CN116188877B - Method and system for detecting and classifying unknown wafer defect categories - Google Patents
Method and system for detecting and classifying unknown wafer defect categories Download PDFInfo
- Publication number
- CN116188877B CN116188877B CN202310438044.1A CN202310438044A CN116188877B CN 116188877 B CN116188877 B CN 116188877B CN 202310438044 A CN202310438044 A CN 202310438044A CN 116188877 B CN116188877 B CN 116188877B
- Authority
- CN
- China
- Prior art keywords
- model
- wafer
- defect
- training
- defects
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30148—Semiconductor; IC; Wafer
-
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Testing Or Measuring Of Semiconductors Or The Like (AREA)
Abstract
The invention discloses a method and a system for detecting and classifying unknown categories of wafer defects, which relate to the field of semiconductor industrial defect detection, and comprise the following steps: converting wafer map data acquired by the wafer production line AOI equipment into data set data for model training by using a wafer map preprocessing module; building a convolutional neural network model based on an incremental learning method; training a convolutional neural network model; parameter compression and dimension reduction are carried out on the original network structure of the model through a TensorRT technology; and deploying the compressed model into an AOI (automated optical inspection) of a wafer production line of a semiconductor factory, and accurately classifying unknown defect types. The system comprises a convolutional neural network structure model based on ResNet 50; online incremental learning PathInt strategy; the loss function adjustment module and the incremental learning model. The invention improves the accuracy and the efficiency of wafer surface defect detection.
Description
Technical Field
The invention relates to the field of semiconductor industrial defect detection, in particular to a method and a system for detecting and classifying unknown categories of wafer defects.
Background
As one of the key raw materials in the modern semiconductor industry, the wafer has become a hot spot technology issue of global widespread attention in the context of rapid increase of chip demands in the fields of artificial intelligence, electronic products, automotive electronics, and the like. Therefore, increasing the overall productivity of the semiconductor industry has become a new development goal. Among them, wafer surface defects are one of the main factors affecting yield. By accurately identifying and classifying the surface defects of the wafer, defective wafers can be prevented from entering subsequent processes, and professionals are helped to locate fault links according to reliable results, and find out faults and optimizable links existing in a manufacturing system.
In recent years, with the continuous progress of wafer production technology, the updating of manufacturing equipment, and complex and fine manufacturing processes, the morphology of wafer surface defects is increasingly irregular and randomly distributed. Compared with the traditional manual detection method, the automatic optical detection (AOI) device greatly shortens the defect identification time and remarkably improves the identification accuracy. However, for unknown defects occurring during wafer fabrication, AOI equipment cannot be accurately classified due to technical limitations.
Therefore, it is highly desirable to adopt deep learning and incremental learning methods in the wafer surface defect detection and classification process to realize rapid and accurate automatic detection and classification of unknown defects. Therefore, the production of defective semiconductor products can be avoided, the manufacturing cost caused by defects is reduced, and the optimization and adjustment of the wafer manufacturing system are facilitated. Ultimately, this will promote an increase in the overall yield and throughput of the semiconductor manufacturing industry.
Disclosure of Invention
The invention aims to provide a method and a system for detecting and classifying unknown categories of wafer defects, which are used for solving the problems in the background art.
In order to achieve the above purpose, the invention adopts the following technical scheme:
a method for detecting and classifying unknown wafer defect categories comprises the following steps:
(a) Converting wafer map data acquired by the wafer production line AOI equipment into data set data for model training by using a wafer map preprocessing module;
(b) Building a convolutional neural network model based on an incremental learning method; the model comprises:
(i) Convolutional neural network structural model based on ResNet 50;
(ii) Online incremental learning PathInt strategy;
(iii) The loss function remains close to the learning parameters of the old defect type;
(iv) The incremental learning model is used for classifying unknown defects;
(c) Training a convolutional neural network model, comprising:
(i) Reworking a basic wafer defect data set, and randomly dividing according to defect types;
(ii) Training models by using multiple data sets, and training and testing the defect type incremental input models;
(iii) Using a lightweight cache to store known defects, and keeping N defect samples with equal quantity in each category;
(iv) Repeating the model training and testing process for a plurality of times until a stable model is obtained;
(d) Parameter compression and dimension reduction are carried out on the original network structure of the model through a TensorRT technology;
(e) And deploying the compressed model into an AOI (automated optical inspection) of a wafer production line of a semiconductor factory, and accurately classifying unknown defect types.
Preferably, step (a) comprises: collecting wafer graphs containing annular scratches, unifying the sizes of all the wafer graphs, and performing image processing on the wafer graphs in an image compression, enhancement and noise reduction mode to obtain a standardized wafer graph containing defects; and adding corresponding labels to the image data according to the obvious characteristics of different defect types to manufacture a wafer defect data set.
Preferably, in step (b) (ii), the weight that has a significant impact on the old defect task classification is constrained, and when learning the second defect type, a learning trajectory is calculated in the parameter space that minimizes the loss function L while remaining in the low-loss region of the first defect type, thereby protecting the performance of the model on the first defect type.
Preferably, in step (c) (iii), the model maintains the loss function close to the learning parameters of the old defect type when the ith defect type is learned.
Preferably, the method is characterized in that:
in step (c) (iv), a lightweight cache is used to save known defects; n defect samples are saved in an equivalent manner in each category, when the ith sample is detected, the defects are added into the corresponding category if the defects are known, and if the defects are unknown, a new defect category is created; when the sample of the j-th defect class is larger than N, deleting the old sample, and keeping that N defect samples exist in equal quantity in each class.
Preferably, the basic wafer defect data set in step (c) (i) is rearranged and split to generate a plurality of new wafer map data sets with different structures.
The invention also discloses a system for detecting and classifying unknown wafer defects, which comprises:
the wafer map preprocessing module is used for converting wafer map data acquired by the wafer production line AOI equipment into data set data for model training;
convolutional neural network model based on incremental learning method;
the model training module is used for processing and inputting the model to perform incremental training and testing;
the TensorRT optimizing module is used for carrying out parameter compression and dimension reduction on the trained model;
the deployment module is used for deploying the optimized model to the AOI equipment of the wafer production line of the semiconductor factory, collecting the defect wafer map in real time, and carrying out image preprocessing and incremental learning.
Preferably, the convolutional neural network model based on the incremental learning method comprises:
convolutional neural network structural model based on ResNet 50;
an online incremental learning PathInt strategy for constraining weights in the model; the technical content of the PathInt strategy can be found in the paper: on Learning the Geodesic Path for Incremental Learning (https:// arxiv. Org/abs/2104.08572);
the loss function adjusting module is used for keeping learning parameters close to the old defect type when the new defect type is learned in an incremental mode;
and an incremental learning model for classification of unknown defects.
Preferably, the model training module includes:
the data set processing module is used for carrying out redoing on the basic wafer defect data set;
the multi-data set training module is used for training models by using various data sets and training and testing the defect type incremental input model;
the lightweight cache module is used for storing known defects, and N defect samples are stored in equal quantity in each category;
the model evaluation module is used for repeating the model training and testing process for a plurality of times and evaluating the classification result of each model.
The invention has the advantages compared with the prior art that:
1. the invention adopts deep learning, artificial intelligence application and computer vision technology based on the incremental learning algorithm, realizes accurate detection and classification of unknown defects of the wafer, and improves the accuracy and efficiency of wafer surface defect detection.
2. Aiming at the problems of unknown defect detection and classification in the wafer manufacturing process, the invention provides an incremental learning-based wafer unknown defect automatic detection and classification method which can more effectively solve the problems of unknown defect detection and classification compared with the existing solutions in the industry.
3. The invention adopts an online incremental learning method to establish a convolutional neural network model, and can quickly achieve the best performance of the model by a newly proposed light-weight caching mechanism and a compound training mode, thereby realizing the efficient learning and classification of unknown defect types.
4. After the trained model is accelerated and optimized, the model can be deployed to AOI equipment of a wafer production line, and defective wafer diagrams are collected in real time and are subjected to incremental learning. By updating the model on line, accurate classification of unknown defect types is achieved, thereby facilitating optimization of the wafer manufacturing process.
5. The automatic detection and classification method for the unknown defects of the wafer is beneficial to improving the yield of wafer products, greatly reduces the manufacturing cost of semiconductors and brings practical economic benefits for the semiconductor industry.
In summary, the wafer surface defect classification method based on incremental learning effectively solves the problems of unknown defect detection and classification in the wafer manufacturing process, improves the yield of wafer products and reduces the manufacturing cost of semiconductors.
Drawings
FIG. 1 is a general flow chart of the method of the present invention;
FIG. 2 is a flow chart of the present invention for establishing a wafer surface defect classification model by incremental learning.
Description of the embodiments
The following describes specific embodiments of the present invention with reference to the drawings.
As shown in fig. 1, the method of the present invention comprises the steps of:
step S1: and converting the wafer map data acquired by the wafer production line AOI equipment into data set data for model training by using a wafer map preprocessing module. And collecting wafer maps containing common defect types such as annular defects, scratches and the like, unifying the sizes of all the wafer maps, and performing image processing on the wafer maps in modes such as image compression, enhancement, noise reduction and the like to obtain a standardized wafer map containing defects. And finally, adding corresponding labels to the image data according to the obvious characteristics of different defect types to manufacture a wafer defect data set.
Step S2: the method comprises the steps of establishing a convolutional neural network model based on an incremental learning method, classifying unknown defect types of a continuously-occurring wafer based on deep learning and computer vision technology, and establishing a wafer surface defect classification model through incremental learning, wherein the steps are shown in fig. 2:
s21: a convolutional neural network structural model based on the res net50 is built for wafer defect class classification.
S22: and adopting an online incremental learning PathInt strategy to restrict the weight which has great influence on the classification of the old defect task in the model, and calculating a learning track which enables the loss function L to reach the minimum value in a parameter space when learning the second defect type, and simultaneously keeping the learning track in a low-loss area of the first defect type, thereby protecting the performance of the model on the first defect type. The calculated path integral can be expressed as:
where g (θ (t)) represents the gradient of the parameter θ at time t during learning, and θ (t) represents the model parameter at time t.
t 1 Indicating the start time of the learning process.
t 2 Indicating the end time of the learning process.
θ (t)' represents the derivative of the parameter θ with respect to t at time t.
C represents a learned trajectory integration path.
dθ and dt represent the integral variables of the left and right integrals, respectively, in the above formula.
S23: on the basis of S22, in order for the model not to forget the old defect type when incrementally learning the new defect type, the model keeps the loss function close to the learning parameters of the old defect type when learning the ith defect type. In this process, the loss function of the problem can be expressed as:
L i representing a loss function comprising i defect types.
L i-1 Representing a loss function containing i-1 defect types.
θ k i-1 The kth parameter on the ith-1 defect type is indicated.
θ k i Representing the kth parameter on the ith defect type.
Ω i k The importance of the weight of the kth parameter on the ith defect type based on path integration is represented.
S24: based on the correlation function and the target of S23, an incremental learning model is built for classification of unknown defects.
Step S3: after the model structure is built, a training model is started, the wafer map data set acquired in the step S1 is manufactured and divided again, and the model is input for incremental training and testing. In the training process, part of old defect type samples are selected and stored in a lightweight cache, and are mixed with new defect sample types and then used in the model updating process. Until the defect types input in an incremental way are accurately classified, training steps of the wafer unknown defect classification model are as follows:
s31: and (3) reworking the basic wafer defect data set in the step (S1), generating a plurality of new wafer map data sets with different structures through different modes such as rearrangement, splitting and the like, and carrying out random division according to defect types.
S32: and training and testing the defect type incremental input model by using a plurality of data sets to train the model, and continuously repeating to optimize the parameters of the network model.
S33: on the basis of S32, S33 uses a lightweight cache to save known defects in order for the model not to forget old defect types when incrementally learning new defect types. N defect samples were saved in equal amounts for each class. When detecting the ith sample, adding to the corresponding class if the defect is known, and creating a new defect class if the defect is unknown. When the sample of the j-th defect class is larger than N, deleting the old sample, and keeping that N defect samples exist in equal quantity in each class.
S34, repeating the model training and testing process for a plurality of times according to the data reserved in the cache in S33, and evaluating the classification result of each model until a stable model is obtained.
Step S4: after model training is completed, parameter compression, dimension reduction and other operations are performed on the original network structure of the model through a TensorRT technology, so that the training and reasoning speed of the network is improved, meanwhile, the robustness of the model and the accuracy of classification results are maintained, and finally, various indexes in model evaluation are enabled to meet the actual requirements in the production scene of the semiconductor industry.
Step S5: the model compressed in the step S4 can be deployed into an AOI (automated optical inspection) of a wafer production line of a semiconductor factory, and after the AOI acquires a defect wafer map in real time, image preprocessing is carried out and the model is input for incremental learning. After the trained model is updated online, the accurate classification of the unknown defect category can be realized.
The wafer surface defect classification method based on incremental learning combines deep learning, artificial intelligence application and computer vision technology to realize accurate detection and classification of unknown defects of the wafer. The trained model can quickly reach the optimal performance through an online incremental learning method, a lightweight caching mechanism and a compound training mode. The optimized and accelerated model is deployed in the AOI equipment of the wafer production line, unknown defect types can be collected and identified in real time, and new unknown defect types are updated rapidly, so that the optimization of the wafer manufacturing process is improved, the product yield is improved, and the semiconductor manufacturing cost is greatly reduced.
Through the specific implementation manner, the wafer surface defect classification method based on incremental learning provided by the invention fully utilizes deep learning, artificial intelligence application and computer vision technology, and provides an effective and accurate solution for detecting and classifying unknown defects in the wafer manufacturing process. The method aims at the problems of unknown defect detection and classification in the wafer manufacturing process, the limitations of the existing solutions in the industry and the current situation that the problems cannot be effectively solved, and provides an automatic wafer unknown defect detection and classification method based on incremental learning, so that an innovative technical solution is provided for the semiconductor manufacturing industry.
The foregoing is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art, who is within the scope of the present invention, should be covered by the protection scope of the present invention by making equivalents and modifications to the technical solution and the inventive concept thereof.
Claims (7)
1. The method for detecting and classifying the unknown wafer defect category is characterized by comprising the following steps:
(a) Converting wafer map data acquired by the wafer production line AOI equipment into data set data for model training by using a wafer map preprocessing module;
(b) Building a convolutional neural network model based on an incremental learning method; the model comprises:
(i) Convolutional neural network structural model based on ResNet 50;
(ii) Online incremental learning PathInt strategy;
(iii) The loss function remains close to the learning parameters of the old defect type;
(iv) The incremental learning model is used for classifying unknown defects;
(c) Training a convolutional neural network model, comprising:
(i) The basic wafer defect data set in the step (a) is reworked, a plurality of new wafer map data sets with different structures are generated in a rearrangement and splitting mode, and random division is carried out according to defect types;
(ii) Training models by using multiple data sets, and training and testing the defect type incremental input models;
(iii) Using a lightweight cache to store known defects, and keeping N defect samples with equal quantity in each category;
(iv) Repeating the model training and testing process for a plurality of times until a stable model is obtained;
(d) Parameter compression and dimension reduction are carried out on the original network structure of the model through a TensorRT technology;
(e) And deploying the compressed model into an AOI (automated optical inspection) of a wafer production line of a semiconductor factory, and accurately classifying unknown defect types.
2. The method of claim 1, wherein step (a) comprises:
collecting wafer graphs containing annular scratches, unifying the sizes of all the wafer graphs, and performing image processing on the wafer graphs in an image compression, enhancement and noise reduction mode to obtain a standardized wafer graph containing defects; and adding corresponding labels to the image data according to the obvious characteristics of different defect types to manufacture a wafer defect data set.
3. The method of claim 1, wherein in step (b) (ii), the weights that have a significant impact on the classification of old defect tasks are constrained, and wherein when learning the second defect type, a learning trajectory is calculated in the parameter space that minimizes the loss function L while remaining in a low loss region of the first defect type, thereby protecting the performance of the model on the first defect type.
4. A method according to claim 3, characterized in that:
in step (c) (iv), when the ith defect type is learned, the model maintains the loss function close to the learning parameters of the old defect type.
5. The method according to claim 4, wherein:
in step (c) (iii), a lightweight cache is used to save known defects; n defect samples are saved in an equivalent manner in each category, when the ith sample is detected, the defects are added into the corresponding category if the defects are known, and if the defects are unknown, a new defect category is created; when the sample of the j-th defect class is larger than N, deleting the old sample, and keeping that N defect samples exist in equal quantity in each class.
6. The method of claim 1, wherein the base wafer defect dataset of step (c) (i) is rearranged and split to generate a plurality of new wafer map datasets having differential structures.
7. A system for detecting and classifying unknown classes of wafer defects, comprising:
the wafer map preprocessing module is used for converting wafer map data acquired by the wafer production line AOI equipment into data set data for model training;
convolutional neural network model based on incremental learning method;
the model training module is used for processing and inputting the model to perform incremental training and testing;
the TensorRT optimizing module is used for carrying out parameter compression and dimension reduction on the trained model;
the deployment module is used for deploying the optimized model to the AOI equipment of the wafer production line of the semiconductor factory, collecting the defect wafer map in real time, and carrying out image preprocessing and incremental learning;
the convolutional neural network model based on the incremental learning method comprises the following steps:
convolutional neural network structural model based on ResNet 50;
an online incremental learning PathInt strategy for constraining weights in the model;
the loss function adjusting module is used for keeping learning parameters close to the old defect type when the new defect type is learned in an incremental mode;
the incremental learning model is used for classifying unknown defects;
the model training module comprises:
the data set processing module is used for carrying out redoing on the basic wafer defect data set; generating a plurality of new wafer map data sets with different structures through rearrangement and splitting modes, and randomly dividing according to defect types;
the multi-data set training module is used for training models by using various data sets and training and testing the defect type incremental input model;
the lightweight cache module is used for storing known defects, and N defect samples are stored in equal quantity in each category;
the model evaluation module is used for repeating the model training and testing process for a plurality of times and evaluating the classification result of each model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310438044.1A CN116188877B (en) | 2023-04-23 | 2023-04-23 | Method and system for detecting and classifying unknown wafer defect categories |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310438044.1A CN116188877B (en) | 2023-04-23 | 2023-04-23 | Method and system for detecting and classifying unknown wafer defect categories |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116188877A CN116188877A (en) | 2023-05-30 |
| CN116188877B true CN116188877B (en) | 2023-08-04 |
Family
ID=86434799
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310438044.1A Active CN116188877B (en) | 2023-04-23 | 2023-04-23 | Method and system for detecting and classifying unknown wafer defect categories |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116188877B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110766660A (en) * | 2019-09-25 | 2020-02-07 | 上海众壹云计算科技有限公司 | Integrated circuit defect image recognition and classification system based on fusion depth learning model |
| CN112967255A (en) * | 2021-03-09 | 2021-06-15 | 暨南大学 | Shield segment defect type identification and positioning system and method based on deep learning |
-
2023
- 2023-04-23 CN CN202310438044.1A patent/CN116188877B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110766660A (en) * | 2019-09-25 | 2020-02-07 | 上海众壹云计算科技有限公司 | Integrated circuit defect image recognition and classification system based on fusion depth learning model |
| CN112967255A (en) * | 2021-03-09 | 2021-06-15 | 暨南大学 | Shield segment defect type identification and positioning system and method based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| Research on Wafer Surface Defect Pattern Detection Method Based on Incremental Learning;Naigong Yu 等;Journal of Physics: Conference Series;第1-6页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116188877A (en) | 2023-05-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109977808B (en) | Wafer surface defect mode detection and analysis method | |
| CN109272500B (en) | Fabric classification method based on adaptive convolutional neural network | |
| CN113763312B (en) | Detection of defects in semiconductor samples using weak labels | |
| US20220050061A1 (en) | Automatic optimization of an examination recipe | |
| CN110503288B (en) | System and method for identifying yield loss reason considering machine interaction | |
| CN114627383B (en) | Small sample defect detection method based on metric learning | |
| CN109146847B (en) | Wafer map batch analysis method based on semi-supervised learning | |
| CN111291814B (en) | Crack identification algorithm based on convolutional neural network and information entropy data fusion strategy | |
| CN115439458A (en) | Industrial image defect target detection algorithm based on depth map attention | |
| CN113222982A (en) | Wafer surface defect detection method and system based on improved YOLO network | |
| CN114581782A (en) | Fine defect detection method based on coarse-to-fine detection strategy | |
| CN114049305A (en) | Distribution line pin defect detection method based on improved ALI and fast-RCNN | |
| CN115713488A (en) | Bridge apparent disease pixel level identification method and system based on instance segmentation | |
| CN116883393A (en) | Metal surface defect detection method based on anchor frame-free target detection algorithm | |
| CN113763364B (en) | Image defect detection method based on convolutional neural network | |
| CN115375635A (en) | Wafer defect identification method | |
| VAN THAN et al. | Performance of different CNN-based models on classification of steel sheet surface defects | |
| CN116188877B (en) | Method and system for detecting and classifying unknown wafer defect categories | |
| CN113869418A (en) | Small sample ship target identification method based on global attention relationship network | |
| CN116630989A (en) | Visual fault detection method and system for intelligent ammeter, electronic equipment and storage medium | |
| CN113724233B (en) | Transformer equipment appearance image defect detection method based on fusion data generation and transfer learning technology | |
| CN115830302A (en) | Multi-scale feature extraction and fusion power distribution network equipment positioning identification method | |
| CN115082650A (en) | Implementation method of automatic pipeline defect labeling tool based on convolutional neural network | |
| Chien et al. | Image-based defect classification for TFT-LCD array via convolutional neural network | |
| Go et al. | A deep learning-based surface defect inspection system for smartphone glass |
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 |