CN117710378B - Wafer surface defect detection method, system and storage medium based on deep learning - Google Patents
Wafer surface defect detection method, system and storage medium based on deep learning Download PDFInfo
- Publication number
- CN117710378B CN117710378B CN202410166653.0A CN202410166653A CN117710378B CN 117710378 B CN117710378 B CN 117710378B CN 202410166653 A CN202410166653 A CN 202410166653A CN 117710378 B CN117710378 B CN 117710378B
- Authority
- CN
- China
- Prior art keywords
- layer
- wafer
- encoder
- data
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 87
- 230000007547 defect Effects 0.000 title claims abstract description 82
- 238000013135 deep learning Methods 0.000 title claims abstract description 30
- 238000003860 storage Methods 0.000 title claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims abstract description 44
- 238000013528 artificial neural network Methods 0.000 claims abstract description 28
- 230000004927 fusion Effects 0.000 claims abstract description 23
- 238000000034 method Methods 0.000 claims abstract description 22
- 230000006870 function Effects 0.000 claims abstract description 15
- 125000004122 cyclic group Chemical group 0.000 claims description 17
- 239000013598 vector Substances 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 12
- 230000002950 deficient Effects 0.000 claims description 7
- 238000011176 pooling Methods 0.000 claims description 6
- 230000002708 enhancing effect Effects 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 4
- 230000007246 mechanism Effects 0.000 claims description 4
- 238000007689 inspection Methods 0.000 claims 3
- 230000006872 improvement Effects 0.000 abstract description 4
- 238000004422 calculation algorithm Methods 0.000 abstract description 3
- 235000012431 wafers Nutrition 0.000 description 69
- 230000008569 process Effects 0.000 description 9
- 238000004519 manufacturing process Methods 0.000 description 6
- 239000013078 crystal Substances 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000005229 chemical vapour deposition Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000011282 treatment Methods 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Testing Or Measuring Of Semiconductors Or The Like (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a wafer surface defect detection method, a system and a storage medium based on deep learning, wherein the defect detection method comprises the following steps: acquiring a wafer image; detecting the wafer image by using the trained neural network detection model, and outputting a defect detection result; the neural network detection model includes a multi-layer encoder-decoder structure, wherein an upper layer encoder-decoder incorporates a modified attention structure that generates a key matrix using multi-layer fusion dataMatrix of valuesThe multi-layer fusion data comprises data generated by a network of the layer and data output by the network before the layer. The novel deep learning algorithm structure is provided, and the detection and identification precision of small targets is improved by the methods of data enhancement, improvement of a neural network structure, application of a novel loss function and the like, so that the detection accuracy of various defects on the surface of a wafer is improved, and the purpose of accurately detecting the defects of a chip wafer is achieved.
Description
Technical Field
The invention belongs to the technical field of chip manufacturing, and relates to a wafer surface defect detection method, a wafer surface defect detection system and a storage medium based on deep learning.
Background
The wafer is the basic material for manufacturing chips, and the processing quality directly determines the yield and quality of subsequent chip products. In the chip production process, the wafer processing technology is complex in process and the production cost is high. The atoms are not present in strict periodic arrangement according to crystal lattice, and may have different kinds and types of defects in the processes of crystal growth, chemical vapor deposition, etc. under the influence of various factors such as process, raw materials, etc. Meanwhile, the surface defects of the wafer are usually very tiny, the shape and the structure are various and complex, and the current detection method is easy to generate the conditions of missing detection, false detection and the like. At present, a scanning electron microscope and an automatic optical system are generally used for detecting defects in a manual mode, so that the cost is high, the efficiency is low, and the accuracy is difficult to guarantee.
In order to improve the accuracy of chip wafer defect detection and the quality of chip devices, a deep learning model is a solution, and application number 2022109542448 discloses a microscopic defect processing method, a device and computer equipment based on image recognition, wherein the deep learning model is constructed based on a decoder-encoder network structure, and an encoder and a decoder fuse feature images with the same size in the two in a splicing mode and carry out defect type statistics on recognition results. The detection and identification precision of the method on the small target is still to be improved.
Disclosure of Invention
The invention aims to provide a wafer surface defect detection method, a wafer surface defect detection system and a storage medium based on deep learning, which are improved in a neural network structure and use a multi-layer encoder-decoder structure, wherein a plurality of layers of encoder-decoders are added with an improved attention structure and are generated by using multi-layer fusion data、/>I.e. generate/>、/>The data of the network layer not only comprises the data generated by the network of the layer, but also comprises the data output by the network before the layer, so that the detection and identification precision of the small target is improved, and the detection accuracy of various defects on the surface of the wafer is improved.
The technical solution for realizing the purpose of the invention is as follows:
A wafer surface defect detection method based on deep learning comprises the following steps:
S01: acquiring a wafer image;
s02: detecting the wafer image by using the trained neural network detection model, and outputting a defect detection result; the neural network detection model includes a multi-layer encoder-decoder structure, wherein an upper layer encoder-decoder incorporates a modified attention structure that generates a key matrix using multi-layer fusion data Value matrix/>The multi-layer fusion data comprises data generated by a network of the layer and data output by the network before the layer.
In a preferred technical solution, the calculation formula of the multi-layer fusion data in step S02 is as follows:
Wherein, 、/>Index for network layer,/>Is the total layer number of the network,/>For the output of the encoder,/>For the output of the decoder,/>Is a feature data stacking operation.
In a preferred technical scheme, the encoder-decoder structure is 4 layers, and the layer 1 and 2 encoders consist of 2 convolution layers and 1 maximum pooling layer; layer 3, 4 encoders add attention-based modules on the basis of layer 1, 2 encoders, each decoder consisting of a modified attention structure, 1 upsampling layer, 2 convolutional layers.
In a preferred technical solution, when training the neural network detection model in step S02, image data enhancement is performed on a wafer defect image of training data, and an image enhancement channel is used to perform preprocessing on the image, where the preprocessing method includes:
Processing the image using the cyclic matrix, and setting a certain line of data in the wafer image as ，/>For the size of the row vector, the circulant matrix/>The formula is as follows:
Will be Acting on/>Obtain the vector/>, after cyclic shiftThe following formula:
Wherein, Is the shift number of bits;
For a pair of images After processing by using the cyclic matrix, different shifted pictures are obtained, and the calculation formula is as follows:
enhancing wafer image data using rotation and random scaling on a cyclic basis, rotation matrix And random scaling matrix/>The following formulas are respectively shown:
Wherein, For the rotation angle,/>、/>Scaling factors in the horizontal and vertical directions of the image, respectively.
In a preferred technical solution, the loss function when training the neural network detection model in step S02 is:
Wherein, Is the area of the wafer disc,/>Is the area of a defective area,/>For the defect index to be used,A predictive confidence score representing the candidate target;
For parameters And (3) carrying out dynamic adjustment, and adaptively reducing the loss contribution of the background in the sample.
The invention also discloses a wafer surface defect detection system based on deep learning, which comprises:
The wafer image acquisition module to be detected acquires a wafer image;
The defect detection module is used for detecting the wafer image by using the trained neural network detection model and outputting a defect detection result; the neural network detection model includes a multi-layer encoder-decoder structure, wherein an upper layer encoder-decoder incorporates a modified attention structure that is generated using multi-layer fusion data 、/>And the matrix is used for integrating the data generated by the network of the layer and the data output by the network before the layer.
In a preferred technical solution, the calculation formula of the multi-layer fusion data in the defect detection module is as follows:
Wherein, 、/>Index for network layer,/>Is the total layer number of the network,/>For the output of the encoder,/>For the output of the decoder,/>Is a feature data stacking operation.
In a preferred technical scheme, the encoder-decoder structure is 4 layers, and the layer 1 and 2 encoders consist of 2 convolution layers and 1 maximum pooling layer; layer 3, 4 encoders add attention-based modules on the basis of layer 1, 2 encoders, each decoder consisting of a modified attention structure, 1 upsampling layer, 2 convolutional layers.
In a preferred technical scheme, the defect detection module further comprises a training module, and the loss function of the training module when training the neural network detection model is as follows:
Wherein, Is the area of the wafer disc,/>Is the area of a defective area,/>For the defect index to be used,A predictive confidence score representing the candidate target;
For parameters And (3) carrying out dynamic adjustment, and adaptively reducing the loss contribution of the background in the sample.
The invention also discloses a computer storage medium, on which a computer program is stored, which when executed, implements the wafer surface defect detection method based on deep learning.
Compared with the prior art, the invention has the remarkable advantages that:
Improved neural network architecture, new deep learning algorithm architecture is proposed using a multi-layer encoder-decoder architecture, wherein several layers of encoder-decoders add improved attention structure, key matrix generation using multi-layer fusion data Value matrix/>I.e. generating a key matrix/>Value matrix/>The data of the network layer not only comprises the data generated by the network of the layer, but also comprises the data output by the network before the layer, so that the detection and identification precision of the small target is improved, and the detection accuracy of various defects on the surface of the wafer is improved.
The detection model is pertinently trained by methods such as data enhancement and new loss function application, so that the detection and identification precision of small targets is further improved, the purpose of improving the accuracy of chip wafer defect detection is achieved, the reliability and stability of chip product production are improved, and the method has wide application market space and economic value.
Drawings
FIG. 1 is a flow chart of a wafer surface defect detection method based on deep learning in accordance with a preferred embodiment;
Fig. 2 is a deep learning network architecture diagram of the present embodiment;
FIG. 3 is a schematic diagram of a CA module configuration;
FIG. 4 is a schematic diagram of an improved attention structure of the present invention;
Fig. 5 is a diagram showing the effect of detecting surface defects of a chip wafer.
Detailed Description
The principle of the invention is as follows: firstly, high-precision imaging is carried out on a wafer by utilizing a microscopic vision technology, and then, feature extraction, detection and identification of the surface defects of the wafer are finished by methods such as image enhancement, neural network detection and identification and the like. In order to achieve the purpose of accurately identifying the wafer defects, a new deep learning algorithm structure is provided, and the detection and identification precision of small targets is improved by the methods of data enhancement, improvement of a neural network structure, application of a new loss function and the like, so that the detection accuracy of various defects on the surface of the wafer is improved, and the purpose of accurately detecting the wafer defects of the chip is achieved.
Example 1:
As shown in fig. 1, a wafer surface defect detection method based on deep learning includes the following steps:
S01: acquiring a wafer image;
s02: detecting the wafer image by using the trained neural network detection model, and outputting a defect detection result; the neural network detection model includes a multi-layer encoder-decoder structure, wherein an upper layer encoder-decoder incorporates a modified attention structure that generates a key matrix using multi-layer fusion data Value matrix/>The multi-layer fusion data comprises data generated by a network of the layer and data output by the network before the layer.
The wafer image acquisition in step S01 may be to image the wafer with high accuracy using microscopic vision techniques. Or other methods, the present embodiment is not limited.
In a preferred embodiment, the calculation formula of the multi-layer fusion data in step S02 is as follows:
Wherein, 、/>Index for network layer,/>Is the total layer number of the network,/>For the output of the encoder,/>For the output of the decoder,/>Is a feature data stacking operation.
In a preferred embodiment, the encoder-decoder structure is 4 layers, and the layer 1, 2 encoder consists of 2 convolutional layers and 1 max-pooling layer; layer 3, 4 encoders add attention-based modules on the basis of layer 1, 2 encoders, each decoder consisting of a modified attention structure, 1 upsampling layer, 2 convolutional layers.
In a preferred embodiment, when training the neural network detection model in step S02, image data enhancement is performed on the wafer defect image of the training data, and the image is preprocessed by using the image enhancement channel, where the preprocessing method includes:
Processing the image using the cyclic matrix, and setting a certain line of data in the wafer image as ，/>For the size of the row vector, the circulant matrix/>The formula is as follows:
Will be Acting on/>Obtain the vector/>, after cyclic shiftThe following formula:
Wherein, Is the shift number of bits;
For a pair of images After processing by using the cyclic matrix, different shifted pictures are obtained, and the calculation formula is as follows:
enhancing wafer image data using rotation and random scaling on a cyclic basis, rotation matrix And random scaling matrix/>The following formulas are respectively shown:
Wherein, For the rotation angle,/>、/>Scaling factors in the horizontal and vertical directions of the image, respectively.
In a preferred embodiment, the loss function of training the neural network detection model in step S02 is:
Wherein, Is the area of the wafer disc,/>Is the area of a defective area,/>For the defect index to be used,A predictive confidence score representing the candidate target;
For parameters And (3) carrying out dynamic adjustment, and adaptively reducing the loss contribution of the background in the sample.
In another embodiment, a computer storage medium has a computer program stored thereon, which when executed implements the deep learning-based wafer surface defect detection method described above. The above detection method is adopted, and will not be described here.
In another embodiment, a wafer surface defect detection system based on deep learning includes:
The wafer image acquisition module to be detected acquires a wafer image;
The defect detection module is used for detecting the wafer image by using the trained neural network detection model and outputting a defect detection result; the neural network detection model includes a multi-layer encoder-decoder structure, wherein an upper layer encoder-decoder incorporates a modified attention structure that generates a key matrix using multi-layer fusion data Matrix of valuesThe multi-layer fusion data comprises data generated by a network of the layer and data output by the network before the layer.
Specifically, the following description will be given by taking a preferred embodiment as an example for the workflow of the wafer surface defect detection system based on deep learning as follows:
Step one: and (5) realizing a wafer surface defect detection deep learning network model.
Step 11: construction of deep learning network model
In the production process of wafers, atoms are not arranged strictly according to the periodicity of crystal lattices, and in the processes of crystal growth, chemical vapor deposition and the like, different types of defects can exist. Meanwhile, the surface defects of the wafer are usually very tiny, the shape and the structure are various and complex, and the condition of missing detection, false detection and the like easily occurs when the detection is carried out by naked eyes.
Therefore, the invention provides a deep learning neural network detection model for semantic segmentation of wafer images and high-precision detection of surface defects, and the model structure is shown in fig. 2. The model designed by the invention mainly comprises an encoder, a decoder and the like, wherein the encoder and the decoder are corresponding and are all 4 layers. The encoder comprises two structures, wherein the layer 1 and 2 encoder consists of 2 convolution layers of 3×3 and 1 max pooling layer of 2×2, and the layer 3 and 4 encoder is added with CA (Coordinate Attention) module based on attention mechanism based on the layer 1 and 2 encoder. The attention mechanism can better extract the image key information, the 1 st layer and the 2 nd layer of the model mainly pay attention to the low-dimensional information of the wafer defect, such as edges, corner points and the like, and the 3 rd layer and the 4 th layer mainly pay attention to the high-dimensional information of the defect along with the deep sampling, so that the detection capability of the high-dimensional information is improved by adding a CA module, and the structure of the CA module is shown in figure 3. Each decoder consists essentially of an improved attention structure, 12 x 2 up-sampling layer, 23 x 3 convolutional layers, wherein in contrast to the conventional attention structure, the multi-layer fusion feature is used as input in the present invention, as shown in fig. 4.
The calculation of the attention structure is shown in formula (1):
（1）
Wherein, 、/>、/>Is a data matrix obtained by multiplying the embedded vector by a parameter matrix,/>Is a query matrix,Is a key matrix,/>For a value matrix, embedding vectors is used to convert the original 2-dimensional image into a series of 1-dimensional vector data,/>Is the dimension of the parameter matrix.
For the activation function, its calculation is as shown in equation (2):
（2）
Wherein the method comprises the steps of Is the total number of types of data,/>Type index,/>Is a natural logarithmic constant,/>Vector composed of related data of a certain type,/>Is a vector index.
In a conventional attention structure, for generating、/>、/>The source data of the matrix are the same, and in order to improve the perceptibility of the wafer image data, the invention uses multi-layer fusion data generation/>、/>I.e. generate/>、/>The data of the network of the layer not only comprises the data generated by the network of the layer, but also comprises the data output by the network before the layer, and the structure of the data is shown in figure 4. Multilayer fusion data used by the attention structure in the present invention/>The calculation of (2) is shown as a formula (3), wherein/>、/>Index for network layer,/>Is the total layer number of the network,/>For the output of the encoder,/>For the output of the decoder,/>Is a feature data stacking operation.
（3）
Step 12: enhancement of chip wafer defect images
The surface defects of the chip wafer have various and tiny characteristics, relevant defect data are difficult to acquire in engineering practice, and the quantity of the defect data is small, so that the defect data are insufficient in deep learning network training. In order to expand the defect picture data and improve the accuracy of defect detection, the invention firstly uses a data enhancement technology to process the wafer image. In order to efficiently complete the enhancement of image data, the invention uses an image enhancement channel to carry out image generation to generate images covering various angles and distances, wherein the image enhancement channel mainly comprises operations such as circulation, rotation, random scaling and the like.
The cyclic operation uses the cyclic matrix to process the image, and one line of data in the wafer image is set as，/>For the size of the row vector, the circulant matrix/>As shown in formula (4), will/>Acting on/>Can obtain the vector/>, after cyclic shiftAs shown in formula (5)/(Is the number of shift bits. Thus for a pair of images/>After processing using the cyclic matrix, a shifted different picture can be obtained, the calculation of which is shown in equation (6).
（4）
（5）
（6）
Enhancing wafer image data using rotation and random scaling on a cyclic basis, rotation matrixAnd random scaling matrix/>As shown in the formula (7) and the formula (8), respectively, wherein/>For the rotation angle,/>、/>Scaling factors in the horizontal and vertical directions of the image, respectively.
（7）
（8）
Step 13: loss function for wafer surface defect detection
Because the wafer defect image is difficult to acquire and the ratio of defects on the wafer background is generally smaller, the problem of sample unbalance is easy to generate in the process of training and reasoning the deep learning network model, and therefore, the invention provides targeted improvement on the basis of the Focal Loss function so as to relieve the problem of sample unbalance of the wafer defects and improve the model detection precision.
The Focal Loss function is a solution to the foreground-background imbalance problem, as shown in equation (9), in whichRepresenting predictive confidence scores for candidate targets,/>Is a parameter of the Focal Loss function.
（9）
In order to relieve the problem of unbalance of wafer defect samples and improve the detection precision of a model, the loss function provided by the inventionAs shown in formula (10), wherein/>Is the area of the wafer disc,/>Is the area of a defective area,/>Is a defect index. By matching parameters/>The dynamic adjustment of (3) adaptively reduces the loss contribution of background in the sample, thereby improving the impact of defective samples.
（10）
Step two: training and detection of deep learning network models.
The training process of the network model comprises the steps of preprocessing a wafer picture, enhancing image data, inputting the enhanced image data into a network, processing the image data by each layer of encoder, inputting a result into a decoder of a corresponding layer for decoding, simultaneously receiving the processing results of the lower layer decoder and each level of encoder of the upper layer of the corresponding encoder by each decoder for decoding, carrying out error counter-propagation according to an improved loss function, and finally completing training. The reasoning about wafer defects is a forward propagation process, and data enhancement and other treatments are not needed, and the treatment of image wafer defects is shown in fig. 5.
The improvement of the present invention in chip wafer defect detection was examined by processing the same dataset using different models, the results are shown in table 1:
Table 1 comparison of different methods for chip wafer defect detection
From the results, the deep learning neural network model provided by the invention has obvious progress in the aspects of precision, speed and the like.
The foregoing examples are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the foregoing examples, and any other changes, modifications, substitutions, combinations, and simplifications that do not depart from the spirit and principles of the present invention should be made therein and are intended to be equivalent substitutes within the scope of the present invention.
Claims (6)
1. The wafer surface defect detection method based on deep learning is characterized by comprising the following steps of:
S01: acquiring a wafer image;
s02: detecting the wafer image by using the trained neural network detection model, and outputting a defect detection result; the neural network detection model includes a multi-layer encoder-decoder structure, wherein an upper layer encoder-decoder incorporates a modified attention structure that generates a key matrix using multi-layer fusion data Value matrix/>The encoder-decoder structure is 4 layers, and the 1 st layer encoder and the 2 nd layer encoder are composed of 2 convolution layers and 1 largest pooling layer; the 3 rd layer encoder and the 4 th layer encoder are added with modules based on attention mechanisms on the basis of the 1 st layer encoder and the 2 nd layer encoder, and each decoder consists of an improved attention structure, 1 upsampling layer and 2 convolution layers; the multi-layer fusion data comprises data generated by a network of the layer and data output by the network before the layer; the calculation formula of the multi-layer fusion data is as follows: /(I)，
Wherein,、/>Index for network layer,/>Is the total layer number of the network,/>For the output of the encoder,/>For the output of the decoder,/>Is a feature data stacking operation.
2. The wafer surface defect detection method based on deep learning according to claim 1, wherein when training the neural network detection model in step S02, image data enhancement is performed on a wafer defect image of training data, and the image is preprocessed by using an image enhancement channel, and the preprocessing method includes:
Processing the image using the cyclic matrix, and setting a certain line of data in the wafer image as ，For the size of the row vector, the circulant matrix/>The formula is as follows:
，
Will be Acting on/>Obtain the vector/>, after cyclic shiftThe following formula:
，
Wherein, Is the shift number of bits;
For a pair of images After processing by using the cyclic matrix, different shifted pictures are obtained, and the calculation formula is as follows:
，
enhancing wafer image data using rotation and random scaling on a cyclic basis, rotation matrix And random scaling matrix/>The following formulas are respectively shown:
，
，
Wherein, For the rotation angle,/>、/>Scaling factors in the horizontal and vertical directions of the image, respectively.
3. The wafer surface defect detection method based on deep learning according to claim 1, wherein the loss function when training the neural network detection model in step S02 is:
，
Wherein, Is the area of the wafer disc,/>Is the area of a defective area,/>Index for defect,/>A predictive confidence score representing the candidate target;
For parameters And (3) carrying out dynamic adjustment, and adaptively reducing the loss contribution of the background in the sample.
4. A wafer surface defect detection system based on deep learning, comprising:
The wafer image acquisition module to be detected acquires a wafer image;
The defect detection module is used for detecting the wafer image by using the trained neural network detection model and outputting a defect detection result; the neural network detection model includes a multi-layer encoder-decoder structure, wherein an upper layer encoder-decoder incorporates a modified attention structure that generates a key matrix using multi-layer fusion data Value matrix/>The encoder-decoder structure is 4 layers, and the 1 st layer encoder and the 2 nd layer encoder are composed of 2 convolution layers and 1 largest pooling layer; the 3 rd layer encoder and the 4 th layer encoder are added with modules based on attention mechanisms on the basis of the 1 st layer encoder and the 2 nd layer encoder, and each decoder consists of an improved attention structure, 1 upsampling layer and 2 convolution layers; the multi-layer fusion data comprises data generated by a network of the layer and data output by the network before the layer; the calculation formula of the multi-layer fusion data is as follows:
，
Wherein, 、/>Index for network layer,/>Is the total layer number of the network,/>For the output of the encoder,/>For the output of the decoder,/>Is a feature data stacking operation.
5. The deep learning based wafer surface defect inspection system of claim 4, wherein the defect inspection module further comprises a training module, the training module training the neural network inspection model to have a loss function of:
，
Wherein, Is the area of the wafer disc,/>Is the area of a defective area,/>Index for defect,/>A predictive confidence score representing the candidate target;
For parameters And (3) carrying out dynamic adjustment, and adaptively reducing the loss contribution of the background in the sample.
6. A computer storage medium having stored thereon a computer program, wherein the computer program when executed implements the deep learning based wafer surface defect detection method of any of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410166653.0A CN117710378B (en) | 2024-02-06 | 2024-02-06 | Wafer surface defect detection method, system and storage medium based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410166653.0A CN117710378B (en) | 2024-02-06 | 2024-02-06 | Wafer surface defect detection method, system and storage medium based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117710378A CN117710378A (en) | 2024-03-15 |
| CN117710378B true CN117710378B (en) | 2024-04-30 |
Family
ID=90150172
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410166653.0A Active CN117710378B (en) | 2024-02-06 | 2024-02-06 | Wafer surface defect detection method, system and storage medium based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117710378B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113344886A (en) * | 2021-06-11 | 2021-09-03 | 长江存储科技有限责任公司 | Wafer surface defect detection method and equipment |
| CN113362320A (en) * | 2021-07-07 | 2021-09-07 | 北京工业大学 | Wafer surface defect mode detection method based on deep attention network |
| CN115311203A (en) * | 2022-06-29 | 2022-11-08 | 上海电力大学 | Method and system for detecting wafer defects based on Transformer |
| CN115409824A (en) * | 2022-09-06 | 2022-11-29 | 长沙理工大学 | Silicon wafer surface defect detection method based on deep convolutional neural network |
| CN116778235A (en) * | 2023-06-12 | 2023-09-19 | 北京石油化工学院 | Wafer surface defect classification method based on deep learning network |
-
2024
- 2024-02-06 CN CN202410166653.0A patent/CN117710378B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113344886A (en) * | 2021-06-11 | 2021-09-03 | 长江存储科技有限责任公司 | Wafer surface defect detection method and equipment |
| CN113362320A (en) * | 2021-07-07 | 2021-09-07 | 北京工业大学 | Wafer surface defect mode detection method based on deep attention network |
| CN115311203A (en) * | 2022-06-29 | 2022-11-08 | 上海电力大学 | Method and system for detecting wafer defects based on Transformer |
| CN115409824A (en) * | 2022-09-06 | 2022-11-29 | 长沙理工大学 | Silicon wafer surface defect detection method based on deep convolutional neural network |
| CN116778235A (en) * | 2023-06-12 | 2023-09-19 | 北京石油化工学院 | Wafer surface defect classification method based on deep learning network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117710378A (en) | 2024-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109829903B (en) | Chip surface defect detection method based on convolution denoising autoencoder | |
| CN111932511B (en) | Electronic component quality detection method and system based on deep learning | |
| CN110570433B (en) | Image semantic segmentation model construction method and device based on generation countermeasure network | |
| CN115880298B (en) | Glass surface defect detection system based on unsupervised pre-training | |
| CN104615911B (en) | Method based on sparse coding and chain study prediction memebrane protein beta barrel trans-membrane regions | |
| CN113673509B (en) | Instrument detection classification method based on image text | |
| CN114757938B (en) | Transformer oil leakage identification method and system | |
| CN116994140A (en) | Cultivated land extraction method, device, equipment and medium based on remote sensing image | |
| CN114862814A (en) | Solar cell panel defect detection method and system, storage medium and terminal | |
| CN116612106A (en) | Method for detecting surface defects of optical element based on YOLOX algorithm | |
| CN113870286A (en) | Foreground segmentation method based on multi-level feature and mask fusion | |
| CN113554630A (en) | Chip surface defect detection method, system, computer device and storage medium | |
| CN110633706B (en) | Semantic segmentation method based on pyramid network | |
| CN115661578A (en) | Industrial defect image generation method | |
| CN117710378B (en) | Wafer surface defect detection method, system and storage medium based on deep learning | |
| CN114882204A (en) | Automatic ship name recognition method | |
| CN114332047A (en) | Construction method and application of surface defect detection model | |
| CN116894839B (en) | Chip wafer defect detection method, device, terminal equipment and storage medium | |
| CN116206248B (en) | Target detection method based on machine learning guide deep learning | |
| CN111595737B (en) | Optical holographic particle field particle point detection method based on three-dimensional branch network | |
| Wang et al. | MeDERT: A metal surface defect detection model | |
| CN116912144A (en) | Data enhancement method based on discipline algorithm and channel attention mechanism | |
| Sun et al. | Underwater Object Detection with Swin Transformer | |
| CN114049939A (en) | Pneumonia CT image generation method based on UNet-GAN network | |
| CN113034432A (en) | Product defect detection method, system, device and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |