CN1823350A - Method of learning a knowledge-based database used in automatic defect classification - Google Patents
Method of learning a knowledge-based database used in automatic defect classification Download PDFInfo
- Publication number
- CN1823350A CN1823350A CN 200480019987 CN200480019987A CN1823350A CN 1823350 A CN1823350 A CN 1823350A CN 200480019987 CN200480019987 CN 200480019987 CN 200480019987 A CN200480019987 A CN 200480019987A CN 1823350 A CN1823350 A CN 1823350A
- Authority
- CN
- China
- Prior art keywords
- defective
- data
- learning method
- button
- adc
- 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.)
- Pending
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
- G01N21/84—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
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8854—Grading and classifying of flaws
Landscapes
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Image Analysis (AREA)
- Testing Or Measuring Of Semiconductors Or The Like (AREA)
Abstract
The invention relates to a method of learning a knowledge-based database used in automatic defect classification. According to said method, the user is spared a series of entries as the system carries out an automatic learn mode, which requires a reduced number of user entries.
Description
Technical field
The present invention relates to a kind of learning method that is used for automatic defect classification based on the data bank of knowledge.
Background technology
When semiconductor was made, wafer or light shield were being processed in a plurality of steps during the processing procedure in regular turn.Along with long-pending volume density increases gradually, then the demand on the quality of formed structure on the wafer is also improved.For the defective that can detect the quality of formed structure and find to exist, then to controlling the quality of the used member of wafer and each step, the demand of accuracy and recyclability also improves.This expression: when producing wafer with a plurality of steps, a kind of to defective reliable-and early identification be particular importance.Therefore, must suitably classify, a kind ofly process fast and detect so that wafer reached to the defective that is produced.
Automatic defect classification formerly " Automatic Defect Classification " (ADC) in, must on the wafer-or light shield on defective carry out the manual sort.Study based on knowledge is therefore very time-consuming.
Summary of the invention
The purpose of this invention is to provide a kind of method, by this can be simply and produce one " ADC-operation " required whole data and data shelves (knowledge-benchmark is aimed at focusing automatically) (ADC-Run) time apace.
Above-mentioned purpose is reached by the method with the 1st feature of claim.
The method advantageous particularly, this is because of can be simply by Leica ADC HP and produce one " ADC-operation " required whole data and data shelves (knowledge-benchmark is aimed at focusing automatically) (ADC-Run) time apace.In this, need to use a part of default data and data shelves.Owing to no longer carry out the manual sort as needing in the previous ADC version to the defective on the wafer, the required time can drop to 50% when then learning a kind of knowledge-benchmark in order to set a kind of new ADC prescription.In addition, can make the quality of knowledge-benchmark obtain improvement by " organizing function in advance " (the Pregrouping Function) that is obtained in the multiple situation, this accuracy to the ADC-operation has direct influence again.ADC HP is a kind of module of independently learning in Leica ADC.The user must give required data in each other step, and confirms that this data and situation change this data when needing.Each other step is shown as the independently page in Leica ADCHP-Dialog (dialogue).The user is to carry out in so-called smart (Wizard)-form to quoting of the page out of the ordinary, that is, via<Back〉and<Next〉button carries out.
The advantage that new mode of learning is had for present mode of learning is: new mode of learning uncomplicated and when carrying out according to correct order with the user the required number of steps that has descended relevant.With regard to mode of learning so far, we need the various defectives of classification in advance.Required whole of new mode of learning are one or more wafers with non-classified defective as much as possible.
Since in some steps must with a kind of video signal control desk (Visual Console) surface interaction effect, then Leica ADC HP-Dialog shows in the pattern mode but demonstrates top (Top-Most).Dialogue (Dialog) can sightless mode and automatically switch or the user can come this dialogue is switched in sightless mode or in visible mode.
The above-mentioned automatic defect classification that is used for comprises following each step based on the learning method of the data bank of knowledge:
-select a kind of covering to look into (review)-data shelves,
-by the user with parameter and data input to the page (50) of mode of learning, these parameters and the data person of being to use be well known,
-initial a kind of aligning-step and a kind of step that light intensity is adjusted,
-by the driving of several defectives automatically the suitable strength of illumination being adjusted and situation is adjusted to best illumination when needing,
-come this detecting is detected according to several examples, wherein make this detection parameters optimization according to image,
-the whole defectives to (respectively) wafer drive automatically, and wherein must detect each other defective and a kind of descriptor (descriptor) and be dispensed to each other defective, and
-descriptor of each defective is analyzed and divided into groups automatically.
The input of parameter and data comprises choosing of already present each element on the semiconductor substrate, and wherein these elements can be memory circuits, logical circuit, non-resistance or the negative crystal circle of resistance is arranged.Parameter of each layer or data comprise a kind of polymeric layer on the wafer, oxide skin(coating), the explanation of contact region or metal level.
The user chooses the lighting system and the type of focusing of at least a object lens that used.Can choose bright field, ultraviolet ray or DUV are as lighting system.The adjustment of (default) decided at the higher level but not officially announced is to use bright field and object lens this bright field to be had 100 times magnification.
Carry out a kind of artificial two-point and aim at, wherein aim at by the operation of desktop with manual type for first.Between first learning period, must store this required automatic data of aligning-shelves automatically.Each alignment point is learnt with three kinds of object lens different magnifications.The adjustment of suitable strength of illumination is that the defective that (randomly) chooses some by is randomly reached.Subsequently selected defective is driven and absorbs the image of each defective, wherein the adjustment of the initial value of the brightness of this illumination and this illumination is to reach according to a kind of assessment of histogram.With regard to the adjustment of suitable strength of illumination, only need to consider these be not more than video image-width and-25% defective of height gets final product.
Consider that 20 defectives adjust the intensity of this illumination.Come pickup image and temporarily storing by the defective that drives automatically on the wafer, until the image of whole defectives absorbed finish till.After whole images had absorbed, these images were presented on the display with the form of minimum object (thumbnail).When required thresholding (threshold) was worth when a plurality of minimum objects surpass focusing, some minimum objects promptly must be given up.The descriptor of each defective makes the minimum object that has been ingested of each defective be divided into a plurality of groups when analyzing and divide into groups automatically.Demonstrate the first new example of the group of having chosen of each defective on the display with the form of minimum object figure.
To describe according to being presented at each embodiment in graphic below the present invention.Reference symbol identical during each is graphic is represented components identical.
Description of drawings
Fig. 1 is the constructional synoptic diagram that a wafer pick-up unit of method of the present invention is installed.
Fig. 2 is ADC HP tool bar (Toolbar)-button, and user's button by this calls out the function that a kind of defective is recognized usefulness automatically.
Fig. 3 is the ADC HP calling list in " ADC "-list (menu).
Fig. 4 is " Leica ADC HP Control Desk " form, and it clearly is incorporated into a part of ADC theme that is present in the last ADC version in one form.
Fig. 5 is a kind of page of mode of learning, and the user calls out this page and therefore opens input data shelves, that is, set a kind of covering and look into-the data shelves.
Fig. 6 is a kind of page of mode of learning, and the user calls out this page and therefore gives a kind of title to a kind of prescription shelves " Recipe File ".
Fig. 7 is a kind of page of mode of learning, and the user can set ADC-knowledge benchmark-data by this.
Fig. 8 is a kind of page of mode of learning, and the user can carry out this study by this and carry out a kind of automatic aligning.
Fig. 9 is a kind of page of mode of learning, and the user can carry out a kind of automatic light adjustment by this.
Figure 10 is a kind of page of mode of learning, and the user can make the setting of detection parameters reach optimization by this.
Figure 11 is the diagram of the minimum object on the screen.
Figure 12 is the diagram of information boxes.
Figure 13 is the diagram of a kind of change susceptibility-dialogue (Change Sensitivity Dialog).
Figure 14 is a kind of form, and it sends a kind of caution to the user.
Figure 15 is a kind of diagram of indicating form when adopting new detecting thresholding (threshold).
Figure 16 is a kind of page of mode of learning, and user's page by this can make a kind of knowledge-benchmark produce automatically.
Figure 17 is the diagram of " the defective sign indicating number is divided (Defect Code Mapping) "-dialogue.
Figure 18 is the diagram of a kind of indication-dialogue.
Figure 19 is the diagram of the dialogue of initial a kind of ADC-operation " ADC-Runs " usefulness.
Figure 20 is the diagram of a kind of finishing (Finish)-dialogue.
Figure 21 is the diagram of a kind of report (Report)-dialogue.
Figure 22 is the diagram of a kind of Easy ADC Reports that has printed off.
1: wafer pick-up unit 2: pedestal
4: scanning table 5: eyepiece
7: microcobjective 8: wafer
9: carrier element 11: display
13: camera 15: electronic unit
16: wafer props up part 25: form
26: study prescription 27: editor's prescription
28: enlarge prescription 29: the operation prescription
Embodiment
Fig. 1 shows the constructional synoptic diagram of wafer pick-up unit 1, in this wafer pick-up unit 1 method of the present invention is installed.A kind of table 4 that scans is installed with integrating with carrying table on the pedestal 2 as wafer 8 usefulness.This scanning table 4 can move in X-coordinate direction He in the Y-coordinate direction.Placing on this scanning table 4 or hook clips the wafer 8 that is about to detect.A kind of observation device (it is preferably and is provided with a kind of microcobjective 7) is connected with pedestal 2 via a carrier element 9.This microcobjective 7 can make this wafer 8 watch under the amplification situation.A plurality of microcobjectives 7 can be located on the spinner (not shown), so that can watch dynamically under different magnifications.Amplifying of wafer 8 and the structure observed can directly be carried out static observation or observe via a display 11 via an eyepiece 5, and this display 11 is to be connected with CCD-camera 13.Be provided with a kind of electronic unit 15 in addition, can reach the function of system automation by this.This electronic unit 15 particularly can be used to control this scanning table 14, and this camera 13 is read and can be used to control this display 11.Usually must be provided with a kind of wafer and prop up part 16, make it can hold this wafer to be detected 8, and this wafer props up part and is fixed detecting in the period.This scanning table 14 forms in the mode that can move in mutually perpendicular X-coordinate direction out of the ordinary and in the Y-coordinate direction.Therefore, each position to be observed can be directed to below the optical axis 7a of microcobjective 7 (Fig. 1) on the wafer 8.
Fig. 2 shows ADC HP tool bar (Toolbar)-button 20, and user's button by this calls out the function that a kind of defective is recognized usefulness automatically.ADC HP-dialogue call out via ADC HP tool bar (Toolbar)-button 20 or via in " ADC "-list or the mother-tool bar 19 of video signal control desk-application (Viscon-Application) 21 in context (the context)-list of ADC-dialogue call out (consulting Fig. 3).Each user (by user-grade " Operator " beginning) can get involved this list-login.Because ADC HP is a kind of option that separates (Option), list-login and can only carry out when ADC HP has also installed then with visual way.This kind option (Option) mode can be protected, and is similar at present via registering-login middle person, and this registers-login by formula is installed and results from the option of having chosen (option).
If in the loaded video signal control desk of a kind of formula (Viscon), then list-login is removed to drive (deactivated) and is demonstrated.
Fig. 4 shows a kind of what is called " Leica ADC HP Control Desk " form, its will be a part of ADC theme that is present in the last ADC version clearly be incorporated in the form and as a kind of output reference with each other module 26,27,28 and 29 of starting.Specifically this form is:
■ " study (Leran) prescription ": learn and set a kind of new ADC-prescription and a kind of knowledge-benchmark, carry out ADC-operation (Run Recipe) subsequently,
■ " editor (Edit) prescription ": be used for handling an already present knowledge-benchmark,
■ " enlarges (Expand) prescription ": be used for enlarging an already present knowledge-benchmark,
■ " operation (Run) prescription ": be used for starting a kind of ADC-operation.
Each other module is provided with a kind of button respectively.It is a kind of in this form of implementation " study prescription "-button 26, " editor's prescription "-button 27, " enlarge prescription "-button 28 and " the operation prescription "-button 29.Promptly carried out each other theme at 26,27,28,29 o'clock at each other button of operation.The theme that has been present in the last ADC-version only is written into herein momently.
■ " editor (Edit) prescription ": after this button 27 was depressed, the user must choose already present knowledge-benchmark-data shelves, and it is by the application of outside " KB Wizard " start, and must show the content of these data shelves.The data shelves can be handled and on the whole knowledge-benchmark be tested at this place.
■ " enlarges (Expand) prescription ": the user chooses already present knowledge-benchmark-data shelves with button 28 and chooses a kind of covering and look into (Review)-data shelves.In service at ADC subsequently, in background, collect new data and temporarily storing.During this end of run, the data that this is temporary and employed knowledge-benchmark (KB) data shelves are by application " KB Wizard " be written into and demonstrate.The user can suitably adopt this new data now in knowledge-benchmark.
■ " operation (Run) prescription ": cover and look into-data shelves and ADC-prescription and the ADC-operation can be started to choose one by choosing this button 29.Automatically to detecting by the selected whole defectives of user and each defective being classified by the knowledge-benchmark data shelves that found in the ADC-prescription.Resulting result is described as again to cover when finishing and looks into-the data shelves.
The theme relevant with the operation of " study prescription "-button 26 will be specified in down.
ADC HP-mode of learning be shown as a kind of non--modal dialog.The data that input that the user must be in 8 steps in regular turn (that is, 8 page on) is required or choose the data shelves.The last page is the result of ADC HP-study.
In this, the user can be by<Back〉and<Next 〉-button 30,31 (smart style) arrives previous step or arrives next procedure (consulting Fig. 5) under the state of reality allows.
Usually, the demonstration of the page out of the ordinary is irrelevant with user's grade (level).Other user's interface-element (it can be seen when development-user's grade) then is an exception (, relevant with user's grade).These elements only can be seen when development-phase place and be removed when discharging (Release)-version or all can not be seen whole user's grades usually.
Fig. 5 shows a kind of page of this mode of learning, and the user calls out this page and therefore opens input data shelves 34, that is, set a kind of covering and look into-the data shelves.This page 33 is called " Open Input File ".Show these data shelves (no path) on the page 33.Utilize a kind of shelves (FileOpen)-button 35 of opening, the user during then for this data shelves input can show various catalogues.When a kind of input shelves had been determined, these input shelves were temporarily opened, but video signal control desk-sequence (sequencer) is inoperative.Use these shelves " EasyADCLearn.vsl " to be encoded into former shelves (script file) with hard (hard).By the shelves of having opened read batch feature (Lotld) of first wafer-, wafer feature (Waferld)-, steps characteristic (Stepld)-and set the required data of feature (Setupld).These shelves are closed again then.Possible standard-adjustment (for example, initial automatically) process therewith has nothing to do or sets original state again for.By a kind of cancellation-button 39, then the user can make this process interrupt.
Fig. 6 is the another kind of page 38 of mode of learning, and the user calls out this page and therefore gives a kind of title to a kind of prescription shelves " Recipe File ".This page 33 is called " Recipe File ".Returning operating in the page 38 of (Back)-button 30 does not allow.When choosing a kind of effective input shelves 37, the operation of next (Next)-button 31 allows.Utilize cancellation (Cancel)-button 39, then the user can make ADC HP mode of learning interrupt.Leica ADC HP prescription shelves are presented in editor's box 40.The previous title composition of having read is combined and show formed shelves names (that last partly is " .vsl ") according to master.Each title composition separates by " _ " symbol (bottom line).
Unsuitable letter is removed and hyphen replaces with bottom line in the formed data shelves name.The user also can change each default title (whole or a part of) according to its default situation.Use the as a result of write out a prescription original paper of shelves (Result Recipe File, the flow process control-data shelves when ADC-moves) usefulness of this shelves " EasyADCRun.vsl " (having finished hard coded) in " Kopie ".The page 38 comprises a plurality of inspection boxes 41,42,43 and 44.Each checks box 41,42, and 43 and 44 usefulness decide the composition of each title.Therefore, use Lotld, Stepld and Setupld are as data shelves decided at the higher level but not officially announced.Formed shelves name (not being as ending with " vsl ") also can be used as the master of other data shelves (aligning-, focusing-shelves or the like) automatically usefulness.Result-data shelves (Result-Dtaen-File) usually with the identical title of input shelves, identical form and describe with standard-result-catalogue.Elected when getting at least one title composition, then allow to use this to return (Back)-button 30 and this next (Next)-button 31.Also allow to use this cancellation (Cancel)-button 39.
Fig. 7 shows the page 50 of this mode of learning, the user who points out ADC knowledge benchmark-data on it.This page 50 is labeled as " ADC Basic Data ".Can be the user that chooses in the hurdle 51 who is called " Structure Type " at memory body " Memory " and logical circuit " Logic " between select.The another kind of selection " blank not structurized wafer " (Bare Wafer) also is possible.In order to determine ADC operation-pattern (repeat-or at random-pattern) or alignment pattern (general automatic aligning or blank wafer are aimed at) automatically, then can carry out corresponding to this kind selection.Decided at the higher level but not officially announced-setting is to set on logical circuit " Logic ".
Be called choosing in the hurdle 52 of " Layer Type " at another, the user can select " whether on wafer, apply a kind of layer or a plurality of layers (" Layer ").Equally, interesting is: which layer will be applied on the wafer.No photoresistance represents that with " w/o Resist " photoresistance is represented (consulting Fig. 7) with " with Resist ".Photoresistance or other layer is applied on the wafer or on the semiconductor substrate." preliminary election or setting decided at the higher level but not officially announced " " time be " w/o Resist ".In the possible setting means of the next one, the user can choose a kind of layer of form.Polymeric layer represents that with " Poly " oxide skin(coating) represents that with " Oxid " contact region represents that with " Contact " metal level is represented with " Metal ".Equally, can choose the flow process that applies of various layer.Therefore, for example, oxide (Oxid) layer can promptly apply before polymkeric substance (Poly) layer and form, and this is to represent with " Before Poly ".The user can be at single metal level (Metal1) when being selected for layer form " metal ", selects between the metal level (n-Metal) that dual metal level (Metal2) and n-weigh." whether existing a kind of main stor(e)y or a kind of subordinate layer " is with deciding this random pattern and focus version.The setting item decided at the higher level but not officially announced of layer is " Poly ", be " Metal1 " when when " Oxid ", being BeforePoly " and at Metal ".Oxide-and metal-lower floor-radio box only choose Shi Caihui in advance and be driven at " Oxid " or " Metal ", otherwise it promptly is shown as and is not subjected to drive (inactive).Choose in the hurdle 53 at next, the user can choose light illumination mode " illuminationmode ".The bright field that is called BF, the radio box that is called the ultraviolet ray of UV and is called dark (Deep) UV or the like place of DUV all can be used by the user.The user is given in operational object lens demonstration in tabular (List) box 54, wherein only shows the object lens of the ADC-pattern that these are suitable for having chosen.With regard to setting Xiang Eryan decided at the higher level but not officially announced, choose bright field " BF " and advise a kind of 100 times-or object lens of littler magnification that have.
Following table (Table I) demonstrates a focus-setting that forms according to selected data.
Layer/ADC-pattern | Focusing-type-Shi | Off-set value when TV-focuses on |
Polymkeric substance | TV-focuses on | 400 |
Polymkeric substance-photoresistance | TV-focuses on | 0 |
Oxide before the polymkeric substance | Laser | - |
Oxide before polymkeric substance-photoresistance | TV-focuses on | 0 |
Oxide behind the polymkeric substance | TV-focuses on | 2000 |
Oxide behind polymkeric substance-photoresistance | TV-focuses on | 2000 |
The contact region | Laser | - |
Contact region-photoresistance | Laser | - |
| TV-focuses on | 1500 |
Metal 1-photoresistance | TV-focuses on | 0 |
| TV-focuses on | 1800 |
Metal 2-photoresistance | TV-focuses on | 0 |
The n-metal | TV-focuses on | 2500 |
N-metal-photoresistance | TV-focuses on | 0 |
When TV-focuses on, use the default value of " TV-Focus (focusing) Flexible 2 ".Return-knob 30, next-button 31 and cancellation-button 39 allow in this form.When depressing<Next〉during button 31, can not see ADC HP-Dialog.
Produce " EasyADCLeran "-the duplicating shelves and adjust some specific actions (aiming at automatically) and data (seizing setting) of shelves.
The shelves that duplicate of being named title to " EasyADCRun.vsl "-shelves (ADC Run Recipe after a while) carry out identical variation.Input shelves are written into adjusted former shelves and start the VisconNT-sequence.These shelves are automatically handled till choosing wafer.Use-and show that this standard-wafer chooses-talk with (Dialog).The establishing criteria mode is chosen whole existing wafer (the setting items decided at the higher level but not officially announced in the former shelves of Easy ADC (ScriptFile)).
Fig. 8 is a kind of page 60 of mode of learning, the user can carry out this study by this and carry out a kind of automatically-or automanual at least aligning.The page 80 is called " Alignment Procedure ".Confirming to carry out wafer-after choosing, be written into first wafer and these shelves are handled till automatic aligning by the user.According to the setting of already present each layer on the wafer, with this corresponding self-aligning mode of learning of starting (semi-automatic aligning or blank wafer are after a while aimed at).The user can carry out a kind of artificial two-point and aim at, and wherein the user is only manually to aim at initial point (operation of desktop is to carry out in Live Video image by rocking bar or by mouse-two buttons (click)) and must be to confirm.Between first learning period, must store the required automatic data of aligning-shelves automatically.Each alignment point is learnt with three kinds of different objective lens magnifications, and wherein the object lens of magnification maximum are preset in by choosing on the page 50 (ADC basic document).
Automatically learn and aim at for second according to first the data that has stored.The ADC-object lens of having chosen are preset by software usually.Must use these object lens, this be because of its for after a while the light adjustment (method of having used: alignment point) need.
If on second alignment point, find first study structure having set, then second palpus around a crystal grain (Die) and " moving " must search this structure to the central point of wafer and at this place.Before this aimed at by a kind of error interrupt, second can center on 6 crystal grain and " moving " at most.In such cases, a kind of indication-form shows and gives the user that this indication-form is pointed out: this aligning has interrupted and wafer must unload.
After this aims at end, video signal control desk-sequence time-out (established time-out action in the former shelves of Easy ADC (ScriptFile) (do not have " information boxes-demonstration ")), ADC HP-talks with can see and demonstrate next page again.When this is aligned in when carrying out or is interrupted by a kind of error, then do not allow next-button 31 presses.When this is aimed at success, then allow next-button 31 presses.Cancellation-button 39 be allow and whole ADC HP mode of learning is interrupted.
Fig. 9 is a kind of page 70 of mode of learning, and the user can carry out a kind of automatic light adjustment by this.This page is called " optical registration (Light Alignment) ".After depressing a kind of " Perform AutomaticLightadjustment "-button 71, picked at random one certain number point of destination (defective is by the data shelves).If there is a kind of numerical value information, then only choose be not more than video image-width and-these defectives of 25% of height.These defectives are driven and absorb some images." lamp brightness "-initial value decides according to the histogram assessed value and adjusts on microscope.This expression: brightness must be adjusted downwards, makes defect image that " crossing control (over controlled) " can not take place.Existing whole color channel all can be used to carry out this discussion and adjusts in corresponding mode.
Then, carry out a kind of automatic light adjustment.Acquired data storage is in knowledge benchmark-data shelves after finishing.Standard mode is to use 20 points (defective) to decide " initial value ", and use this light to adjust " aligning "-method.
Show a kind of state box 72 " ProgressControlBox " and a kind of information box 73 " Read-Only EditBox " on the page 70.In light adjustment automatically " LightAdjustment " during its progress be presented in the state box 72.State this paper (text) is presented in the information box 73 when finishing or when failure.When carry out light when adjusting this return-button 30 do not allow to press.When the light adjustment has stopped, allow this return-button 30 presses.Wafer is removed and shows this page 50 " ADC Basic Data ".When light adjustment success, allow next button 31 to press.When carrying out this light adjustment, allow this cancellation-button to press, and therefore make opened whole data shelves close and remove.
Figure 10 is a kind of page 80 of mode of learning, and the user can make the setting of detection parameters reach optimization by this.This page 80 is called " Optimize ADC Detection ".By the button 81 of a kind of being called " StartOptimization ", to start this process.This best-of-breed functionality should be guaranteed: standard value that focusing is used and detection parameters act on the selected wafer.If situation is not so, then the user still can make default standard value change again.After depressing this button 81, make video signal control desk-sequence starting, choose-and start each defective and the picked-up of carrying out image.This paper on the button 81 changes over " Stop Optimization ".Progress during image capture is presented in the state box 82.The user can be interrupted this process by depressing this button again.If absorbed required all images, then these images are being presented on the screen with the form of minimum object when showing in the dialogue (Dialog) next time next time.Use 10 defectives (having finished hard coded) to carry out optimization.By registering-be written into or develop-user's grade, this number is changed.When detecting-optimization is carried out, do not allow to depress this and return-button 30.When detecting-optimization is carried out, do not allow to depress this next-button 31.When detecting-optimization is carried out, do not allow to depress this cancellation-button 39.
By depressing<Start Optimization 〉-button 81, then video signal control desk-sequence starts again, and button this paper changes over " Stop Optimization " and chooses the defective of a preset number of actual wafer.
Start each defective and therefore trigger a kind of special ADC-action (action).Must pickup image in this action, one developed detect each defective of side on the customary path of ADC-(routine) of finishing and temporarily storing each image when the picked-up action of all images has been finished till.The progress of this process shows by the state box.During image capture, the user can be interrupted this process by press push button again.
Figure 11 shows a plurality of minimum objects 91 that are presented on the display 11
1, 91
2, 91
3..., 91
nIf absorbed whole images, then ADC HP-Dialog switches in sightless mode and image is presented in minimum object-dialogue (Thumbnail-Dialog) 90 (entire image is presented on the screen).This video signal control desk-sequence can suspend when this time point.
Minimum object-dialogue 90 is divided into 92, the three districts 93,91, the second districts, first district and the 4th district 94 basically.First district 91 comprises the tabular of level, wherein minimum object 91
1, 91
2, 91
3..., 91
nShow with detecting mark and defect characteristic (Defect-1d).This moment, selected image was presented in second district 92 with the resolution of 640 * 480 picture point at most.If this image exists, then each reference picture can dwindle and be presented in the 3rd district 93.By the mousebutton of defect image below, vernier touch-control and/or browse-button 95 then can make actual image select to change.
Flaw labeling can be via<Hide Defect Detection 〉-button 96 eliminates and can import again again.Browse-button 95 is used for choosing and shows the next one-or previous defect image.<Hide DefectDetection 〉-button 96 forms with a kind of bifurcation thixotroping button (toggle button), and therefore make the detecting mark can with as seen-or sightless mode switch.
Focus Difference-Defect/Reference button 97 can show a kind of information boxes 86 (consulting Figure 12).The defect image of having chosen when confirming (with already present reference picture) must be given up, that is, and and by eliminating in the display.If it is (decided at the higher level but not officially announced: as 30%), then must to make focus value change (that is, switch to by laser that TV-focuses on or make the change of TV-focusings-off-set value with the step value of 500 nanometers) that the imperfect image " Bad Focus-Bilder " given up surpasses the interior door limit value.Various data are reset and received to each defective then.In this, close above-mentioned minimum object-dialogue (Thumbnail Dialog) 90 and in scanning process ADC HP-Dialog show again.
A kind of Wrong Defect Detect ion-button 86 can make the detecting threshold value of the image of having chosen be determined again.In this, show a kind of new dialogue 80 (consulting Figure 10).
A kind of renewal (Refresh)-button can make the mean value of the threshold value of all images in the tabular be obtained, and whole detectings is all recomputated with this mean value.Reconstitute this tabular then.The image of " Autothreshold " (=1) does not consider to be used for trying to achieve this mean value.
A kind of (default)-button 88 decided at the higher level but not officially announced can make whole variation of the detection parameters of all images be eliminated.This tabular reconstitutes with original value.
Finish this dialogue with a kind of Apply-button 89, calculate the mean value of this threshold value and (global) detection parameters that this mean value is used as wide area.The image of " Autothreshold " (=1) does not consider to be used for trying to achieve this mean value.Finish this best dialogue (Optimize-Dialog), ADC HP-Dialog switches in visible mode again and new total detecting threshold value is written in above-mentioned knowledge-benchmark.
This cancellation (Cancel)-button 39 cuts out this best dialogue (Optimize-Dialog) and ADCHP-Dialog is shown.Whole changes are given up..
Figure 13 is the diagram of a kind of change susceptibility-dialogue (Change Sensitivity Dialog) 100.This dialogue 100 is used for the best of detecting-susceptibility " Detection-Threshold " of the defect image that obtains to have chosen and sets.This defect image 101 is presented at affiliated detecting threshold value in the dialogue 100 in centre.If use the Auto-Sensing threshold value indirectly, then set a kind of 50% value this moment by previous setting (on the page 50 " ADC Basic Data ").
Utilize two buttons 102 that the susceptibility of detecting is descended or raising.The defect image 101 that is presented at central authorities shows the eigenwert of this detecting with present selected susceptibility.This value is presented in this detecting image 101.One image that has dwindled 103 appears at the other and variation this detecting when demonstrating susceptibility and having diminished of this defective pattern 101 again at left.Equally, an image that has dwindled 103 appears at the other and variation this detecting when demonstrating susceptibility and having improved of this defective pattern 101 again right-hand.Click an image that has dwindled or by the pressing down of image below button 102 by mousebutton, then Shi Ji susceptibility can change over this value and this image is presented at central authorities now.Then the left and right-hand variation are calculated again.
A kind of<Hide Defect Detection 〉-button 105 forms with a kind of bifurcation thixotroping button (togglebutton).Above-mentioned detecting mark can with as seen-or sightless mode switch.
The slip chi 106 of a kind of being called " susceptibility step level size (a Sensitivity Step Size) " is used for making the size of the changing value of susceptibility to change when this button 102 of operation.A kind of deletion image (Delete Image)-button 107 is used for giving up the defective of other assessment usefulness.Remove in the tabular of this defective by the best-dialogue (Optimization-Dialog).Finish this dialogue and user then and arrive previous dialogue.
Figure 14 is the diagram of form 10, and it sends a kind of caution to the user.If " error detection " (Wrong Detection) image of having given up surpass the interior door limit value (decided at the higher level but not officially announced: 30%), then (automatically) choose and drive new defective and receive various data.A kind of Apply-button 107 initial this kind application.Figure 15 is a kind of diagram of indicating form 110 when adopting new detecting thresholding (threshold) value.This indication form is pointed out the user: by adopting new detecting threshold value, then the detecting of all other image also can change.By in the best-dialogue 90<Refresh-the pressing down of button 87, then new value can be used in the image of all other.
By<YES 〉-affirmation of button 111, arrive previous dialogue 100 again with detecting threshold value and the user who adopts the center image viewing area.
The operation that changes (Change) by this cancellation-button 39 in the dialogue 100 arrives previous the best-dialogue 90 (consulting Figure 11) again to give up the change and the user that have all carried out.
Figure 16 is a kind of page of mode of learning, and user's page by this can make a kind of knowledge-benchmark produce automatically.Utilize a kind of beginning data collection (Start Collecting Data)-button 121, with required whole data of whole defectives of receiving the wafers of all having chosen and storing.
This state shows in state box 122 and information box 123 and gives the user.The whole numbers (for example, " 267 " in " 750 ") that show the defective that " to handle " in the information box 123.When data reception-process when carrying out, this returns-operation of button 30 is unallowed.When data reception-process when carrying out, this next-operation of button 31 is unallowed equally.When data reception-process when carrying out, the operation of this cancellation-button 39 is unallowed.Then, when the operation of this cancellation-button 39 had allowed, then all opened data shelves were closed and are removed.
Above-mentioned flow process is as follows: video signal control desk-sequence is initial again and choose whole defective of input shelves.In first step, the defective on the wafer is activated, pickup image produces each descriptor and each descriptor is stored on the defective in ADC-Result-Daten.The image of this defective is storing by following setting:
" Write to Archive File " (writing in the former shelves)
" All Images " (whole images)
" Image Compression " (compression of images): be
" Leica-ImageStore " (Leica-image storage): not
In second step, video signal control desk-sequence suspends (before these output shelves store) on basket shape (Basket) plane.
In third step, produce formed group of a plurality of set (" pre-combination (Pregrouping ")) by descriptor.
In the 4th step, this pre-combination attempts only to produce maximum 20 groups.These groups that are less than 2 examples are rejected.The group that is caused temporarily is copied in knowledge-benchmark like this, wherein the defective sign indicating number of each group and " Symptom " for the first time " by continuous programming code " (1,2,3 ... or EasyClass1, EasyClass2, EasyClass3 ...).
Show that in the 5th step a kind of dialogue 130 is to divide (" Defect CodeMapping ") (consulting Figure 17) to defective.This kind " Defect Code Mapping "-dialogue 130 shows with first form 131, the second forms, 132, the three forms 133 and the 4th form 134 basically.In first form 131 each group that is produced is shown a kind of classification-image (icon) in the 4th step.In second form 132, the image of the first new example of the group of having chosen is presented in minimum object-viewing area.In the 3rd form 133, show actual defective sign indicating number-Biao (table).By choosing a defective sign indicating number and depressing one<Map 〉-button 135, make this yard corresponding to the kind of having chosen.Make the image modification of this kind, this image can obtain a kind of hook symbol 136 of green and show corresponding defective sign indicating number-this paper this moment.This kind divided " Mapping " and also can be undertaken by clicking of a kind of pair of button among defective sign indicating number-Biao.By depressing<Delete Group 〉-button 137, the group that has then shown in fact promptly is labeled as " being about to eliminate ".Corresponding classification-image can be received a kind of redness " * " numbers 138.
The bifurcation thixotroping button 139 of a kind of being called " Optimize Image Display " can make flaw labeling section on every side demonstrate with the original size of image example under depressed state.If this flaw labeling is too big in the image example, then this viewing area needn't change.By depressing bifurcation thixotroping button 139 again, entire image is shown diminish again.When whole defective groups (group) is handled (that is, made image or be labeled as defective to be eliminated), then operate a kind of applying (Apply)-button 129 and allow.
In the 6th step, by confirming to attempt to make each decreased number of each other example of the group of imaging (be not and given up) as the group of mark to be eliminated.Therefore this needs, and makes the specific group with a lot of defectives can not occupy the major part of knowledge-benchmark and defective and is preferably and belongs to this type.This result is used in knowledge-benchmark, so the user arrives the ADC mode of learning.This cancellation-button 39 of operation on the display 11 of indication shown in Figure 180-dialogue 140.The operation this<YES-button 141 after, this division (" Mapping ") and whole ADC mode of learning are promptly interrupted.
Figure 19 is the diagram of the dialogue 150 of initial a kind of ADC-operation " ADC-Runs " usefulness.Utilize a kind of Start ADC Run-button 151, then after depressing this button 151, the classification of whole defectives of the wafer of having chosen promptly becomes " off-line " (offline promptly, needn't drive again).This kind classification is to reach by the ADC-knowledge benchmark of reality.Dialogue 150 comprises a kind of state box 152 and a kind of information box 153.The sum (for example, " in 750 123 ") that can show the defective that to classify by state box 153.
When off-line type (Offline) ADC moved, operating this, to return a button 30 be unallowed.When off-line type (Offline) ADC moves, operate this next-button 31 is unallowed.Equally, when off-line type (Offline) ADC moved, it was unallowed operating this cancellation-button 39.When all opened data shelves are closed and removed, then operate this cancellation-button 39 and allow.When next-when button 31 was depressed, this sequence started again.Write this result (Result)-data shelves and this sequence and finish automatically, this moment, all opened data shelves were closed.
Figure 20 is the diagram of a kind of finishing (Finish)-dialogue 160.One output shelves are provided with an information box 161.This information box 161 is used for showing the data shelves that stored, only shows the shelves name.
Easy ADC VSL shelves are presented among the Read-Only EditBox162 equally.In information box 162, show " ADC-Run " shelves that produced, only show the shelves name.Complete path is presented in the top (Tooltip).
All the number of defective (Total Defects) is presented in the information box 163.Can read the sum of whole defectives in the information box 163.
The defective of having found (Defects Detected) is presented in the information box 164.The detecting again (Redetection) of the defective that recently shows with percentage is presented in the information box 166 equally.Then, demonstrate the absolute number and the number percent of the defective of being detected in the ADC mode.LED 149 shows " whether the value of this number percent surpasses a predefined value " with color mode.If the value of this number percent surpasses a predefined value, then LED 149 sends green glow, otherwise LED 149 sends ruddiness.
The number of classified defective (Defects Classified) is presented to be read in (Read)-information box 165.The value of the number percent of classifiable defective (Classifiability) is presented in the information box 167 equally.Then, demonstrate the absolute number and the number percent of the defective of being classified in the ADC mode.LED 148 shows with color mode " whether the value of this number percent surpass a predefined value ".If the value of this number percent surpasses a predefined value, then LED 148 sends green glow, otherwise LED 149 sends ruddiness.
Show a kind of report-dialogue 170 (Figure 21) by operation a kind of report (Report)-button 147, it is relevant with the user.Only by user's grade " slip-stick artist " show a kind of report that has enlarged.The end that a kind of finishing (Finish)-button 146 operations are reached allows.This kind report-dialogue 170 is relevant with the user.
Figure 21 is the diagram of a kind of report (Report)-dialogue 170, and wherein data demonstrates with having enlarged in information box 171.The data that has shown is a grade information: (P) output shelves names (+path), prescription information: (P), ADC HP Recipe (prescription) shelves names (+path), knowledge-benchmark shelves name (+path), automatically aim at shelves names (+path), focus type " laser " or " video signal (VIDEO) " with setting of seizing (GrabSetup) shelves names (+path), knowledge base information (A) is (P), the object lens that used, the control methods of having used, type of focus, the aperture of having used, the light intensity of having used, statistics information (P), the number of wafer, the sum of defective, the number of classified defective, the number of ADC kind, the defective in each kind (representing), the number of the defective of representing with absolute number/number percent of having detected with matrix form, the number of the classified defective of representing with absolute number/number percent, the number of the classification of each ADC-defect kind (P), performance (performance) information: (A) (P), accuracy, purity, confusion matrix (A) are (P) and defective tabular (A).Classified table comprises the following data in each data set: groove (s1ot) number, incident (Event) number, manual sort, ADC-classification and the ADC-classification with trust (Confidence) value.Therefore, only send hundred of the junior three and login item.(A) expression: these data only just can be seen in the report that has enlarged.(P) be used for representing that these data can print off.
Above-mentioned report dialogue 170 is provided with a kind of printing (Print)-button 171.Show the printing advance notice of ADC HP Reports on the display 11.This kind printing can be reached by the standard printer.This kind printing is carried out with lined formula, and this is not show fully or print off because of path major part in high form.
Figure 22 is the diagram of a kind of Easy ADC Reports 180 that has printed off.Utilize a kind of Save As-button 173 (consulting Figure 21) when operating, then ADC HP Reports can store into a kind of this document (with TXT as the ending).
Claims (14)
1. learning method that is used in the automatic defect classification based on the data bank of knowledge is characterized in that it may further comprise the steps:
Select a kind of covering and look into (review)-data shelves,
By the user with parameter and data input to the page (50) of mode of learning, these parameters and the data person of being to use know,
Initial a kind of aligning-step and a kind of step that light intensity is adjusted,
By the driving of several defectives automatically the suitable strength of illumination being adjusted and situation is adjusted to best illumination when needing,
Come this detecting is detected according to several examples, wherein make this detection parameters optimization according to image,
Automatically the whole defectives to (respectively) wafer drive, and wherein must detect each other defective and a kind of descriptor (descriptor) and be dispensed to each other defective, and
The descriptor of each defective is analyzed and divided into groups automatically.
2, learning method according to claim 1, wherein the input of parameter and data comprises choosing of already present each element on the semiconductor substrate, wherein these elements can be memory circuits, logical circuit, non-resistance or the negative crystal circle of resistance is arranged.
3, learning method according to claim 2, wherein parameter of each layer or data comprise a kind of polymeric layer on the wafer, oxide skin(coating), the explanation of contact region or metal level.
4, learning method according to claim 1, wherein the user chooses the lighting system and the type of focusing of at least a object lens that used.
5, learning method according to claim 4 wherein can be chosen bright field, and ultraviolet ray or deep UV (DUV) are as lighting system.
6, learning method according to claim 4, the adjustment of (default) wherein decided at the higher level but not officially announced are to use bright field and object lens this bright field to be had 100 times magnification.
7, learning method according to claim 4, wherein carrying out a kind of artificial two-point aims at, wherein aim at by the operation of desktop with manual type for first, between first learning period, must store this required automatic data of aligning-shelves automatically, each alignment point is learnt with three kinds of object lens different magnifications.
8, learning method according to claim 4, wherein the adjustment of Zhao Ming suitable strength is that the defective that (randomly) chooses some by is randomly reached, subsequently selected defective is driven and absorbs the image of each defective, wherein the adjustment of the initial value of the brightness of this illumination and this illumination is to reach according to a kind of assessment of histogram.
9, learning method according to claim 8, wherein with regard to the adjustment of suitable strength of illumination, only need to consider these be not more than video image-width and-25% defective of height gets final product.
10, learning method according to claim 8 considers that wherein 20 defectives adjust the intensity of this illumination.
11, learning method according to claim 1 is wherein come pickup image by the defective that drives automatically on the wafer and is temporarily being stored, until the image of whole defectives absorbed finish till.
12, learning method according to claim 11, after wherein whole images had absorbed, these images were presented on the display (11) with the form of minimum object.
13, learning method according to claim 12, when wherein required thresholding (threshold) was worth when a plurality of minimum objects surpass focusing, some minimum objects promptly must be given up.
14, learning method according to claim 1, wherein the descriptor of each defective makes the minimum object that has been ingested of each defective be divided into a plurality of groups when analyzing and divide into groups automatically, demonstrates the first new example of the group of having chosen of each defective on the display with the form of minimum object figure.
Applications Claiming Priority (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
DE10331646.9 | 2003-07-12 | ||
DE10331646 | 2003-07-12 | ||
DE102004022717.9 | 2004-05-07 |
Publications (1)
Publication Number | Publication Date |
---|---|
CN1823350A true CN1823350A (en) | 2006-08-23 |
Family
ID=34041817
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN 200480019987 Pending CN1823350A (en) | 2003-07-12 | 2004-06-03 | Method of learning a knowledge-based database used in automatic defect classification |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN1823350A (en) |
DE (1) | DE102004022717B4 (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10650508B2 (en) | 2014-12-03 | 2020-05-12 | Kla-Tencor Corporation | Automatic defect classification without sampling and feature selection |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE102012019033A1 (en) | 2012-09-27 | 2014-03-27 | Chris Kroos | Method for outputting and archiving data in e.g. personal computer, involves adjusting document management system (DMS) in configuration wizard/DMS wizard, such that output of data with bar code is provided in order to archive data |
CN111047561A (en) * | 2019-11-22 | 2020-04-21 | 国网江西省电力有限公司电力科学研究院 | Method for identifying cracks of shed tortoise of composite insulator |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6246787B1 (en) * | 1996-05-31 | 2001-06-12 | Texas Instruments Incorporated | System and method for knowledgebase generation and management |
US6512842B1 (en) * | 1999-04-09 | 2003-01-28 | Advanced Micro Devices | Composition based association engine for image archival systems |
-
2004
- 2004-05-07 DE DE102004022717A patent/DE102004022717B4/en not_active Expired - Fee Related
- 2004-06-03 CN CN 200480019987 patent/CN1823350A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10650508B2 (en) | 2014-12-03 | 2020-05-12 | Kla-Tencor Corporation | Automatic defect classification without sampling and feature selection |
Also Published As
Publication number | Publication date |
---|---|
DE102004022717B4 (en) | 2006-09-21 |
DE102004022717A1 (en) | 2005-02-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP4065893B1 (en) | Defect detection device, defect detection method, information processing device, information processing method, and program thereof | |
JP4102842B1 (en) | Defect detection device, defect detection method, information processing device, information processing method, and program thereof | |
CN105486687B (en) | Touch panel inspection apparatus and method | |
US8581976B2 (en) | Method and apparatus for reviewing defects of semiconductor device | |
CN1854718A (en) | Visual inspection apparatus | |
JP4617970B2 (en) | Defect inspection apparatus and defect inspection method | |
CN1188776C (en) | Apparatus for determining image, method and medium | |
KR101955268B1 (en) | Defect observation device and defect observation method | |
JP2001168160A (en) | System for inspecting semiconductor wafer | |
CN1690678A (en) | Image analysis method, image analysis program and pixel evaluation system having the sames | |
CN1904676A (en) | Laser repairing device | |
CN1402322A (en) | Monitoring system and method for monitoring process condition change by electron beam | |
JP2008139201A (en) | Apparatus and method for detecting defect, apparatus and method for processing information, and its program | |
JP2002310962A (en) | Sorting method and observation method for image and apparatus therefor | |
CN1261819C (en) | Gray mask defect checking method | |
US7013222B2 (en) | Wafer edge inspection data gathering | |
CN1152748A (en) | Image expansion device for computer system | |
US20120249770A1 (en) | Method for automatically focusing a microscope on a predetermined object and microscope for automatic focusing | |
CN114270179A (en) | Defective portion identification device and defective portion identification method | |
CN116921260A (en) | Light-emitting chip sorting method and device | |
JP2011112431A (en) | Discriminating method and discriminating device for front and back defect of color filter | |
CN1300994A (en) | Device and method for image pick-up | |
CN1823350A (en) | Method of learning a knowledge-based database used in automatic defect classification | |
JP2009258069A (en) | Inspection apparatus and inspection method of foreign matters within hole of spinneret | |
CN1484165A (en) | File information input apparatus, input method, input program and recording medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C12 | Rejection of a patent application after its publication | ||
RJ01 | Rejection of invention patent application after publication |