CN114998330A - Unsupervised wafer defect detection method, unsupervised wafer defect detection device, unsupervised wafer defect detection equipment and storage medium - Google Patents
Unsupervised wafer defect detection method, unsupervised wafer defect detection device, unsupervised wafer defect detection equipment and storage medium Download PDFInfo
- Publication number
- CN114998330A CN114998330A CN202210894978.1A CN202210894978A CN114998330A CN 114998330 A CN114998330 A CN 114998330A CN 202210894978 A CN202210894978 A CN 202210894978A CN 114998330 A CN114998330 A CN 114998330A
- Authority
- CN
- China
- Prior art keywords
- wafer
- unsupervised
- detection data
- self
- label
- 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.)
- Granted
Links
Images
Classifications
-
- 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
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/75—Systems in which material is subjected to a chemical reaction, the progress or the result of the reaction being investigated
- G01N21/76—Chemiluminescence; Bioluminescence
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/95—Investigating the presence of flaws or contamination characterised by the material or shape of the object to be examined
- G01N21/9501—Semiconductor wafers
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10064—Fluorescence image
-
- 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
- G06T2207/30—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)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Chemical & Material Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Pathology (AREA)
- Immunology (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Biochemistry (AREA)
- Analytical Chemistry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Quality & Reliability (AREA)
- Signal Processing (AREA)
- Theoretical Computer Science (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Plasma & Fusion (AREA)
- Testing Or Measuring Of Semiconductors Or The Like (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The invention belongs to the technical field of wafer detection, and discloses an unsupervised wafer defect detection method, device, equipment and storage medium. The method comprises the following steps: acquiring an initial wafer detection data set; utilizing a preset unsupervised self-encoder to make a pseudo label for the label-free wafer detection data in the initial wafer detection data set, and generating a target wafer detection data set; training a preset classifier according to the target wafer detection data set to obtain a trained target classifier; and when the wafer data to be detected is received, detecting the wafer data to be detected by using the target classifier to obtain a wafer defect detection result. By the mode, the pseudo labels are made for the label-free wafer detection data, the label data and the pseudo label data are used for training the target classifier, so that the target classifier learns different defect characteristics, the unknown defect types can be identified, and the detection precision of wafer defects is improved.
Description
Technical Field
The invention relates to the technical field of wafer detection, in particular to an unsupervised wafer defect detection method, device, equipment and storage medium.
Background
In the Micro LED display technology, multiple fault types are generated in the wafer (wafer) production process due to multiple influences such as differences in preparation equipment, production process, material design, and the like, and there is no coupling between the fault types. Wafer detection data is large in amount and free of labels, and wafer detection data is large in amount and cannot be obtained easily. In the existing wafer detection method, a relatively stable detection model can be trained only by using a large amount of label data, but the acquired label-free data is difficult to use, and only a few defect types can be learned, so that the detection accuracy is not high in unknown defect types.
The above is only for the purpose of assisting understanding of the technical aspects of the present invention, and does not represent an admission that the above is prior art.
Disclosure of Invention
The invention mainly aims to provide an unsupervised wafer defect detection method, device, equipment and storage medium, and aims to solve the technical problems that a relatively stable detection model can be trained only by using a large amount of label data in the conventional wafer detection method, the acquired label-free data is difficult to use, and only a few defect types can be learned, so that the detection precision is not high in the unknown defect types.
In order to achieve the above object, the present invention provides an unsupervised wafer defect detecting method, which comprises the following steps:
acquiring an initial wafer detection data set;
utilizing a preset unsupervised self-encoder to make a pseudo label for the label-free wafer detection data in the initial wafer detection data set, and generating a target wafer detection data set;
training a preset classifier according to the target wafer detection data set to obtain a trained target classifier;
and when the wafer data to be detected is received, detecting the wafer data to be detected by using the target classifier to obtain a wafer defect detection result.
Optionally, the preset unsupervised self-encoder and the preset arbiter form an anti-self-encoding network;
the generating of the target wafer detection data set by using the preset unsupervised auto-encoder to make the pseudo-label for the unlabeled wafer detection data in the initial wafer detection data set comprises:
inputting unlabeled wafer inspection data in the initial wafer inspection dataset into the countering self-coding network;
predicting the label of the label-free wafer detection data by using a preset unsupervised self-encoder in the antagonistic self-encoding network, and constraining the prediction result of the preset unsupervised self-encoder by using a preset discriminator in the antagonistic self-encoding network to obtain a pseudo label corresponding to the label-free wafer detection data;
and generating a target wafer detection data set according to the pseudo label and the initial wafer detection data set.
Optionally, before the preset unsupervised auto-encoder is used to make a pseudo tag for the unlabeled wafer inspection data in the initial wafer inspection dataset and generate the target wafer inspection dataset, the method further includes:
and performing preliminary training on an initial discriminator according to the labeled wafer detection data in the initial wafer detection data set to obtain a preset discriminator.
Optionally, the preset unsupervised auto-encoder is formed by a stack of multiple auto-encoders, wherein the multiple auto-encoders are provided with different activation functions.
Optionally, the generating a target wafer inspection dataset by using a preset unsupervised auto-encoder to make a pseudo tag for the unlabeled wafer inspection data in the initial wafer inspection dataset includes:
predicting labels of the label-free wafer detection data in the initial wafer detection data set by using a plurality of self-encoders to obtain a plurality of prediction results;
and determining a pseudo label corresponding to the label-free wafer detection data according to the plurality of prediction results, and generating a target wafer detection data set.
Optionally, after the labels of the unlabeled wafer inspection data in the initial wafer inspection dataset are respectively predicted by using a plurality of self-encoders to obtain a plurality of prediction results, the method further includes:
determining prediction accuracies respectively corresponding to the plurality of self-encoders;
deleting the self-encoders with the prediction accuracy lower than a preset threshold value from the self-encoders to obtain a plurality of residual self-encoders;
and determining a pseudo label corresponding to the label-free wafer detection data according to the prediction results corresponding to the residual self-encoders to generate a target wafer detection data set.
Optionally, the determining, according to the prediction results corresponding to the multiple remaining self-encoders, the pseudo label corresponding to the unlabeled wafer inspection data to generate a target wafer inspection data set includes:
determining a weight value corresponding to each residual self-encoder according to the prediction accuracy corresponding to the residual self-encoders;
and determining a pseudo label corresponding to the label-free wafer detection data according to the prediction results corresponding to the residual self-encoders and the weight values, and generating a target wafer detection data set.
In addition, in order to achieve the above object, the present invention further provides an unsupervised wafer defect inspection apparatus, which includes:
the acquisition module is used for acquiring an initial wafer detection data set;
the label generation module is used for making a pseudo label for the label-free wafer detection data in the initial wafer detection data set by using a preset unsupervised self-encoder to generate a target wafer detection data set;
the training module is used for training a preset classifier according to the target wafer detection data set to obtain a trained target classifier;
and the detection module is used for detecting the wafer data to be detected by using the target classifier when receiving the wafer data to be detected to obtain a wafer defect detection result.
In addition, in order to achieve the above object, the present invention further provides an unsupervised wafer defect inspection apparatus, including: a memory, a processor, and an unsupervised wafer defect detection program stored on the memory and executable on the processor, the unsupervised wafer defect detection program configured to implement an unsupervised wafer defect detection method as described above.
In addition, to achieve the above object, the present invention further provides a storage medium having an unsupervised wafer defect detecting program stored thereon, wherein the unsupervised wafer defect detecting program, when executed by a processor, implements the unsupervised wafer defect detecting method as described above.
The method comprises the steps of obtaining an initial wafer detection data set; utilizing a preset unsupervised self-encoder to make a pseudo label for the label-free wafer detection data in the initial wafer detection data set, and generating a target wafer detection data set; training a preset classifier according to a target wafer detection data set to obtain a trained target classifier; and when receiving the wafer data to be detected, detecting the wafer data to be detected by using the target classifier to obtain a wafer defect detection result. By the mode, the pseudo labels are made for the label-free wafer detection data, the label data and the pseudo label data are used for training the target classifier, so that the target classifier learns different defect characteristics, the unknown defect types can be identified, and the detection precision of wafer defects is improved.
Drawings
FIG. 1 is a schematic structural diagram of an unsupervised wafer defect inspection apparatus for a hardware operating environment according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart illustrating a method for unsupervised wafer defect inspection according to a first embodiment of the present invention;
FIG. 3 is a flowchart illustrating an unsupervised wafer defect inspection method according to a second embodiment of the present invention;
FIG. 4 is a schematic diagram of a model layout of an unsupervised wafer defect inspection method according to an embodiment of the present invention;
FIG. 5 is a flowchart illustrating a method for unsupervised wafer defect inspection according to a third embodiment of the present invention;
FIG. 6 is a schematic diagram of an integration strategy of an unsupervised wafer defect detection method according to an embodiment of the present invention;
FIG. 7 is a block diagram of an unsupervised wafer defect inspection apparatus according to a first embodiment of the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Referring to fig. 1, fig. 1 is a schematic structural diagram of an unsupervised wafer defect detecting apparatus in a hardware operating environment according to an embodiment of the present invention.
As shown in fig. 1, the unsupervised wafer defect inspection apparatus may include: a processor 1001, such as a Central Processing Unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to implement connection communication among these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a Wireless interface (e.g., a Wireless-Fidelity (Wi-Fi) interface). The Memory 1005 may be a Random Access Memory (RAM) or a Non-Volatile Memory (NVM), such as a disk Memory. The memory 1005 may alternatively be a storage device separate from the processor 1001.
Those skilled in the art will appreciate that the configuration shown in fig. 1 does not constitute a limitation of an unsupervised wafer defect inspection apparatus and may include more or fewer components than shown, or some components in combination, or a different arrangement of components.
As shown in fig. 1, a memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and an unsupervised wafer defect inspection program.
In the unsupervised wafer defect inspection apparatus shown in fig. 1, the network interface 1004 is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 of the unsupervised wafer defect detecting apparatus may be disposed in the unsupervised wafer defect detecting apparatus, and the unsupervised wafer defect detecting apparatus calls the unsupervised wafer defect detecting program stored in the memory 1005 through the processor 1001 and executes the unsupervised wafer defect detecting method provided by the embodiment of the present invention.
An embodiment of the present invention provides an unsupervised wafer defect detecting method, and referring to fig. 2, fig. 2 is a schematic flow chart of a first embodiment of the unsupervised wafer defect detecting method according to the present invention.
In this embodiment, the unsupervised wafer defect detection method includes the following steps:
step S10: an initial wafer inspection dataset is obtained.
It should be understood that the main implementation body of this embodiment is unsupervised wafer defect detection equipment, and the unsupervised wafer defect detection equipment may be equipment such as a computer and a server, and may also be other equipment with inference capability, which is not limited in this embodiment.
It should be noted that the initial wafer inspection data set includes a small amount of labeled wafer inspection data and a large amount of unlabeled wafer inspection data, and in a specific implementation, the wafer inspection data are photoluminescence images corresponding to a plurality of defective wafers acquired by using a Photoluminescence (PL) technology, and a detection frame is framed in the corresponding photoluminescence images in advance according to a known defect type and corresponding defect type information is labeled to obtain the labeled wafer inspection data.
Step S20: and utilizing a preset unsupervised self-encoder to make a pseudo label for the label-free wafer detection data in the initial wafer detection data set, and generating a target wafer detection data set.
It should be understood that the preset unsupervised self-encoder infers the hidden structural features inside the data by using the unlabeled wafer detection data, and distinguishes the wafer detection data of different defect types, so as to make pseudo labels for a large amount of unlabeled wafer detection data, and generate a target wafer detection data set according to a small amount of labeled wafer detection data and a large amount of pseudo label wafer detection data. In a specific implementation, the predetermined unsupervised self-encoder distinguishes data differences between different defect types according to a spatial scale distance function, such as a mean square error function (MSE) or a root mean square error function (RMSE), where the mean square error function is expressed as:
the root mean square error function is expressed as:
preferably, the preset unsupervised self-encoder and the preset discriminator are arranged to form an anti-self-encoding network, and in the process of training and learning a large amount of unlabeled wafer detection data by the preset unsupervised self-encoder, the preset unsupervised self-encoder and the preset discriminator transmit weights through back propagation and are mutually restricted. Specifically, the preset discriminator is used for discriminating the authenticity of the pseudo label, so that whether the training of the preset unsupervised self-encoder is correct or not is determined, the pseudo label generated by the preset unsupervised self-encoder is as close to the result of the preset discriminator as possible, and the quality of the pseudo label generated by the self-encoder is improved.
Preferably, the preset discriminator is initialized in advance according to a small amount of labeled wafer detection data, so that the preset discriminator can master partial characteristics of real data, rapid fitting during model training is facilitated, and back propagation gradient dissipation is prevented.
On the other hand, in the embodiment, a plurality of depth self-encoders are arranged to be stacked to form a preset unsupervised self-encoder, thereby effectively avoiding the randomness of the network, and each depth self-encoder corresponds to an activation function and provides the nonlinear modeling capability. And aiming at the output results of the multiple depth self-encoders, combining the output results according to a preset integration strategy so as to determine the prediction result of the preset unsupervised self-encoders, and manufacturing pseudo labels for the label-free wafer detection data according to the prediction result.
Step S30: and training a preset classifier according to the target wafer detection data set to obtain the trained target classifier.
It should be understood that the target wafer inspection dataset includes a small amount of labeled wafer inspection data and a large amount of pseudo-labeled wafer inspection data, and the preset classifier is trained through the target wafer inspection dataset, so that the preset classifier learns the features for distinguishing the defect types. The preset classifier of this embodiment may be a convolutional network classifier, the convolutional network classifier is used to detect the wafer image, output a prediction result, determine a corresponding loss function value according to the prediction result and the tag information, perform iterative training until the loss function value is lower than a certain value or the current iteration number is greater than the maximum iteration number, perform defect stratification on the target wafer detection data set through the convolutional network classifier to achieve defect classification, and obtain a trained convolutional network classifier.
Step S40: and when the wafer data to be detected is received, detecting the wafer data to be detected by using the target classifier to obtain a wafer defect detection result.
It should be noted that the wafer data to be detected is a photoluminescence image acquired by a Photoluminescence (PL) technology, and is detected by using a trained target classifier, so as to determine whether the currently acquired data has defects, and determine a corresponding defect type.
In the embodiment, an initial wafer detection data set is obtained; utilizing a preset unsupervised self-encoder to make a pseudo label for the label-free wafer detection data in the initial wafer detection data set, and generating a target wafer detection data set; training a preset classifier according to the target wafer detection data set to obtain a trained target classifier; and when receiving the wafer data to be detected, detecting the wafer data to be detected by using the target classifier to obtain a wafer defect detection result. By the mode, the pseudo labels are made for the label-free wafer detection data, the label data and the pseudo label data are used for training the target classifier, so that the target classifier learns different defect characteristics, the unknown defect types can be identified, and the detection precision of wafer defects is improved.
Referring to fig. 3, fig. 3 is a flowchart illustrating an unsupervised wafer defect inspection method according to a second embodiment of the present invention.
Based on the first embodiment, the preset unsupervised self-encoder and the preset discriminator in the unsupervised wafer defect detection method of the embodiment form an anti-self-encoding network;
the step S20 includes:
step S201: inputting unlabeled wafer inspection data in the initial wafer inspection data set into the countering self-coding network.
Step S202: and predicting the label of the label-free wafer detection data by using a preset unsupervised self-encoder in the antagonistic self-encoding network, and constraining the prediction result of the preset unsupervised self-encoder by using a preset discriminator in the antagonistic self-encoding network to obtain a pseudo label corresponding to the label-free wafer detection data.
It should be understood that, in the embodiment, the discriminator is used in combination with the self-encoder to form the countering self-encoding network, and the countering self-encoding network performs classification prediction on a large amount of unlabeled wafer inspection data, so as to distinguish the unlabeled wafer inspection data into different defect types, and the discriminator is used to improve the quality of generating the pseudo labels by the self-encoder. The weight is transmitted by the aid of back propagation in the training process of the preset unsupervised self-encoder and the preset discriminator, and mutual restriction is achieved. The discriminator determines whether the self-encoder training is correct or not by discriminating the authenticity of the pseudo label, and the pseudo label generated by the self-encoder is as close as possible to the result of the discriminator, so that the corresponding pseudo label is made for each non-label wafer detection data.
Step S203: and generating a target wafer detection data set according to the pseudo label and the initial wafer detection data set.
In a specific implementation, a pseudo label corresponding to the label-free wafer inspection data in the initial wafer inspection data set is marked, and a target wafer inspection data set including label wafer inspection data and pseudo label wafer inspection data is generated.
Further, before the step S20, the method further includes: and performing preliminary training on an initial discriminator according to the labeled wafer detection data in the initial wafer detection data set to obtain a preset discriminator.
It should be noted that, the discriminator cannot initially discriminate whether the data is true, so in this embodiment, before training the anti-self-coding network, the discriminator is initially learned, and the discriminator is initially trained by using a small amount of labeled wafer detection data, so that the discriminator can grasp partial characteristics of true data, which is beneficial to fast fitting during model training and prevents back propagation gradient dissipation.
Referring to fig. 4, fig. 4 is a schematic diagram of a model layout of an embodiment of an unsupervised wafer defect detection method according to the present invention, in this embodiment, a small number of labeled wafer data sets are used to initialize an unsupervised hierarchical learner, pseudo-label data sets are created for a large number of wafer data sets by stacking an unsupervised self-encoder, a confrontation discriminator is trained to improve the quality of pseudo labels generated by the self-encoder, and finally a convolutional network classifier is trained by using a large number of pseudo-label data and labeled data, so as to improve the detection accuracy.
In the embodiment, an initial wafer detection data set is obtained; inputting the unlabeled wafer detection data in the initial wafer detection data set into a counterself-coding network; predicting the label of the label-free wafer detection data by using a preset unsupervised self-encoder in the antagonistic self-encoding network, and constraining the prediction result of the preset unsupervised self-encoder by using a preset discriminator in the antagonistic self-encoding network to obtain a pseudo label corresponding to the label-free wafer detection data; generating a target wafer detection data set according to the pseudo label and the initial wafer detection data set; training a preset classifier according to the target wafer detection data set to obtain a trained target classifier; and when the wafer data to be detected is received, detecting the wafer data to be detected by using the target classifier to obtain a wafer defect detection result. Through the mode, the discriminator is arranged to restrain the prediction result of the unsupervised self-encoder, the generation quality of the pseudo label is improved, the labeled data and the pseudo label data are used for training the target classifier, the target classifier learns different defect characteristics, the unknown defect types can be recognized, and the training precision and the recognition precision of the target classifier are further improved.
Referring to fig. 5, fig. 5 is a flowchart illustrating an unsupervised wafer defect inspection method according to a third embodiment of the present invention.
Based on the first embodiment, the preset unsupervised self-encoder in the unsupervised wafer defect detecting method of the present embodiment is formed by stacking a plurality of self-encoders, wherein the plurality of self-encoders are set with different activation functions.
The step S20 includes:
step S204: and respectively predicting the labels of the label-free wafer detection data in the initial wafer detection data set by using a plurality of self-encoders to obtain a plurality of prediction results.
It can be understood that, in order to improve generalization performance and enable the model to learn deep features of input data, in the present embodiment, multiple depth Automatic Encoders (AEs) are stacked to form a preset unsupervised auto-encoder, which can effectively avoid randomness of the network. The self-encoders are respectively provided with different activation functions, and the activation functions are used for providing nonlinear modeling capability.
In a particular implementation, the activation function is represented as
Training sample
Determining hidden variables by activating functions for input data x
Specifically, it is expressed by the following formula:
wherein the training purpose of the self-encoder is to optimize the parameter set
The reconstruction error is minimized. For the loss function of the data reconstruction error, the present embodiment is set as a traditional mean square error loss function (MSE) and a cross entropy function, where the cross entropy cost function exhibits a faster convergence speed and a stronger global optimization capability. In the case of a training sample for unlabeled m-dimensional data, the cross-entropy cost function is defined as:
preferably, considering the advantage of the activation function as a sparse processing mode in learning useful features of the model, in this embodiment, the regularized cross entropy in the automatic encoders of different activation functions rewrites the loss function of an unlabeled m-dimensional training sample into:
it should be noted that, the stacked multiple autoencoders are trained according to the above loss function, so that the multiple autoencoders perform feature layering on a large amount of unlabeled wafer inspection data.
Step S205: and determining a pseudo label corresponding to the label-free wafer detection data according to the plurality of prediction results to generate a target wafer detection data set.
It should be understood that an integration strategy is set to integrate the plurality of prediction results, a pseudo label corresponding to each non-label wafer detection data is determined, and a target wafer detection data set is generated according to the labeled wafer detection data and the pseudo label wafer detection data. Optionally, the integrated metering policy is a majority voting policy.
Further, after the step S205, the method further includes: determining prediction accuracies respectively corresponding to the plurality of self-encoders; deleting self-encoders with prediction accuracy lower than a preset threshold value from the self-encoders to obtain a plurality of residual self-encoders; and determining a pseudo label corresponding to the label-free wafer detection data according to the prediction results corresponding to the residual self-encoders to generate a target wafer detection data set.
In a specific implementation, since the majority voting policy has a defect that all the individual models have the same weight, for multiple types of defect data, the data is easily interfered by some network layer to be useless, preferably, the present embodiment sets a preset threshold for screening the stacked multiple autoencoders, where the preset threshold may be an accuracy threshold obtained by using the majority voting policy. Referring to fig. 6, fig. 6 is an integrated strategy diagram of an embodiment of the unsupervised wafer defect detection method of the present invention, in this embodiment, the accuracy judgment is performed according to a preset threshold, only a single network with an accuracy exceeding the threshold is considered in the training, other networks are discarded, and a final output result is determined according to a prediction result of an auto-encoder with an accuracy exceeding the preset threshold. Optionally, the prediction results of the remaining self-encoders are averaged to obtain a final prediction result, so as to determine a pseudo label corresponding to each non-label wafer detection data, and generate a target wafer detection data set.
Further, the determining, according to the prediction results corresponding to the multiple residual encoders, the pseudo label corresponding to the unlabeled wafer inspection data to generate a target wafer inspection data set includes: determining a weight value corresponding to each residual self-encoder according to the prediction accuracy corresponding to the residual self-encoders; and determining a pseudo label corresponding to the label-free wafer detection data according to the prediction results corresponding to the residual self-encoders and the weight values, and generating a target wafer detection data set.
It should be noted that, in this embodiment, a corresponding weight value is assigned to a network output meeting a requirement according to a corresponding prediction accuracy, and a comprehensive diagnosis result of each sample is determined according to the weight value. In a particular implementation, multiple self-encoders in a stack are trained multiple times in order to maintain a stable combined diagnostic result.
In the embodiment, an initial wafer detection data set is obtained; respectively predicting the labels of the label-free wafer detection data in the initial wafer detection data set by using a plurality of self-encoders to obtain a plurality of prediction results; determining a pseudo label corresponding to the label-free wafer detection data according to the plurality of prediction results to generate a target wafer detection data set; training a preset classifier according to the target wafer detection data set to obtain a trained target classifier; and when receiving the wafer data to be detected, detecting the wafer data to be detected by using the target classifier to obtain a wafer defect detection result. Through the mode, the unsupervised self-encoder formed by stacking the self-encoders is used for generating the pseudo labels, the influence of network randomness on the generation of the pseudo labels is avoided, the generation quality of the pseudo labels is improved, the labeled data and the pseudo label data are used for training the target classifier, the target classifier learns different defect characteristics, the unknown defect types can be identified, and the training precision and the identification precision of the target classifier are further improved.
In addition, an embodiment of the present invention further provides a storage medium, where an unsupervised wafer defect inspection program is stored on the storage medium, and when executed by a processor, the unsupervised wafer defect inspection program implements the unsupervised wafer defect inspection method described above.
Since the storage medium adopts all technical solutions of all the embodiments, at least all the beneficial effects brought by the technical solutions of the embodiments are achieved, and no further description is given here.
Referring to fig. 7, fig. 7 is a block diagram illustrating a structure of an unsupervised wafer defect inspection apparatus according to a first embodiment of the present invention.
As shown in fig. 7, an unsupervised wafer defect detecting apparatus according to an embodiment of the present invention includes:
an obtaining module 10 is configured to obtain an initial wafer inspection dataset.
And a label generating module 20, configured to use a preset unsupervised auto-encoder to generate a pseudo label for the unlabeled wafer inspection data in the initial wafer inspection dataset, so as to generate a target wafer inspection dataset.
And the training module 30 is configured to train a preset classifier according to the target wafer detection data set, so as to obtain a trained target classifier.
And the detection module 40 is configured to detect the wafer data to be detected by using the target classifier when the wafer data to be detected is received, so as to obtain a wafer defect detection result.
It should be understood that the above is only an example, and the technical solution of the present invention is not limited in any way, and in a specific application, a person skilled in the art may set the technical solution as needed, and the present invention is not limited in this respect.
In the embodiment, an initial wafer detection data set is obtained; utilizing a preset unsupervised self-encoder to make a pseudo label for the label-free wafer detection data in the initial wafer detection data set, and generating a target wafer detection data set; training a preset classifier according to the target wafer detection data set to obtain a trained target classifier; and when the wafer data to be detected is received, detecting the wafer data to be detected by using the target classifier to obtain a wafer defect detection result. By the method, the pseudo labels are made for the label-free wafer detection data, the target classifier is trained by using the labeled data and the pseudo label data, so that the target classifier learns different defect characteristics, the unknown defect types can be identified, and the detection precision of the wafer defects is improved.
It should be noted that the above-mentioned work flows are only illustrative and do not limit the scope of the present invention, and in practical applications, those skilled in the art may select some or all of them according to actual needs to implement the purpose of the solution of the present embodiment, and the present invention is not limited herein.
In addition, the technical details that are not described in detail in this embodiment may refer to the unsupervised wafer defect detection method provided in any embodiment of the present invention, and are not described herein again.
In one embodiment, the predetermined unsupervised self-encoder and the predetermined discriminator form a counteracting self-encoding network;
the tag generation module 20 is further configured to input the non-tag wafer inspection data in the initial wafer inspection data set to the countermeasure self-encoding network; predicting the label of the label-free wafer detection data by using a preset unsupervised self-encoder in the antagonistic self-encoding network, and constraining the prediction result of the preset unsupervised self-encoder by using a preset discriminator in the antagonistic self-encoding network to obtain a pseudo label corresponding to the label-free wafer detection data; and generating a target wafer detection data set according to the pseudo label and the initial wafer detection data set.
In an embodiment, the training module 30 is further configured to perform a preliminary training on an initial discriminator according to the labeled wafer detection data in the initial wafer detection data set, so as to obtain a preset discriminator.
In an embodiment, the preset unsupervised auto-encoder is formed by stacking a plurality of auto-encoders, wherein the plurality of auto-encoders are provided with different activation functions.
In an embodiment, the label generating module 20 is further configured to predict labels of the unlabeled wafer inspection data in the initial wafer inspection data set by using a plurality of self-encoders, respectively, to obtain a plurality of prediction results; and determining a pseudo label corresponding to the label-free wafer detection data according to the plurality of prediction results to generate a target wafer detection data set.
In an embodiment, the tag generating module 20 is further configured to determine prediction accuracies corresponding to the plurality of self-encoders, respectively; deleting the self-encoders with the prediction accuracy lower than a preset threshold value from the self-encoders to obtain a plurality of residual self-encoders; and determining a pseudo label corresponding to the label-free wafer detection data according to the prediction results corresponding to the residual self-encoders to generate a target wafer detection data set.
In an embodiment, the tag generating module 20 is further configured to determine a weight value corresponding to each of the plurality of remaining self-encoders according to a prediction accuracy corresponding to each of the plurality of remaining self-encoders; and determining a pseudo label corresponding to the label-free wafer detection data according to the prediction results corresponding to the residual self-encoders and the weight values, and generating a target wafer detection data set.
Further, it is to be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solution of the present invention or portions thereof that contribute to the prior art may be embodied in the form of a software product, where the computer software product is stored in a storage medium (e.g. Read Only Memory (ROM)/RAM, magnetic disk, optical disk), and includes several instructions for enabling a terminal device (e.g. a mobile phone, a computer, a server, or a network device) to execute the method according to the embodiments of the present invention.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (10)
1. An unsupervised wafer defect detection method, comprising:
acquiring an initial wafer detection data set;
utilizing a preset unsupervised self-encoder to make a pseudo label for the label-free wafer detection data in the initial wafer detection data set, and generating a target wafer detection data set;
training a preset classifier according to the target wafer detection data set to obtain a trained target classifier;
and when the wafer data to be detected is received, detecting the wafer data to be detected by using the target classifier to obtain a wafer defect detection result.
2. The unsupervised wafer defect detecting method of claim 1, wherein the predetermined unsupervised self-encoder and the predetermined discriminator form a robust self-encoding network;
the generating of the target wafer detection data set by using the preset unsupervised auto-encoder to make the pseudo-label for the unlabeled wafer detection data in the initial wafer detection data set comprises:
inputting unlabeled wafer inspection data in the initial wafer inspection dataset into the countering self-coding network;
predicting the label of the label-free wafer detection data by using a preset unsupervised self-encoder in the antagonistic self-encoding network, and constraining the prediction result of the preset unsupervised self-encoder by using a preset discriminator in the antagonistic self-encoding network to obtain a pseudo label corresponding to the label-free wafer detection data;
and generating a target wafer detection data set according to the pseudo label and the initial wafer detection data set.
3. The unsupervised wafer defect detection method of claim 2, wherein before generating the target wafer inspection data set, the method further comprises:
and performing preliminary training on an initial discriminator according to the labeled wafer detection data in the initial wafer detection data set to obtain a preset discriminator.
4. The unsupervised wafer defect detection method of claim 1, wherein the preset unsupervised self-encoder is formed by a stack of a plurality of self-encoders, wherein the plurality of self-encoders are provided with different activation functions.
5. The unsupervised wafer defect detection method of claim 4, wherein generating the target wafer inspection dataset by creating pseudo-labels for the unlabeled wafer inspection data in the initial wafer inspection dataset using a predetermined unsupervised self-encoder comprises:
predicting labels of the label-free wafer detection data in the initial wafer detection data set by using a plurality of self-encoders to obtain a plurality of prediction results;
and determining a pseudo label corresponding to the label-free wafer detection data according to the plurality of prediction results, and generating a target wafer detection data set.
6. The unsupervised wafer defect inspection method of claim 5, wherein after the labels of the unlabeled wafer inspection data in the initial wafer inspection data set are predicted by the plurality of self-encoders respectively, and a plurality of prediction results are obtained, the method further comprises:
determining prediction accuracies respectively corresponding to the plurality of self-encoders;
deleting the self-encoders with the prediction accuracy lower than a preset threshold value from the self-encoders to obtain a plurality of residual self-encoders;
and determining a pseudo label corresponding to the label-free wafer detection data according to the prediction results corresponding to the residual self-encoders to generate a target wafer detection data set.
7. The unsupervised wafer defect inspection method of claim 6, wherein said determining the pseudo label corresponding to the unlabeled wafer inspection data according to the prediction results corresponding to the plurality of residual autoencoders to generate a target wafer inspection data set, comprises:
determining a weight value corresponding to each residual self-encoder according to the prediction accuracy corresponding to the residual self-encoders;
and determining a pseudo label corresponding to the label-free wafer detection data according to the prediction results corresponding to the residual self-encoders and the weight values, and generating a target wafer detection data set.
8. An unsupervised wafer defect inspection device, comprising:
the acquisition module is used for acquiring an initial wafer detection data set;
the label generation module is used for making a pseudo label for the label-free wafer detection data in the initial wafer detection data set by using a preset unsupervised self-encoder to generate a target wafer detection data set;
the training module is used for training a preset classifier according to the target wafer detection data set to obtain a trained target classifier;
and the detection module is used for detecting the wafer data to be detected by using the target classifier when receiving the wafer data to be detected to obtain a wafer defect detection result.
9. An unsupervised wafer defect inspection apparatus, the apparatus comprising: a memory, a processor, and an unsupervised wafer defect detection program stored on the memory and executable on the processor, the unsupervised wafer defect detection program configured to implement the unsupervised wafer defect detection method of any of claims 1-7.
10. A storage medium having an unsupervised wafer defect inspection program stored thereon, the unsupervised wafer defect inspection program when executed by a processor implementing the unsupervised wafer defect inspection method of any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210894978.1A CN114998330B (en) | 2022-07-28 | 2022-07-28 | Unsupervised wafer defect detection method, unsupervised wafer defect detection device, unsupervised wafer defect detection equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210894978.1A CN114998330B (en) | 2022-07-28 | 2022-07-28 | Unsupervised wafer defect detection method, unsupervised wafer defect detection device, unsupervised wafer defect detection equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114998330A true CN114998330A (en) | 2022-09-02 |
| CN114998330B CN114998330B (en) | 2022-11-01 |
Family
ID=83020972
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210894978.1A Active CN114998330B (en) | 2022-07-28 | 2022-07-28 | Unsupervised wafer defect detection method, unsupervised wafer defect detection device, unsupervised wafer defect detection equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114998330B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115631178A (en) * | 2022-11-03 | 2023-01-20 | 昆山润石智能科技有限公司 | Automatic wafer defect detection method, system, equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107316049A (en) * | 2017-05-05 | 2017-11-03 | 华南理工大学 | A kind of transfer learning sorting technique based on semi-supervised self-training |
| CN111832570A (en) * | 2020-07-02 | 2020-10-27 | 北京工业大学 | Image semantic segmentation model training method and system |
| CN114240882A (en) * | 2021-12-16 | 2022-03-25 | 深圳市商汤科技有限公司 | Defect detection method and device, electronic equipment and storage medium |
| CN114299034A (en) * | 2021-12-30 | 2022-04-08 | 杭州海康威视数字技术股份有限公司 | Defect detection model training method, defect detection method and device |
-
2022
- 2022-07-28 CN CN202210894978.1A patent/CN114998330B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107316049A (en) * | 2017-05-05 | 2017-11-03 | 华南理工大学 | A kind of transfer learning sorting technique based on semi-supervised self-training |
| CN111832570A (en) * | 2020-07-02 | 2020-10-27 | 北京工业大学 | Image semantic segmentation model training method and system |
| CN114240882A (en) * | 2021-12-16 | 2022-03-25 | 深圳市商汤科技有限公司 | Defect detection method and device, electronic equipment and storage medium |
| CN114299034A (en) * | 2021-12-30 | 2022-04-08 | 杭州海康威视数字技术股份有限公司 | Defect detection model training method, defect detection method and device |
Non-Patent Citations (2)
| Title |
|---|
| DEYI ZENG: "Anomaly Detection by Unsupervised Adversarial Generative Self-labelling Autoencoder", 《2022 IEEE INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE AND COMPUTER APPLICATIONS (ICAICA)》 * |
| 爱知菜: "提高分类器准确率的几种方法总结", 《HTTPS://WWW.CNBLOGS.COM/RAV009/ARCHIVE/2013/05/30/5131154.HTML》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115631178A (en) * | 2022-11-03 | 2023-01-20 | 昆山润石智能科技有限公司 | Automatic wafer defect detection method, system, equipment and storage medium |
| CN115631178B (en) * | 2022-11-03 | 2023-11-10 | 昆山润石智能科技有限公司 | Automatic wafer defect detection method, system, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114998330B (en) | 2022-11-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111368896B (en) | Hyperspectral remote sensing image classification method based on dense residual three-dimensional convolutional neural network | |
| Li et al. | LGM-Net: Learning to generate matching networks for few-shot learning | |
| CN108052911B (en) | Deep learning-based multi-mode remote sensing image high-level feature fusion classification method | |
| CN110443143B (en) | Multi-branch convolutional neural network fused remote sensing image scene classification method | |
| Springenberg et al. | Improving deep neural networks with probabilistic maxout units | |
| TW201734955A (en) | Generating simulated output for a specimen | |
| CN110046706B (en) | Model generation method and device and server | |
| Porto et al. | ML‐morph: A fast, accurate and general approach for automated detection and landmarking of biological structures in images | |
| JP6107531B2 (en) | Feature extraction program and information processing apparatus | |
| CN114332578A (en) | Image anomaly detection model training method, image anomaly detection method and device | |
| CN115170449B (en) | Multi-mode fusion scene graph generation method, system, equipment and medium | |
| CN112633310A (en) | Method and system for classifying sensor data with improved training robustness | |
| Wang et al. | Hand-drawn electronic component recognition using deep learning algorithm | |
| CN114998330B (en) | Unsupervised wafer defect detection method, unsupervised wafer defect detection device, unsupervised wafer defect detection equipment and storage medium | |
| CN110263808B (en) | Image emotion classification method based on LSTM network and attention mechanism | |
| Yamada et al. | Guiding labelling effort for efficient learning with georeferenced images | |
| CN115062779A (en) | Event prediction method and device based on dynamic knowledge graph | |
| CN116402777B (en) | Power equipment detection method and system based on machine vision | |
| CN109101984B (en) | Image identification method and device based on convolutional neural network | |
| CN112529025A (en) | Data processing method and device | |
| CN116108363A (en) | Incomplete multi-view multi-label classification method and system based on label guidance | |
| CN115858388A (en) | Test case priority ordering method and device based on variation model mapping chart | |
| CN113033716B (en) | Image mark estimation method based on confrontation fusion crowdsourcing label | |
| CN114510592A (en) | Image classification method and device, electronic equipment and storage medium | |
| Canchila et al. | Hyperparameter Optimization and Importance Ranking in Deep Learning–Based Crack Segmentation |
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 |