Defect self-adaptive detection method and system combined with feedback regulation

Abstract

The application discloses a defect self-adaptive detection method and system combined with feedback regulation and relates to the technical field of semiconductors. The method comprises the following steps: collecting a target surface image of a target wafer, performing feature extraction, and generating a target dot array geometric structure; randomly extracting a first number of atomic points, and obtaining K positioning dot array geometric structures; analyzing the loss degree, performing iterative optimization, and generating a target positioning dot array geometric structure; obtaining a target detection positioning image; performing positioning offset identification, and obtaining an offset identification result; moving an electric displacement table, and obtaining a target measurement result when the offset identification result meets preset requirements; and analyzing the target measurement result by using a defect identification unit, and obtaining a target detection result of the target wafer. The technical problem that the detection efficiency and accuracy are relatively low due to insufficient positioning accuracy in traditional semiconductor wafer defect detection is solved, and the technical effect of improving the detection efficiency and accuracy is achieved by optimizing wafer positioning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor technology, and in particular to a defect adaptive detection method and system combined with feedback regulation. BACKGROUND

[0002] In today's rapid development of the semiconductor industry, wafers as the core material of integrated circuit manufacturing, its quality is directly related to the performance and reliability of the final product. With the continuous improvement of chip integration and increasingly fine manufacturing process, the small defects on the wafer surface may have a serious impact on the final product. Therefore, efficient and accurate defect detection on the wafer surface becomes an indispensable part of the semiconductor manufacturing process. However, the traditional semiconductor wafer defect detection method has obvious limitations in positioning accuracy, such as image processing technology detection, although it realizes automatic detection to a certain extent, but in the face of complex structure characteristics of wafer surface, its positioning accuracy often cannot meet the demand of high-precision detection. This deficiency leads to limited detection efficiency and accuracy, especially in the face of small and difficult to identify defects, the traditional method often cannot meet the requirements. SUMMARY

[0003] The present application provides a defect adaptive detection method and system combined with feedback regulation, which solves the technical problem of low detection efficiency and accuracy caused by insufficient positioning accuracy in traditional semiconductor wafer defect detection.

[0004] In view of the above problems, the present application provides a defect adaptive detection method and system combined with feedback regulation.

[0005] In a first aspect of the present application, a defect adaptive detection method combined with feedback regulation is provided, the method comprising:

[0006] The target surface image of the target wafer is collected, and the target surface image is input into a structure feature extraction network layer for feature extraction to generate a target dot array geometry of the target wafer; a first number of atomic points are randomly extracted from the target dot array geometry multiple times, K positioning dot array geometries are obtained according to the distribution positions of the extracted atomic points; the loss degree between the K positioning dot array geometries and the target dot array geometry is analyzed by using a loss amount analysis function, and the K positioning dot array geometries are iteratively optimized according to the analysis result to generate a target positioning dot array geometry, wherein the target positioning dot array geometry includes a plurality of positioning atomic points and a plurality of positioning coordinates; the target positioning dot array geometry is stored in a visual alignment unit, and a target wafer placed on a sampling table is transferred to a detection platform of an electric displacement table by using a mechanical arm, an optical camera in the visual alignment unit is used to collect an image of the target wafer to obtain a target detection positioning image; a plurality of target detection positioning atomic points in the target detection positioning image are extracted based on the plurality of positioning coordinates, and a positioning offset identification is performed according to the plurality of target detection positioning atomic points and the plurality of positioning atomic points to obtain an offset identification result, wherein the offset identification result includes a target offset amount and a target offset direction; the target offset amount and the target offset direction are transmitted to a feedback control unit, the electric displacement table is moved by using the feedback control unit, when the offset identification result meets a preset requirement, a 3D line spectrum measurement of the target wafer is performed by using an optical profiler to obtain a target measurement result; the target measurement result is analyzed by using a defect identification unit to obtain a target detection result of the target wafer.

[0007] In a second aspect of the present application, a defect adaptive detection system combined with feedback regulation is provided, which comprises:

[0008] The target point array acquisition module is configured to collect a target surface image of a target wafer, input the target surface image into a structure feature extraction network layer for feature extraction, and generate a target point array geometry of the target wafer; the positioning point array acquisition module is configured to randomly extract a first number of atomic points from the target point array geometry multiple times, obtain K positioning point array geometries according to the distribution positions of the extracted atomic points; the optimization module is configured to analyze the loss degree between the K positioning point array geometries and the target point array geometry by using a loss amount analysis function, iteratively optimize the K positioning point array geometries according to the analysis result, and generate a target positioning point array geometry, wherein the target positioning point array geometry includes multiple positioning atomic points and multiple positioning coordinates; the image acquisition module is configured to store the target positioning point array geometry into a visual alignment unit, transfer the target wafer placed on a sampling table to a detection platform of an electric displacement table by using a mechanical arm, acquire an image of the target wafer by using an optical camera in the visual alignment unit, and obtain a target detection positioning image; the offset identification module is configured to extract multiple target detection positioning atomic points in the target detection positioning image based on the multiple positioning coordinates, identify a positioning offset based on the multiple target detection positioning atomic points and the multiple positioning atomic points, and obtain an offset identification result, wherein the offset identification result includes a target offset amount and a target offset direction; the measurement module is configured to transmit the target offset amount and the target offset direction to a feedback control unit, move the electric displacement table by using the feedback control unit, call an optical profiler to perform 3D line spectrum measurement on the target wafer after positioning when the offset identification result meets a preset requirement, and obtain a target measurement result; and the analysis module is configured to analyze the target measurement result by using a defect identification unit, and obtain a target detection result of the target wafer.

[0009] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0010] The target surface image of the target wafer is collected, and the target surface image is input into a structure feature extraction network layer for feature extraction to generate a target dot lattice geometry of the target wafer. Next, a first number of atomic points are randomly extracted from the target dot lattice geometry multiple times, and K positioning dot lattice geometries are obtained according to the distribution positions of the extracted atomic points. Then, the loss degree between the K positioning dot lattice geometries and the target dot lattice geometry is analyzed by using a loss amount analysis function, and the K positioning dot lattice geometries are iteratively optimized according to the analysis result to generate a target positioning dot lattice geometry, wherein the target positioning dot lattice geometry includes a plurality of positioning atomic points and a plurality of positioning coordinates. Then, the target positioning dot lattice geometry is stored in a visual alignment unit, and a target wafer placed on a sampling table is transferred to a detection platform of an electric displacement table by using a mechanical arm. An optical camera in the visual alignment unit is used to collect an image of the target wafer to obtain a target detection positioning image. Further, a plurality of target detection positioning atomic points in the target detection positioning image are extracted based on the plurality of positioning coordinates, and a positioning offset identification is performed according to the plurality of target detection positioning atomic points and the plurality of positioning atomic points to obtain an offset identification result, wherein the offset identification result includes a target offset amount and a target offset direction. The target offset amount and the target offset direction are transmitted to a feedback control unit, and the electric displacement table is moved by using the feedback control unit. When the offset identification result meets a preset requirement, a 3D line spectrum measurement of the positioned target wafer is performed by using an optical profiler to obtain a target measurement result. Finally, the target measurement result is analyzed by using a defect identification unit to obtain a target detection result of the target wafer. The technical problem of low detection efficiency and accuracy caused by insufficient positioning accuracy in traditional semiconductor wafer defect detection is solved, and the technical effect of improving the detection efficiency and accuracy is achieved by optimizing the wafer positioning. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0012] Figure 1 The defect adaptive detection method flowchart combined with feedback regulation provided by the embodiments of the present application is shown.

[0013] Figure 2 The defect adaptive detection system structure diagram combined with feedback regulation provided by the embodiments of the present application is shown.

[0014] Explanation of reference signs: target dot array acquisition module 11, positioning dot array acquisition module 12, optimization module 13, image acquisition module 14, offset identification module 15, measurement module 16, analysis module 17. DETAILED DESCRIPTION

[0015] The present application provides a defect adaptive detection method and system combined with feedback regulation, which solves the technical problem of low detection efficiency and accuracy caused by insufficient positioning accuracy in traditional semiconductor wafer defect detection.

[0016] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0017] It should be noted that the terms "comprise" and "have" are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server comprising a series of steps or units does not have to be limited to those clearly listed steps or units, but can include other steps or modules that are not clearly listed or inherent to these processes, methods, products or devices.

[0018] Embodiment one, as shown in the present application provides a defect adaptive detection method combined with feedback regulation, wherein the method comprises: Figure 1 Acquiring a target surface image of a target wafer and inputting the target surface image into a structure feature extraction network layer for feature extraction to generate a target dot array geometry of the target wafer.

[0019] Using a high-precision image acquisition device to acquire an image of the target surface of the target wafer, and inputting the acquired target surface image into a structure feature extraction network layer for feature extraction. After feature extraction, the network layer outputs a target dot array geometry containing detailed feature information of the target wafer surface. The target dot array geometry is composed of multiple atomic points, each atomic point represents a feature point on the wafer surface, and the atomic points also have corresponding coordinate information, which constitutes the target dot array geometry of the target wafer.

[0020] Further, it includes:

[0021]

[0022] ​obtaining a feedforward neural network; collecting a plurality of sample target surface images and a plurality of sample point array geometrical structures as construction data, and identifying the plurality of sample point array geometrical structures; training the feedforward neural network using the construction data, and supervising the training process according to the identified plurality of sample point array geometrical structures until convergence, thereby obtaining the trained structure feature extraction network layer.

[0023] Preferably, a feedforward neural network is obtained as a basic model of the structure feature extraction network layer. The feedforward neural network is a simple neural network structure in which information only flows in one direction, from the input layer to the output layer, without loops or cross-layer connections. A plurality of sample target surface images and a plurality of sample point array geometrical structures corresponding to the images are collected as construction data for training the feedforward neural network. In order to effectively supervise the network learning process during training, the plurality of sample point array geometrical structures collected need to be identified, which usually includes feature information of the target wafer surface, the location and type of defects, etc. Through identification, it can be ensured that the network can learn the correct feature extraction and classification method. The feedforward neural network is trained using the construction data. During the training process, the feedforward neural network learns how to extract useful features from the input target surface images and generate outputs similar to the sample point array geometrical structures. The training process is supervised according to the identified plurality of sample point array geometrical structures. Specifically, the output of the feedforward neural network is compared with the identified sample point array geometrical structures, and the difference (i.e. loss) between them is calculated. Then, the parameters of the network are adjusted according to the loss to reduce the future output difference. This process is repeated until the difference between the output of the network and the sample point array geometrical structures reaches a preset threshold (i.e. convergence). After the training process converges, a trained structure feature extraction network layer is obtained, which can effectively extract useful features from the target surface images of the target wafer and generate accurate point array geometrical structures.

[0024] Randomly extracting a first number of atomic points from the target point array geometrical structure multiple times, and obtaining K positioning point array geometrical structures according to the distribution positions of the extracted atomic points.

[0025] Randomly extracting a first number of atomic points from the target point array geometrical structure multiple times, and obtaining K positioning point array geometrical structures according to the distribution positions of the extracted atomic points. The first number is pre-set and depends on specific detection requirements and accuracy requirements. The purpose of random extraction is to obtain a plurality of different positioning point array geometrical structures, thereby increasing the robustness and accuracy of detection. For each set of extracted atomic points, a positioning point array geometrical structure is generated according to its distribution position. The positioning point array geometrical structure contains the extracted atomic points and their coordinate information. The above random extraction and generation process is repeated K times, thereby obtaining K different positioning point array geometrical structures.

[0026] The loss amount analysis function is used to analyze the loss degree between the K positioning lattice geometries and the target lattice geometry, and the K positioning lattice geometries are iteratively optimized according to the analysis result to generate a target positioning lattice geometry, wherein the target positioning lattice geometry includes a plurality of positioning atomic points and a plurality of positioning coordinates.

[0027] The loss amount analysis function is used to analyze the loss degree between the K positioning lattice geometries and the target lattice geometry, and the K positioning lattice geometries are iteratively optimized according to the analysis result to generate a target positioning lattice geometry, wherein the target positioning lattice geometry includes a plurality of positioning atomic points and a plurality of positioning coordinates.

[0028] Further, it includes:

[0029] The loss amount analysis function is used to analyze the loss degree between the K positioning lattice geometries and the target lattice geometry, and the K positioning lattice geometries are iteratively optimized according to the analysis result to generate a target positioning lattice geometry, wherein the target positioning lattice geometry includes a plurality of positioning atomic points and a plurality of positioning coordinates.

[0030] Preferably, the loss amount analysis function is used to calculate the loss amount between the K positioning point array geometries and the target point array geometry, to obtain K positioning loss amounts; the minimum value is found from the K positioning loss amounts, and the positioning point array geometry corresponding to the minimum value is the current leading structure; for the remaining K-1 positioning point array geometries, a preset adjustment mode is used to adjust the atomic points in the remaining K-1 positioning point array geometries, the preset adjustment mode being to take the position of the atomic point in the leading structure as the adjustment direction, and to adjust the atomic points in the remaining K-1 positioning point array geometries according to a preset adjustment step; after the K-1 positioning point array geometries are adjusted, the loss amount analysis function is used again to calculate K-1 adjustment positioning loss amounts between the K-1 adjusted positioning point array geometries and the target point array geometry; the K-1 adjustment positioning loss amounts are compared with the positioning loss amount corresponding to the leading structure, and if an adjustment positioning loss amount is smaller than the positioning loss amount of the leading structure, the adjustment positioning point array geometry corresponding to the adjustment positioning loss amount is set as the new leading structure; this process is iterated continuously until a preset iteration number is met or the loss amount converges to below a certain threshold; after multiple iterations, the leading structure obtained finally is the target positioning point array geometry.

[0031] Further, comprising:

[0032] When the minimum value in the K-1 adjustment positioning loss amounts is greater than or equal to the positioning loss amount corresponding to the leading structure, the leading structure is iteratively updated according to a certain probability;

[0033] When the minimum value in the K-1 adjustment positioning loss amounts is smaller than the positioning loss amount corresponding to the leading structure, the adjustment positioning point array geometry corresponding to the minimum value in the K-1 adjustment positioning loss amounts is updated as the leading structure, and the adjustment is continued;

[0034] When a preset iteration number is met, the leading structure corresponding to the minimum value of the positioning loss amount in the iteration process is taken as the target positioning point array geometry.

[0035] Preferably, the minimum value in the K-1 adjusted positioning loss amounts is compared with the positioning loss amount corresponding to the leading structure, if the minimum value in the K-1 adjusted positioning loss amounts is greater than or equal to the positioning loss amount corresponding to the leading structure, it indicates that the current leading structure is still the best in terms of loss amount, in order to increase diversity or avoid falling into local optimum, the leading structure can be randomly disturbed or fine-tuned according to a certain probability (for example, a small random number or a probability decreasing according to the number of iterations) to try to find a better solution; if the minimum value in the K-1 adjusted positioning loss amounts is less than the positioning loss amount corresponding to the leading structure, it indicates that there is an adjusted positioning point array geometry structure that is better than the current leading structure in terms of loss amount, at this time, the adjusted positioning point array geometry structure corresponding to the minimum value in the K-1 adjusted positioning loss amounts is updated as the leading structure, and the adjustment is continued; according to the updated leading structure, the remaining K-1 positioning point array geometry structures are adjusted again, and the loss amount is recalculated; when the preset number of iterations is met, the iteration is stopped, at this time, the leading structure corresponding to the minimum positioning loss amount in the iteration process is taken as the target positioning point array geometry structure.

[0036] Further, the loss amount analysis function is:

[0037] ;

[0038] wherein, is the positioning loss amount, is the target point array geometry structure, is the positioning point array geometry structure.

[0039] The loss amount analysis function calculates the similarity or correlation between two point sets (the target point array geometry structure and the positioning point array geometry structure), and normalizes them considering their module length, represents the point set in the target point array geometry structure, represents the point set in the positioning point array geometry structure.

[0040] The target positioning point array geometry structure is stored in the visual alignment unit, and a target wafer placed on a sampling table is transferred to a detection platform of an electric displacement table by using a mechanical arm, an optical camera in the visual alignment unit is used to collect images of the target wafer, and a target detection positioning image is obtained.

[0041] The target positioning lattice geometry structure (including a plurality of positioning atomic points and corresponding positioning coordinates) obtained after iterative optimization is stored in the visual alignment unit. The target wafer is placed on the sampling table. The control system of the mechanical arm sends instructions to make the mechanical arm grab the target wafer on the sampling table. The mechanical arm is accurately operated to transfer the target wafer from the sampling table to the detection platform of the electric displacement table. The electric displacement table can provide accurate position control and movement ability to ensure that the target wafer can be accurately placed at the predetermined position on the detection platform. After the target wafer is placed on the detection platform, the optical camera in the visual alignment unit is activated. The information in the target positioning lattice geometry structure is used to adjust the position and parameters (such as focal length, exposure time, etc.) of the camera to ensure that the feature points on the target wafer can be clearly photographed. The image of the target wafer is acquired by the camera to obtain the target detection positioning image.

[0042] A plurality of target detection positioning atomic points in the target detection positioning image are extracted based on the plurality of positioning coordinates. Positioning offset identification is performed according to the plurality of target detection positioning atomic points and the plurality of positioning atomic points to obtain an offset identification result, wherein the offset identification result includes a target offset and a target offset direction.

[0043] A plurality of target detection positioning atomic points are extracted from the target detection positioning image according to a plurality of positioning coordinates. The coordinates of the extracted target detection positioning atomic points are matched with the coordinates of the plurality of positioning atomic points stored in advance. For each pair of matched target detection positioning atomic points and positioning atomic points, the coordinate difference, i.e., the offset, in the image coordinate system is calculated. The offsets of all matched pairs are summarized to obtain an overall offset by methods such as averaging or least squares. According to the calculated overall offset, the offset direction of the target wafer relative to the expected position can be determined. The calculated target offset and target offset direction are combined into an offset identification result.

[0044] Further, it includes:

[0045] According to a plurality of positioning coordinates, a one-to-one mapping is performed on the plurality of target detection positioning atomic points and the plurality of positioning atomic points to obtain a plurality of atomic point mapping groups. The plurality of offset amounts and the plurality of offset directions of the plurality of atomic point mapping groups are calculated based on the plurality of positioning atomic points. The mode of the plurality of offset directions is taken as the target offset direction. The plurality of offset amounts are centrally searched and analyzed to determine the target offset amount.

[0046] Preferably, according to the plurality of positioning coordinates, a plurality of target detection positioning atomic points in a target detection positioning image are searched, and the plurality of target detection positioning atomic points and the plurality of positioning atomic points are one-to-one mapped to form a plurality of atomic point mapping groups; for each atomic point mapping group, an offset between a position of the target detection positioning atomic point in an image coordinate system and a position of the corresponding positioning atomic point in a theoretical coordinate system is calculated, the offset can be obtained by calculating a vector difference between the two coordinate points, and an offset direction can be obtained by calculating a directional cosine of the vector; a statistical method is used to determine the most likely offset direction, and the offset direction with the highest occurrence frequency (mode) is selected as the target offset direction; a plurality of offsets are centrally searched and analyzed, such as an average method (calculating an average value of all offsets) and a median method (selecting a median of the offsets), to determine the target offset.

[0047] Further, comprising:

[0048] The median, the first quartile and the third quartile of the plurality of offsets are calculated; a plurality of central offsets among the plurality of offsets between the first quartile and the third quartile are counted; and a mean value of the plurality of central offsets is calculated to generate the target offset.

[0049] Preferably, the median, the first quartile and the third quartile of the plurality of offsets are calculated, the first quartile is an offset at the 25% position after sorting all offsets, and the third quartile is an offset at the 75% position after sorting all offsets, if the total number of offsets cannot be divided by 4, the first quartile and the third quartile are usually the average value of the two offsets closest to these percentage positions, and if the total number of offsets can be divided by 4, the first quartile and the third quartile directly take the offsets at the corresponding positions; a plurality of central offsets among the plurality of offsets between the first quartile and the third quartile are counted, all central offsets are added together, and then divided by their number to obtain an average value, which is the target offset.

[0050] The target offset and the target offset direction are transmitted to a feedback control unit, the electric displacement table is moved by using the feedback control unit, when the offset recognition result meets a preset requirement, a 3D line spectrum measurement of the target wafer after positioning is performed by using an optical profiler to obtain a target measurement result.

[0051] The calculated target offset amount and target offset direction are transmitted to a feedback control unit, and the feedback control unit controls the electric displacement table to move accordingly according to the received target offset amount and target offset direction, so as to adjust the position of the target wafer. After the electric displacement table moves, offset identification is performed again to verify whether the position of the target wafer meets the preset requirement. When the offset identification result meets the preset requirement (i.e., the target wafer is correctly positioned), the optical profiler is called to perform 3D line spectrum measurement on the positioned target wafer. The optical profiler obtains 3D line spectrum data of the target wafer and generates a target measurement result.

[0052] The target measurement result generated by the optical profiler is imported into the defect identification unit, and the defect identification unit is used to analyze the target measurement result, so as to identify any potential defect or abnormality on the target wafer and generate a target detection result of the target wafer. The defect identification unit can be constructed based on a defect identification algorithm, so as to analyze the target measurement result and identify the potential defect on the target wafer.

[0053] The target measurement result generated by the optical profiler is imported into the defect identification unit, and the defect identification unit is used to analyze the target measurement result, so as to identify any potential defect or abnormality on the target wafer and generate a target detection result of the target wafer. The defect identification unit can be constructed based on a defect identification algorithm, so as to analyze the target measurement result and identify the potential defect on the target wafer.

[0054] In summary, the embodiments of the present application have at least the following technical effects:

[0055] The target surface image of the target wafer is collected, the target surface image is input into a structure feature extraction network layer for feature extraction, and a target dot array geometry of the target wafer is generated. Next, a first number of atomic points are randomly extracted from the target dot array geometry multiple times, and K positioning dot array geometries are obtained according to the distribution positions of the extracted atomic points. Then, the loss degree between the K positioning dot array geometries and the target dot array geometry is analyzed by using a loss amount analysis function, and the K positioning dot array geometries are iteratively optimized according to the analysis result to generate a target positioning dot array geometry, wherein the target positioning dot array geometry includes a plurality of positioning atomic points and a plurality of positioning coordinates. Then, the target positioning dot array geometry is stored in a visual alignment unit, and the target wafer placed on the sampling table is transferred to the detection platform of the electric displacement table by using the mechanical arm. The optical camera in the visual alignment unit is used to collect images of the target wafer to obtain a target detection positioning image. Further, a plurality of target detection positioning atomic points in the target detection positioning image are extracted based on the plurality of positioning coordinates, and a positioning offset identification is performed according to the plurality of target detection positioning atomic points and the plurality of positioning atomic points to obtain an offset identification result, wherein the offset identification result includes a target offset amount and a target offset direction. The target offset amount and the target offset direction are transmitted to a feedback control unit, and the electric displacement table is moved by using the feedback control unit. When the offset identification result meets the preset requirement, the 3D line spectrum measurement of the positioning completed target wafer is performed by using the optical profiler to obtain a target measurement result. Finally, the target measurement result is analyzed by using a defect identification unit to obtain a target detection result of the target wafer. The technical problem of low detection efficiency and accuracy caused by insufficient positioning accuracy in traditional semiconductor wafer defect detection is solved, and the technical effect of improving the detection efficiency and accuracy is achieved by optimizing the wafer positioning.

[0056] In the second embodiment, based on the same inventive concept as the defect adaptive detection method combined with feedback adjustment in the foregoing embodiments, as shown in the drawings, the present application provides a defect adaptive detection system combined with feedback adjustment. The system and method embodiments in the present application are based on the same inventive concept. The system includes: Figure 2

[0057] The target dot array acquisition module 11 is configured to collect a target surface image of a target wafer, input the target surface image into a structure feature extraction network layer for feature extraction, and generate a target dot array geometry of the target wafer.

[0058] The positioning dot array acquisition module 12 is configured to randomly extract a first number of atomic points from the target dot array geometry multiple times, and obtain K positioning dot array geometries according to the distribution positions of the extracted atomic points.

[0059] ​An optimization module 13 is configured to analyze the loss degree between the K positioning point array geometries and the target point array geometry by using a loss amount analysis function, and iteratively optimize the K positioning point array geometries according to the analysis result to generate a target positioning point array geometry, wherein the target positioning point array geometry includes a plurality of positioning atomic points and a plurality of positioning coordinates.

[0060] An image acquisition module 14 is configured to store the target positioning point array geometry into a visual alignment unit, and transfer a target wafer placed on a sampling table to a detection platform of an electric displacement table by using a mechanical arm, acquire an image of the target wafer by using an optical camera in the visual alignment unit, and obtain a target detection positioning image.

[0061] An offset identification module 15 is configured to extract a plurality of target detection positioning atomic points in the target detection positioning image based on the plurality of positioning coordinates, identify a positioning offset according to the plurality of target detection positioning atomic points and the plurality of positioning atomic points, and obtain an offset identification result, wherein the offset identification result includes a target offset amount and a target offset direction.

[0062] A measurement module 16 is configured to transmit the target offset amount and the target offset direction to a feedback control unit, move the electric displacement table by using the feedback control unit, and call an optical profiler to perform 3D line spectrum measurement on the target wafer with completed positioning when the offset identification result meets a preset requirement, and obtain a target measurement result.

[0063] An analysis module 17 is configured to analyze the target measurement result by using a defect identification unit, and obtain a target detection result of the target wafer.

[0064] Further, the target point array acquisition module 11 is configured to perform the following method:

[0065] Acquire a feedforward neural network, acquire a plurality of sample target surface images and a plurality of sample point array geometries as construction data, and identify the plurality of sample point array geometries, train the feedforward neural network by using the construction data, and supervise the training process according to the identified plurality of sample point array geometries until convergence, and obtain the trained structure feature extraction network layer.

[0066] Further, the optimization module 13 is configured to perform the following method:

[0067] The loss amount analysis function is used to calculate the loss amount of the K positioning point lattice geometries and the target point lattice geometry respectively, to generate K positioning loss amounts; the positioning point lattice geometry corresponding to the minimum value in the K positioning loss amounts is taken as a leading structure, and the remaining K-1 positioning point lattice geometries are adjusted according to a preset adjustment mode to obtain K-1 adjusted positioning point lattice geometries, wherein the preset adjustment mode is to adjust the atomic points in the remaining K-1 positioning point lattice geometries according to a preset adjustment step length, with the position of the atomic point in the leading structure as the adjustment direction; the loss amount analysis function is used again to calculate K-1 adjusted positioning loss amounts of the K-1 adjusted positioning point lattice geometries and the target point lattice geometry; the K-1 adjusted positioning loss amounts and the positioning loss amount corresponding to the leading structure are compared in size, and the leading structure is iteratively updated according to the comparison result until a preset iteration number is met, to obtain the target positioning point lattice geometry.

[0068] Further, the optimization module 13 is configured to perform the following method:

[0069] When the minimum value in the K-1 adjusted positioning loss amounts is greater than or equal to the positioning loss amount corresponding to the leading structure, the leading structure is iteratively updated according to a certain probability; when the minimum value in the K-1 adjusted positioning loss amounts is less than the positioning loss amount corresponding to the leading structure, the adjusted positioning point lattice geometry corresponding to the minimum value in the K-1 adjusted positioning loss amounts is updated as the leading structure, and the adjustment is continued; when a preset iteration number is met, the leading structure corresponding to the minimum value of the positioning loss amount in the iteration process is taken as the target positioning point lattice geometry.

[0070] Further, the optimization module 13 is configured to perform the following method:

[0071] The loss amount analysis function is: ; wherein, is the positioning loss amount, is the target point lattice geometry, is the positioning point lattice geometry.

[0072] Further, the offset identification module 15 is configured to perform the following method:

[0073] According to the plurality of positioning coordinates, the plurality of target detection positioning atomic points and the plurality of positioning atomic points are one-to-one mapped to obtain a plurality of atomic point mapping groups; the plurality of offset amounts and the plurality of offset directions of the plurality of atomic point mapping groups are calculated based on the plurality of positioning atomic points; the mode of the plurality of offset directions is taken as a target offset direction; the plurality of offset amounts are centrally searched and analyzed to determine the target offset amount.

[0074] Further, the offset identification module 15 is configured to perform the following method:

[0075] Calculate the median, the first quartile and the third quartile of the plurality of offsets; count a plurality of concentrated offsets between the first quartile and the third quartile of the plurality of offsets; and calculate the mean of the plurality of concentrated offsets to generate the target offset.

[0076] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The above describes a specific embodiment of the present application. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0077] The above only describes the preferred embodiments of the present application and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0078] The present application is only an exemplary description of the present application, and should be considered as covering any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A method of defect adaptive detection with feedback regulation, characterized in that, The method comprises: collecting a target surface image of a target wafer and inputting the target surface image into a structure feature extraction network layer for feature extraction to generate a target dot array geometry of the target wafer; randomly extracting a first number of atomic dots from the target dot array geometry multiple times, and obtaining K positioning dot array geometries according to the distribution positions of the extracted atomic dots; analyzing the loss degree between the K positioning dot array geometries and the target dot array geometry by using a loss amount analysis function, and iteratively optimizing the K positioning dot array geometries according to the analysis result to generate a target positioning dot array geometry, wherein the target positioning dot array geometry comprises a plurality of positioning atomic dots and a plurality of positioning coordinates; storing the target positioning dot array geometry into a visual alignment unit, transferring the target wafer placed on a sampling table to a detection platform of an electric displacement table by using a mechanical arm, collecting an image of the target wafer by using an optical camera in the visual alignment unit to obtain a target detection positioning image; extracting a plurality of target detection positioning atomic dots in the target detection positioning image based on the plurality of positioning coordinates, identifying a positioning offset according to the plurality of target detection positioning atomic dots and the plurality of positioning atomic dots to obtain an offset identification result, wherein the offset identification result comprises a target offset amount and a target offset direction; transmitting the target offset amount and the target offset direction to a feedback control unit, moving the electric displacement table by using the feedback control unit, and when the offset identification result meets a preset requirement, calling an optical profiler to perform 3D line spectrum measurement on the target wafer after positioning to obtain a target measurement result; analyzing the target measurement result by using a defect identification unit to obtain a target detection result of the target wafer.

2. The method for defect adaptive detection with feedback regulation as claimed in claim 1, wherein, It comprises: obtaining a feedforward neural network; collecting a plurality of sample target surface images and a plurality of sample dot array geometries as construction data, and identifying the plurality of sample dot array geometries, training the feedforward neural network by using the construction data, and supervising the training process according to the identified plurality of sample dot array geometries until convergence to obtain the trained structure feature extraction network layer.

3. The method for defect adaptive detection with feedback regulation as claimed in claim 1, wherein, It comprises: calculating the loss amount of the K positioning dot array geometries and the target dot array geometry by using the loss amount analysis function respectively to generate K positioning loss amounts; taking the positioning dot array geometry corresponding to the minimum value in the K positioning loss amounts as a leading structure, adjusting the remaining K-1 positioning dot array geometries according to a preset adjustment mode to obtain K-1 adjusted positioning dot array geometries, wherein the preset adjustment mode is to adjust the atomic dots in the remaining K-1 positioning dot array geometries according to a preset adjustment step size with the positions of the atomic dots in the leading structure as the adjustment direction; calculating K-1 adjusted positioning loss amounts of the K-1 adjusted positioning dot array geometries and the target dot array geometry by using the loss amount analysis function again; The K-1 adjustment positioning loss amounts are compared with the positioning loss amount corresponding to the leading structure, and the leading structure is iteratively updated according to a comparison result until a preset iteration number is met, so as to obtain the target positioning point array geometry.

4. The method for defect adaptive detection with feedback regulation as claimed in claim 3, wherein, Comprise: When the minimum value in the K-1 adjustment positioning loss amounts is greater than or equal to the positioning loss amount corresponding to the leading structure, the leading structure is iteratively updated according to a certain probability; When the minimum value in the K-1 adjustment positioning loss amounts is less than the positioning loss amount corresponding to the leading structure, the adjustment positioning point array geometry corresponding to the minimum value in the K-1 adjustment positioning loss amounts is updated as the leading structure, and the adjustment is continued; When the preset iteration number is met, the leading structure corresponding to the minimum value of the positioning loss amount in the iteration process is taken as the target positioning point array geometry.

5. The method for defect adaptive detection with feedback adjustment as claimed in claim 1, wherein, The loss amount analysis function is: ； wherein, is the positioning loss amount, is the target point array geometry, is the positioning point array geometry.

6. The method for defect adaptive detection with feedback adjustment of claim 1, wherein, Comprise: According to a plurality of positioning coordinates, the plurality of target detection positioning atomic points and the plurality of positioning atomic points are one-to-one mapped to obtain a plurality of atomic point mapping groups; The plurality of offset amounts and the plurality of offset directions of the plurality of atomic point mapping groups are calculated based on the plurality of positioning atomic points; The mode in the plurality of offset directions is taken as the target offset direction; The target offset amount is determined by centrally searching and analyzing the plurality of offset amounts.

7. The method for defect adaptive detection with feedback regulation as claimed in claim 6, wherein, Comprise: The median, the first quartile and the third quartile of the plurality of offset amounts are calculated; The plurality of central offset amounts located between the first quartile and the third quartile in the plurality of offset amounts are counted; The target offset amount is generated by mean calculation on the plurality of central offset amounts.

8. A system for detecting defects with feedback regulation, characterized in that The system for implementing the defect adaptive detection method with feedback regulation according to any one of claims 1-7, the system comprises: A target point array acquisition module, the target point array acquisition module is used for collecting a target surface image of a target wafer, and inputting the target surface image into a structure feature extraction network layer for feature extraction to generate a target point array geometry of the target wafer; A positioning point array acquisition module, the positioning point array acquisition module is used for randomly extracting a first number of atomic points from the target point array geometry multiple times, and obtaining K positioning point array geometries according to the distribution positions of the extracted atomic points; An optimization module, the optimization module is used for analyzing the loss degree between the K positioning point array geometries and the target point array geometry by using a loss amount analysis function, and iteratively optimizing the K positioning point array geometries according to an analysis result to generate a target positioning point array geometry, wherein the target positioning point array geometry comprises a plurality of positioning atomic points and a plurality of positioning coordinates; An image acquisition module, the image acquisition module is used for storing the target positioning point array geometry into a visual alignment unit, and transferring a target wafer placed on a sampling table to a detection platform of an electric displacement table by using a mechanical arm, and acquiring an image of the target wafer by using an optical camera in the visual alignment unit to obtain a target detection positioning image; An image acquisition module, the image acquisition module is used for storing the target positioning point array geometry into a visual alignment unit, and transferring a target wafer placed on a sampling table to a detection platform of an electric displacement table by using a mechanical arm, and acquiring an image of the target wafer by using an optical camera in the visual alignment unit to obtain a target detection positioning image; An offset identification module is configured to extract a plurality of target detection positioning atomic points in the target detection positioning image based on the plurality of positioning coordinates, perform positioning offset identification according to the plurality of target detection positioning atomic points and the plurality of positioning atomic points, and obtain an offset identification result, wherein the offset identification result includes a target offset amount and a target offset direction. A measurement module is configured to transmit the target offset amount and the target offset direction to a feedback control unit, move the electric displacement stage by using the feedback control unit, and when the offset identification result meets a preset requirement, call an optical profiler to perform 3D line spectrum measurement on the target wafer after positioning, and obtain a target measurement result. An analysis module is configured to analyze the target measurement result by using a defect identification unit, and obtain a target detection result of the target wafer.

Images