A general semiconductor defect detection method and system

Abstract

This invention discloses a general semiconductor defect detection method and system, relating to the field of semiconductor technology. The method includes: acquiring multiple defect images of a general semiconductor collected by a detection device at different detection channels, and preprocessing the multiple defect images to obtain a target defect image; using a pre-trained instance segmentation model to perform defect detection on the target defect image obtained after processing at each detection channel to obtain the basic defect features of the general semiconductor at each detection channel; fusing the basic defect features of the general semiconductor at each detection channel according to a preset rule to obtain fused defect features, and finally determining the defect type of the general semiconductor based on the fused defect features. This invention solves the problem of poor accuracy in semiconductor defect detection in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of semiconductor technology, and in particular to a general semiconductor defect detection method and system. Background Technology

[0002] In semiconductor manufacturing, defect detection is a crucial step in ensuring product quality. Due to differences in processes and equipment, defect morphologies vary significantly between different semiconductor plants. In particular, the same defect can exhibit vastly different appearances under different detection channels. Traditional detection algorithms often employ a single-channel detection method, which suffers from poor detection accuracy. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a general semiconductor defect detection method and system, which aims to solve the problem of poor accuracy in general semiconductor defect detection in the prior art.

[0004] This invention provides a general semiconductor defect detection method, the method comprising:

[0005] Multiple defect images of a general semiconductor collected by the testing equipment in different testing channels are acquired, and the multiple defect images are preprocessed to obtain the target defect image;

[0006] A pre-trained instance segmentation model is used to perform defect detection on the target defect images obtained after acquisition and processing of each detection channel, so as to obtain the basic defect features of general semiconductors under each detection channel.

[0007] The basic defect features of the general semiconductor under each detection channel are fused according to preset rules to obtain fused defect features, and then the defect type of the general semiconductor is obtained based on the fused defect features.

[0008] Furthermore, in the aforementioned general semiconductor defect detection method, the training process of the instance segmentation model is as follows:

[0009] A semiconductor defect annotation dataset is constructed to obtain a training dataset, which contains a predetermined number of images and their corresponding annotation information.

[0010] Perform multi-dimensional preprocessing on the images in the training dataset;

[0011] Construct a YOLO neural network;

[0012] The preprocessed training dataset is input into the YOLO neural network for deep learning training to obtain an instance segmentation model.

[0013] Furthermore, in the aforementioned general semiconductor defect detection method, the step of fusing the basic defect features of the general semiconductor under each detection channel according to a preset rule to obtain fused defect features includes:

[0014] By transforming coordinates, the positional overlap of defects in different detection channels is calculated to achieve positional matching of defects in different detection channels;

[0015] By comparing the similarity of the contour parameters and color features of defects, morphological matching of defects in different detection channels can be achieved.

[0016] For directional defects, the orientation and angle consistency are compared to achieve defect posture matching in different detection channels;

[0017] For defects in different matching detection channels, cross-channel duplicate annotations are eliminated, and the features of the same defect in different detection channels are merged to generate a unified fused defect feature.

[0018] Furthermore, in the aforementioned general semiconductor defect detection method, the step of using a pre-trained instance segmentation model to perform defect detection on the target defect image obtained after acquisition and processing for each detection channel, in order to obtain the basic defect features of the general semiconductor under each detection channel, includes:

[0019] The target defect image is segmented into multiple target defect image sub-regions according to preset rules, and a pre-trained instance segmentation model is used to detect defects in the target defect image sub-regions to obtain basic sub-defect features.

[0020] Based on location information, the basic sub-defect features within the same detection channel are spliced ​​together to obtain the basic defect features of general semiconductors.

[0021] Furthermore, in the aforementioned general semiconductor defect detection method, the location matching includes:

[0022] Take the defect set of one of the detection channels As the baseline set, the defect set of another detection channel. The set to be matched;

[0023] Calculate the mask intersection-union ratio (CUIR) of the defect masks from two different detection channels in the standard coordinate system.

[0024]

[0025] If only bounding boxes are available, calculate the bounding box intersection-union ratio (IU / UK), and simultaneously introduce center distance weights:

[0026]

[0027] in, Let Euclidean distance be the center of the defects in two different detection channels. The attenuation coefficient is... These are the center distance weights, , These are masks for defects in two different detection channels;

[0028] Set position matching threshold Only retain of Candidate pair.

[0029] Furthermore, in the aforementioned general semiconductor defect detection method, the basic defect characteristics include:

[0030] Basic categories and confidence levels, where confidence level characterizes the reliability of the detection;

[0031] Location features;

[0032] Morphological features, including contour features, are selected based on the defect size to determine whether to extract the contour.

[0033] Color features, including mean brightness, extreme values, and contrast.

[0034] Furthermore, in the aforementioned general semiconductor defect detection method, the step of preprocessing multiple defect images to obtain the target defect image includes:

[0035] Lighting / color standardization: unifies image brightness, contrast, and color distribution, eliminating the effects of uneven lighting;

[0036] Scale standardization: Adjust the defect image to a preset scale to ensure that the defect size is represented under a unified benchmark;

[0037] Coordinate system standardization: Establish a unified coordinate system and a unified coordinate reference for the location of defects.

[0038] Another object of the present invention is to provide a universal semiconductor defect detection system, the system comprising:

[0039] The acquisition module is used to acquire multiple defect images of general semiconductors collected by the detection equipment in different detection channels, and to preprocess the multiple defect images to obtain the target defect image;

[0040] The recognition module is used to perform defect detection on the target defect image obtained after acquisition and processing of each detection channel using a pre-trained instance segmentation model, so as to obtain the basic defect features of the general semiconductor under each detection channel.

[0041] The detection module is used to fuse the basic defect features of the general semiconductor under each detection channel according to preset rules to obtain fused defect features, and then obtain the defect type of the general semiconductor based on the fused defect features.

[0042] Another object of the present invention is to provide a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0043] Another object of the present invention is to provide an electronic device including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the method described above.

[0044] This invention acquires multiple defect images of a general-purpose semiconductor from different detection channels using a detection device, and preprocesses these images to obtain a target defect image. A pre-trained instance segmentation model is then used to detect defects in the target defect image obtained from each detection channel, yielding the basic defect features of the general-purpose semiconductor under each detection channel. These basic defect features are then fused according to preset rules to obtain fused defect features, and the defect type of the general-purpose semiconductor is determined based on these fused features. This invention avoids the problems of single-channel acquisition, which lacks comprehensive multi-dimensional defect features and leads to missed or false detections due to insignificant or interfered defect features under a single channel. Ultimately, it achieves accurate, efficient, and universal semiconductor defect detection, solving the problem of poor accuracy in general-purpose semiconductor defect detection in existing technologies. Attached Figure Description

[0045] Figure 1 This is a flowchart of the general semiconductor defect detection method in the first embodiment of the present invention;

[0046] Figure 2 This is a structural block diagram of a general semiconductor defect detection system in the third embodiment of the present invention.

[0047] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0048] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0049] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0051] Example 1

[0052] Please see Figure 1 The figure shows a general semiconductor defect detection method in the first embodiment of the present invention, the method including steps S10 to S12.

[0053] Step S10: Acquire multiple defect images of a general semiconductor collected by the detection device in different detection channels, and preprocess the multiple defect images to obtain the target defect image.

[0054] The general semiconductor can include various semiconductor materials such as epitaxy, substrate, and patterned wafers; different detection channels can include bright field, dark field, DIC, PL, etc. Using multiple detection (imaging) channels can improve the overall defect coverage. After acquiring multiple defect images of the general semiconductor, the acquired images are preprocessed to reduce randomness.

[0055] Specifically, the steps for preprocessing multiple defect images to obtain the target defect image include:

[0056] Lighting / color standardization: unifies image brightness, contrast, and color distribution, eliminating the effects of uneven lighting;

[0057] Scale standardization: Adjust the defect image to a preset scale to ensure that the defect size is represented under a unified benchmark;

[0058] Coordinate system standardization: Establish a unified coordinate system and a unified coordinate reference for the location of defects.

[0059] Step S11: Use a pre-trained instance segmentation model to perform defect detection on the target defect image obtained after acquisition and processing of each detection channel, so as to obtain the basic defect features of the general semiconductor under each detection channel.

[0060] The basic defect characteristics include:

[0061] Basic categories (such as dot, rectangle, triangle, scratch, etc.) and confidence levels, where confidence level characterizes the reliability of the detection;

[0062] Location features, such as coordinates or dimensions;

[0063] Morphological features, including contour features, are selected based on defect size to determine whether to extract the contour. For example, small defects (such as dots or small rectangles) do not require a contour, while large or specific-shaped defects require contour extraction. The contour area, perimeter, aspect ratio, major axis / minor axis, etc., are then calculated.

[0064] Color features, including average brightness, extreme values, and contrast, characterize the degree of severity of defects.

[0065] For example, the training process of an instance segmentation model is as follows:

[0066] A semiconductor defect annotation dataset is constructed to obtain a training dataset, which contains a predetermined number of images and their corresponding annotation information.

[0067] Perform multi-dimensional preprocessing on the images in the training dataset;

[0068] Construct a YOLO neural network;

[0069] The preprocessed training dataset is input into the YOLO neural network for deep learning training to obtain an instance segmentation model.

[0070] The dataset should include a multi-channel image dataset covering at least typical semiconductor defects. The dataset should contain no fewer than 6000 images and corresponding pixel-level annotations. The annotations must include the bounding box coordinates of the defects, the mask region, and the defect category label. The dataset should be divided into a training set, a validation set, and a test set in a 7:2:1 ratio. The training set should contain image data from different detection channels, and the number of samples for each type of defect should be balanced using an oversampling strategy to ensure that the number of samples for rare defects (such as short circuits and bond breaks) is no less than 500.

[0071] The training set images undergo multi-dimensional preprocessing: First, an image alignment algorithm is used to correct the spatial offset of different detection channels, ensuring the consistency of defect locations across multiple channels. Second, Gaussian filtering is used to remove noise, and adaptive histogram equalization is employed to enhance the contrast between defects and the background. Data augmentation operations include: geometric transformations such as ±15° rotation, 5%-20% scaling, and horizontal / vertical flipping of the images. Simultaneously, random brightness (±10%) and contrast (±15%) adjustments are added to simulate process noise (such as Gaussian noise and salt-and-pepper noise) tailored to semiconductor defect characteristics. All augmentation operations must be performed synchronously on the multi-channel images to maintain feature consistency.

[0072] The model employs a YOLO neural network or an improved instance segmentation model based on the Mask R-CNN framework. The Mask R-CNN feature extraction network uses an ISD backbone network comprising four dense blocks and four transition layers. This network enhances the representation of small defect features through a dynamic growth rate feature concatenation mechanism, and the model parameter size is controlled to within 1 / 268 of the traditional ResNet to meet edge deployment requirements. The model output layer contains three branches: a classification branch outputting defect category probabilities, a bounding box regression branch outputting defect location coordinates, and a mask branch outputting a 28×28 pixel defect shape mask. Initial weights are initialized using weights pre-trained on the ImageNet dataset, while the mask and classification branches employ the Xavier initialization strategy.

[0073] The training process is divided into two phases: the first phase (1-50 epochs) uses a frozen backbone network transfer learning approach, training only the head network with a learning rate of 0.001; the second phase (51-200 epochs) unfreezes all network layers, reducing the learning rate to 0.0001. The loss function uses a multi-task weighted loss: cross-entropy loss for classification, SmoothL1 loss for bounding box regression, and binary cross-entropy loss for masking, with a weight ratio of 1:1:2. Small defects (less than 50 pixels) are given an additional 1.5x weight in the masking loss. Training uses the SGD optimizer with a momentum parameter of 0.9, a weight decay coefficient of 0.0001, and a batch size of 8. Model performance is evaluated on the validation set every 5 epochs.

[0074] The model is evaluated using a dedicated metric system for semiconductor defect detection: recall (≥99.99%), precision (≥99%), mask intersection-over-union ratio (mIoU ≥0.9), and F1 score (≥99.5%) are calculated on the validation set. An early stopping mechanism is triggered when the validation set loss does not decrease for 10 consecutive epochs, and training is optimized using a learning rate decay strategy (decreasing to 1 / 10 of the current value each time). Evaluation on the test set must cover image data from different detection channels to ensure that the model's recall fluctuation in cross-channel scenarios does not exceed 0.5%. Finally, the model weights with the best overall performance on the test set are saved as the pre-trained instance segmentation model.

[0075] In addition, in some optional embodiments of the present invention, the step of using a pre-trained instance segmentation model to perform defect detection on the target defect image obtained after acquisition and processing of each detection channel, so as to obtain the basic defect features of the general semiconductor under each detection channel, includes:

[0076] The target defect image is segmented into multiple target defect image sub-regions according to preset rules, and a pre-trained instance segmentation model is used to detect defects in the target defect image sub-regions to obtain basic sub-defect features.

[0077] Based on location information, the basic sub-defect features within the same detection channel are spliced ​​together to obtain the basic defect features of general semiconductors.

[0078] In actual detection, a defect image is often divided into multiple small regions for block-by-block detection; multiple independent image frames may be generated in the same channel, and the detection results of each frame exist separately; the purpose of stitching is to integrate the detection results of multiple frames or multiple regions in the same channel to form a complete set of defect features without repetition.

[0079] Step S12: The basic defect features of the general semiconductor under each detection channel are fused according to preset rules to obtain fused defect features, and then the defect type of the general semiconductor is obtained based on the fused defect features.

[0080] The basic defect features are obtained by preprocessing multi-channel general semiconductor defect images to obtain target defect images, and then being detected and extracted by a pre-trained instance segmentation model. These features cover key individual-level information about defects in each channel. Then, they are fused according to preset rules. These preset rules are logic or algorithms based on the requirements of general semiconductor defect detection. Through these rules, the originally isolated multi-channel basic features are integrated to form fused defect features that can integrate multi-dimensional information and make up for the one-sidedness of single-channel features. Finally, the defect type of general semiconductor is obtained based on the fused defect features.

[0081] Specifically, defects that do not meet the requirements can be directly filtered out based on the wafer fab's hard requirements (such as defect size threshold and contrast lower limit); or machine learning models (such as decision trees and boosting trees) can be used to learn feature parameters (such as defect size distribution and morphology probability) based on sample statistical results to achieve automatic classification.

[0082] In summary, the general semiconductor defect detection method in the above embodiments of the present invention acquires multiple defect images of a general semiconductor collected by a detection device in different detection channels, and preprocesses these multiple defect images to obtain a target defect image. A pre-trained instance segmentation model is used to perform defect detection on the target defect image obtained after processing by each detection channel to obtain the basic defect features of the general semiconductor under each detection channel. The basic defect features of the general semiconductor under each detection channel are fused according to preset rules to obtain fused defect features, and then the defect type of the general semiconductor is obtained based on the fused defect features. This avoids the problems of single detection channel acquiring limited information, failing to comprehensively capture multi-dimensional defect features, and defects having insignificant features or being interfered with under a single channel, leading to missed or false detections. Ultimately, it achieves accurate, efficient, and universal semiconductor defect detection. It solves the problem of poor accuracy in general semiconductor defect detection in existing technologies.

[0083] Example 2

[0084] This embodiment also proposes a general semiconductor defect detection method. The difference between the general semiconductor defect detection method in this embodiment and the general semiconductor defect detection method in Embodiment 1 is as follows:

[0085] The step of fusing the basic defect features of the general semiconductor under each detection channel according to a preset rule to obtain the fused defect features includes:

[0086] By transforming coordinates, the positional overlap of defects in different detection channels is calculated to achieve positional matching of defects in different detection channels;

[0087] By comparing the similarity of the contour parameters and color features of defects, morphological matching of defects in different detection channels can be achieved.

[0088] For directional defects, the orientation and angle consistency are compared to achieve defect posture matching in different detection channels;

[0089] For defects in different matching detection channels, cross-channel duplicate annotations are eliminated, and the features of the same defect in different detection channels are merged to generate a unified fused defect feature.

[0090] First, the spatial coordinate system of different detection channels is unified by coordinate transformation. Then, the degree of overlap of defect positions in each channel is calculated to determine whether the defects detected in different channels correspond to the same entity in spatial position, thereby achieving defect position matching.

[0091] Next, the similarity of the contour parameters (such as perimeter, area, shape complexity, etc.) and color features (such as grayscale value, hue distribution, etc.) of defects in different channels is compared to confirm whether these defects are the same defect in different channels from the perspective of appearance morphology, so as to achieve morphological matching.

[0092] For directional defects (such as linear scratches, directional cracks, etc.), the correspondence between defects detected in different channels in spatial attitude is further verified by comparing the consistency of their orientation and angle, thus achieving attitude matching.

[0093] After matching the location, shape, and posture, the features of different channels belonging to the same defect are identified. At this point, it is necessary to eliminate the duplicate annotations of the same defect in different channels (to avoid the same defect being counted multiple times), and integrate the unique features of the defect in each channel (such as surface detail features that one channel excels at, internal structural features that another channel excels at, etc.) to finally generate a unified fused defect feature that contains multi-dimensional information and can fully reflect the essence of the defect.

[0094] For example, location matching includes:

[0095] Take the defect set of one of the detection channels As the baseline set, the defect set of another detection channel. The set to be matched;

[0096] Calculate the mask intersection-union ratio (CUIR) of the defect masks from two different detection channels in the standard coordinate system.

[0097]

[0098] If only bounding boxes are available, calculate the bounding box intersection-union ratio (IU / UK), and simultaneously introduce center distance weights:

[0099]

[0100] in, Let Euclidean distance be the center of the defects in two different detection channels. The attenuation coefficient is... These are the center distance weights, , These are masks for defects in two different detection channels;

[0101] Set position matching threshold Only retain of Candidate pair.

[0102] First, from the defect sets of two different detection channels, one is selected as the baseline set and the other as the set to be matched. Next, the positional matching degree of the defects in the two channels is calculated. If the defect has a mask (pixel-level outline), the intersection-union ratio (IUU) of the two defect masks is calculated in a unified standard coordinate system (i.e., the ratio of the overlapping area of ​​the mask to the total area of ​​the mask, used to measure the degree of positional overlap). If the defect only has a bounding box (rectangular box label), the IUU of the bounding box is calculated, and the center distance weight is introduced for adjustment. The Euclidean distance (straight-line distance) between the centers of the two defects and the attenuation coefficient jointly affect the center distance weight. Through this weight, the influence of the distance between the defect centers on the matching degree can be further considered on the basis of the IUU (the closer the center, the higher the weight, and the better the overall matching degree). Then, a positional matching threshold is set, and only those defect candidate pairs whose matching degree (after comprehensive calculation of IUU and center distance weight) exceeds the threshold are retained. This determines the corresponding defects in the two detection channels, that is, the positional matching is achieved.

[0103] In summary, the general semiconductor defect detection method in the above embodiments of the present invention acquires multiple defect images of a general semiconductor collected by a detection device in different detection channels, and preprocesses these multiple defect images to obtain a target defect image. A pre-trained instance segmentation model is used to perform defect detection on the target defect image obtained after processing by each detection channel to obtain the basic defect features of the general semiconductor under each detection channel. The basic defect features of the general semiconductor under each detection channel are fused according to preset rules to obtain fused defect features, and then the defect type of the general semiconductor is obtained based on the fused defect features. This avoids the problems of single detection channel acquiring limited information, failing to comprehensively capture multi-dimensional defect features, and defects having insignificant features or being interfered with under a single channel, leading to missed or false detections. Ultimately, it achieves accurate, efficient, and universal semiconductor defect detection. It solves the problem of poor accuracy in general semiconductor defect detection in existing technologies.

[0104] Example 3

[0105] Please see Figure 2 The figure shows a general semiconductor defect detection system proposed in the third embodiment of the present invention, the system comprising:

[0106] The acquisition module 100 is used to acquire multiple defect images of general semiconductors collected by the detection equipment in different detection channels, and to preprocess the multiple defect images to obtain the target defect image;

[0107] The recognition module 200 is used to perform defect detection on the target defect image obtained after acquisition and processing of each detection channel using a pre-trained instance segmentation model, so as to obtain the basic defect features of the general semiconductor under each detection channel.

[0108] The detection module 300 is used to fuse the basic defect features of the general semiconductor under each detection channel according to preset rules to obtain fused defect features, and then obtain the defect type of the general semiconductor based on the fused defect features.

[0109] The functions or operation steps implemented by the above modules are largely the same as those in the above method embodiments, and will not be repeated here.

[0110] Example 4

[0111] In another aspect, the present invention provides a readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of the method described in any one of Embodiments 1 to 2 above.

[0112] Example 5

[0113] In another aspect, the present invention provides an electronic device, the electronic device including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of any one of the methods described in Embodiments 1 to 2 above.

[0114] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0115] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequential list of executable instructions for implementing logical functions, and can be embodied in any computer-readable storage medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable storage medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0116] More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable storage media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0117] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0118] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0119] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A general semiconductor defect detection method characterized by, The method comprises: acquiring multiple defect images of a general semiconductor collected by a detection device in different detection channels, and pre-processing the multiple defect images to obtain target defect images; using a pre-trained instance segmentation model to perform defect detection on the target defect images obtained after processing in each detection channel to obtain basic defect features of the general semiconductor under each detection channel; fusing the basic defect features of the general semiconductor under each detection channel according to a preset rule to obtain fused defect features, and then obtaining a defect type of the general semiconductor according to the fused defect features; the step of fusing the basic defect features of the general semiconductor under each detection channel according to a preset rule to obtain fused defect features comprises: through coordinate conversion, calculating the position coincidence degree of defects in different detection channels to realize position matching of defects in different detection channels; comparing the profile parameters and color features of the defects to realize morphology matching of defects in different detection channels; for defects with directionality, comparing the consistency of their orientations and angles to realize attitude matching of defects in different detection channels; for matched defects in different detection channels, eliminating cross-channel repeated labeling of different detection channels, merging the features of the same defect in different detection channels, and generating unified fused defect features.

2. The general semiconductor defect detection method according to claim 1, wherein The training process of the instance segmentation model is: constructing a semiconductor defect labeling data set to obtain a training data set, wherein the semiconductor defect labeling data set contains a preset number of images and corresponding labeling information; performing multi-dimensional preprocessing on the images in the training data set; constructing a YOLO neural network; inputting the multi-dimensionally preprocessed training data set into the YOLO neural network for deep learning training to obtain the instance segmentation model.

3. The universal semiconductor defect detection method of claim 1, wherein, The step of using the pre-trained instance segmentation model to perform defect detection on the target defect images obtained after processing in each detection channel to obtain basic defect features of the general semiconductor under each detection channel comprises: segmenting the target defect images into multiple target defect image sub-regions according to a preset rule, and using the pre-trained instance segmentation model to perform defect detection on the target defect image sub-regions to obtain basic sub-defect features; based on position information, splicing the basic sub-defect features within the same detection channel to obtain the basic defect features of the general semiconductor.

4. The universal semiconductor defect detection method of claim 1, wherein, Position matching includes: a set of defects of one of the detection channels a set of defects of the other detection channel to be matched; Computing a mask of defects of both different detection channels under a mask intersection ratio in a standard coordinate system if there is only a bounding box, calculating the intersection over union of the bounding box while introducing a center distance weight: wherein is the Euclidean distance of the centers of the defects of the two different detection channels, is the decay coefficient, are the center distance weights, respectively, , are the masks of the defects of the two different detection channels, respectively; Setting a position matching threshold , only keep candidate pairs of .

5. The universal semiconductor defect detection method of claim 4, wherein, The basic defect features include: basic categories and confidence, wherein the confidence represents detection reliability; position features; morphology features, wherein the morphology features include profile features, and whether to extract the profile is selected according to defect size; color features, wherein the color features include brightness mean value, extreme value, and contrast.

6. The universal semiconductor defect detection method of claim 1, wherein, The step of pre-processing the multiple defect images to obtain target defect images comprises: illumination / color standardization: unifying image brightness, contrast, and color distribution, and eliminating the influence of uneven illumination; scale standardization: adjusting the defect images to a preset scale to ensure that the defect size is represented under a unified benchmark; coordinate system standardization: establishing a unified coordinate system to unify the coordinate benchmark of the defect position.

7. A general semiconductor defect detection system characterized by, A system for implementing the general semiconductor defect detection method of any one of claims 1 to 6, the system comprising: a collection module configured to obtain multiple defect images of a general semiconductor collected by a detection device in different detection channels, and to pre-process the multiple defect images to obtain target defect images; an identification module configured to perform defect detection on the target defect images obtained after processing in each detection channel by using a pre-trained instance segmentation model, to obtain basic defect features of the general semiconductor in each detection channel; a detection module configured to fuse the basic defect features of the general semiconductor in each detection channel according to a preset rule to obtain fused defect features, and to obtain a defect type of the general semiconductor according to the fused defect features.

8. A readable storage medium, having stored thereon a computer program, characterized in that, The program, when executed by a processor, implements the steps of the method of any one of claims 1 to 6.

9. An electronic device, comprising: A computer program product, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, wherein the processor implements the steps of the method of any one of claims 1 to 6 when executing the program.

Images