A method and system for defect detection of a semiconductor sample

Abstract

The application discloses a kind of defect detection method and system of semiconductor sample, it is related to semiconductor technical field, it solves the technical problem of low efficiency of part defect identification on semiconductor surface caused by the inherent detection scheme of the existing semiconductor surface defect detection;Including: obtaining the first image group of semiconductor sample to be detected, each detection image in first image group is sequentially input into defect identification model to obtain its corresponding identification type and confidence;The defect identification model is obtained by artificial intelligence model training;Based on the identification type and confidence of each detection image, a secondary acquisition scheme is generated;The secondary acquisition scheme includes a plurality of second camera coordinates and second shooting directions;Obtain the second image group of semiconductor sample to be detected;Based on the detection image in first image group and second image group, generate an identification result;Increase the accuracy of semiconductor surface defect detection.

Description

Technical Field

[0001] This application belongs to the field of semiconductor technology, specifically a defect detection method and system for semiconductor samples. Background Technology

[0002] Semiconductor defects are mainly classified into surface defects and internal defects. Surface defects include scratches, foreign matter, cracks, bubbles, and stains, while internal defects include grain fracture, grain dislocation, metal wire fracture, and metal wire misalignment. Semiconductor defects are characterized by diversity, complexity, randomness, and minute size, posing challenges to detection. Semiconductor defect detection methods are mainly divided into contact detection and non-contact detection. Contact detection uses a probe to contact the semiconductor surface or interior, determining the presence of defects through electrical or force signals.

[0003] Existing semiconductor surface defect detection methods often involve placing the semiconductor at a specific location, taking pictures of its surface with a camera, and then using a trained model to identify and analyze the images to determine defects in the semiconductor device. However, image acquisition of the semiconductor is often limited to a set location. Since semiconductor defects exhibit different characteristics in different directions, image acquisition and recognition from different directions will result in different recognition confidence levels or results. These defects are not obvious at these set locations, which leads to low accuracy in identifying such defects. Therefore, a defect detection method for semiconductor samples is needed. Summary of the Invention

[0004] This application aims to solve at least one of the technical problems existing in the prior art; to this end, this application proposes a defect detection method and system for semiconductor samples to solve the technical problem of low efficiency in identifying some defects on the semiconductor surface caused by the inherent detection scheme of existing semiconductor surface defect detection.

[0005] To achieve the above objectives, a first aspect of this application provides a method for defect detection of a semiconductor sample, comprising: Acquire a first image set of a semiconductor sample to be tested, the first image set including detection images of the semiconductor sample to be tested captured by a camera at several set positions; Each detected image in the first image group is sequentially input into the defect recognition model to obtain its corresponding recognition type and confidence level; the defect recognition model is trained through an artificial intelligence model. A secondary acquisition scheme is generated based on the recognition type and confidence level of each detected image; the secondary acquisition scheme includes several second camera coordinates and a second shooting direction; Acquire a second image set of the semiconductor sample to be tested, the second image set including detection images of the semiconductor to be tested in several predetermined directions in the secondary acquisition scheme; Recognition results are generated based on the detected images in the first and second image groups.

[0006] This application acquires a first image set of the semiconductor sample to be tested, and sequentially inputs each detection image in the first image set into a defect recognition model to obtain its corresponding recognition type and confidence level; generates a secondary acquisition scheme based on the recognition type and confidence level of each detection image; acquires a second image set of the semiconductor sample to be tested, and generates recognition results based on the detection images in the first and second image sets; and performs secondary detection by analyzing the first recognition results, so that some defects with low reliability detected at a set location can be targeted for secondary detection, thereby improving the accuracy of defect recognition; and further increasing the accuracy of semiconductor surface defect detection.

[0007] In conjunction with the first aspect above, in one possible implementation, acquiring the first image set of the semiconductor sample to be detected includes: A three-dimensional coordinate system is constructed with the center of the semiconductor sample to be tested as the center. Since the robotic arm that places the semiconductor sample to be tested is in a fixed position during testing, that is, the center of the semiconductor sample to be tested is placed in a fixed position, the three-dimensional coordinate system is generally fixed. Several set first camera coordinates and first shooting directions are obtained. The semiconductor sample to be tested is imaged in the camera coordinates with the acquisition direction to obtain the detection image corresponding to the first camera coordinates and the first shooting direction. The detection images at each set first camera coordinate and in the first shooting direction are integrated into a first image group.

[0008] In conjunction with the first aspect above, in one possible implementation, the defect identification model is obtained through training an artificial intelligence model, including: Acquire several semiconductor samples corresponding to various defect types, perform image acquisition on each semiconductor sample to obtain several detection images, and generate several training data and test data based on the detection images; The AI ​​model is trained using training images; the trained AI model is tested using test images to obtain an AI model whose input is the detection image and whose output is the corresponding defect type; the defect type is output as the recognition type, and the confidence level corresponding to the defect type is also output, finally obtaining a defect recognition model whose input is the detection image and whose output is the recognition type and its corresponding confidence level.

[0009] In conjunction with the first aspect above, in one possible implementation, the generation of several training data and test data based on the detected image includes: The detection images are labeled with their corresponding recognition types, and the detection images and their corresponding recognition types are integrated into training data and test data.

[0010] In conjunction with the first aspect above, in one possible implementation, the generation of the secondary acquisition scheme based on the recognition type and confidence level of each detected image includes: Obtain the location of the defect corresponding to each identification type on the semiconductor to be inspected; integrate the identification types and confidence levels with the same location into a defect identification group; If the maximum confidence score in the defect identification group is greater than the set confidence score threshold, then the identification type is marked as a defect type; otherwise, a secondary acquisition scheme is generated based on the confidence score in the defect identification group. The secondary acquisition schemes corresponding to each defect identification group are obtained sequentially.

[0011] In conjunction with the first aspect above, in one possible implementation, the secondary acquisition scheme is generated based on the confidence level in the defect identification group, including: When there is only one confidence level in the defect identification group, the first camera coordinates corresponding to the confidence level are obtained, and the first camera coordinates are used as the detection coordinates. The detection coordinates are the center coordinates of the camera during the second image acquisition. Several second camera coordinates and corresponding second shooting directions are generated based on the detection coordinates. When there are multiple confidence levels in the defect identification group, the first camera coordinates corresponding to each confidence level are obtained, the detection coordinates are calculated based on each first camera coordinate and the confidence level, and several second camera coordinates and corresponding second shooting directions are generated based on the detection coordinates. The coordinates of each second camera and its corresponding second shooting direction are integrated into a secondary acquisition scheme.

[0012] In conjunction with the first aspect above, in one possible implementation, the calculation of the detection coordinates based on the coordinates and confidence levels of each first camera includes: Obtain the coordinates of each first camera and its corresponding confidence score, convert the coordinates of each first camera into spherical coordinates, and label them as follows. And label its corresponding confidence level as ZDi; where i is the number of the first camera coordinates; using the formula:

[0013] Calculated detection coordinates Where i = 1, 2, ..., I; I is the total number of coordinates of the first camera.

[0014] In conjunction with the first aspect above, in one possible implementation, generating several second camera coordinates and corresponding second shooting directions based on the detection coordinates includes: The detection coordinates are used as the second camera coordinates of the target, and the direction of the detection coordinates pointing to the origin is used as the second shooting direction of the target; Obtain the set reference offset angle; obtain a set of reference directions with the angle between them and the second shooting direction of the target as the reference offset angle, and randomly select a set number of directions from the set of reference directions as the reference second shooting direction; For example, the detection coordinates are transformed into a three-dimensional Cartesian coordinate system to obtain the detection coordinates, which are also the target second camera coordinates:

[0015] The direction of the detected coordinates pointing to the origin is taken as the second shooting direction of the target; that is, the direction of the second camera coordinates pointing to the origin of the target is the second shooting direction of the target; that is, the vector. The direction;

[0016] The second shooting direction is referenced as a vector. The direction; ;vector The following conditions must be met:

[0017] Where ω is the reference offset angle; Calculate the distance between the target second camera coordinates and the origin; select the coordinates of a point on the reference second shooting direction that is equidistant from the origin as the reference second camera coordinates; Sequentially obtain the coordinates of the reference second camera corresponding to each reference second shooting direction; The second camera coordinates include the target second camera coordinates and the reference second camera coordinates; the second shooting direction includes the target second shooting direction and the reference second shooting direction.

[0018] In conjunction with the first aspect above, in one possible implementation, generating the recognition result based on the detected images in the first image group and the second image group includes: Obtain the defect type identified based on the first image group, and the detection image corresponding to the defect type; Extract several detection images from the second image group; input the detection images into the defect recognition model to obtain their corresponding recognition type and confidence level; Determine whether the highest confidence level in the second identification group is greater than the set confidence level threshold; if yes, then the identification type corresponding to the highest confidence level is taken as the defect type corresponding to the second image group; if no, then the defect type is marked as unidentified. Sequentially acquire the defect type corresponding to each second image group and its corresponding detection images; The various defect types and their corresponding detection images are integrated into the recognition results.

[0019] Another aspect of this application provides a defect detection system for semiconductor samples, including: an image acquisition module, a data processing module, and a data storage module; The image acquisition module acquires a first image group and a second image group of the semiconductor sample to be tested; both the first image group and the second image group contain several test images of the semiconductor sample to be tested acquired by cameras at several set positions. The data processing module: sequentially inputs each detected image in the first image group into the defect recognition model to obtain its corresponding recognition type and confidence level; the defect recognition model is trained through an artificial intelligence model; and generates a secondary acquisition scheme based on the recognition type and confidence level of each detected image; the secondary acquisition scheme includes several second camera coordinates and a second shooting direction; A second image set of the semiconductor sample to be tested is acquired, the second image set including detection images of the semiconductor sample to be tested in several predetermined directions in the secondary acquisition scheme; a recognition result is generated based on the detection images in the first image set and the second image set; The data storage module is used to store data such as the first image group, the second image group, and the recognition results.

[0020] Compared with the prior art, the beneficial effects of this application are: This application acquires a first image set of the semiconductor sample to be tested, and sequentially inputs each detection image in the first image set into a defect recognition model to obtain its corresponding recognition type and confidence level; generates a secondary acquisition scheme based on the recognition type and confidence level of each detection image; acquires a second image set of the semiconductor sample to be tested, and generates recognition results based on the detection images in the first and second image sets; and performs secondary detection by analyzing the first recognition results, so that some defects with low reliability detected at a set location can be targeted for secondary detection, thereby improving the accuracy of defect recognition; and further increasing the accuracy of semiconductor surface defect detection. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the method steps for defect detection of semiconductor samples in this application; Figure 2This is a schematic diagram of the module connections of the defect detection system for semiconductor samples in this application. Detailed Implementation

[0023] The technical solutions of this application will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0024] Please see Figure 1 The first aspect of this application provides a defect detection method for semiconductor samples, comprising: The first image group of the semiconductor sample to be tested is acquired. The first image group includes detection images of the semiconductor sample to be tested captured by the camera at several set positions. The set positions are fixed positions set by the user, such as the front, back, left, right, top, and bottom positions of the semiconductor sample to be tested. Each detected image in the first image group is sequentially input into the defect recognition model to obtain its corresponding recognition type and confidence level; the defect recognition model is trained by an artificial intelligence model; the recognition type is the possible defect type corresponding to the defective part in the detected image identified by the trained artificial intelligence model, and the confidence level can be understood as the accuracy of the recognition of the corresponding defect type; A secondary acquisition scheme is generated based on the recognition type and confidence level of each detection image. The secondary acquisition scheme includes several second camera coordinates and a second shooting direction. When the accuracy of identifying a certain defect on the detection image acquired at the set acquisition position is low, the defect at the corresponding position is further confirmed by setting a secondary acquisition to increase the accuracy of defect identification. Acquire a second image set of the semiconductor sample to be tested, the second image set including detection images of the semiconductor to be tested in several predetermined directions in the secondary acquisition scheme; The recognition result is generated based on the detected images in the first image group and the second image group; the recognition result is the final recognition result of the semiconductor to be detected, including normal and various defects.

[0025] Because defects on semiconductors exhibit different characteristics in different directions, image acquisition and recognition from different directions will result in different recognition confidence levels or results. It is understandable that the recognition effect will be best for images of defect locations acquired at a particular angle. This embodiment acquires a first image set of the semiconductor sample to be tested, and sequentially inputs each detection image in the first image set into a defect recognition model to obtain its corresponding recognition type and confidence level. A secondary acquisition scheme is generated based on the recognition type and confidence level of each detection image. A second image set of the semiconductor sample to be tested is acquired, and recognition results are generated based on the detection images in the first and second image sets. By analyzing the first recognition results and performing secondary detection, defects with low reliability detected at set locations can be targeted for secondary detection, improving the accuracy of defect recognition and further increasing the accuracy of semiconductor surface defect detection.

[0026] In one possible implementation, acquiring the first image set of the semiconductor sample to be detected includes: A three-dimensional coordinate system is constructed with the center of the semiconductor sample to be tested as the center. Since the robotic arm that places the semiconductor sample to be tested is in a fixed position during testing, that is, the center of the semiconductor sample to be tested is placed in a fixed position, the three-dimensional coordinate system is generally fixed. Several set first camera coordinates and first shooting directions are obtained. The semiconductor sample to be tested is imaged in the camera coordinates with the acquisition direction to obtain the detection image corresponding to the first camera coordinates and the first shooting direction. The detection images at each set first camera coordinate and in the first shooting direction are integrated into a first image group.

[0027] For example, in this embodiment, there are six sets of first camera coordinates and first shooting directions: the first camera coordinates are (0, 0, z), and the first shooting direction is the negative Z-axis; the first camera coordinates are (0, 0, -z), and the first shooting direction is the positive Z-axis; the first camera coordinates are (x, 0, 0), and the first shooting direction is the negative X-axis; the first camera coordinates are (-x, 0, 0), and the first shooting direction is the positive X-axis; the first camera coordinates are (0, y, 0), and the first shooting direction is the negative Y-axis; the first camera coordinates are (0, y, 0), and the first shooting direction is the positive Y-axis; x, y, and z are set positive numbers. It can be understood that the three-dimensional coordinate system that can be constructed here can be any one of the three-dimensional Cartesian coordinate system and spherical coordinate system. This embodiment takes the three-dimensional Cartesian coordinate system as an example. If other coordinate systems are used, it is only necessary to convert the corresponding coordinates to the corresponding coordinate system through existing coordinate transformation methods.

[0028] It is understandable that the first camera coordinates correspond one-to-one with the first shooting direction, and the first shooting direction is the direction from the corresponding first camera coordinates as the origin to the origin of the three-dimensional coordinate system.

[0029] In one possible implementation, the defect identification model is trained using an artificial intelligence model, including: Acquire several semiconductor samples corresponding to various defect types, perform image acquisition on each semiconductor sample to obtain several detection images, and generate several training data and test data based on the detection images; The AI ​​model is trained using training images; the trained AI model is then tested using inspection images, resulting in an AI model whose input is the detection image and whose output is the corresponding defect type; the defect type is output as the recognition type, along with the corresponding confidence level, ultimately yielding a defect recognition model whose input is the detection image and whose output is the recognition type and its corresponding confidence level; the AI ​​model includes recurrent neural network models, etc.; it can be understood that when the defect type is a specific defect, its corresponding recognition type is the corresponding defect type, such as scratches, cracks, and chipped edges; when the defect type is none, its corresponding recognition type is normal; the training data includes images with no defect type.

[0030] In one possible implementation, the generation of several training and testing data based on the detected image includes: The detection images are labeled with their corresponding recognition types, and the detection images and their corresponding recognition types are integrated into training data and test data.

[0031] In one possible implementation, the secondary acquisition scheme generated based on the recognition type and confidence level of each detected image includes: Obtain the location of the defect corresponding to each identification type on the semiconductor to be inspected; integrate the identification types and confidence levels with the same location into a defect identification group; If the maximum confidence score in the defect identification group is greater than the set confidence score threshold, then the identification type is marked as a defect type; otherwise, a secondary acquisition scheme is generated based on the confidence score in the defect identification group. The secondary acquisition schemes corresponding to each defect identification group are obtained sequentially.

[0032] In one possible implementation, the secondary acquisition scheme is generated based on the confidence level in the defect identification group, including: When there is only one confidence level in the defect identification group, the first camera coordinates corresponding to the confidence level are obtained, and the first camera coordinates are used as detection coordinates. The detection coordinates are the center coordinates of the camera when the second image acquisition is performed. Based on the detection coordinates, several second camera coordinates and corresponding second shooting directions are generated. When there are multiple confidence levels in the defect identification group, the first camera coordinates corresponding to each confidence level are obtained, the detection coordinates are calculated based on each first camera coordinate and the confidence level, and several second camera coordinates and corresponding second shooting directions are generated based on the detection coordinates. The coordinates of each second camera and its corresponding second shooting direction are integrated into a secondary acquisition scheme.

[0033] In one possible implementation, the detection coordinates are calculated based on the coordinates and confidence levels of each first camera, including: Obtain the coordinates of each first camera and its corresponding confidence score, convert the coordinates of each first camera into spherical coordinates, and label them as follows. And label its corresponding confidence level as ZDi; where i is the number of the first camera coordinates; using the formula:

[0034] Calculated detection coordinates Where i = 1, 2, ..., I; I is the total number of coordinates of the first camera.

[0035] In this embodiment, the confidence level and its corresponding first camera coordinates are used to calculate the detection coordinates using the formula described above. When the confidence level in a certain camera coordinate direction is much higher than that in other directions, it indicates that the defect is more obvious in the shooting direction corresponding to that camera coordinate. In this case, the closer the final detection coordinates are to that camera coordinate, the greater the probability of obtaining obvious defect images by re-acquiring images around that camera coordinate, thus improving the accuracy of subsequent identification. When the confidence levels in each camera coordinate direction are similar, it indicates that the defect is more obvious in the area enclosed by each camera coordinate. In this case, the detection coordinates inside this area are generated based on the confidence level, and images are re-acquired, thus improving the probability of obtaining obvious defect images, thus improving the accuracy of subsequent identification.

[0036] In one possible implementation, several second camera coordinates and corresponding second shooting directions are generated based on the detection coordinates, including: The detection coordinates are used as the second camera coordinates of the target, and the direction of the detection coordinates pointing to the origin is used as the second shooting direction of the target; Obtain a set reference offset angle; the reference offset angle is manually set, and in this embodiment, the set reference offset angle is 10°; obtain a set of reference directions whose included angle with the target second shooting direction is the reference offset angle, and randomly select a set number of directions from the reference direction set as reference second shooting directions; in this embodiment, by setting an angle limit between reference directions, a set number of directions are randomly selected from the reference direction set as reference second shooting directions, that is, arbitrarily selecting one reference direction as the reference second shooting direction, and the reference directions in the reference direction set whose included angle with the reference second shooting direction is the set angle limit are also used as reference second shooting directions, and iterating to finally obtain a set number of reference second shooting directions; the specific number is determined by the individual. In this embodiment, the set number is 10. In another embodiment, the distance between the starting points of the unit vectors in the reference direction is set as a constraint. A set number of directions are randomly selected from the set of reference directions as reference second shooting directions. That is, any reference direction is selected as the reference second shooting direction, and the starting point coordinates of the unit vector of the reference second shooting direction are obtained. It can be understood that the ending point coordinates of the unit vector are the origin. The starting point coordinates of the unit vectors corresponding to each reference direction in the set of reference directions are obtained. The distance between the starting point coordinates and the starting point coordinates of the unit vectors corresponding to the reference second shooting direction and the reference direction with the set distance constraint are also used as reference second shooting directions. The process is iterated to finally obtain a set number of reference second shooting directions. The specific number is set manually.

[0037] For example, the detection coordinates are transformed into a three-dimensional Cartesian coordinate system to obtain the detection coordinates, which are also the target second camera coordinates:

[0038] The direction of the detected coordinates pointing to the origin is taken as the second shooting direction of the target; that is, the direction of the second camera coordinates pointing to the origin of the target is the second shooting direction of the target; that is, the vector. The direction;

[0039] The second shooting direction is referenced as a vector. The direction; ;vector The following conditions must be met:

[0040] Where ω is the reference offset angle; substituting the reference offset angle yields several vectors. ; to convert each vector The direction is used as the second shooting direction; The following conditions must be met:

[0041] Calculate the distance between the target second camera coordinates and the origin; select the coordinates of a point on the reference second shooting direction that is equidistant from the origin as the reference second camera coordinates; that is, there is a vector. The coordinates of the reference second camera corresponding to the determined second shooting direction; Sequentially obtain the coordinates of the reference second camera corresponding to each reference second shooting direction; The second camera coordinates include the target second camera coordinates and the reference second camera coordinates; the second shooting direction includes the target second shooting direction and the reference second shooting direction. It can be understood that the distances between each reference second camera coordinate and the target second camera coordinates are equal, and they are distributed around the target second camera coordinates.

[0042] In one possible implementation, generating the recognition result based on the detected images in the first image group and the second image group includes: Obtain the defect type identified based on the first image group, and the detection image corresponding to the defect type; Extract several detection images from the second image group; input the detection images into the defect recognition model to obtain their corresponding recognition type and confidence level; Determine whether the highest confidence level in the second identification group is greater than the set confidence level threshold; if yes, then the identification type corresponding to the highest confidence level is taken as the defect type corresponding to the second image group; if no, then the defect type is marked as unidentified. Sequentially acquire the defect type corresponding to each second image group and its corresponding detection images; The various defect types and their corresponding detection images are integrated into the recognition results.

[0043] Please see Figure 2 Another aspect of this application provides a defect detection system for semiconductor samples, including: an image acquisition module, a data processing module, and a data storage module; The image acquisition module acquires a first image group and a second image group of the semiconductor sample to be tested; both the first image group and the second image group contain several test images of the semiconductor sample to be tested acquired by cameras at several set positions. The data processing module: sequentially inputs each detected image in the first image group into the defect recognition model to obtain its corresponding recognition type and confidence level; the defect recognition model is trained through an artificial intelligence model; and generates a secondary acquisition scheme based on the recognition type and confidence level of each detected image; the secondary acquisition scheme includes several second camera coordinates and a second shooting direction; A second image set of the semiconductor sample to be tested is acquired, the second image set including detection images of the semiconductor sample to be tested in several predetermined directions in the secondary acquisition scheme; a recognition result is generated based on the detection images in the first image set and the second image set; The data storage module is used to store data such as the first image group, the second image group, and the recognition results.

[0044] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0045] How this application works: This application acquires a first image set of the semiconductor sample to be tested, and sequentially inputs each detection image in the first image set into a defect recognition model to obtain its corresponding recognition type and confidence level; generates a secondary acquisition scheme based on the recognition type and confidence level of each detection image; acquires a second image set of the semiconductor sample to be tested, and generates recognition results based on the detection images in the first and second image sets; and performs secondary detection by analyzing the first recognition results, so that some defects with low reliability detected at a set location can be targeted for secondary detection, thereby improving the accuracy of defect recognition; and further increasing the accuracy of semiconductor surface defect detection.

[0046] The above embodiments are only used to illustrate the technical methods of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of this application without departing from the spirit and scope of the technical methods of this application.

Claims

1. A method for defect detection in semiconductor samples, characterized in that, include: Construct a three-dimensional coordinate system centered on the center of the semiconductor sample to be tested; A first image set of a semiconductor sample to be tested is acquired. The first image set includes detection images of the semiconductor sample to be tested captured by a camera at several set first camera coordinates and a first shooting direction. The first shooting direction is the direction from the origin of the corresponding first camera coordinate to the origin of the three-dimensional coordinate system. Each detected image in the first image group is sequentially input into the defect recognition model to obtain its corresponding recognition type and confidence level; the defect recognition model is trained by an artificial intelligence model; the recognition types and confidence levels with the same location identified from each detected image are integrated into a defect recognition group; If the maximum confidence level in the defect identification group is not greater than a set confidence level threshold, a secondary acquisition scheme is generated, wherein generating the secondary acquisition scheme includes: Obtain the coordinates of each first camera and its corresponding confidence score, convert the coordinates of each first camera into spherical coordinates, and label them as follows. And label its corresponding confidence level as ZDi; where i is the number of the first camera coordinates; using the formula: Calculated detection coordinates Where i = 1, 2, ..., I; I is the total number of coordinates of the first camera; the secondary acquisition scheme is generated based on the detection coordinates; the secondary acquisition scheme includes several second camera coordinates and a second shooting direction; Acquire a second image set of the semiconductor sample to be tested, the second image set including detection images of the semiconductor to be tested in several predetermined directions in the secondary acquisition scheme; Recognition results are generated based on the detected images in the first and second image groups.

2. The defect detection method for a semiconductor sample according to claim 1, characterized in that, The defect identification model is trained using an artificial intelligence model and includes: Acquire several semiconductor samples corresponding to various defect types, perform image acquisition on each semiconductor sample to obtain several detection images, and generate several training data and test data based on the detection images; The AI ​​model is trained using training images; the trained AI model is tested using test images to obtain an AI model whose input is the detection image and whose output is the corresponding defect type; the defect type is output as the recognition type, and the confidence level corresponding to the defect type is also output, finally obtaining a defect recognition model whose input is the detection image and whose output is the recognition type and its corresponding confidence level.

3. The defect detection method for a semiconductor sample according to claim 2, characterized in that, The generation of several training and testing data based on the detected images includes: The detection images are labeled with their corresponding recognition types, and the detection images and their corresponding recognition types are integrated into training data and test data.

4. The defect detection method for a semiconductor sample according to claim 1, characterized in that, The secondary acquisition scheme generated based on the recognition type and confidence level of each detected image includes: Obtain the location of the defect corresponding to each identification type on the semiconductor to be inspected; integrate the identification types and confidence levels with the same location into a defect identification group; If the maximum confidence score in the defect identification group is greater than the set confidence score threshold, then the identification type is marked as a defect type; otherwise, a secondary acquisition scheme is generated based on the confidence score in the defect identification group. The secondary acquisition schemes corresponding to each defect identification group are obtained sequentially.

5. The defect detection method for a semiconductor sample according to claim 4, characterized in that, The secondary acquisition scheme is generated based on the confidence level in the defect identification group, including: When there is only one confidence level in the defect identification group, the first camera coordinates corresponding to the confidence level are obtained and used as the detection coordinates; based on the detection coordinates, several second camera coordinates and corresponding second shooting directions are generated. When there are multiple confidence levels in the defect identification group, the first camera coordinates corresponding to each confidence level are obtained, the detection coordinates are calculated based on each first camera coordinate and the confidence level, and several second camera coordinates and corresponding second shooting directions are generated based on the detection coordinates. The coordinates of each second camera and its corresponding second shooting direction are integrated into a secondary acquisition scheme.

6. The defect detection method for a semiconductor sample according to claim 5, characterized in that, The process of generating several second camera coordinates and corresponding second shooting directions based on the detection coordinates includes: The detection coordinates are used as the second camera coordinates of the target, and the direction of the detection coordinates pointing to the origin is used as the second shooting direction of the target; Obtain the set reference offset angle; obtain a set of reference directions with the angle between them and the second shooting direction of the target as the reference offset angle, and randomly select a set number of directions from the set of reference directions as the reference second shooting direction; Calculate the distance between the target second camera coordinates and the origin; select the coordinates of a point on the reference second shooting direction that is equidistant from the origin as the reference second camera coordinates; Sequentially obtain the coordinates of the reference second camera corresponding to each reference second shooting direction; The second camera coordinates include the target second camera coordinates and the reference second camera coordinates; the second shooting direction includes the target second shooting direction and the reference second shooting direction.

7. The defect detection method for a semiconductor sample according to claim 1, characterized in that, The recognition result is generated based on the detected images in the first image group and the second image group, including: Obtain the defect type identified based on the first image group, and the detection image corresponding to the defect type; Extract several detection images from the second image group; input the detection images into the defect recognition model to obtain their corresponding recognition type and confidence level; Determine whether the highest confidence level in the second identification group is greater than the set confidence level threshold; if yes, then the identification type corresponding to the highest confidence level is taken as the defect type corresponding to the second image group; if no, then the defect type is marked as unidentified. Sequentially acquire the defect type corresponding to each second image group and its corresponding detection images; The various defect types and their corresponding detection images are integrated into the recognition results.

8. A defect detection system for semiconductor samples, based on the defect detection method for semiconductor samples according to any one of claims 1 to 7, characterized in that, include: Image acquisition module, data processing module, and data storage module; The image acquisition module acquires a first image group and a second image group of the semiconductor sample to be tested. The first image group and the second image group each contain a number of detection images of the semiconductor sample to be tested, captured by cameras at several set positions. The data processing module: sequentially inputs each detected image in the first image group into the defect recognition model to obtain its corresponding recognition type and confidence level; The defect recognition model is trained using an artificial intelligence model; a secondary acquisition scheme is generated based on the recognition type and confidence level of each detected image; the secondary acquisition scheme includes several second camera coordinates and a second shooting direction; A second image set of the semiconductor sample to be tested is acquired, the second image set including detection images of the semiconductor sample to be tested in several predetermined directions in the secondary acquisition scheme; a recognition result is generated based on the detection images in the first image set and the second image set; The data storage module is used to store the first image group, the second image group, and the recognition results.

Images