CN111754480B - Crystal back defect map retrieval and early warning method, storage medium and computer equipment - Google Patents
Crystal back defect map retrieval and early warning method, storage medium and computer equipment Download PDFInfo
- Publication number
- CN111754480B CN111754480B CN202010577051.6A CN202010577051A CN111754480B CN 111754480 B CN111754480 B CN 111754480B CN 202010577051 A CN202010577051 A CN 202010577051A CN 111754480 B CN111754480 B CN 111754480B
- Authority
- CN
- China
- Prior art keywords
- defect map
- crystal
- back defect
- database
- crystal back
- 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
- 230000007547 defect Effects 0.000 title claims abstract description 284
- 239000013078 crystal Substances 0.000 title claims abstract description 175
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000013528 artificial neural network Methods 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims description 21
- 238000012545 processing Methods 0.000 claims description 10
- 230000009467 reduction Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 4
- 238000010586 diagram Methods 0.000 abstract description 15
- 230000008569 process Effects 0.000 abstract description 13
- 238000001514 detection method Methods 0.000 abstract description 10
- 235000012431 wafers Nutrition 0.000 description 36
- 238000004590 computer program Methods 0.000 description 6
- 238000004519 manufacturing process Methods 0.000 description 6
- 239000004065 semiconductor Substances 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 238000011161 development Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 101100453960 Drosophila melanogaster klar gene Proteins 0.000 description 2
- 238000011109 contamination Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000001459 lithography Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000000465 moulding Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000000206 photolithography Methods 0.000 description 1
- 229910021420 polycrystalline silicon Inorganic materials 0.000 description 1
- 229920005591 polysilicon Polymers 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004626 scanning electron microscopy Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30148—Semiconductor; IC; Wafer
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Quality & Reliability (AREA)
- Testing Or Measuring Of Semiconductors Or The Like (AREA)
Abstract
The invention provides a crystal back defect detection and early warning method, a storage medium and computer equipment, which are characterized in that a known crystal back defect map in a database is trained through a self-encoder neural network to obtain a crystal back defect map retrieval model, the high-dimensional characteristics of the known crystal back defect map are extracted by utilizing the crystal back defect retrieval model, and a first coding database is generated by coding the high-dimensional characteristics; for a crystal back defect map to be searched, searching second coded data of the model by utilizing the crystal back defect map, and searching the second coded data in the first coded database by utilizing a nearest neighbor algorithm; and obtaining whether the wafer back defect map to be searched is unknown defect early warning or known defect type according to the existence or non-existence of the second encoded data in the first encoded database. Therefore, the labor cost is saved, the judging time is greatly shortened, the efficiency is improved, the searching precision is improved based on the existing known crystal back defect diagram, the omission ratio and the false alarm rate are reduced, and important guarantee is provided for improving the yield of the subsequent process.
Description
Technical Field
The present invention relates to a method for manufacturing semiconductor integrated circuits, and more particularly, to a method for searching and early warning a wafer back defect map, a storage medium, and a computer device.
Background
In the semiconductor chip manufacturing process, with the continuous improvement of the integration level of the semiconductor chip, the size of the semiconductor device is smaller and smaller, and the molding requirement of the device is stricter and stricter. In particular, the photolithography process requires not only good uniformity of crystal planes of the wafer, but also that the wafer back cannot have defects such as contamination, imprinting, and the like. However, it is difficult to avoid the occurrence of various back of the wafer defect patterns at the beginning of new technology nodes and/or new products because back of the wafer defect patterns may occur for various reasons, such as the ease of marking the back of the wafer when the processing tool contacts a contact feature on the back of the wafer, such as a back side fixture. These imprints may form specific pattern residues with the subsequent process, and these pattern residues may form ring-shaped polysilicon (Poly) residues, thereby affecting the lithography process on the crystal plane, forming defocusing (defocus), and seriously affecting the product yield. Therefore, there is a need for timely and accurate discovery of wafer back contamination or defects during semiconductor manufacturing.
In the prior art, the most commonly used method for detecting the crystal back defect map comprises the following steps: yield engineers distinguish the shape, size, composition, etc. of the back-of-the-crystal defect map by Scanning Electron Microscopy (SEM) to find its cause. As can be appreciated, this method suffers from the following drawbacks: firstly, by means of manual judgment, not only is a yield engineer responsible for detection required to have abundant experience, but also the time and the labor are consumed, and the efficiency is low; secondly, the situation that misjudgment or omission occurs is difficult to avoid by manual operation. And once erroneous judgment or omission occurs, the yield is possibly low, and even the wafer is scrapped.
Therefore, the back defect map searching and early warning method is provided, so that the back defect map can be accurately and rapidly judged automatically, and the reason for generating the back defect map is rapidly found to be a technical problem to be solved by the person skilled in the art.
It should be noted that the information disclosed in this background section is only for enhancement of understanding of the general background of the invention and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person skilled in the art.
Disclosure of Invention
The invention aims to provide a crystal back defect map retrieval and early warning method, a storage medium and computer equipment, which are used for solving the technical problems of low retrieval efficiency, missed judgment and misjudgment in the crystal back defect map retrieval in the prior art.
In order to achieve the above purpose, the present invention is realized by the following technical scheme: a method for searching and early warning a crystal back defect map comprises the following steps:
s1: training a self-coding neural network by taking a known crystal back defect map in a known crystal back defect map database as a training sample set until a crystal back defect map retrieval model is obtained; respectively carrying out feature coding on each known crystal back defect map by using the crystal back defect map retrieval model to obtain a first coding database;
s2: performing feature coding on the wafer back defect map to be searched by using the wafer back defect map model to obtain second coded data;
s3: judging whether a candidate crystal back defect map of the crystal back defect map to be searched exists in the known crystal back defect map database or not by searching the second coding data in the first coding database by utilizing a nearest neighbor algorithm;
if so, judging the defect type of the wafer back defect map to be searched according to the defect type of the candidate wafer back defect map;
if the defect map does not exist, giving an early warning that the defect map of the back of the crystal to be searched is an unknown defect;
the wafer back defect map retrieval model meets the following conditions, each piece of first coded data in the first coded database is subjected to feature decoding by using the wafer back defect map retrieval model to obtain a reconstructed wafer back defect map data set, and the error of each reconstructed wafer back defect map and the corresponding known wafer back defect map is in a first preset threshold range; wherein the feature decoding is an inverse of the feature encoding.
Optionally, before step S1, performing image normalization processing on the original picture of each known crystal back defect map to obtain normalized pictures with uniform size and identical channels;
in step S1, the step of using the known back of crystal defect map in the known back of crystal defect map database as a training sample set includes using all the normalized pictures as the training sample set.
Optionally, the wafer back defect map retrieval model includes an input layer, a residual layer, and an output layer, the residual layer being respectively connected to the input layer and the output layer, wherein,
the input layer comprises a plurality of first convolution layers;
the residual layers comprise a plurality of residual sublayers, each residual sublayer is obtained by superposing at least one convolution block and at least one identification block, wherein each convolution block comprises a plurality of second convolution layers and a convolution shortcut; each identification block comprises a plurality of third convolution layers;
the output layer comprises a plurality of full connection layers.
Optionally, the first convolution layer is two,
and/or
The number of the residual sub-layers is three, each residual sub-layer comprises a convolution block and an identification block, each convolution block comprises three second convolution layers and a convolution shortcut, and each identification block comprises three third convolution layers;
and/or
The number of the full connection layers is four.
Optionally, in step S1, the using the back-of-crystal defect map retrieval model performs feature encoding on each of the known back-of-crystal defect maps to obtain a first encoded database, where the feature encoding method includes the steps of,
s11: the input layer receives the training sample set and reduces the dimension of the training sample set to obtain a first characteristic dimension data set with a characteristic dimension of a first dimension and a depth of a first depth;
s12: the residual layer receives the first characteristic dimension data set, and performs dimension reduction on the first characteristic dimension data set to obtain a second characteristic dimension data set with a characteristic dimension of a second dimension and a depth of a second depth;
s13: the output layer receives the second feature dimension data set, reduces the dimension of the second feature dimension data set to obtain a third feature dimension data set with a third dimension of feature dimension and a third depth of depth, and the third feature dimension data set is used as the first coding database.
Optionally, the second dimension is one eighth of the first dimension and/or the second depth is eight times the first depth.
Optionally, in step S3, a method for determining whether the candidate back defect map of the back defect map to be searched exists in the known back defect map database by searching the second encoded data in the first encoded database by using a nearest neighbor algorithm, includes,
calculating Euclidean distance, hamming distance or cosine similarity between the first coded data and the second coded data in the first coded data set to obtain similarity between the crystal back defect map to be searched and the known crystal back defect map, thereby obtaining a candidate crystal back defect map list of the crystal back defect map to be searched;
rearranging the candidate crystal back defect map list according to the size of the similarity, and if the candidate crystal back defect map with the similarity within a second preset threshold value range exists, judging that the candidate crystal back defect map of the crystal back defect map to be searched exists in the known crystal back defect map database; otherwise, judging that the candidate crystal back defect map of the crystal back defect map to be searched does not exist in the known crystal back defect map database.
Optionally, if it is determined in step S3 that there is no candidate back defect map corresponding to the known back defect map database, performing the following steps,
and (3) using the crystal back defect map to be searched as a known crystal back defect map to amplify the known crystal back defect map database, and upgrading the crystal back defect map search model and the first coding database by using the same method in the step S1.
Based on the same inventive concept, the invention also provides a computer readable storage medium, wherein the computer readable storage medium stores computer executable instructions, and when the computer executable instructions are executed, the method for searching and early warning the crystal back defect map is realized.
Based on the same inventive concept, the invention further provides a computer device comprising a processor and a storage device, the processor being adapted to implement instructions, the storage device being adapted to store instructions adapted to be loaded by the processor and to perform the back of crystal defect map retrieving and early warning method of any one of claims 1 to 8.
Compared with the prior art, the wafer back defect detection and early warning method provided by the invention has the following beneficial effects:
according to the back of crystal defect detection and early warning method provided by the invention, a large number of existing known back of crystal defect graphs in a database are trained through a self-encoder neural network, and the generated back of crystal defect graph retrieval model can rapidly and accurately extract the high-dimensional characteristics of the known back of crystal defect graphs and encode the high-dimensional characteristics to generate a first encoding database; for the newly generated crystal back defect map to be searched, the crystal back defect map searching model is utilized to generate characteristic second coding data as well, and the nearest neighbor algorithm is utilized to search the second coding data in the first coding database; if the second coded data exist in the first coded database, giving an unknown defect early warning to the crystal back defect map to be searched, so that related technicians can trace the root of the problem; if the defects exist, the defect type of the crystal back defect map to be searched is determined according to the defect type of the corresponding known crystal back defect map in the known crystal back defect map database, so that the current manual judgment mode is changed into full-automatic judgment, the labor cost is saved, the judgment time is greatly shortened, the efficiency is improved, the searching precision is improved based on the existing known crystal back defect map, the omission ratio and the false alarm ratio are reduced, and important guarantee is provided for improving the yield of the subsequent process.
Further, for the back defect map to be searched which does not exist in the known back defect map database, the back defect map to be searched is further used as the known back defect map to amplify the known back defect map database, and the same method as the step S1 is used for upgrading the back defect map search model and the first coding database. Therefore, the crystal back defect retrieval model has good openness, can be updated and expanded continuously along with the discovered defect types of the crystal back defect map, is further suitable for continuous development of new technology nodes and rapid development of related equipment technologies, and has good adaptability.
Because the readable storage medium and the computer equipment provided by the invention belong to the same conception with the crystal back defect map retrieval and early warning method, the invention has at least the same beneficial effects and is not repeated.
Drawings
FIG. 1 is a schematic flow chart of a method for searching and pre-warning a wafer back defect map according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a back defect map search model in FIG. 1;
FIG. 3 is a schematic diagram of a data processing process of a backside defect map retrieving and pre-warning method in FIG. 1;
FIG. 3A is a schematic diagram of the feature encoding process of FIG. 3;
FIG. 3B is a schematic diagram illustrating the result of one implementation of the nearest neighbor algorithm in FIG. 3;
FIG. 4 is a diagram of a candidate back of crystal defect map and a picture of the back of crystal defect map to be inspected corresponding to FIG. 3B;
FIG. 5 is a schematic diagram showing the relationship between the number of similar known back-of-crystal defect maps and the related detection index in one embodiment of FIG. 3;
wherein reference numerals are as follows:
100-input layer, 200-residual layer, 300-output layer.
Detailed Description
In order to make the objects, advantages and features of the present invention more apparent, the method for retrieving and pre-warning the defect map of the back of the body, the storage medium and the computer device according to the present invention will be described in further detail below with reference to the accompanying drawings. It is apparent that the method described herein comprises a series of steps, and that the order of the steps presented herein is not necessarily the only order in which the steps may be performed, and that some of the described steps may be omitted and/or some other steps not described herein may be added to the method. Further, the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In order to facilitate understanding of the present invention, the sources of the database of known back-of-crystal defect maps will be briefly described before describing embodiments of the present invention. The Klarity Defect system is widely used as a Defect monitoring system in the manufacture and production of semiconductors. After the Defect detection machine detects the Defect signals, further inspection is needed on the Defect inspection machine for the wafers with the out-of-standard Defect number, specific photos of the defects are obtained and classified, after a yield engineer searches the Defect reasons of the wafers, the original photos of the wafer back Defect map and the Defect reasons (types) are stored into a Klar Defect server through a Klar Defect system, so that the known wafer back Defect map database not only comprises known wafer back Defect maps, but also comprises Defect types corresponding to each known wafer back Defect map.
In order to shorten the retrieval efficiency of the crystal back defect map and reduce the omission factor and the omission factor, the inventor provides a crystal back defect map retrieval and early warning method based on a known crystal back defect map database through intensive research and continuous practice. The crystal back defect map retrieval and early warning method is based on a known crystal back defect map database and a self-coding neural network, so that the retrieval efficiency of crystal back defects is shortened, and the omission ratio of crystal back defect retrieval in the prior art are reduced. For ease of understanding, the core concept of the self-encoding neural network is briefly described as follows: the neural network model is constructed to comprise a model input layer, a model hidden layer and a model output layer, wherein the model is a process of collecting a picture, compressing the picture by the neural network after receiving the picture, and finally recovering from the compressed picture, namely, decompressing the picture of the self-coding model after being compressed. When compressing, the quality of original picture is reduced (dimension reduction), when decompressing, the original picture is restored by using the file with small information content but containing all information. Now we assume that the information a in the model input layer is decompressed into the hidden layer to obtain a, then the a of the model hidden layer is compared with the a of the input layer to obtain a prediction error, then reverse transfer is performed, then the accuracy of self-coding is gradually improved, and the data a of one part obtained in the middle hidden layer after training for a period of time is the essence of source data. Self-coding is an unsupervised learning, and the first half of self-coding (dimension reduction), i.e. the coding process, can obtain the main characteristics of the source data.
Referring to fig. 1 and fig. 3, an embodiment of a back defect map retrieving and early warning method according to the present invention includes the following steps:
s1: training to obtain a crystal back defect map retrieval model and using the crystal back defect map retrieval model to obtain a first coding database: training a self-coding neural network by taking a known crystal back defect map in a known crystal back defect map database as a training sample set until a crystal back defect map retrieval model is obtained; and respectively carrying out feature coding on each known crystal back defect map by using the crystal back defect map retrieval model to obtain a first coding database.
S2: and carrying out feature coding on the wafer back defect map to be searched by utilizing the wafer back defect map model to obtain second coded data.
S3: and judging whether the candidate crystal back defect map of the crystal back defect map to be searched exists in the known crystal back defect map database by searching the second coding data in the first coding database by utilizing a nearest neighbor algorithm. If so, judging the defect type of the wafer back defect map to be searched according to the defect type of the candidate wafer back defect map; if the defect map does not exist, giving out that the defect map to be searched is an unknown defect early warning. The wafer back defect map retrieval model meets the following conditions, each piece of first coded data in the first coded database is subjected to feature decoding by using the wafer back defect map retrieval model to obtain a reconstructed wafer back defect map data set, and the error of each reconstructed wafer back defect map and the corresponding known wafer back defect map is in a first preset threshold range; wherein the feature decoding is an inverse of the feature encoding.
Preferably, in one embodiment, before step S1, the method further includes performing image normalization processing on the original pictures of each of the known crystal back defect maps to obtain normalized pictures with uniform size and identical channels. Therefore, in step S1, the taking the known back of crystal defect map in the database of known back of crystal defect maps as a training sample set includes taking all the normalized pictures as the training sample set. For example, if the original picture of the crystal back defect map is 1024×1024 pictures with RGB three channels, in one mode, after the image normalization processing is performed, the obtained normalized picture is a 128×128 single channel picture, and then the normalized picture is input into the self-coding neural network in a tensor form.
Preferably, in one embodiment, referring to fig. 3A, in step S1, the feature encoding is performed on each of the known crystal back defect maps by using the crystal back defect map retrieval model, so as to obtain a first encoded database, where the feature encoding method includes the following steps:
s11: the input layer 100 receives the training samples, and reduces the dimensions of the training sample set to obtain a first feature dimension data set with a feature dimension of a first dimension and a depth of a first depth.
S12: the residual layer 200 receives the first feature dimension data set, and performs dimension reduction on the first feature dimension data set to obtain a second feature dimension data set with a feature dimension of a second dimension and a depth of a second depth.
S13: the output layer 200 receives the second feature dimension data set, and performs dimension reduction on the second feature dimension data set to obtain a third feature dimension data set with a feature dimension being a third dimension. Until the third feature dimension dataset is taken as the first encoding database. According to the above description, since the back-of-crystal defect map retrieval model needs to meet the following conditions, performing feature decoding on each first encoded data in the first encoded database by using the back-of-crystal defect map retrieval model to obtain a reconstructed back-of-crystal defect map data set, wherein an error between each reconstructed back-of-crystal defect map and the known back-of-crystal defect map corresponding to the reconstructed back-of-crystal defect map is within a first preset threshold range; thus, the above-described processes of step S11, S12, and S13 feature encoding, that is, the process of model construction.
Preferably, referring to fig. 2, in one embodiment, the back-of-crystal defect map retrieval model includes an input layer 100, a residual layer 200, and an output layer 300, where the residual layer 200 connects the input layer 100 and the output layer 200, respectively. Wherein the input layer 100 includes a plurality of first convolution layers; the residual layer 200 comprises a plurality of residual sublayers, each residual sublayer is obtained by superposing at least one convolution block and at least one identification block, wherein each convolution block comprises a plurality of second convolution layers and a convolution shortcut; each identification block comprises a plurality of third convolution layers; the output layer 300 includes a number of fully connected layers. Preferably, referring to fig. 2, there are two first convolution layers, three residual sublayers, each residual sublayer includes a convolution block and an identification block, each convolution block includes three second convolution layers and one convolution shortcut, and each identification block includes three third convolution layers; the number of the full connection layers is four. Obviously, the foregoing description of the preferred embodiment is only implemented, and not the limitation of the present invention, it is understood that the number of the first convolution layers, the number of the residual sub-layers, the number of the convolution blocks, the number of the second convolution layers, and the number of the third convolution layers may be set according to the actual situation, but are all within the scope of the present invention. Further, in one embodiment, after passing through the input layer 100, the data in the training sample set is reduced to a first feature dimension data set with a first dimension of 32×32 and a first depth of 64; the residual layer 200 receives the first feature dimension dataset, after passing through three residual sublayers, after the convolution blocks of each residual sublayer, each convolution block reduces the feature dimension of the data in the first feature dimension dataset to one half of the original, and each identification block does not change the feature dimension of the data in the first feature dimension dataset, thereby, each time the data passes through each residual sublayer, the feature dimension of the data in the first feature dimension dataset is reduced to one half of the original, the depth feature is doubled, and therefore, the feature dimension is 32×32, the feature dimension is reduced to 4×4 after the first feature dimension dataset with the first depth of 64 passes through the residual layer 200, and the feature dimension is reduced to 1024 second feature dimension dataset with the second depth of 1024. The output layer 300 performs data feature unidimensional on the second feature dimension data set, and after passing through four fully connected layers, the feature dimension is reduced to 100, that is, the third feature dimension data set with the feature dimension of 100 is the first coding database.
Preferably, in one embodiment, referring to fig. 3B to fig. 5, the method for determining whether the candidate back of crystal defect map of the back of crystal defect map to be searched exists in the known back of crystal defect map database by searching the second coding data in the first coding database by using a nearest neighbor algorithm includes calculating a euclidean distance between the first coding data and the second coding data in the first coding data set, so as to obtain a similarity between the back of crystal defect map to be searched and the known back of crystal defect map, thereby obtaining a list of candidate back of crystal defect maps of the back of crystal defect map to be searched; rearranging the candidate crystal back defect map list according to the size of the similarity, and if the candidate crystal back defect map with the similarity within a second preset threshold value range exists, judging that the candidate crystal back defect map of the crystal back defect map to be searched exists in the known crystal back defect map database; otherwise, judging that the candidate crystal back defect map of the crystal back defect map to be searched does not exist in the known crystal back defect map database. As shown in fig. 3B, it is obvious that, in other embodiments, the similarity between the back defect map to be searched and the known back defect map may be obtained by calculating the hamming distance or cosine similarity between the first encoded data and the second encoded data in the first encoded data set. Fig. 4 is a schematic diagram of a picture of the back defect map to be inspected and a list of the most similar 4 candidate back defect maps corresponding to fig. 3B; fig. 5 is a schematic diagram of a relationship between the number of similar known crystal back defect graphs and related detection indexes in one embodiment of fig. 3, and it can be seen from the graph that, by using the method for detecting and early warning crystal back defects provided by the invention, no missing detection exists (the missing detection rate is 0%), the false alarm rate is far lower than 20%, and the detected defects are abnormal defects and are 100% recalled. It should be noted that the foregoing description is merely illustrative of the preferred embodiment, and not restrictive to the present invention, the nearest neighbor algorithm may use a K nearest neighbor basic algorithm, or may use an improved algorithm thereof, and in other embodiments, the similarity between the back-of-crystal defect map to be searched and the back-of-crystal defect map may be obtained by calculating a manhattan distance between the first encoded data and the second encoded data in the first encoded data set.
With continued reference to fig. 1, in one embodiment, if it is determined in step S3 that there is no corresponding candidate back-of-crystal defect map in the known back-of-crystal defect map database, then the following steps are performed,
s4: and (3) using the crystal back defect map to be searched as a known crystal back defect map to amplify the known crystal back defect map database, and upgrading the crystal back defect map search model and the first coding database by using the same method in the step S1.
Therefore, according to the crystal back defect detection and early warning method provided by the invention, the existing large number of known crystal back defect graphs in the database are trained through the self-encoder neural network, and the generated crystal back defect graph retrieval model can rapidly and accurately extract the high-dimensional characteristics of the known crystal back defect graphs and encode the high-dimensional characteristics to generate a first encoding database; for the newly generated crystal back defect map to be searched, the crystal back defect map searching model is utilized to generate characteristic second coding data as well, and the nearest neighbor algorithm is utilized to search the second coding data in the first coding database; if the second coded data exist in the first coded database, giving an unknown defect early warning to the crystal back defect map to be searched, so that related technicians can trace the root of the problem; if the defects exist, the defect type of the crystal back defect map to be searched is determined according to the defect type of the corresponding known crystal back defect map in the known crystal back defect map database, so that the current manual judgment mode is changed into full-automatic judgment, the labor cost is saved, the judgment time is greatly shortened, the efficiency is improved, the searching precision is improved based on the existing known crystal back defect map, the omission ratio and the false alarm ratio are reduced, and important guarantee is provided for improving the yield of the subsequent process.
Further, for the back defect map to be searched which does not exist in the known back defect map database, the back defect map to be searched is further used as the known back defect map to amplify the known back defect map database, and the same method as the step S1 is used for upgrading the back defect map search model and the first coding database. Therefore, the crystal back defect retrieval model has good openness, can be updated and expanded continuously along with the discovered defect types of the crystal back defect map, is further suitable for continuous development of new technology nodes and rapid development of related equipment technologies, and has good adaptability.
Based on the same inventive concept, still further, a computer readable storage medium is provided, where computer executable instructions are stored, and when the computer executable instructions are executed, the steps of the method for retrieving and pre-warning a crystal back defect map are implemented, and specific steps are described above and are not repeated herein.
Based on the same inventive concept, still further, a computer device is provided according to yet another embodiment of the present invention, where the computer device includes a processor and a storage device, where the processor is adapted to implement instructions, and where the storage device is adapted to store a plurality of instructions, where the instructions are adapted to be loaded by the processor and to execute the method for retrieving and pre-warning a crystal back defect map according to any one of the foregoing embodiments.
From the description of the implementations above, those skilled in the art will appreciate that embodiments of the invention may be provided as a method, system, or computer program product. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects but in many cases the former is a preferred embodiment. Based on such understanding, the portions of the technical solutions of the present invention that contribute to the prior art can be embodied in the form of a computer software product stored on a computer readable storage medium, including but not limited to disk storage, CD-ROM, optical storage, etc.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Because the readable storage medium and the computer device provided by the embodiment of the invention belong to the same conception with the crystal back defect map searching and early warning method, the method has at least the same beneficial effects and is not repeated one by one.
The foregoing is merely a preferred embodiment of the present invention and is not intended to limit the scope of the present invention, which includes but is not limited to the configurations set forth in the foregoing description. Various modifications and variations of the embodiments of the invention will be apparent to those skilled in the art in light of the foregoing description. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (8)
1. The method for searching and early warning the crystal back defect map is characterized by comprising the following steps of:
s1: training a self-coding neural network by taking a known crystal back defect map in a known crystal back defect map database as a training sample set until a crystal back defect map retrieval model is obtained; respectively carrying out feature coding on each known crystal back defect map by using the crystal back defect map retrieval model to obtain a first coding database;
s2: performing feature coding on the wafer back defect map to be searched by using the wafer back defect map model to obtain second coded data;
s3: judging whether a candidate crystal back defect map of the crystal back defect map to be searched exists in the known crystal back defect map database or not by searching the second coding data in the first coding database by utilizing a nearest neighbor algorithm;
if so, judging the defect type of the wafer back defect map to be searched according to the defect type of the candidate wafer back defect map;
if the defect map does not exist, giving an early warning that the defect map of the back of the crystal to be searched is an unknown defect;
the wafer back defect map retrieval model meets the following conditions, each piece of first coded data in the first coded database is subjected to feature decoding by using the wafer back defect map retrieval model to obtain a reconstructed wafer back defect map data set, and the error of each reconstructed wafer back defect map and the corresponding known wafer back defect map is in a first preset threshold range; wherein the feature decoding is an inverse of the feature encoding;
the crystal back defect map retrieval model comprises an input layer, a residual layer and an output layer, wherein the residual layer is respectively connected with the input layer and the output layer, and the input layer comprises a plurality of first convolution layers; the residual layers comprise a plurality of residual sublayers, each residual sublayer is obtained by superposing at least one convolution block and at least one identification block, wherein each convolution block comprises a plurality of second convolution layers and a convolution shortcut; each identification block comprises a plurality of third convolution layers; the output layer comprises a plurality of full-connection layers;
the feature encoding method comprises the following steps:
s11: the input layer receives the training sample set and reduces the dimension of the training sample set to obtain a first characteristic dimension data set with a characteristic dimension of a first dimension and a depth of a first depth;
s12: the residual layer receives the first characteristic dimension data set, and performs dimension reduction on the first characteristic dimension data set to obtain a second characteristic dimension data set with a characteristic dimension of a second dimension and a depth of a second depth;
s13: the output layer receives the second feature dimension data set, reduces the dimension of the second feature dimension data set to obtain a third feature dimension data set with a third dimension of feature dimension and a third depth of depth, and the third feature dimension data set is used as the first coding database.
2. The method for retrieving and pre-warning a crystal back defect map according to claim 1, wherein before step S1, further comprising performing image normalization processing on an original picture of each known crystal back defect map to obtain normalized pictures with uniform size and identical channels;
in step S1, the step of using the known back of crystal defect map in the known back of crystal defect map database as a training sample set includes using all the normalized pictures as the training sample set.
3. The method for retrieving and pre-warning a defect map of a wafer back according to claim 1, wherein,
the number of the first convolution layers is two,
and/or
The number of the residual sub-layers is three, each residual sub-layer comprises a convolution block and an identification block, each convolution block comprises three second convolution layers and a convolution shortcut, and each identification block comprises three third convolution layers;
and/or
The number of the full connection layers is four.
4. The method of claim 1, wherein the second dimension is one eighth of the first dimension and/or the second depth is eight times the first depth.
5. The method for retrieving and pre-warning a back-of-crystal defect map according to claim 1, wherein in step S3, the method for determining whether the candidate back-of-crystal defect map of the back-of-crystal defect map to be retrieved exists in the known back-of-crystal defect map database by retrieving the second encoded data in the first encoded database using a nearest neighbor algorithm comprises,
calculating Euclidean distance, hamming distance or cosine similarity between the first coded data and the second coded data in the first coded data set to obtain similarity between the crystal back defect map to be searched and the known crystal back defect map, thereby obtaining a candidate crystal back defect map list of the crystal back defect map to be searched;
rearranging the candidate crystal back defect map list according to the size of the similarity, and if the candidate crystal back defect map with the similarity within a second preset threshold value range exists, judging that the candidate crystal back defect map of the crystal back defect map to be searched exists in the known crystal back defect map database; otherwise, judging that the candidate crystal back defect map of the crystal back defect map to be searched does not exist in the known crystal back defect map database.
6. The method for retrieving and pre-warning a back defect map according to claim 1, wherein if it is determined in step S3 that there is no candidate back defect map corresponding to the known back defect map database, performing the following steps,
and (3) using the crystal back defect map to be searched as a known crystal back defect map to amplify the known crystal back defect map database, and upgrading the crystal back defect map search model and the first coding database by using the same method in the step S1.
7. A computer-readable storage medium having stored thereon computer-executable instructions, wherein the computer-executable instructions, when executed, implement the back-of-crystal defect map retrieval and pre-warning method of any one of claims 1 to 6.
8. A computer device comprising a processor adapted to implement instructions and a storage device adapted to store instructions adapted to be loaded by the processor and to perform the back of crystal defect map retrieval and early warning method of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010577051.6A CN111754480B (en) | 2020-06-22 | 2020-06-22 | Crystal back defect map retrieval and early warning method, storage medium and computer equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010577051.6A CN111754480B (en) | 2020-06-22 | 2020-06-22 | Crystal back defect map retrieval and early warning method, storage medium and computer equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111754480A CN111754480A (en) | 2020-10-09 |
| CN111754480B true CN111754480B (en) | 2024-04-16 |
Family
ID=72676432
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010577051.6A Active CN111754480B (en) | 2020-06-22 | 2020-06-22 | Crystal back defect map retrieval and early warning method, storage medium and computer equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111754480B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112907542B (en) * | 2021-02-24 | 2024-05-03 | 上海华力集成电路制造有限公司 | Crystal back defect detection method, storage medium and computer equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109885709A (en) * | 2019-01-08 | 2019-06-14 | 五邑大学 | A kind of image search method, device and storage medium based on from the pre- dimensionality reduction of coding |
| CN109977808A (en) * | 2019-03-11 | 2019-07-05 | 北京工业大学 | A kind of wafer surface defects mode detection and analysis method |
| CN110083729A (en) * | 2019-04-26 | 2019-08-02 | 北京金山数字娱乐科技有限公司 | A kind of method and system of picture search |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10999606B2 (en) * | 2019-01-08 | 2021-05-04 | Intel Corporation | Method and system of neural network loop filtering for video coding |
-
2020
- 2020-06-22 CN CN202010577051.6A patent/CN111754480B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109885709A (en) * | 2019-01-08 | 2019-06-14 | 五邑大学 | A kind of image search method, device and storage medium based on from the pre- dimensionality reduction of coding |
| CN109977808A (en) * | 2019-03-11 | 2019-07-05 | 北京工业大学 | A kind of wafer surface defects mode detection and analysis method |
| CN110083729A (en) * | 2019-04-26 | 2019-08-02 | 北京金山数字娱乐科技有限公司 | A kind of method and system of picture search |
Non-Patent Citations (2)
| Title |
|---|
| 北斗电文编码与解码的研究;易晨晖;何蓉;;电子世界(17);全文 * |
| 基于栈式卷积自编码的哈希图像检索研究;周纤;邱奕敏;吴振宇;;电视技术(10);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111754480A (en) | 2020-10-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| TW202217664A (en) | Pattern grouping method based on machine learning | |
| KR101331249B1 (en) | Method and apparatus for manufacturing data indexing | |
| CN112381788B (en) | Part surface defect increment detection method based on double-branch matching network | |
| CN106971953B (en) | Error detection method in manufacturing process | |
| US9142014B2 (en) | System and method for identifying systematic defects in wafer inspection using hierarchical grouping and filtering | |
| CN111754480B (en) | Crystal back defect map retrieval and early warning method, storage medium and computer equipment | |
| CN112381790A (en) | Abnormal image detection method based on depth self-coding | |
| Dey et al. | Deep learning-based defect classification and detection in SEM images | |
| CN111179263A (en) | Industrial image surface defect detection model, method, system and device | |
| CN114800041B (en) | Cutter state monitoring method and monitoring device thereof | |
| KR102623389B1 (en) | Method, apparatus and program for checking consistency among a plurality of manufacturing process equipment based on sensor data | |
| US5483636A (en) | Automated diagnosis using wafer tracking databases | |
| CN115082650A (en) | Implementation method of automatic pipeline defect labeling tool based on convolutional neural network | |
| KR102182678B1 (en) | Method and appratus for predicting fault pattern using multi-classifier based on feature selection method in semiconductor manufacturing process | |
| CN115423737A (en) | Wafer map defect detection method based on multiple principal component analysis networks | |
| US11423529B2 (en) | Determination of defect location for examination of a specimen | |
| TW202213191A (en) | Error cause estimation device and estimation method | |
| Aydin et al. | A deep learning-based hybrid approach to detect fastener defects in real-time | |
| Abdullah et al. | Identifying significant features for neural network-based particle contamination prediction in plasma etching process | |
| TWI834916B (en) | Machine learning-based defect detection of a specimen | |
| CN116342422A (en) | Defect identification method based on wafer map denoising | |
| US20230411224A1 (en) | System and method for detecting defects on a wafer and related non-transitory computer-readable medium | |
| CN116188877B (en) | Method and system for detecting and classifying unknown wafer defect categories | |
| US20230022057A1 (en) | Method for retrieving images from database | |
| CN117593745A (en) | Photoetching process window analysis method based on image segmentation and convolutional neural network |
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 |