specific control instructions for the controllable light source unit, and combine with the image acquisition unit (such as a line array camera with 12K resolution) to collect the corresponding wafer images to obtain a second image set. In this way, the second image set is a reliable supplement to the first image set.

[0058] In summary, compared with the prior art, based on the first defect distribution, this application retrieves the historical wafer production database, screens out multiple historical wafer samples, eliminates the first defect distribution, obtains the potential defect distribution, extracts the first potential defect points and their corresponding coordinates and defect types from the potential defect distribution, then retrieves and obtains multiple groups of candidate illumination configuration items with detection rate identifiers according to the first potential defect type, selects the candidate illumination configuration item with the highest detection rate as the first preferred illumination configuration item and adds it to the second light source scheme, and finally configures the controllable light source unit and combines with the image acquisition unit to collect images of the target wafer again to obtain a second image set. In this way, the potential defects are effectively identified through the second light source scheme, which is a reliable supplement to the first defect distribution, improving the accuracy and precision of defect detection.

[0059] S40: According to the second image set, identify multiple second defect points of the target wafer, label the defect types of each second defect point, and combine with the first defect distribution to generate the surface defect distribution of the target wafer.

[0060] In the embodiment of this application, first, according to the same method as in S20, "According to the first image set, identify multiple first defect points of the target wafer and generate a first defect distribution", identify multiple second defect points of the target wafer according to the second image set, label the defect types of each second defect point, and generate a second defect distribution.

[0061] Further, by combining the first defect distribution and the second defect distribution, the surface defect distribution of the target wafer is generated. Among them, the combination of the first defect distribution and the second defect distribution can calibrate the global coordinate system through the Mark points on the wafer edge (accuracy ±0.5μm), and map the first defect distribution and the second defect distribution to the three-dimensional structure model uniformly. In this way, the accurate surface defect distribution of the target wafer is obtained.

[0062] In summary, the embodiments of the present application at least have the following technical effects: Compared with the prior art, according to the raw material characteristics and production process characteristics of the target wafer, the present application retrieves multiple sample defect distributions, then obtains multiple historical light source schemes corresponding to the multiple sample defect distributions from the historical wafer production database, and then selects the illumination configuration items with a historical configuration frequency greater than the preset configuration frequency from the multiple historical light source schemes, summarizes them to form the first light source scheme. Finally, based on the obtained first light source scheme, the controllable light source unit is configured, and the image acquisition unit is combined to collect images of the target wafer to obtain the first image set. In this way, by screening multiple illumination configurations in the historical similar data and collecting images accordingly, the intelligent reuse of historical data is realized, the efficiency and accuracy of image acquisition are improved, and necessary support is provided for subsequent defect analysis.

[0063] Secondly, based on the first image set, through the wafer defect recognition network, multiple first defect points and their corresponding coordinates and defect types of multiple target wafer images are identified, and finally marked on the three-dimensional structure of the target wafer to generate the first defect distribution. In this way, the distribution of most defects on the wafer surface is obtained, and a necessary basis is provided for obtaining the subsequent second defect distribution.

[0064] Thirdly, based on the first defect distribution, the historical wafer production database is retrieved, multiple historical wafer samples are screened out, and the first defect distribution is removed to obtain the potential defect distribution. The first potential defect points and their corresponding coordinates and defect types are extracted from the potential defect distribution, and then according to the first potential defect type, multiple groups of candidate illumination configuration items with detection rate marks are retrieved, and the candidate illumination configuration item with the highest detection rate is selected as the first preferred illumination configuration item and added to the second light source scheme. Finally, the controllable light source unit is configured, and the image acquisition unit is combined to collect images of the target wafer again to obtain the second image set. In this way, the potential defects are effectively identified through the second light source scheme, which is a reliable supplement to the first defect distribution, and the accuracy and precision of defect detection are improved.

[0065] Finally, according to the second image set, multiple second defect points of the target wafer are identified, the defect types of each second defect point are marked, and combined with the first defect distribution, the surface defect distribution of the target wafer is generated. In this way, the accurate surface defect distribution of the target wafer is obtained.

[0066] Through the above technical solution, the present application dynamically matches the illumination configuration items from the historical data according to the raw material characteristics and production process characteristics of the target wafer, and fully considers potential defects, significantly improving the detection rate and accuracy of wafer surface defect detection.

[0067] Embodiment 2, as Figure 3 shown, based on the same inventive concept as the artificial intelligence-based wafer surface defect detection method provided in Embodiment 1, the embodiment of the present invention further provides an artificial intelligence-based wafer surface defect detection system, including: An initial light source module 11, configured to obtain a first light source scheme, configure the controllable light source unit, and combine with the image acquisition unit to perform image acquisition on the target wafer to obtain a first image set; A defect analysis module 12, configured to identify multiple first defect points of the target wafer according to the first image set and generate a first defect distribution; An optimized light source module 13, configured to generate a second light source scheme based on the first defect distribution, reconfigure the controllable light source unit, and combine with the image acquisition unit to perform image acquisition on the target wafer again to obtain a second image set; An integration output module 14, configured to identify multiple second defect points of the target wafer according to the second image set, label the defect types of each of the second defect points, and combine with the first defect distribution to generate the surface defect distribution of the target wafer.

[0068] Among them, the initial light source module 11 is specifically configured to: Obtain the raw material characteristics and production process characteristics of the target wafer; Based on the raw material characteristics and the production process characteristics, retrieve the historical wafer production database, screen out multiple sample wafers, and extract the sample defect distributions of the multiple sample wafers to obtain multiple sample defect distributions; Generate the first light source scheme of the target wafer according to the multiple sample defect distributions.

[0069] Specifically, the "generating the first light source scheme of the target wafer according to the multiple sample defect distributions" includes: Obtain multiple historical light source schemes corresponding to the multiple sample defect distributions from the historical wafer production database, each of the historical light source schemes includes multiple illumination configuration items, and each illumination configuration item includes wavelength parameters, angle parameters, polarization parameters, and illumination mode parameters used by the controllable light source unit in a single image acquisition; Form the first light source scheme according to the multiple historical light source schemes.

[0070] Further, the step of "forming the first light source scheme according to the multiple historical light source schemes" includes: Taking the union of the lighting configuration items in the multiple historical light source schemes to obtain a lighting configuration set; Traversing the lighting configuration set, sequentially extracting each lighting configuration item, and statistically calculating the historical configuration frequency of each lighting configuration item in the multiple historical light source schemes; Selecting the lighting configuration items with historical configuration frequencies greater than the preset configuration frequency, and summarizing them to form the first light source scheme.

[0071] Among them, the defect analysis module 12 is specifically configured to: Invoking the wafer defect recognition network, sequentially inputting the multiple first images in the first image set into the wafer defect recognition network to obtain multiple first defect points, the coordinate positions of each first defect point in the target wafer, and the defect types corresponding to each first defect point; Obtaining the three-dimensional structure of the target wafer; Based on the coordinate positions of each first defect point in the target wafer, labeling the multiple first defect points and the defect types corresponding to each first defect point on the three-dimensional structure to generate the first defect distribution.

[0072] Further, the construction steps of the "wafer defect recognition network" include: Collecting historical wafer images in the historical wafer production database to form a sample wafer image set; Performing image defect annotation on the sample wafer image set to generate a sample wafer defect set and a sample defect type set; Using machine learning to generate a defect presence recognition branch based on the sample wafer image set and the sample wafer defect set, and generating a defect type judgment branch based on the sample wafer image set and the sample defect type set; Connecting and configuring the defect presence recognition branch and the defect type judgment branch to obtain the wafer defect recognition network.

[0073] Among them, the optimized light source module 13 is specifically configured to: According to the first defect distribution, retrieving the historical wafer production database and screening out multiple historical wafer samples; Extracting the sample complete defect distributions of the multiple historical wafer samples, and taking the union of the sample complete defect distributions of the multiple historical wafer samples to form an overall predicted defect distribution; Deleting the first defect distribution from the overall predicted defect distribution to obtain a potential defect distribution; Generating the second light source scheme according to the potential defect distribution.

[0074] Further, the step of "generating the second light source scheme according to the potential defect distribution" includes: Extracting a first potential defect point from the potential defect distribution, and extracting the first potential coordinates and the first potential defect type of the first potential defect point; According to the first potential defect type, retrieving a defect-illumination mapping database to obtain multiple groups of candidate illumination configuration items corresponding to the first potential defect type, and each candidate illumination configuration item has a detection rate identifier; Arranging the multiple groups of candidate illumination configuration items in descending order according to the detection rate identifier, and selecting the candidate illumination configuration item with the highest detection rate as the first preferred illumination configuration item; Adding the first preferred illumination configuration item to the second light source scheme.

[0075] Among them, the integration and output module 14 is specifically configured to: Identifying multiple second defect points of the target wafer according to the second image set, labeling the defect types of each second defect point, and combining the first defect distribution to generate the surface defect distribution of the target wafer.

[0076] In summary, the embodiments of the present application at least have the following technical effects: Compared with the prior art, the present application first obtains a first light source scheme through the initial light source module, configures the controllable light source unit, and combines the image acquisition unit to perform image acquisition on the target wafer to obtain a first image set, realizing the intelligent reuse of historical data, improving the efficiency and accuracy of image acquisition, and providing necessary support for subsequent defect analysis. Secondly, through the defect analysis module, according to the first image set, multiple first defect points of the target wafer are identified, a first defect distribution is generated, the distribution status of most defects on the wafer surface is obtained, and a necessary basis for obtaining the subsequent second defect distribution is provided. Thirdly, through the optimized light source module, it is used to generate a second light source scheme based on the first defect distribution, reconfigure the controllable light source unit, and combine the image acquisition unit to perform image acquisition on the target wafer again to obtain a second image set, effectively identifying potential defects, which is a reliable supplement to the first defect distribution, and improving the accuracy and precision of defect detection. Finally, through the integration and output module, it is used to identify multiple second defect points of the target wafer according to the second image set, label the defect types of each second defect point, and combine the first defect distribution to generate the surface defect distribution of the target wafer, obtaining the accurate surface defect distribution of the target wafer. In this way, the detection rate and accuracy of wafer surface defect detection are improved.

[0077] It should be noted that in the above embodiments, the descriptions of the respective embodiments have their own emphases. For the parts not described in detail in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0078] Those skilled in the art will understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0079] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for realizing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0080] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that realize the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0081] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for realizing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0082] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts.

[0083] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalent technologies, the present invention also intends to include these changes and modifications.

Claims

1. An artificial intelligence-based method for detecting wafer surface defects, characterized in that, Applied to a wafer surface defect detection device, the wafer surface defect detection device includes a controllable light source unit and an image acquisition unit; the method includes: Obtain a first light source scheme, configure the controllable light source unit, and combine with the image acquisition unit to perform image acquisition on the target wafer to obtain a first image set; According to the first image set, identify multiple first defect points of the target wafer and generate a first defect distribution; Based on the first defect distribution, generate a second light source scheme, reconfigure the controllable light source unit, and combine with the image acquisition unit to perform image acquisition on the target wafer again to obtain a second image set; According to the second image set, identify multiple second defect points of the target wafer, label the defect types of each of the second defect points, and combine with the first defect distribution to generate the surface defect distribution of the target wafer.

2. The method according to claim 1, wherein Obtaining a first light source scheme includes: Obtain the raw material characteristics and production process characteristics of the target wafer; Based on the raw material characteristics and the production process characteristics, retrieve the historical wafer production database, screen out multiple sample wafers, and extract the sample defect distributions of the multiple sample wafers to obtain multiple sample defect distributions; Generate the first light source scheme of the target wafer according to the multiple sample defect distributions.

3. The method according to claim 2, characterized in that, Generating the first light source scheme of the target wafer according to the multiple sample defect distributions includes: Obtain multiple historical light source schemes corresponding to the multiple sample defect distributions from the historical wafer production database. Each of the historical light source schemes includes multiple lighting configuration items, and each lighting configuration item includes wavelength parameters, angle parameters, polarization parameters, and illumination mode parameters used by the controllable light source unit in a single image acquisition; Form the first light source scheme according to the multiple historical light source schemes.

4. The method according to claim 3, characterized in that Forming the first light source scheme according to the multiple historical light source schemes includes: Take the union of the lighting configuration items in the multiple historical light source schemes to obtain a lighting configuration set; Traverse the lighting configuration set, sequentially extract each lighting configuration item, and count the historical configuration frequencies of each lighting configuration item in the multiple historical light source schemes; Select the lighting configuration items with historical configuration frequencies greater than the preset configuration frequency and summarize them to form the first light source scheme.

5. The method according to claim 1, wherein Identifying multiple first defect points of the target wafer according to the first image set and generating a first defect distribution includes: Invoke the wafer defect recognition network, sequentially input multiple first images in the first image set into the wafer defect recognition network to obtain multiple first defect points, the coordinate positions of each first defect point in the target wafer, and the defect types corresponding to each first defect point; Obtain the three-dimensional structure of the target wafer; Based on the coordinate positions of each first defect point in the target wafer, label the multiple first defect points and the defect types corresponding to each first defect point on the three-dimensional structure to generate the first defect distribution.

6. The method according to claim 5, characterized in that The construction steps of the wafer defect recognition network include: Collect historical wafer images in the historical wafer production database to form a sample wafer image set; Perform image defect annotation on the set of sample wafer images to generate a set of sample wafer defects and a set of sample defect types; Adopt machine learning to generate a defect presence recognition branch based on the set of sample wafer images and the set of sample wafer defects, and generate a defect type judgment branch based on the set of sample wafer images and the set of sample defect types; Connect and configure the defect presence recognition branch and the defect type judgment branch to obtain the wafer defect recognition network.

7. The method according to claim 1, wherein Generate a second light source scheme based on the first defect distribution, including: According to the first defect distribution, retrieve the historical wafer production database and screen out multiple historical wafer samples; Extract the sample complete defect distributions of the multiple historical wafer samples, and take the union of the sample complete defect distributions of the multiple historical wafer samples to form an overall predicted defect distribution; Exclude the first defect distribution from the overall predicted defect distribution to obtain a potential defect distribution; Generate the second light source scheme according to the potential defect distribution.

8. The method according to claim 7, wherein Generate the second light source scheme according to the potential defect distribution, including: Extract first potential defect points from the potential defect distribution, and extract the first potential coordinates and the first potential defect types of the first potential defect points; According to the first potential defect type, retrieve the defect-light mapping database to obtain multiple groups of candidate lighting configuration items corresponding to the first potential defect type, and each candidate lighting configuration item has a detection rate identifier; Arrange the multiple groups of candidate lighting configuration items in descending order according to the detection rate identifier, and select the candidate lighting configuration item with the highest detection rate as the first preferred lighting configuration item; Add the first preferred lighting configuration item to the second light source scheme.

9. An artificial intelligence-based wafer surface defect detection system, characterized in that, For executing the method according to any one of claims 1-8, including: An initial light source module for obtaining a first light source scheme, configuring the controllable light source unit, and combining with the image acquisition unit to perform image acquisition on a target wafer to obtain a first image set; A defect analysis module for identifying multiple first defect points of the target wafer according to the first image set and generating a first defect distribution; An optimized light source module for generating a second light source scheme based on the first defect distribution, reconfiguring the controllable light source unit, and combining with the image acquisition unit to perform image acquisition on the target wafer again to obtain a second image set; An integration output module for identifying multiple second defect points of the target wafer according to the second image set, annotating the defect types of the second defect points, and combining with the first defect distribution to generate the surface defect distribution of the target wafer.

Images