CN103733322A - Method for providing a prediction model for crack detection and method for crack detection on a semiconductor structure - Google Patents
Method for providing a prediction model for crack detection and method for crack detection on a semiconductor structure Download PDFInfo
- Publication number
- CN103733322A CN103733322A CN201280038885.3A CN201280038885A CN103733322A CN 103733322 A CN103733322 A CN 103733322A CN 201280038885 A CN201280038885 A CN 201280038885A CN 103733322 A CN103733322 A CN 103733322A
- Authority
- CN
- China
- Prior art keywords
- crackle
- semiconductor structure
- descriptor
- measurement
- point
- 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
- 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
- G01N21/62—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light
- G01N21/63—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light optically excited
- G01N21/64—Fluorescence; Phosphorescence
- G01N21/6489—Photoluminescence of semiconductors
-
- 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
- G01N21/93—Detection standards; Calibrating baseline adjustment, drift correction
-
- 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
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/001—Industrial image inspection using an image reference approach
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01L—SEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
- H01L22/00—Testing or measuring during manufacture or treatment; Reliability measurements, i.e. testing of parts without further processing to modify the parts as such; Structural arrangements therefor
- H01L22/10—Measuring as part of the manufacturing process
- H01L22/12—Measuring as part of the manufacturing process for structural parameters, e.g. thickness, line width, refractive index, temperature, warp, bond strength, defects, optical inspection, electrical measurement of structural dimensions, metallurgic measurement of diffusions
-
- 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
- G06T2207/10048—Infrared 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
Abstract
The invention relates to a method for providing a prediction model for crack detection on a semiconductor structure that is a photovoltaic solar cell, a precursor of a photovoltaic solar cell in the production process, more particularly a semiconductor material for producing a solar cell, the method comprising the following steps: A) provision of a reference semiconductor structure having at least one crack; B) provision of crack data for the at least one crack, the crack data comprising geometric position data regarding the position of the crack on the reference semiconductor structure; C) spatially resolved scanning of the reference semiconductor structure by spatially resolved measurement of a plurality of spatial measurement points of the photoluminescence generated in the semiconductor structure and/or spatially resolved measurement of IR absorption of the semiconductor structure, and D) creation of a prediction model by training a learning algorithm on the basis of the spatially resolved measurement data acquired in step C and the crack data provided in step B, the training of the learning algorithm comprising the following steps: D1) creation of at least one descriptor for at least one spatial descriptor point by specifying or determining a test region for the descriptor point and creating the descriptor on the basis of the measurement data inside the test region, which descriptor is a feature vector and/or a feature distribution and/or a feature histogram, and D2) training the learning algorithm by means of the descriptor and the crack data. The invention further relates to a method and a device for crack detection.
Description
Technical field
The present invention relates to according to the method that the crack detection forecast model on semiconductor structure is provided of claim 1 with according to the method for the crackle on the detection semiconductor structure of claim 2, this semiconductor structure is the prime thing of the photovoltaic solar cell in photovoltaic solar cell, manufacture process or the semi-conducting material that is particularly useful for manufacturing such photovoltaic solar cell.
Background technology
The photovoltaic solar cell consisting of semi-conducting material is used to electromagnetic radiation to be converted into electric energy already.In typical photovoltaic solar cell, semi-conducting material has occupied the suitable share of solar cell total manufacturing cost.Thereby the very favourable such as polysilicon of material of use cost, in addition, an object of solar cell research is to reduce photovoltaic solar cell manufactures the thickness of semiconductor wafer used, reduces thus material cost.
But occur thus following danger, that is, crackle has adversely affected semi-conducting material stability.Such crackle for example may be because the fault in material in silicon crystal block forms, and because the load in the manufacture of wafer raw material occurs, the mechanical stress during because of defect and wafer sawing occurs, and the mechanical stress during because of transportation and carrying operation occurs.In addition, semi-conducting material bears Mechanical loading and heat load in manufacture process.
Once semi-conducting material is broken in the course of processing, just occurs sky high cost, because especially must stop, manufacturing.Although this semi-conducting material that has crackle but also experience whole manufacture process also likely causes the remarkable power loss when photovoltaic solar cell is used in application subsequently.
Therefore cause, people are in the urgent need to the method for the crackle for detection of in semiconductor structure, and this semiconductor structure is that above-mentioned semiconductor source material is as the solar cell prime thing in semiconductor wafer, manufacture process for the manufacture of solar cell or finished product solar cell.Because generally having micron order, crackle extends size, therefore it is also referred to as micro-crack.
Knownly utilize so-called IR printing opacity to take to detect crackle.For this reason, semiconductor structure is accepted printing opacity detection as measured in ccd video camera by imaging method by the radiation in infra-red range.Because the different absorption characteristics within the scope of IR, therefore user can find crackle by naked eyes in the image so obtaining.Known equally the crack detection of utilizing luminescence generated by light: for this reason, in semiconductor structure, produce luminescence generated by light and carry out measurement of spatial resolution by imaging method as ccd video camera.In the using priciple of photoluminescence measurement, be known and for example at the proceedings (2007 of the 22nd European photovoltaic solar meeting, Milan, Italy) in Trupke, T " for the luminescence imaging progress of the sign of silicon wafer and solar cell " has description.But effective noise spot of now conventionally crackle and other effect for example cannot being recombinated distinguishes.
In addition, the crack detection of utilizing resonance ultrasonic vibration (RUV) is known and for example at Monastyrskyi, A. wait in people's " for the resonance ultrasonic vibration of the online crack detection of silicon wafer and solar cell " (proceedings of the 33rd PVSC, 2008) and have description.
Summary of the invention
Task of the present invention is, provide a kind of reliably, semiconductor structure crack detecting method that especially can commercial Application, this semiconductor structure is photovoltaic solar cell prime thing in photovoltaic solar cell, manufacture process or the raw material for the manufacture of photovoltaic solar cell.
This task is by solving according to the method that the forecast model that on semiconductor structure, crack detection is used is provided of claim 1 and according to the method for the crackle on the detection semiconductor structure of claim 2.In claim 3 to 13, find the favourable execution mode of the inventive method.The present invention also completes by the device for detection of the crackle on semiconductor structure according to claim 14.The statement hereby all authority being required is quoted and is included specification in.
The present invention is based on applicant's following understanding, that is, learning algorithm is applicable to set up the forecast model that on semiconductor structure, crack detection is used, thereby carries out crack detection in conjunction with forecast model.Therefore, method of the present invention and device of the present invention are different from principle in photovoltaic field crack detecting method used so far, this be because for the first time by the machine learning by learning algorithm for crack detection.
As mentioned above, crack detecting method of the present invention and crack detection device of the present invention relate to photovoltaic field.Correspondingly at this with hereinafter use term " semiconductor structure " to call raw material for the manufacture of solar cell as silicon wafer, especially polycrystalline silicon wafer, also photovoltaic solar cell prime thing and the finished product solar cell at the arbitrary operation place in manufacture process in order to address.
According to method of the present invention, that be provided for the forecast model of the crack detection on semiconductor structure, comprise following methods step:
In method step A, provide with reference to semiconductor structure, this has at least one crackle with reference to semiconductor structure.
In method step B, the crackle data about at least one crackle are provided, this crackle packet containing with crackle at this with reference to the relevant geometric position data in the position on semiconductor structure.
This semiconductor structure is generally planar element, thereby generally can describe crackle about the attitude at the front of semiconductor structure or the back side, in other words, and generally by two-dimensional geometry data.Now within the scope of the present invention, crackle data comprise multiple points and/or the line of describing crack growth, and/or at least two lines that crackle data comprise star fracture are in this crossing Crack Center.
Also within the scope of the present invention, with reference to semiconductor structure, be the semiconductor structure with crackle, this crackle utilizes another mensuration measured about geometric position.Also feasible as an alternative or supplement, by delineation or similar approach, on the geometric position of regulation, give with reference to semiconductor structure and add crackle.
In method step C, carry out the measurement of spatial resolution with reference to semiconductor structure.For this reason, for multiple local measurement points, carry out the measurement of spatial resolution of the luminescence generated by light producing and/or the measurement of spatial resolution of semiconductor structure INFRARED ABSORPTION (IR absorption) in semiconductor structure.For carrying out spatial discrimination formation method, known measurement mechanism especially ccd video camera own can be used to measurement of spatial resolution.Importantly for multiple local measurement points, carry out measurement of spatial resolution.Measurement point covers lip-deep at least one region that comprises crackle of semiconductor structure.
In method step D, set up forecast model.For this reason, be combined in the measurement of spatial resolution data of obtaining in method step C and the crackle data that provide carried out the training of learning algorithm in method step B.
Be different from known crack detecting method, especially based on physical model, according to measuring-signal, do not distinguish measurement point and other measurement point corresponding to crackle.Replace, in conjunction with measurement data and crackle data about with reference to semiconductor structure known the description of existing crackle is carried out to the training (Trainieren) of learning algorithm.Therefore regulation does not show explanation and standard (that is, the clearly statement of this description) crackle and that with regard to measuring-signal, distinguish as effective geometry of recombinating with other element, but trains to form by learning algorithm.
The training of learning algorithm now comprises the following steps:
In method step D1, set up at least one descriptor at least one partial descriptions symbol point.Now, set or determine a test zone, and set up this descriptor according to the measurement data in this test zone for this descriptor point, this descriptor is that characteristic vector and/or feature distribute and/or feature histogram.
Thereby this test zone comprises at least one subset according to the measurement point of method step C.Also within the scope of the present invention be that this test zone comprises the whole measurement points according to method step C.
To this, descriptor point forms a partial points of described descriptor, thereby in conjunction with characteristic vector and/or feature distribution and/or feature histogram and be configured for the description of partial descriptions symbol point in the situation that considering test zone.The measurement data of the feature of characteristic vector and/or feature distribution and/or feature histogram based on obtaining in method step C.Now within the scope of the present invention be that this feature directly comes from measurement data.But especially advantageously this feature is further processing and/or the especially structure description of association of multiple measurement data.
In method step D2, utilize this descriptor and this crackle data training study algorithm.
This training can be carried out according to known known learning algorithm of mode utilization itself itself.Also within the scope of the present invention, only utilize a descriptor point and corresponding descriptor and/or only utilize a crackle and corresponding geometric data to train this learning algorithm.But advantageously, utilize the descriptor of multiple descriptor points and corresponding foundation to carry out described training.Or and especially also advantageously, utilize multiple crackles and each self-corresponding crackle data to carry out described method.
The training of learning algorithm can be carried out according to known mode own as mentioned above.Now importantly, crackle data based on predetermined and know whether on this descriptor point have crackle.It is vital that this information forms for the classification realizing by learning algorithm in training.
Within the scope of the present invention, multiplely with reference to semiconductor structure, be used to train this learning algorithm.Now not necessarily needing eachly has crackle with reference to semiconductor structure, and this is to training, to be also significant because there is no the descriptor point of crackle and descriptor.But one of them must have at least one crackle as described above with reference to semiconductor structure.
Method for detection of the crackle on semiconductor structure according to the present invention comprises following methods step:
In method step A, provide semiconductor structure.
Be different from reference to semiconductor structure, the semiconductor structure for carrying out crack detection certainly do not know in advance crackle geometric position in other words semiconductor structure have flawless information actually.
In method step B, provide forecast model, this forecast model is to set up by the training of learning algorithm, preferably according to the method for claim 1 or its preferred implementation, sets up.
Multiple local measurement points of the luminescence generated by light producing in semiconductor structure by measurement of spatial resolution in method step C and/or absorb this semiconductor structure of measurement of spatial resolution by the IR of this semiconductor structure of measurement of spatial resolution.Method step C thereby can be designed to identical with the method step C according to claim 1 or its favourable execution mode.Especially advantageously, when setting up forecast model and in crack detection, carry out identical metering system, be all in other words the measurement of spatial resolution of luminescence generated by light or be all IR absorb measurement of spatial resolution.
In method step D, at least one toposcopy point, determine on this checkpoint, whether there is crackle.This toposcopy point is at the lip-deep point of semiconductor structure and preferably corresponding to according to measurement point described in one of them of method step C.
Described definite following methods step that comprises:
In method step D1, for this checkpoint, set up at least one descriptor, for the descriptor of checkpoint, setting and determining a test zone, and setting up this descriptor in conjunction with the measurement data in test zone, this descriptor is that characteristic vector and/or feature distribute and/or feature histogram.
Therefore, method step D1 in crack detection is designed to similar according to the method step D1 in the forecast model foundation of claim 1, wherein, in claim 1, set up for the descriptor of a corresponding descriptor point with in claim 2 and set up the descriptor for a corresponding checkpoint.Before all and descriptor constitution and implementation mode subsequently thereby can at least similarly paid attention to according to the method step D1 of claim 1 with aspect the method step D1 of claim 2.
Therefore, to crack detection also importantly, for this checkpoint, set up a descriptor, this descriptor because of but be that characteristic vector and/or feature distribute and/or the feature of feature histogram form, thereby for the characteristic of this checkpoint, under the measurement data of considering in the test zone obtaining in method step C, be mapped in this descriptor.
Whether in method step D2, utilize this descriptor and forecast model to determine has crackle on this checkpoint.
Thereby, the feature of crack detecting method of the present invention is, need only for example by carrying out, according to the method for claim 1, provide a forecast model, according to known mode itself, with luminescence generated by light or IR absorptiometry form, carry out according to the measurement of method step C subsequently, and can form and utilize forecast model to determine on checkpoint, whether to have crackle by descriptor.
Now within the scope of the present invention be only a checkpoint, to carry out determining according to the above-mentioned of method step D2.Being preferably in multiple checkpoints carries out described definite.Thereby within the scope of the present invention, in method step C, at this, carry out the each described measurement point of measuring and be successively in succession elected to be as checkpoint.
Result, utilize crack detecting method of the present invention to guarantee whether definite semiconductor structure has the geometric position of crackle and definite crackle and/or extend size, without deriving by supposition, predetermined or physical model to be scheduled to according to the relation between the measurement data of method step C and abstract crackle characteristic.
Applicant's research shows, according to the formation of the forecast model of claim 1 and according to the use of claim 2 outstanding be applicable to detect the crackle in semiconductor structure.According to crack detecting method of the present invention thereby compared with known crack detecting method, significantly improve.Especially can when carrying out according to the method for claim 1, use in photoluminescence measurement method situation in method step C separately associative learning Algorithm for Training that the high differentiation of accuracy at target of for example recombinating between resulting structure at crackle and other geometry is provided, at this, not necessarily predetermine the standard of distinguishing crack structtire and other structure, but only by the training of learning algorithm, form.
The inventive method also has following advantage, can be divided into the foundation of forecast model and by forecast model for crack detection:
Therefore, for example can by accurate preparation, with reference to semiconductor structure and/or in conjunction with the training of many crackles and crackle data, be carried out accurately training study algorithm and the forecast model so forming is offered to user to use by manufacturer.
Therefore, not necessarily need user side also to carry out learning algorithm training.User also can be directly by the forecast model of being trained by manufacturer for crack detection.
In crack detecting method of the present invention, be preferably in method step E and carry out crackle reconstruction, in rebuilding, crackle determines the geometric data of crackle presentation.Thereby based on a checkpoint corresponding to crackle, rebuild, at least to determine the part of crackle and best whole geometrical representation. for this reasonEspecially advantageously, in method step E, for a partial reconstruction region around checkpoint, for the each measurement point in reconstruction area, determine orientation and by relatively carrying out to determine the measurement point corresponding to crackle by pattern identification means with similitude mulle and/or linearity pattern.Above-mentioned preferred implementation is especially when the use of photoluminescence measurement method and especially favourable during the use in polysilicon semiconductor structure.
The following understanding of this preferred implementation of the inventive method based on applicant, that is, the crackle in semiconductor structure generally has mulle.Crackle reconstruction is carried out in thus can utilization itself known mulle and/or the combination of the identification of linearity pattern and the method for reconstruction.
As mentioned above, within the scope of the present invention be, in method step D when setting up forecast model and/or when crack detection successively in succession according to multiple partial points as descriptor point or checkpoint, now for example can adopt all local measurement point measured in various method steps C.
But advantageously utilize key point to determine that method limits the quantity of descriptor point and/or checkpoint.Such method is also referred to as point of interest (POI) and determines method.Especially advantageously, in order to determine descriptor point and/or checkpoint, adopt key point search method, preferably adopt according to SIFT method, GLOH method and/or SURF method or definite for for example Tuscany oscilloscope of known method of feature detection or Harris's oscilloscope or key point similarly, that derive or similarly one of them method of method.Said method is own known and is for example described below:
" A performance evulation of local descriptors " (IEEE Transactions on Pattern Analysis and Machine Intelligence of-GLOH:Krystian Mikolajczyk and Cordelia Schmid, 10,27,1615-1630 page, 2005).
-SIFT:Lowe, David G.(1999) " Object recognition from local scale-invariant features " (international computer prospect conference proceedings 2. 1150-1157 pages,
doi:10.1109/ICCV.1999.790410.http://doi.ieeecs.org/10.1109/ICCV.1999.790410）。
-SURF:Bay, H., Tuytelaars, T., Gool, " SURF:Speeded Up Robust Features " (proceedings of the 9th prospective capabilities in hardware European Convention, 2006.05) of L.V..
" A combined corner and edge detector " (the 4th Alvey Vision Conference proceedings, 1988, the 147-151 pages) of-Harris:C.Harris and M.Stephens.
Applicant's research shows, key point is determined and especially advantageously utilized the controlled filter in orientation to determine described at least one key point.Advantageously utilize be particularly useful for computation and measurement data derivative preferably the filter of second dervative carry out describedly determining.Applicant's research shows, the bar chart filter especially consisting of Gaussian filter second dervative is applicable to key point and determines.
In order to select key point, especially from all measurement points, start to determine that key point in whole region within the scope of the present invention.Also within the scope of the present invention, predetermine in semiconductor structure lip-deep finite geometry region and/or the subset of the measurement point of measuring in method step C, determine therein key point.
When determining key point, the different qualities of regional area or measurement point and feature can be used for determining associated picture part, the distribution of the presentation of for example measure signal intensity or measure signal intensity value.This feature for example can describe and be used to determine this key point by definite quilt of factor factor as known in variance, mean value, symmetry or its itself.
Also can adopt feature presentation Ru Jiao, limit, line or circular configuration about geometrical representation.Such feature presentation for example can the filter response when with filter determine and/or by multiple filter banks and perhaps spinoff preliminary treatment and further process and determine.The example of available filters is that Suo Baier filter, Puli tie up special filter, Gaussian filter, Difference of Gaussian filter, Gauss Laplce.Use tensor for example Harris's oscilloscope, hessian determinant also within the scope of the present invention, or use controllable filter, use for example Gabor of small echo and Haar small echo also within the scope of the present invention.And, use Susan's Corner Detection device or canny edge detection device also within the scope of the present invention.
Applicant's research shows, the filtration that utilizes anisotropic bar chart filter is very favourable, this filter is the second dervative of Gaussian filter and for example at " Contour and Texture Analysis for Image Segmentation " (the prospective capabilities in hardware International Periodicals 43(1) of J.Malik, 7-27 page, 2001) in, there is description.
Now especially advantageously use the filter with different spaces orientation.Now, positive filter response is associated in the different spaces orientation situation of filter, preferably cumulative.Thereby, realization is about rotating S set O(2) Haar integration shape, as for example " Anwendung von Invarianzprinzipien zur Merkmalgewinnung in derMustererkennung Dissertation " (technology university, port, hamburger at H.Schulz-Mirbach, 1995.02.10, Nr.372, VDI publishes) as described in.
Applicant's research shows, the value obtaining thus especially Crack Center increase and be rotate constant.Thereby, thus can plain mode according to rotating, determine Crack Center, that is, determine the actual trend of crackle on survey map separately.Obtain thus following advantageous effects, that is, the crackle during with measurement or the rotation of semiconductor structure have nothing to do and obtain value of a size when analyzing, thereby are detected equally.
In addition, use there are different scales filter also within the scope of the present invention.Obtain thus with respect to crack size and change insensitive result.
Now especially advantageously, in the interpretation of result of different scale ranks, be considered for estimating the size for the treatment of consideration of regional: thereby corresponding high result refers to high scale rank, thereby the larger crackle of indication, thereby compared with the result of result when compared with low scale rank, advantageously according to larger image section, determine key point, crack detection and/or crackle reconstruction.
By key point, determine and obtain following advantage generally, the restricted number in potential crack region can be arrived to obviously less key point, thereby can obtain the speed-raising of the method.
Also within the scope of the present invention, first utilize other method of measurement to determine key point.In the case, the testing result that for example IR measures is used to determine key point in a conventional manner.Different method of measurement also can combine mutually: thereby the key point of for example measuring according to IR is definite within the scope of the present invention, and subsequently, analyzed for crack detection for the photoluminescence measurement of original key point of measuring acquisition with IR.
Preferably determining according to SIFT method, GLOH method, HOG method, LESH method and/or SURF method and/or at least one method of producing in conjunction with the feature of test zone in the method modification of descriptor according to the descriptor in the method step D1 of claim 1 and/or claim 2.As an alternative or supplement and advantageously, in order to set up descriptor, adopt the maximal filter response of adjustable and the preferred above-mentioned bar chart filter of rotating filter.LESH method itself is also known and for example at Sarfraz, S., Hellwich, O. in " Head Pose Estimation in Face Recognition across Pose Scenarios ", there is description (VISAPP2008 meeting, prospective capabilities in hardware theory and application message conference, Portugal Madeira, 235-242 page, the best student papers prize of 2008.01()).
Especially advantageously, according to the mode that comprises following methods step, set up descriptor:
In method step D1.1 for each partial points of test zone, preferably by determining for the parameter of each measurement point of this test zone, determine amplitude and orientation, this parameter further preferably responds and orientation for the gradient of the measured value of each measurement point or for the maximal filter of the controllable filter of each measurement point.
At least selected key point for test zone in method step D1.2 is set up feature histogram.
Now especially advantageously, also determine the main orientation of crackle and carry out the correction of feature histogram according to the main orientation of crackle, to obtain, rotating constant feature histogram.
Thereby when setting up, descriptor carries out the description in conjunction with the test zone around descriptor point or the checkpoint of feature.Feature is for example described and can be comprised of the isometric chart of some presentation of one or more features, feature distribution or its parameter or feature.Now can the known statement of the feature according to the descriptor in own known algorithm SIFT, GLOH, HOG, LESH or SURF of employing itself.The basis of setting up descriptor is definite measurement data in method step C always.
In above-mentioned preferred implementation, advantageously between method step D1.1 and D1.2, determining the main orientation of crackle and in method step D1.2, carrying out by setting up feature histogram with the measurement point in scanning algorithm scanography district.In this employing, according to the own known scanning algorithm of GLON and/or SIFT or its variant, be especially favourable.
In order to set up descriptor, therefore can adopt certain scanning and/or the division of image local.Especially can employing itself known to the scanning of algorithm known.
For setting up descriptor, carry out feature selecting, just as known in SIFT, GLOH, HOG, LESH or SURF algorithm.
Applicant's research shows, at this, adopts as described above the maximal filter response of anisotropic bar chart filter also especially favourable.Replace as the gradient orientation known from SIFT and GLOH and the use of gradient amplitude, thereby advantageously use orientation and the amplitude of the bar chart filter with maximal filter response.
Within the scope of the present invention be that preliminary treatment or aftertreatment feature are described when descriptor is set up.Especially can be constant by realize as described above rotation according to the main orientation correction of determined crackle descriptor.And can obtain the stability with respect to changes in contrast or noise by the convergent-divergent of descriptor and/or leveling.
Be preferably according to comprising following methods step in the method step D1 of claim 1 and/or claim 2 and determine this test zone:
First, as especially realized determining of at least one local key point about as described in claim 5 and 6 before.
Subsequently, determine the measurement point in this test zone, in the measurement point regulation for test zone, extend size and dimension how much.Preferably for test zone, set rectangle, ellipse or the border circular areas that its central point is key point.
In a favourable execution mode of the inventive method, the geometric position of crystal boundary is set and/or determines by measurement of spatial resolution method as the crystal boundary data in semiconductor structure, and is weighted according to crystal boundary data, wherein,
-when determining key point, as mentioned above, the partial points being positioned on crystal boundary is not selected as key point, or there is the weight lower than all the other partial points when determining key point, and/or
While setting up descriptor in-method step D when setting up forecast model and/or when crack detection, do not consider to be positioned at the partial points on crystal boundary, or consider to be positioned at the partial points on crystal boundary with the weight lower than all the other partial points, and/or
-in crackle is rebuild, as mentioned above, do not consider to be positioned at the partial points on crystal boundary or consider to be positioned at the partial points on crystal boundary with the weight lower than all the other partial points.
Thereby especially in application the inventive method, when being used for polycrystalline silicon wafer, provide this preferred implementation.As long as crystal boundary geometric position is known or the measuring equipment based on available can be simply definite, easily makeing mistakes property during crack detection thereby can again be reduced, its way is that crystal boundary positional information is used in the methods of the invention as described above.
For the learning algorithm of setting up forecast model or execution crack detection, it can be known learning algorithm itself.Especially, the use of neuron net or Bayes classifier or core engine within the scope of the present invention.Applicant's research shows, it is especially favourable using algorithm of support vector machine, and it is according to being this SVMs is identified for the feature description of dividing " cracking " class and " not cracking " class best hyperplane in order to complete classification task.Obtain thus remarkable advantage, because division face is preferably between class.The feature in class circle is not described does not have adverse effect to the alignment for the boundary line.In order to complete classification task by division face, data are mapped to a high-dimensional space by so-called " core skill ".This way itself is known and for example at Bernhard
" Learning with Kernels:Support Vector Machines, Regularization, Optimization; and Beyond " (" adaptive computation and machine learning ", MIT publishing house, Cambridge of Alex Smola, MA, 2002, ISBN0-262-19475-9) in have description.
As mentioned above, by learning algorithm, in descriptor feature statement with whether there is the relation between the Given information of crackle to be used to form this forecast model in certain partial points.Subsequently, can in conjunction with forecast model when the crack detection by unknown image section or the descriptor for this reason forming be categorized as crackle or flawless.
In crack detecting method of the present invention, advantageously in method step E, carrying out as mentioned above crackle rebuilds.The architectural characteristic of peripheral region, corresponding checkpoint that now, can be based in paid close attention to test zone is rebuild crack structtire.The architectural characteristic that belongs to the partial points of crackle is now opened with the architectural characteristic boundary that does not belong to crackle.Architectural characteristic can be position, density value, gradient and filter response in the case.
Applicant's research shows, advantageously in the star of supposition crackle or linear structure situation, carries out pattern identification.
Especially advantageously, in linearity or star fracture situation, so design is rebuild, that is, first preferably as described in above when key point search is described, determine a Crack Center, and limit the central point as test zone.
Subsequently, extract the architectural characteristic of this test zone measurement point.The description of one corresponding points of the mulle of the known pattern identification means of the description of key point measurement point around and basis itself compares.Demonstrate the structure example similar to the region of mulle and belong to crackle as therefore the structure in orientation belongs to crackle also high probability thereby can be assessed as.For the similitude of computation structure characteristic, can advantageously realize additional weight, for example, by each partial points density value, gradient or corresponding filter response.In the closely similar situation partial points that judges, whether belong to crackle.This judgement for example can the known sluggish method based on Crack Center of utilization itself complete.
Applicant's research shows, the orientation of structure and density are particularly useful for description scheme characteristic.
For computer azimuth, utilize the filtration of above-mentioned anisotropic bar chart filter in different azimuth to be also proved to be favourable here.Now advantageously, adopt maximal filter response and orientation thereof as structural information.
Accompanying drawing explanation
Below with reference to embodiment and accompanying drawing, further feature of the present invention and preferred implementation are described, in figure:
Fig. 1 has illustrated the luminescence generated by light picture that roughly has between two parties the polycrystalline silicon wafer of crackle at component in a), wherein, draws out the partial enlarged drawing of crackle peripheral region at component in b);
Fig. 2 be by bar chart filter for comprise crackle region time the view of maximal filter response;
Fig. 3 has illustrated around ungraded crackle orientation histogram in a) and the orientation histogram after leveling has been shown at component in b) at component;
Fig. 4 shows crackle reconstructed results.
Embodiment
Describe by reference to the accompanying drawings according to the method for the forecast model that is provided for crack detection of the present invention with according to the embodiment of crack detecting method of the present invention.
Polycrystalline silicon wafer is as with reference to semiconductor structure, and it is roughly square, and it has the about 15.6cm length of side and approximately 180 μ m thickness.This polycrystalline silicon wafer is to manufacture the typical raw material that photovoltaic solar cell is used.
For as form this semiconductor structure with reference to semiconductor structure, utilizing automatic percussion device is medially roughly that so-called punching point drops to semiconductor structure from regulation drop height by a metal tip, thereby forms crackle and known X coordinate and the Y coordinate of the geometric position Crack Center that rum point on wafer overlaps with metal tip.
In method step A, provide preparation like this with reference to semiconductor structure.In addition, known as mentioned above the location coordinates of Crack Center and thereby as crackle data, provide in method step B.
In method step C, carry out the measurement of spatial resolution with reference to semiconductor structure.Now, the known equipment for photoluminescence measurement of employing itself: by utilizing the semiconductor optical of electromagnetic irradiation to excite to produce electron-hole pair.Because regrouping process, by sending luminescence generated by light with reference to semiconductor structure, it is by silicon ccd video camera measurement of spatial resolution.Therefore, as the result of method step C, for many, be the netted measurement point being scattered in reference to semiconductor structure surface, have respectively a measured value, this measured value is corresponding to the intensity of the luminescence generated by light of obtaining for these measurement points or be at least associated with intensity.
Measurement result as shown in Figure 1a.The clear high spatial resolution of seeing (1024 × 1024) individual point.Roughly see between two parties at first star fracture structure.Crack structtire peripheral region indicates by black rectangle.Fig. 1 b illustrates the partial enlarged drawing of this rectangle, and this crack structtire in rectangle (in drawn black circle) is high-visible.
But Fig. 1 a also illustrates that with 1b multiple other structures have similar measured value, thereby the method for the measurement point corresponding to crackle and all the other measurement points being distinguished in conjunction with measuring-signal is proposed to be strict with.
Thereby, in this embodiment of the inventive method, in method step D, by the learning algorithm of the crackle data that are combined in measurement of spatial resolution data definite in method step C and provide in step B, train to set up a forecast model:
Now feasible, for each measurement point, set up respectively a descriptor and train subsequently this learning algorithm as described above.
But in embodiment described herein, first carry out the detection at potential crack center to search key point, thereby significantly limited processing duration and processing cost, because only key point is carried out to subsequent step.
Can adopt in principle in the case said method to search key point (so-called point of interest POI).But applicant's research shows, because crackle has star structure, therefore use anisotropy bar chart filter to search key point, be especially favourable.Adopt in the present embodiment the second dervative of anisotropic Gaussian filter.This filter is orientated to set up according to different azimuth, at this, is according to such orientation:
After measured value screening, the positive filter response of these bar chart filters is added up.Especially in being crackle crosspoint or region, Crack Center obtains thus high filter response.In order further to limit potential Crack Center, extract such measurement point, that is, its cumulative filter response exceeds defined threshold, preferably according to following formula 1:
Here, A(x, y) be when when having different azimuth orientation with the measurement point of coordinate (x, y), utilize the positive filter response of the image of bar chart filter to amount to.Distribute a binary bit value to each measurement point, it elects 1 as, as long as be more than or equal to the threshold value of the maximal filter response of being multiplied by all measured values for the filter response of this measurement point.In all the other situations, distribute 0 value.Therefore by selecting threshold value c to regulate sensitivity, c selects greatlyr, and the key point obtaining is fewer, otherwise or.Threshold value c dependence experience is obtained, and preferably so obtains, and comprises all Crack Center.The fact shows, Crack Center generally has very large value.
Fig. 2 illustrates Measurement and Data Processing, at the cumulative filter response separately for each pixel this illustrate.Know and see that Crack Center has very large value.
Then the structure, being associated is preferably reduced to a partial points.This realizes by morphological operations according to known mode own, and wherein, target perforation is contracted to a little.As a result, the point of remainder is key point thereupon, and wherein, each described key point is Crack Center potentially.
To each key point, in method step D1, set up descriptor respectively.Thereby each key point is successively selected as descriptor point in succession, is intended for respectively each test zone of descriptor point and sets up descriptor.
Two modification below declarative description symbol being set up: the first is set up by the descriptor of gradient, it two is to set up by the descriptor of bar chart filter.
While setting up descriptor by gradient in the present embodiment the first flexible program, set up similarly with known GLOH method own.First, consider gradient orientation.But because the noise in measured value significantly impacts gradient, therefore first by adding Gaussian filter to carry out measured value leveling.This way itself is known.By using Gaussian filter derivative convolution measurement data in x direction and y direction, measurement data is leveled, and obtains the gradient in x direction and y direction simultaneously.Obtain thus orientation and amplitude for the gradient vector of desired locations (x, y).The orientation usable levels [0,2 π] of two-dimensional space represents.
Therefore,, in the method step D1 of the present embodiment, for each key point, first limit a test zone.This be defined as in the present example take key point as the center of circle circle take 40 pixels as radius (measurement point).Subsequently, for the each measurement point in test zone, in method step B1.1, obtain as mentioned above gradient orientation and gradient amplitude.
In order to calibrate gradient orientation according to key point surrounding structure, determine first as follows the orientation of key point surrounding structure:
The gradient orientation of obtaining in advance in test zone is drawn in orientation histogram, and it for example has 36 bars, that is, 36 independent sortings, a uniform azimuthal bin is contained in each classification.Fig. 3 a shows this division result.Now may there is following situation, between peaked two bars, have a bar with smaller value having.Preferably this orientation with smaller value is not considered for the calculating of key point orientation.Therefore advantageously, by known method own, flatten this histogram.Fig. 3 b illustrates this leveling result.
In order to obtain and to rotate constant result now, that is, for the constant descriptor of the rotation of key point, make test zone with definite before key point azimuth rotation.Descriptor is set up can be indirectly by realizing according to the gradient orientation in key point orientation and the correction of descriptor.
Based on this, key point descriptor is for especially carrying out definite key point according to the measurement data of the measurement point in test zone.
Therefore, determined up to now such key point, its orientation is calculated and according to key point orientation, test zone is rotated.This way is own known and for example in (Lowe, " Distinctive Image features from scale-invariant keypoints " (prospective capabilities in hardware International Periodicals, 2004: the 91-110 pages) of D.G.), has description.
Then, for each key point, set up a descriptor now:
This can similarly realize with known GLOH method own.
The simple possibility mode that descriptor is set up be for example to set up characteristic vector according to gradient orientation.Each orientation or azimuthal bin indicate a feature.From there is gradient amplitude cumulative in this orientation, obtain the presentation of the feature in certain orientation.
Except foregoing description symbol is set up, can also the known sampling of employing itself.In the sampling of this utmost point, be especially favourable because crackle to be searched generally have with further from measurement point compared with the gradient amplitude of higher close Crack Center.Thereby the utmost point sampling that its center overlaps with key point is especially favourable.Another advantage is can be polar coordinates by simple conversion according to the above-mentioned test zone rotation in key point orientation.
Further improvement and GLOH that descriptor is set up similarly realize by test zone is divided into zones of different.For every sub regions, set up a characteristic vector.So, by the combination of the characteristic vector of different subregions, obtain descriptor.Obtain thus the additional information about regional structure, this additional information is highly beneficial to difference crackle and other structure.
Applicant's research shows, in above-mentioned employing gradient, sets up in a modification of descriptor, advantageously the filter response of above-mentioned bar chart filter is used for calculating descriptor.For this reason, replace ground, gradient orientation to select the orientation of bar chart filter, it produces maximal filter response.
Similar to above-mentioned way, replace gradient amplitude ground to select the respective strengths of maximal filter response.The difference of modification when this modification and employing gradient especially shows magnitude matrix: in the modification of employing gradient, crackle edge has maximum, and in the modification of filter response that adopts bar chart filter, itself has maximum crackle.
In principle, the modification that this descriptor forms and the difference of above-mentioned modification are only the selection of input: replace the Grad of key point test zone around, adopt now the filter response of anisotropic bar chart filter.Remaining similar process carries out.
Irrelevant with selected flexible program, then in method step D2, utilize set up descriptor and the crackle data that provide are carried out training study algorithm in method step B.
Applicant's research shows especially can reliably adopt SVMs.
The result of this embodiment of the inventive method because of but the forecast model that utilizes the training of SVMs to obtain.
In a second embodiment, will an embodiment of crack detecting method of the present invention be described now.
At this, adopt above-mentioned forecast model.
In method step A, provide semiconductor structure, and provide above-mentioned forecast model in method step B.
Then, in method step C, carry out the measurement of spatial resolution of semiconductor structure.The spatial discrimination photoluminescence measurement of this method step C preferably and in the first embodiment, that basis is described there similarly carries out.
As a result, for multiple local measurement points of semiconductor structure, have measurement data, described measurement data is associated with the current local strength of produced luminescence generated by light.
In method step D, at least one toposcopy point, determine on checkpoint, whether there is crackle.
Now also feasible in principle, for example for each described measurement point, carry out such determining.
But particularly advantageously, also in crack detection, first determine as mentioned above key point and only for determined key point, carry out above-mentioned inspection.Determining of key point now can be carried out as described in about the first embodiment.
For each checkpoint, in method step D1, set up descriptor, in this descriptor, for checkpoint, be scheduled to or definite test zone.The foundation of descriptor is to realize previously in the identical mode described in the first embodiment.
Be different from the first embodiment, in crack detection, in method step D2, so use this forecast model, that is, by descriptor and forecast model, determine on this checkpoint, whether there is crackle.Thereby can classify by this forecast model, in this classification, corresponding key point is divided into the point that belongs to crackle or the point that does not belong to crackle.
In second embodiment of the inventive method described herein, carry out subsequently rebuilding for the crackle that is classified as the key point that belongs to crackle in method step D.
Crackle is rebuild the known sluggish method of utilization itself and is realized: in the key point that is generally Crack Center, start the original shapes that measurement point ground rebuilds crackle one by one.Reconstruction is based on following supposition, that is, crackle is star expansion and determines around crackle and the similitude of mulle based on this.If measurement point is positioned at the bar structure expanding from Crack Center, therefore the measurement point around Crack Center belongs to this crackle high probability so.Therefore advantageously, for the reconstruction of each partial points, consider as described above orientation and the filter response of bar chart filter.
Crack Center measurement point orientation and mulle orientation around compared.Now preferably can be used as similarity degree calculates cosine between differential seat angle and carrys out weighting by the amplitude of each filter response.
Preferably also check the morphological character because of sluggish image that occur, that obtain.Especially advantageously, the area of crackle, orientation and length are considered for being associated with wafer steadiness.
Fig. 4 shows the result that this crackle is rebuild.
Claims (14)
1. one kind for being provided for the method for forecast model of the crack detection on semiconductor structure, described semiconductor structure is the prime thing of the photovoltaic solar cell in photovoltaic solar cell, manufacture process, in particular for manufacturing the semi-conducting material of solar cell
The method comprises following methods step:
A. provide with reference to semiconductor structure, this has at least one crackle with reference to semiconductor structure;
B., crackle data about described at least one crackle are provided, this crackle packet containing with crackle at this with reference to the relevant geometric position data in the position on semiconductor structure;
C. by spatial discrimination formula measure multiple local measurement points of the luminescence generated by light producing in semiconductor structure and/or by the IR of this semiconductor structure of measurement of spatial resolution absorb measurement of spatial resolution this with reference to semiconductor structure, and
D. the crackle data that provide in method step B by the measurement data according to the spatial discrimination of obtaining in method step C and foundation are carried out the training of learning algorithm and are set up forecast model, and wherein, the training of this learning algorithm comprises following methods step:
D1. by following manner, set up at least one descriptor at least one partial descriptions symbol point,, for this descriptor point, set or a definite test zone, and set up this descriptor according to the measurement data in this test zone, this descriptor is that characteristic vector and/or feature distribute and/or feature histogram, and
D2. utilize this descriptor and these crackle data to train this learning algorithm.
2. the method for detection of the crackle on semiconductor structure, this semiconductor structure is photovoltaic solar cell, photovoltaic solar cell prime thing in manufacture process, in particular for manufacturing the semi-conducting material of solar cell, the method comprises following methods step:
A., semiconductor structure is provided;
B., forecast model is provided, and this forecast model is to set up by the training of learning algorithm, preferably according to claim 1, sets up;
Multiple local measurement points of the luminescence generated by light C. producing in semiconductor structure by measurement of spatial resolution and/or absorb this semiconductor structure of measurement of spatial resolution by the IR of this semiconductor structure of measurement of spatial resolution;
D. at least one toposcopy point, determine on this checkpoint, whether there is crackle, comprise following method step:
D1. by following manner, set up at least one descriptor of this checkpoint,, for this checkpoint, set or determine a test zone and set up this descriptor in conjunction with the measurement data in this test zone, this descriptor is that characteristic vector and/or feature distribute and/or feature histogram, and
Whether D2. utilize this descriptor and forecast model to determine has crackle in this checkpoint.
3. method according to claim 2, is characterized in that, when detecting crackle according to method step D, in method step E, according to the mode of the geometric data of obtaining crackle characteristic, carries out crackle reconstruction.
4. method according to claim 3, it is characterized in that, in method step E, for a partial reconstruction region around checkpoint, for the each measurement point in this reconstruction area, determine orientation and by relatively carrying out to determine the measurement point corresponding to this crackle by pattern identification means with similitude mulle.
5. according to the method one of aforementioned claim Suo Shu, it is characterized in that, according to this at least one descriptor point in the method step D1 of claim 1 and/or utilizing key point method for searching to determine according to this at least one checkpoint in the method for claim 2, best according at least one method in SIFT method, GLOH method and/or SURF method.
6. method according to claim 5, it is characterized in that, this at least one key point definite be utilize can control azimuth filter realize, preferably utilize bar chart filter, this bar chart filter especially adopts the derivative of Gaussian filter, preferably second dervative.
7. according to the method one of aforementioned claim Suo Shu, it is characterized in that, being to determine according at least one method in SIFT method, GLOH method, HOG method, LESH method and/or SURF method according to the descriptor in the method step D1 of claim 1 and/or claim 2, and/or
In order to set up this descriptor, use maximal filter response and the orientation of the preferably bar chart filter of scalable and rotating filter.
8. method according to claim 7, is characterized in that, according to the mode that comprises following methods step, sets up this descriptor:
D1.1 is for each partial points of this test zone, preferably by determining for the parameter of each measurement point of this test zone, determine amplitude and orientation, this parameter further preferably responds and orientation for the gradient of the measured value of each measurement point or for the maximal filter of the controllable filter of each measurement point;
D1.2 sets up feature histogram at least selected key point of this test zone.
9. method according to claim 8, is characterized in that, also determines the main orientation of crackle, and according to the main orientation of this crackle, revises this feature histogram and rotate constant feature histogram to obtain.
10. the method described in one of according to Claim 8 to 9, it is characterized in that, the main orientation of crackle determines between method step D1.1 and D1.2, and this feature histogram is that measurement point by scanning in this test zone according to the scanning algorithm of the flexible program of GLOH and/or SIFT and/or GLOH or SIFT is realized in method step D1.2.
11. according to the method one of aforementioned claim Suo Shu, it is characterized in that, according to the mode that comprises following methods step according to determining this test zone in the method step D1 of claim 1 and/or claim 2:
According to one of claim 5 to 6, determine at least one local key point,
The predetermined sizes of extending for how much that are combined in key point test zone are around determined the measurement point in this test zone, according to optimal way, in conjunction with its central point, are that the rectangle, ellipse of this key point or the predetermined geometry extension sizes of circular test zone are determined the measurement point in this test zone.
12. according to the method one of aforementioned claim Suo Shu, it is characterized in that, the geometric position of crystal boundary is set and/or determines by measurement of spatial resolution method as the crystal boundary data in semiconductor structure, and
So according to crystal boundary data, be weighted,
-when determining according to the key point of claim 5, the partial points being positioned on crystal boundary is not selected as key point, or to the partial points being positioned on crystal boundary, give the weight lower than all the other partial points when determining key point, and/or
-when setting up according to the descriptor in the method step D of claim 1 and/or claim 2, do not consider to be positioned at the partial points on crystal boundary, or consider to be positioned at the partial points on crystal boundary with the weight lower than all the other partial points, and/or
-when rebuilding according to the crackle of claim 3, do not consider to be positioned at the partial points on crystal boundary or consider to be positioned at the partial points on crystal boundary with the weight lower than all the other partial points.
13. according to the method one of aforementioned claim Suo Shu, it is characterized in that, this learning algorithm is neuron net or Bayes classifier or especially preferably algorithm of support vector machine of the learning algorithm based on core engine.
14. 1 kinds for detecting the device of crackle on semiconductor structure, and this semiconductor structure is the prime thing of photovoltaic solar cell or the photovoltaic solar cell in manufacture process, comprises semiconductor source material,
This device comprises measuring unit that the IR of the luminescence generated by light that produces at semiconductor structure for measurement of spatial resolution and/or this semiconductor structure of measurement of spatial resolution absorbs and for the analytic unit in conjunction with detected crackle by the determined measurement of spatial resolution data of this measuring unit, it is characterized in that, this analytic unit is configured to carry out crack detection according to one of claim 2 to 13.
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE102011105182A DE102011105182A1 (en) | 2011-06-17 | 2011-06-17 | A method of providing a predictive model for crack detection and a method of crack detection on a semiconductor structure |
| DE102011105182.5 | 2011-06-17 | ||
| PCT/EP2012/061496 WO2012172073A1 (en) | 2011-06-17 | 2012-06-15 | Method for providing a prediction model for crack detection and method for crack detection on a semiconductor structure |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103733322A true CN103733322A (en) | 2014-04-16 |
| CN103733322B CN103733322B (en) | 2017-11-24 |
Family
ID=46384360
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201280038885.3A Expired - Fee Related CN103733322B (en) | 2011-06-17 | 2012-06-15 | The method of crack detection forecast model and the method for the crackle on detection semiconductor structure are provided |
Country Status (4)
| Country | Link |
|---|---|
| CN (1) | CN103733322B (en) |
| DE (2) | DE102011105182A1 (en) |
| HU (1) | HUP1400302A1 (en) |
| WO (1) | WO2012172073A1 (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105869164A (en) * | 2016-03-28 | 2016-08-17 | 国网浙江省电力公司宁波供电公司 | Method and system for detecting on/off state of switch |
| CN106814086A (en) * | 2016-12-07 | 2017-06-09 | 青岛海尔股份有限公司 | liner cracking analysis method |
| CN107389697A (en) * | 2017-07-10 | 2017-11-24 | 北京交通大学 | A kind of crack detection method based on half interactive mode |
| CN108986086A (en) * | 2018-07-05 | 2018-12-11 | 福州大学 | The detection of typographical display panel inkjet printing picture element flaw and classification method and its device |
| CN110325923A (en) * | 2017-02-24 | 2019-10-11 | Asml荷兰有限公司 | It measures the method for variation, check system, computer program and computer system |
| CN110431395A (en) * | 2017-03-13 | 2019-11-08 | 通用电气公司 | Fatigue crack growth prediction |
| CN110495899A (en) * | 2018-05-16 | 2019-11-26 | 西门子医疗有限公司 | The method for determining the method and apparatus of geometry calibration and determining associated data |
| CN112956035A (en) * | 2018-09-21 | 2021-06-11 | 弗劳恩霍夫应用研究促进协会 | Method for processing images of semiconductor structures and method for process characterization and process optimization by means of semantic data compression |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102015119360A1 (en) | 2015-11-10 | 2017-05-11 | Albert-Ludwigs-Universität Freiburg | Method and device for testing the contacting quality of an electrical contact between a solar cell and a contacting unit |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH07201946A (en) * | 1993-12-28 | 1995-08-04 | Hitachi Ltd | Manufacture of semiconductor device and apparatus for manufacture the same, testing of the same and testing apparatus |
| DE19914115A1 (en) * | 1998-04-20 | 1999-11-04 | Gfai Ges Zur Foerderung Angewa | Error analysis of polycrystalline wafer, solar cell, and solar module |
| US6539106B1 (en) * | 1999-01-08 | 2003-03-25 | Applied Materials, Inc. | Feature-based defect detection |
| US7096207B2 (en) * | 2002-03-22 | 2006-08-22 | Donglok Kim | Accelerated learning in machine vision using artificially implanted defects |
-
2011
- 2011-06-17 DE DE102011105182A patent/DE102011105182A1/en not_active Withdrawn
-
2012
- 2012-06-15 CN CN201280038885.3A patent/CN103733322B/en not_active Expired - Fee Related
- 2012-06-15 DE DE112012002509.1T patent/DE112012002509A5/en active Pending
- 2012-06-15 WO PCT/EP2012/061496 patent/WO2012172073A1/en active Application Filing
- 2012-06-15 HU HU1400302A patent/HUP1400302A1/en unknown
Non-Patent Citations (2)
| Title |
|---|
| DEMANT M,ET AL: "Analysis of luminescence images applying pattern recognition techniques", 《25TH EUROPEAN PHOTOVOLTAIC SOLAR ENERGY CONFERENCE AND EXHIBITION AND 5TH WORLD CONFERENCE ON PHOTOVOLTAIC ENERGY CONVERSION》 * |
| LI BIN,ET AL: "Automatic inspection of surface crack in solar cell images", 《CHINESE CONTROL AND DECISION CONFERENCE》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105869164A (en) * | 2016-03-28 | 2016-08-17 | 国网浙江省电力公司宁波供电公司 | Method and system for detecting on/off state of switch |
| CN106814086A (en) * | 2016-12-07 | 2017-06-09 | 青岛海尔股份有限公司 | liner cracking analysis method |
| CN106814086B (en) * | 2016-12-07 | 2019-12-27 | 青岛海尔股份有限公司 | Inner container cracking analysis method |
| CN110325923A (en) * | 2017-02-24 | 2019-10-11 | Asml荷兰有限公司 | It measures the method for variation, check system, computer program and computer system |
| CN110325923B (en) * | 2017-02-24 | 2021-08-27 | Asml荷兰有限公司 | Method of measuring a change, inspection system, computer program and computer system |
| CN110431395A (en) * | 2017-03-13 | 2019-11-08 | 通用电气公司 | Fatigue crack growth prediction |
| CN107389697A (en) * | 2017-07-10 | 2017-11-24 | 北京交通大学 | A kind of crack detection method based on half interactive mode |
| CN110495899A (en) * | 2018-05-16 | 2019-11-26 | 西门子医疗有限公司 | The method for determining the method and apparatus of geometry calibration and determining associated data |
| CN110495899B (en) * | 2018-05-16 | 2023-08-22 | 西门子医疗有限公司 | Method and device for determining geometric calibration and method for determining associated data |
| CN108986086A (en) * | 2018-07-05 | 2018-12-11 | 福州大学 | The detection of typographical display panel inkjet printing picture element flaw and classification method and its device |
| CN112956035A (en) * | 2018-09-21 | 2021-06-11 | 弗劳恩霍夫应用研究促进协会 | Method for processing images of semiconductor structures and method for process characterization and process optimization by means of semantic data compression |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103733322B (en) | 2017-11-24 |
| DE102011105182A1 (en) | 2012-12-20 |
| DE112012002509A5 (en) | 2014-12-04 |
| HUP1400302A1 (en) | 2014-10-28 |
| WO2012172073A1 (en) | 2012-12-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103733322A (en) | Method for providing a prediction model for crack detection and method for crack detection on a semiconductor structure | |
| US20170076174A1 (en) | Image-based feature detection using edge vectors | |
| CN103824093B (en) | It is a kind of based on KFDA and SVM SAR image target's feature-extraction and recognition methods | |
| CN107705328A (en) | Balance probe location for 3D alignment algorithms selects | |
| CN109977886A (en) | Shelf vacancy rate calculation method and device, electronic equipment, storage medium | |
| CN107122737A (en) | A kind of road signs automatic detection recognition methods | |
| CN104915673B (en) | A kind of objective classification method and system of view-based access control model bag of words | |
| CN110210448B (en) | Intelligent face skin aging degree identification and evaluation method | |
| CN102621150A (en) | Airplane skin damage identification method based on gray level co-occurrence matrix and support vector machine | |
| Pan et al. | Object-based and supervised detection of potholes and cracks from the pavement images acquired by UAV | |
| CN103745197B (en) | A kind of detection method of license plate and device | |
| CN107358214A (en) | Polarization SAR terrain classification method based on convolutional neural networks | |
| TW201220039A (en) | Inspection recipe generation and inspection based on an inspection recipe | |
| CN104463240B (en) | A kind of instrument localization method and device | |
| CN109711466A (en) | A kind of CNN hyperspectral image classification method retaining filtering based on edge | |
| Cicconet et al. | Mirror symmetry histograms for capturing geometric properties in images | |
| CN107527035A (en) | Earthquake damage to building information extracting method and device | |
| He et al. | Robust illumination invariant texture classification using gradient local binary patterns | |
| CN106326902B (en) | Image search method based on conspicuousness structure histogram | |
| Gui et al. | Object-based crack detection and attribute extraction from laser-scanning 3D profile data | |
| Lin et al. | Identification of broken rice kernels using image analysis techniques combined with velocity representation method | |
| CN102201038B (en) | Method for detecting P53 protein expression in brain tumor | |
| CN102722718B (en) | Method for classifying cells | |
| CN108446652A (en) | Polarimetric SAR image terrain classification method based on dynamic texture feature | |
| Dawood et al. | Texture image classification with improved weber local descriptor |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20171124 |